JOURNAL OF LATEX CLASS FILES, VOL. 11, NO.

4, DECEMBER 2012 1

Software Module Clustering: An In-Depth

Literature Analysis
Qusay I. Sarhan, Bestoun S. Ahmed, Miroslav Bures, and Kamal Z. Zamli

Abstract—Software module clustering is an unsupervised learning method used to cluster software entities (e.g.,
classes, modules, or files) with similar features. The obtained clusters may be used to study, analyze, and understand the
software entities’ structure and behavior. Implementing software module clustering with optimal results is challenging.
Accordingly, researchers have addressed many aspects of software module clustering in the past decade. Thus, it is
essential to present the research evidence that has been published in this area. In this study, 143 research papers from
arXiv:2012.01057v1 [cs.SE] 2 Dec 2020

well-known literature databases that examined software module clustering were reviewed to extract useful data. The
obtained data were then used to answer several research questions regarding state-of-the-art clustering approaches,
applications of clustering in software engineering, clustering processes, clustering algorithms, and evaluation methods.
Several research gaps and challenges in software module clustering are discussed in this paper to provide a useful
reference for researchers in this field.

Index Terms—Systematic literature study, software module clustering, clustering applications, clustering algorithms,
clustering evaluation, clustering challenges.

1 I NTRODUCTION ware clustering is being defined as the pro-

cess of decomposing large software systems
C LUSTERING (also called cluster analysis) is
an unsupervised data mining technique
that groups a set of data points into several
into smaller, manageable, meaningful (highly
cohesive), independent (loosely coupled), and
feature-oriented (share common features) sub-
clusters [1]. When several points fall inside a
systems [3], [4]. These subsystems may con-
cluster, they are similar in some features. Mea-
tain entities/artifacts (e.g., classes, modules, or
surement of the similarity and dissimilarity
files) of similar features. A software module
relies on the extent to which the data points
clustering approach that accomplishes this task
share the same features. Clustering has been
can have a substantial impact and practical
employed in many important fields of study
benefits, especially for developers working on
and applications, including software engineer-
legacy systems with documentation that is out-
ing, information retrieval, machine learning,
dated or nonexistent. Clustering in software
pattern recognition, and statistics [2].
engineering can be used in many applications,
In the context of software engineering, soft-
such as architecture recovery [5], code clone
detection [6], and poor design detection [7].
• Q. Sarhan is with the Department of Computer Science, Univer- With several algorithms and studies pub-
sity of Duhok, Duhok, Kurdistan Region, Iraq
E-mail: [email protected]
lished in the literature, software module clus-
• B. Ahmed is with the Department of Mathematics and Computer tering has become an active research area. Al-
Science, Karlstad University, 651 88 Karlstad, Sweden though many module clustering approaches
E-mail: [email protected]
• M. Bures is with the Department of Computer Science, Faculty
have been proposed and applied, they have dif-
of Electrical Engineering, Czech Technical University in Prague, ficulties meeting the current development and
Karlovo nam. 13, 121 35 Praha 2, Czech Republic advancement needs in software and its various
E-mail: [email protected]
• K. Zamli is with the IBM Center of Excellence, Faculty of Com-
applications. For example, there is a lack of
puter Systems and Software Engineering, University Malaysia experimental studies on clustering a system
Pahang, 26 300 Gambang, Kuantan, Pahang, Malaysia developed using more than one programming
E-mail: [email protected]
language [8] or a system that performs some of
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 2

its tasks by invoking ready-made web services in the past decade is lacking. A few survey/re-
available on the Internet/network [9], [10]. view studies on software module clustering
This paper presents a comprehensive sys- have been found in the literature. Shtern and
tematic literature study to structure and cat- Tzerpos [11] provided an overview of different
egorize the state-of-the-art research evidence software clustering methods and their applica-
related to software module clustering during tions in software engineering. The study high-
the past decade. The study defines several re- lighted some approaches for the evaluation of
search questions (RQs) that cover many aspects software clustering results. It also presented
of the field and then identifies relevant papers some research challenges to be addressed to
and their results. It concludes by discussing improve software clustering results. In [12],
future research opportunities in the area. In this the basic concepts and necessity of software
process, a systematic method is used to collect module clustering were presented briefly. The
and analyze the related published research. authors also described different metaheuristic
The remainder of this paper is structured search techniques that have been applied to
as follows: Section 2 presents the motivation the software module clustering problem in the
and an overview of related work. Section 3 maintenance phase of the software develop-
describes in detail the research methodology ment life cycle. In [13], the authors presented
used to conduct this study. Section 4 presents different search-based approaches to software
the results and outcomes. Section 5 discusses clustering that have been classified into sev-
the issues of validity. Finally, Section 6 presents eral categories (mono-objective, multiobjective,
the conclusions of the study. and many-objective) based on the number of
clustering quality criteria. Furthermore, the ad-
vantages and disadvantages of each category
2 M OTIVATION AND R ELATED W ORK are presented briefly. In [14] and its extended
Software module clustering is an important version [15], the authors described search-
topic of research in software engineering. Al- based optimization techniques and their appli-
though it started in the 1990s, software module cations in different software engineering do-
clustering research has experienced more mo- mains. They briefly introduced software mod-
mentum and attention in the past decade. This ularization and refactoring as clustering prob-
momentum and attention are reflected in the lems that can be addressed using several search
dramatic increase in the number of publication. algorithms. Additionally, they presented some
Among the many factors leading to this in- research challenges with search algorithms, in-
creased attention, there have been two leading cluding determining suitable stopping criteria
factors in the past decade. First is the dramatic to terminate the search and issues related to
increase in software application size due to the visualizing the search results. The authors in
newly added functionalities and features they [16] also dedicated parts of their study to
provide. This, in turn, has led to an increase performing software modularization and refac-
in the number of modules for these applica- toring using search-based optimization tech-
tions. Here, software module clustering is a niques. They briefly introduced a number of
good approach to manage and maintain this algorithms in this respect, including NSGA-II
kind of application. Second, the advancement and PCA-NSGA-II. Additionally, a number of
of artificial intelligence (AI) methods (such as evaluation metrics, such as coupling, cohesion,
data mining, clustering, optimization, and ma- and MQ (modularization quality) have been
chine learning methods) in the past decade mentioned.
has played a substantial role in increasing the The aforementioned studies were not fo-
research activity related to software module cused on conducting a dedicated literature
clustering. analysis of software module clustering: some
Although it has been an active research area aspects of software module clustering are cov-
since the 1990s, a systematic literature study ex- ered as part of other related topics. By contrast,
ploring research on software module clustering our paper presents an in-depth and systematic
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 3

the questions defined in this study. Finally,

Identification
the fifth stage is the results reporting. The
1 following subsections elaborate on the afore-
Define research objectives
Define research questions Search mentioned stages in detail.
2
Follow PICO method
Select search terms
Selection Select search synonyms 3.1 Identification of the Research Need and
3 Link search synonyms with OR
Link search terms with AND Scope
Apply inclusion/exclusion criteria Design search string
Apply snowballing Apply on databases
Conduct quality evaluation
Accepted papers
All papers 3.1.1 Research Objectives
Extraction This study aims to provide a comprehensive
4 analysis of publications on software module
Specify extraction form clustering in the past decade by identifying, ex-
Documentation Extract data
amining, and categorizing state-of-the-art con-
5
Data analysis
Results
tributions. Only papers from the past decade
Document results
were considered to keep the review focused
on recent works on the topic. Notable changes
Fig. 1: Stages of the used literature analysis in software development have occurred over
study. the past decade. Software systems migrated
from simple architectures, such as monolithic
and two-tiered architectures to multitier (also
analysis with a detailed research methodology called n-tier or multilayered) architectures. Ad-
to examine different software module cluster- ditionally, development approaches changed
ing aspects, such as applications, algorithms, from structured to object-oriented to service-
tools, target systems, evaluation methods, and oriented. In other words, software develop-
possible research gaps. ment continues to change. Software developers
use vast amounts of source code to under-
stand large and complex systems before mak-
3 R ESEARCH M ETHODOLOGY ing changes, which is not feasible. Thus, soft-
This study was based on the guidelines of con- ware clustering has been employed to provide
ducting systematic mapping studies provided automated assistance to recover the abstract
by [17], the guidelines for conducting literature structure of systems and understand them.
review studies provided by [18], and other
studies. These studies aimed to investigate soft- 3.1.2 Research Questions
ware engineering in a particular context other
than software module clustering (e.g., [19], [20], To achieve the aim of this study and reflect its
[21]). Figure 1 shows the five-stage systematic objectives, several RQs related to the topic were
process used in this research. defined. Each defined RQ addresses a specific
The first stage is defining the research scope aspect of software module clustering.
and questions. The main research problem of • RQ1. What is the number and distribution
the study was identified, and several questions of publications on software module clus-
were formed to address the problem. The sec- tering over the last decade?
ond stage is conducting the search process in • RQ2. Which researchers, organizations,
which a search strategy to select the primarily and countries are active in software mod-
published papers was specified. The third stage ule clustering research?
is screening the published papers obtained • RQ3. What are software module clustering
from the previous stage via several filtering applications and how these applications
methods. The fourth stage is data extraction, are distributed?
where the selected papers are carefully ana- • RQ4. What is the standard process of soft-
lyzed, and useful data are extracted to answer ware module clustering?
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 4

• RQ5. What are the software systems used in the keywords, as this study is a general
as targets for the experiments of software
analysis of clustering in software engineering.
module clustering? Table 2 presents the keywords associated with
• RQ6. What are the current factbase each part of the method in the final search
sources, their types, their forms, and ex-
string. Notably, the search terms used in the
traction tools used in software module final search string were linked with one another
clustering? based on the steps described in [22]. Here,
• RQ7. What are the most used similarity the "OR" Boolean operator was used to link
measures in the software module cluster- synonyms or related search terms to the topic
ing? of this study and the "AND" Boolean operator
• RQ8. What are the most used algorithms, was employed to link the main search terms.
their types, and standard stop conditions? Different combinations of search strings tried
• RQ9. What are the most used tools for to construct the final one since the term "clus-
visualizing clustering results? tering" is used in other research areas such as
• RQ10. What are the metrics used to eval- data mining, image processing, and statistics.
The final search string was the one that meets
uate clustering results and the current ap-
proaches to obtaining the gold/expert de-the following two criteria.
composition? • The search string that returns the most
• RQ11. What are the potential future re- relevant studies.
search directions on software module clus- • The search string that returns the maxi-
tering? mum number of the identified pilot set.
For this study, a pilot set of 25 papers has
3.2 Search Strategy been selected based on our experience and
initial research review.
3.2.1 Literature Sources
As an example, the search string Try 2 in
In this study, five standard online databases Table 3 was excluded because the returned
were selected as sources that index the liter- result is incomplete compared to the search
ature of software engineering and computer string Try 5.
science [17]. These sources are presented in
Table 1. 3.3 Paper Selection
TABLE 1: The used database sources to explore 3.3.1 Inclusion/Exclusion Criteria
the literature. To decide whether a paper is relevant to the
Source URL scope of this research, a set of criteria, which
IEEE Xplore https://ieeexplore.ieee.org/ are presented below, for inclusion and exclu-
Elsevier ScienceDirect https://sciencedirect.com/ sion was considered.
ACM Digital Library https://dl.acm.org/
Scopus https://scopus.com/ • Inclusion criteria are:
SpringerLink https://link.springer.com/ – Papers published online from 2008-
2019. Papers related directly to soft-
ware module clustering. This is en-
3.2.2 Search String sured by reading the title of each ob-
The population, intervention, comparison, and tained paper. However, the abstract or
outcomes (PICO) method [18] was employed full-text reading has been also applied
to identify related studies. Here, population when the title reading was not enough.
(P) refers to the applied area of clustering, This criterion filtered most of the pa-
intervention (I) refers to the process or pro- pers out.
cedure used to solve the clustering problem
and outcomes (O) refers to the outcomes of • Exclusion criteria are:
the work and research of clustering in software – Papers not published in English are
engineering. Comparison (C) is not considered excluded since English is the prevalent
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 5

TABLE 2: PICO related keywords.

PICO method Keywords used
Population (P) (software OR project OR system OR application OR program OR module OR component
OR service OR source code OR package OR file OR function OR class)
Intervention (I) (clustering OR cluster analysis OR partitioning OR grouping OR splitting OR structuring
OR modularizing OR constructing OR composing OR categorizing OR classifying)
Comparison (C) No comparisons considered in the keywords
Outcomes (O) (methodology OR algorithm OR technique OR approach OR method OR tool OR improve-
ment OR evaluation OR similarity measurement OR application OR metric OR problem OR
challenge OR limitation)

TABLE 3: Search strings piloted on IEEE Xplore.

# Try Search string # Results # Missing studies
Try 1 ((software) AND (clustering) AND (problem)) 355 15
Try 2 ((software OR module OR component) AND (clustering OR cluster 450 11
analysis) AND (algorithm OR technique OR approach OR problem))
Try 3 ((software OR system OR program OR module OR component OR 631 8
source code OR package OR file) AND (clustering OR cluster analysis
OR partitioning OR grouping OR splitting OR structuring OR catego-
rizing OR classifying) AND (methodology OR algorithm OR technique
OR approach OR method OR tool OR improvement OR problem OR
challenge OR limitation))
Try 4 ((software OR system OR program OR module OR component OR 811 3
source code OR package OR file) AND (clustering OR cluster analysis
OR partitioning OR grouping OR splitting OR structuring OR catego-
rizing OR classifying) AND (methodology OR algorithm OR technique
OR approach OR method OR tool OR improvement OR measurement
OR application OR problem OR challenge OR limitation))
Try 5 ((software OR project OR system OR application OR program OR 905 0
module OR component OR service OR source code OR package OR file
OR function OR class) AND (clustering OR cluster analysis OR par-
titioning OR grouping OR splitting OR structuring OR modularizing
OR constructing OR composing OR categorizing OR classifying) AND
(methodology OR algorithm OR technique OR approach OR method
OR tool OR improvement OR evaluation OR similarity measurement
OR application OR metric OR problem OR challenge OR limitation))

language used in the scientific peer- cluded because the content of such
reviewing global community. publications is eventually presented in
– Papers without accessible full text. peer-reviewed venues, which have al-
– Papers not formally peer-reviewed ready been considered in our study.
(gray literature and books). – Papers that are published as surveys
– Papers not published electronically. are filtered out because they do not
– The duplicated papers were excluded actually bring new technical contribu-
from the list. Authors sometimes pub- tions to software module clustering.
lish expanded versions of their confer-
ence papers to journal venues. Such
papers share most of the material and 3.3.2 Snowballing
considering them both would affect The snowballing [23] search method was ap-
the quality of this study. To overcome plied to the remaining papers to reduce the
this issue, duplicated papers are iden- possibility of missing critical related papers.
tified by comparing paper titles, ab- In this method, each research paper’s list of
stracts, and contents. When the du- references is examined in terms of the previ-
plication is confirmed, the least recent ously applied inclusion/exclusion criteria. The
publication is removed. process was then recursively applied to newly
– Master and Ph.D. dissertations are ex- identified papers.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 6

Apply search string on literature databases Result = 3100

-2989 Result = 160

Apply inclusion/exclusion criteria

+6 Result = 166
Apply snowballing search technique

-17 Result = 149

Conduct quality evaluation

-6 Result = 143
Repeat paper selection process

Fig. 2: Results of the paper selection process.

3.3.3 Quality Evaluation Fig. 3: Publication per year.

For quality evaluation, each paper must be
evaluated to determine whether sufficient in- represent each RQ. The following subsections
formation can be extracted from it. Papers that present and discuss the results based on each
did not provide answers to the following two RQ.
questions were excluded:
• Is the process of software module cluster-
4.1 Distribution of Publications (RQ1)
ing described in detail?
• Are there experimental results and evalu-
4.1.1 Publication Frequency
ation for the software module clustering As previously mentioned, 143 papers pub-
process? lished over the past decade (2008-2019) were
included in this study. The selected papers
The number of included and excluded pa- were analyzed to determine their frequency
pers at each stage of the paper selection process and evolution, as shown in Figure 3.
is shown in Figure 2. Besides, all the identified The figure also shows that the average num-
papers and their references, full names, and ber of publications per year is approximately
publication years are listed in Table 12. 12. Additionally, interest in the topic has in-
creased in the past three years, which indicates
3.4 Data Extraction the successful application of module cluster-
ing to solve problems in software engineering
In this phase, the data of the considered studies along with the exponential growth of software
were extracted and analyzed to answer the applications in number, size, and complexity.
defined RQs. The data obtained in this phase One reason for this growth is that software has
were stored in an Excel file with different fields no limited lifetime: software code is constantly
created for this purpose. Each data extraction changed to meet user needs. Thus, developers
field has a data item and a value, as presented are always in need of tools and approaches
in Table 4. Notably, a reliable data extraction to ease the process of software maintenance.
method was followed in this stage: the data Software module clustering helps considerably
were extracted first by the first author and in this context. Another valid reason is that
then double-checked by the other authors sep- software development is constantly changing
arately. with the development of new technologies.
In the past few years, AI and the Internet
4 R ESULTS of Things (IoT) have become trending tech-
The selected papers were carefully analyzed to nologies [24] impacting software development.
answer the RQs. Here, a short title is used to The complexity of software systems based on
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 7

TABLE 4: Data extraction form.

Data Item Value Relevant RQ
ID Integer paper ID number None
Title Paper title None
Year Paper publication year RQ1
Type Paper publication type RQ1
Venue Publication venue name RQ1
Author Name Name of the author(s) RQ2
Author Organization Name of the organization for each participated author/co-author RQ2
Author Country Name of the country for each participated author/co-author RQ2
Clustering Applications Applications of software module clustering RQ3
Clustering Process The standard steps of software module clustering RQ4
Target Systems The software systems used in the experimental testing RQ5
Factbase Extraction The sources, fact types, and tools used for factbase extraction RQ6
Similarity Measures The similarity measures used in the clustering process RQ7
Clustering Algorithms The algorithms used in clustering and their types RQ8
Visualization Tools The tools used for visualizing clustering results RQ9
Evaluation Metrics The metrics used for clustering results evaluation RQ10
Future Research The possible future research directions RQ11

such technologies is increasing dramatically as

advanced features are employed. As a result,
many legacy systems have been transformed
to cope with this change. Understanding those
systems before integrating them with new tech-
nologies requires an in-depth analysis, which
may be achieved by employing software clus-
tering.
4.1.2 Publication Venue
Figure 4 shows that the considered papers,
namely, 92 are conference papers, 43 are journal
papers, 4 are symposium papers, and 4 are Fig. 4: Publication ratio per each venue.
workshop papers were published in various
venues. This result also shows that only 30%
of the considered papers have reached the ma-
turity of a journal publication, indicating that
software module clustering is a very young or
even immature research area [25], [26]. More-
over, few conference papers were published
as book chapters. For these papers, consider-
ation was given to their original venues, that
is conferences. Figure 5 shows the publication
number per year by venue type.
The most active and top journals, confer-
ences, symposiums, and workshop venues that
publish papers on software module clustering
can be determined by analyzing the publica-
tions. Abbreviations are used in this paper
instead of full names. Figure 6 shows the ac-
tive journals in which the considered studies Fig. 5: Publication number per each venue.
were published. The full names of the jour-
nals are presented in Table 13. The figure also
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 8

Fig. 7: Number of published papers vs. confer-

ence name.

Fig. 6: Number of published papers vs. journal

name.

shows that the most active and top journals are

"Inf. Softw. Technol.", "J. Syst. Softw.", "IEEE
Trans. Softw. Eng.", "Procedia Comput. Sci.",
"Soft Comput.", and "IET Softw.". Notably, 44%
of the journal papers were published in the
top six journals, whereas the other 56% were
published in individual journals.
Figure 7 shows the active conferences that
published papers in software module cluster-
ing. The full names of the conferences can be
found in Table 14. The most active and top
conferences are the International Conference on Fig. 8: Active researchers based on the pub-
Program Comprehension (ICPC), Conference lished papers.
on Software Maintenance and Reengineering
(CSMR), and Working Conference on Reverse
Engineering (WCRE). Notably, approximately ever, the most active researchers are those who
15% of the conference papers were published at
have more than two published papers either
these top three conferences. If other conferences
as a main author or coauthor. The ranking of
that publish two or three papers are consid- active researchers is shown in Figure 8. The
ered, then approximately 37% of the conference
ranking shows that "Jitender Kumar Chhabra"
papers were published by annual conferences. and "Giuseppe Scanniello" are the top two
The largest number, approximately 63%, of theresearchers in this field, with nine and eight
published conference papers were at individual
published papers, respectively. These authors
conferences, represented as "Others" in Figure
participated in approximately 12% (17/143) of
7. all publications.
Table 5 shows the ranking of the active
4.2 Active Researchers, Organizations, and countries and organizations, including the
Countries (RQ2) name of country, organization, participating
Many researchers are interested and involved researchers, reference to the published papers,
in software module clustering research. How- and the total number of papers.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 9

TABLE 5: Countries, organizations, and researchers active in researching software module

clustering.
Country Organization Author(s) Published Papers Total
India National Institute of Technology Jitender Kumar Chhabra, Amarjeet [27], [28], [29], [30], 10
Kurukshetra Prajapati, and Amit Rathee [31], [32], [33], [34],
[35], [36]
Italy University of Basilicata Giuseppe Scanniello [37], [38], [39], [40], 8
[41], [42], [43], [44]
Iran University of Tabriz Habib Izadkhah, Hamid Masoud, [31], [45], [46], [47], 7
and Ayaz Isazadeh [], [48], [49]
Pakistan City University of Science and In- Rashid Naseem [7], [50], [51], [52], 5
formation Technology [53]
Pakistan Quaid-i-Azam University Onaiza Maqbool, Adeel Ahmed, [7], [50], [52], [54], 5
and A.Q. Abbasi [55],
Romania Babes, -Bolyai University Istvan Gergely Czibula and [3], [56], [57], [58] 4
Gabriela Czibula
Canada Carleton University Samuel Ajila, Chung-Horng Lung, [59], [60], [61] 3
and Abdulaziz Alkhalid
Italy University of Naples Federico II Anna Corazza and Sergio Di Mar- [37], [41], [44] 3
tino
Malaysia Universiti Tun Hussein Onn Mustafa Mat Deris and Rashid [51], [52] , [55] 3
Malaysia Naseem
China Xidian University Jing Liu [62], [63], [64] 3
Italy University of Salerno Michele Risi and Genoveffa Tortora [39], [42], [65] 3
Iran Islamic Azad University Masoud Kargar [31], [], [49] 3
Pakistan University of Peshawar Sara Shahzad [52] , [53], [55] 3
Saudi Arabia King Abdullah University of Sci- Abdulaziz Alkhalid [66], [67] 2
ence and Technology
Pakistan Shaheed Benazir Bhutto University Siraj Muhammad [50], [54] 2
Brazil UNIRIO - Universidade Federal do Márcio de O. Barros [68], [69] 2
Estado do Rio de Janeiro
England University of Manchester Christos Tjortjis [2], [70], 2
Greece International Hellenic University Christos Tjortjis [71], [72] 2
China Jiangxi Normal Universit Linhui Zhong, Nengwei Zhang, [73], [74] 2
and Jing Xia
Malaysia University of Malaya Chun Yong Chong and Sai Peck Lee [75], [76] 2

The active countries in the published papers

can also be extracted from the information
presented in Table 5. Such data can be obtained
based on the organizational affiliation of the
authors/coauthors. Figure 9 shows the coun-
tries that are most active in publishing papers
on software module clustering and the share of
papers published by each country with respect
to the total number of publications.
The most active countries were those that
published more than one research paper. Fig. 9: Active countries.
Such countries produced approximately 41%
(58/143) of the total publications. India and
Italy were the two most active countries, with 4.3 Applications of Software Module Clus-
10 and 9 published papers, respectively. Those tering (RQ3)
papers published from India were from a col-
laboration between the authors "Jitender Ku- An in-depth analysis of the selected papers
mar Chhabra," "AmarjeetPrajapati," and "Amit revealed that software module clustering appli-
Rathee." cations can be classified into 14 areas (A1-A14).
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 10

The figure clearly shows that "Junit," "jEdit,"

and "JHotDraw" are the most commonly used
systems in the experimental tests, accounting
Fig. 10: Software module clustering process. for approximately 16% (23/143), 13% (18/143),
and 11% (16/143) of the published papers,
respectively. In addition, the top three systems
This result is presented in Table 6 which shows are written in Java. The reason for this fact
the papers in each application area. From the could be that Java is still the most widely
identified application areas, it is clear that used programming language according to
"information recovery (A1)" and "restructuring the TIOBE Index [164]. For the experimental
(A2)" are the top two applications of software clustering tests, we recommend selecting the
module clustering with 59 and 44 published target systems that are well-known to the
papers, respectively. developer and researcher communities. They
are free and open-source, written in widely
4.4 Software Module Clustering Process used programming languages, and updated
(RQ4) frequently.
Answering RQ4 helps to determine the de-
tailed steps of the software module clustering B. Factbase source selection (RQ6)
process. Moreover, it helps to identify the al- Any software clustering process starts with
gorithms, techniques, and tools used in each the construction of a factbase. As mentioned,
step of the process. The ideal software module the factbase contains information on the target
clustering process includes five main steps, as software system, such as software entities (e.g.,
shown in Figure 10. Here, the process ends classes and variables) and their relationships
when reaching a stopping criterion such as a (e.g., inheritance and method calls). Such infor-
maximum number of iterations or a desired mation can be extracted from various sources.
number of clusters. The following subsections Table 7 presents in detail the different types
describe the steps of the process. of factbase sources. After constructing the fact-
base, a software clustering algorithm can be
4.4.1 Factbase Extraction applied to group entities from the factbase into
The input that a software clustering algorithm useful subsystems. Many clustering methods
expects is often called a factbase as it combine facts extracted from different sources
contains facts (e.g., relationships between to obtain reasonable results at the cost of more
software entities) extracted from the target complex data processing [82].
software system [163]. In this step, the
extracted factbase should consist of sufficient
information about the target software system
to ensure meaningful clustering [70]. Factbase
extraction includes target software system
selection, factbase source selection, filtering
and preprocessing, entity selection, and feature
selection.

A. Target software system selection (RQ5)

Before starting the software clustering pro-
cess, target software systems must be specified.
The analysis of the selected studies reveals
that open-source software systems written in
Java and C/C++ have been the focus in the
literature. Figure 11 shows the most commonly
used target software systems.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 11

TABLE 6: Application areas of software module clustering.

ID Application Areas Published Papers Total
A1 To support information recovery such as abstrac- [77], [5], [29], [30], [31], [35], [38], [39], [42], 59
tion levels (e.g., modules or design patterns as [44], [45], [], [48], [49], [51], [52], [53], [59],
low-level abstraction and architectures as high- [60], [61], [63], [64], [65], [69], [73], [75],
level abstraction). Information recovery is essen- [76], [78], [79], [80], [81], [82], [83], [84],
tial for better software understanding, maintain- [85], [86], [87], [88], [89], [90], [91], [92],
ing, and updating. [93], [94], [95], [96], [97], [98], [99], [100],
[101], [102], [103], [104], [105], [106], [107],
[108], [109]
A2 To increase software maintainability by reducing [3], [7], [27], [32], [33], [34], [37], [47], [50], 44
the complexity of its modules via decoupling, [55], [56], [57], [58], [66], [67], [68], [70],
restructuring, or refactoring. [72], [74], [110], [111], [112], [113], [114],
[115], [116], [117], [118], [119], [120], [54],
[121], [122], [123], [124], [125], [126], [127],
[128], [129], [130], [131], [132], [133]
A3 To identify software components that implement [4], [28], [36], [41], [46], [62], [134], [135], 16
related functionality (a component is a strongly [136], [137], [138], [139], [140], [141], [142],
related set of software entities such as files, [143]
classes, and modules). Software component iden-
tification is important to promote the reuse of
recovered components from existing software sys-
tems as building blocks for newly developed
systems.
A4 To identify the fault/defect proneness of software [43], [71], [144], [145], [146], [147], [148] 7
modules that require further examination and
maintenance.
A5 To find concept locations (also called change [40], [149], [150] 3
propagation or change impact analysis) in source
code. Finding concept locations in source code
helps to identify where changes are to be made
in response to new requirements.
A6 To evaluate a software system’s maintainability [2], [151], [152] 3
via determining the entities that are more difficult
to maintain than others.
A7 To map components found in the source code [153], [154] 2
onto machines found in a distributed environ-
ment. Here, the number of clusters is created
based on the number of machines in the used
distributed environment.
A8 To identify duplicate code (also called code clone). [6], [155], [156] 3
A code clone is a piece of code that appears at
least twice in a software system due to copy-
paste activity or reusability of existing code. Code
clone detection helps improve the performance
of software and reduce its maintenance cost and
effort.
A9 To identify groups of similar code changes. A [157] 1
code change is a sequence of edit operations that
turns an original code (old version) into its mod-
ified version (new version). The identified code
changes can serve as inputs for some software
tools such as recommendation and bug fixing
tools.
A10 To identify different types of software architec- [158] 1
tures that are used in heterogeneous systems.
A11 To understand locations within an AOP system [159] 1
that could cause or exhibit aspect interference
problems. Thus, it is essential to develop an
interference-free AOP system.
A12 To discover program topoi (which are summaries [160] 1
of the main capabilities of a program).
A13 To generate microservice candidates from a [161] 1
monolithic software system.
A14 To assess the initial partition of a software system [162] 1
(assess the extent to which a certain partition
allows its parts to evolve independently).
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 12

Fig. 12: Number of published papers vs. the

factbase source.

Fig. 11: Number of published papers vs. the

targeted software systems. Fig. 13: Number of published papers vs. the
factbase extracted.
 JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012
TABLE 7: Factbase sources.

Factbase Factbase Extracted Factbase Type Major Drawbacks

Sources
Source code • Structural dependencies (also called syntactic information) between Static • It is language-dependent. • May be mismatched to a product version of
software entities (e.g., function calls, class references, class inheritance, binary modules because of errors, and updates • Comments and naming
file inclusions, package usages, and global variable access). Often repre- conventions may not be used or followed by software engineers. • Extracting
sented as directed graphs or matrices. • Domain or naming information domain knowledge is known to be a difficult task due to the existence of
(also called lexical information or semantic information) found in com- noise in the data. • It deals with structural relationships rather than behavioral
ments, variable names, object names, method names, class names, file relationships (dynamic interactions among system entities). • Not available
names, and source code statements with some software systems. • If the software system is extremely large. The
number of entities involved in clustering would be so large that it could be
unmanageable to track. • Natural language processing techniques (stemming)
to extract naming information is very time-consuming
Binary code • Compilation parameters, linkage parameters, and symbol table that Static • It is language-dependent. • It is a compiler dependent. Thus, compiling a
(executable contain relationships between components and modules of a software source code via two different compilers may result in two different binary codes.
files) system. This is because each compiler has its building parameters and nature of work.
Bytecode • Class dependencies. • Names of variables and classes Static It is language-dependent. • Not available with some software systems.
Documentation • System’s structure or design. • System’s functional and non-functional Static • Facts are often incomplete, outdated, unsynchronized, vague, or even missing
requirements. due to long term maintenance or personnel turnover. Thus, they may not
reflect the current state of the system. Also, documentation could be completely
missing.
Human exper- • Knowledge of requirement documents. • High-level design. • Knowl- Static • On the same software system, different experts may construct different
tise edge of architectural styles. software decomposition.
Configuration • Declaration of used resources. • Security permissions. Static • Unavailable because they are stored in middleware configuration files that
files are not part of the the software systems’ source or binary code.
Dynamic infor- • Object construction and destruction. • Sources of exceptions and er- Dynamic • It is language-dependent. • Often, the source code does not remain untouched.
mation rors. • Method entry and exit. • Method invocations/calls. • Execution
logs (structured execution outputs that capture the process of building,
deploying and testing various scenarios) • Performance counters and
statistics such as number of threads, size of buffers, number of network
connections, CPU and memory usage, number of component instances,
average, maximum, and minimum response time.
Data files • Such as database, text, and reports files used as input or output Dynamic • Some software systems do not deal with input and output files.
that contain information about which classes or modules perform file
operations (e.g., create, read, and write).
Files organiza- • The organization of software classes, files, folders, packages and their Organizational • Often, software files and folders are not organized well.
tion physical locations may provide information about their responsibilities.
Human organi- • The structure of the system. Usually, a developer is responsible for Organizational • In medium and large software systems, it is not identified well due to
zation associated components. developers turn out.
Evolutionary • File ownership, file timestamps, file versions, commit logs, persons Evolutionary • It does not cover the essential parts of the software system unless huge
(historical) making changes, maintenance activities, and bug report information evolutionary data has to be dealt with. Thus, it is usually difficult to deal with
information extracted from software repositories, version control/management sys- due to the issue of size. • The lack of evolutionary data formatting.
tems, bug/issue tracking systems, or release documents. • Also, it tells
us how, when, and why a given software entity was modified.

13
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 14

Figure 12 shows that "source code", "docu- document and within the whole document
mentation", "dynamic information", and "byte- collection (the target software system).
code" are the most commonly used sources for • Software metrics: They are quantitative
factbase extraction, constituting approximately measures that enable software engineers
82% (117/143), 3% (4/143), 2% (3/143), and and managers to understand the target
3% (2/143) of the total published papers, re- software system. The number of code lines
spectively. In addition, many papers combined per class, number of methods per class,
two sources for factbase extraction. In this re- and depth of inheritance level are possible
spect, "source code and evolutionary informa- metrics.
tion" is the most commonly used combination, • Extended dependency graph: It is a graph
accounting for approximately 6% (8/143) of the that combines logical/static relationships
total publications. and evolutionary relationships among soft-
The factbases are extracted from the sources ware entities. The evolutionary relation-
in different forms. Analysis of the selected ships of the targeted software system rep-
papers reveals that the "dependency graph," resent the changes applied to its source
"vector-space model," "software metrics," and files over time. Currently, many version
"extended dependency graph" are the most control systems, such as the concurrent
commonly used forms of the factbase. Figure 13 versions system (CVS) and Git, store these
shows the analysis results of the published pa- changes.
pers. The figure also shows that approximately Notably, researchers typically use tools,
57% (81/143), 32% (46/143), 8% (12/143), which are mostly open-source Java programs,
and 3% (4/143) of the total published papers to perform factbase extraction. Table 8 presents
considered "dependency graph", "vector-space all the tools that have been used in the selected
model", "software metrics", and "extended de- studies, along with their links.
pendency graph", respectively. The following
points describe these factbase forms in detail: C. Filtering and preprocessing
• Dependency graph: It is a graph represen- Filtering is a useful preprocessing phase
tation of the target software system. The in any clustering process to identify and
nodes in the dependency graph represent remove unnecessary textual and nontextual
software entities, whereas the edges rep- information that has been extracted from
resent the logical/static relationships be- comments and source codes. Textual
tween entities. In some cases, edges are information can be meaningless words,
weighted to denote the degree of depen- such as words with less than three characters,
dency. Once the dependency graph is ex- language keywords, or common English
tracted, many characteristics of the target words that are not usually useful for a
software system can be discovered, such as search [39]. Textual information can also
the independence degree of the software be library classes or header files used in
entities based on their relationships. multiple modules and made available across
• Vector-space model: It is used to cap- the implementations of a programming
ture the relative importance of terms language. These classes have to be eliminated;
(e.g., class name, function name, object otherwise, they tend to group many classes
name, and variable name) in a docu- in a single large cluster around them [38].
ment, e.g., class file, and program file. Nontextual information can be operators,
In the vector-space model, a document symbols, special characters, and punctuations.
is represented by a vector of terms ex- Preprocessing can be implemented in the form
tracted from the document with associated of a normalization procedure. The attributes
weights (which can often be computed us- of the software entities can be a mixture of
ing the term frequency-inverse document numerical and categorical data. Here, with
frequency (TF-IDF) method [40]) repre- the help of normalization, all the attributes
senting the importance of the terms in the are treated equally [100]. Preprocessing can
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 15

TABLE 8: Factbase Extraction Tools.

Tool URL
Dependency Finder http://depfind.sourceforge.net/
Class Dependency Analyzer (CDA) http://www.dependency-analyzer.org/
ASM http://asm.ow2.org
CScout https://www.spinellis.gr/cscout/
Bunch https://wiki.eecs.yorku.ca/project/cluster/tools
Sotograph https://www.hello2morrow.com/products/sotograph
Structure101 https://structure101.com/
Understand 2.0 https://scitools.com/
JRipples https://marketplace.eclipse.org/content/jripples
E-Quality http://smart.cs.itu.edu.tr/tools/equality/
Javassist http://www.javassist.org/
Visual Paradigm https://www.visual-paradigm.com/
Doxygen http://www.doxygen.nl/
Jdeps https://docs.oracle.com/javase/8/docs/technotes/tools/unix/jdeps.html
Roslyn https://github.com/dotnet/roslyn
Stan4J http://www.stan4j.com/
PomWalker https://github.com/raux/PomWalker
SrcML.NET https://github.com/abb-iss/SrcML.NET

also be conducted in the form of dimensional humans.

reduction, which is used to reduce the size of • Formal features: These include class in-
the input vectors [148]. Filtering is an essential vocations, method invocations, and entity
phase to facilitate further processing and avoid relationships. Relationships can be catego-
the risk of decreasing the clustering quality rized into two types [66]:
[82]. – Direct relationships: Represent an im-
mediate connection between two en-
D. Entity selection tities e.g., if function f1 calls another
Software entities are the input to any soft- function f2 , then f1 and f2 are directly
ware clustering technique. Therefore, entities related.
to be clustered must be identified beforehand. – Indirect relationships: Represent the
The selection of entities depends mainly on the proportion of common features that
aim of the clustering technique [66]. Table 9 two entities share e.g., if functions f1
presents two examples in this respect. and f2 both call function f3 , then f1
Low-level entities represent the functionality and f2 are indirectly related to each
of the target software system much more other.
clearly than high-level entities. However,
some software systems are large and contain The formal features have a direct impact
enormous numbers of functions that make on system behaviors. For example, if a
the use of functions in the clustering process class changes its invocation from one class
inappropriate. In such cases, the clustering of to another, then changes in the system’s
high-level entities is preferable. behavior should be expected. Extracting
formal features, however, can be more
E. Feature selection complicated, as parsing rules need to be
Each software entity has a set of features. The applied. This process becomes even more
features of a class entity, for example, can be complicated when only partial information
broadly divided into two types [4], [93]: is available.
• Nonformal features: These include file Feature selection aims to prepare the features
naming convention, class creation date, of all software entities for the next step as a fea-
number of functions, number of variables, ture matrix (also called the entity-feature ma-
number of lines of codes, and comments. trix) [83]. A feature matrix is a two-dimensional
Nonformal features do not directly affect matrix where the rows represent software en-
system behaviors. Also, they can be easily tities and the columns represent their features.
extracted, interpreted, and understood by The value of each cell in the matrix is either 0,
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 16

TABLE 9: Entities selection.

Clustering aim Input entities Entities abstraction level
Software comprehension Functions and their call statements Low-level (also called detailed level)
Architecture recovery Software classes, packages, modules, and files High-level (also called architectural level)

which indicates the absence of a feature, or 1,

which indicates the presence of a feature [137].
Table 10 presents a feature matrix of software
containing 5 entities and 4 features. Feature F1
is present in all the entities, while feature F3 is
absent in all the entities. Features F2 and F4 are
present or absent in the entities.

TABLE 10: A simple feature matrix.

F1 F2 F3 F4
E1 1 0 0 1
E2 1 0 0 0
E3 1 1 0 1
E4 1 1 0 0
E5 1 1 0 1
Fig. 14: Number of published papers vs. simi-
larity/dissimilarity measure.
Some clustering approaches apply weight-
ing schemes (e.g., binary weighting, absolute
weighting, and relative weighting) to the fea- However, the authors in [163] introduced a
tures to represent the significance of each one. method to help in this respect.
Thus, the connection strength between a pair The analysis of the considered published pa-
of entities can be calculated [4]. pers in the study reveals the following general
categories of clustering algorithms used in the
4.4.2 Similarity Calculation (RQ7) literature:
Similarity measures are used with software • Hard clustering: In this type, a software
module clustering to determine the most simi- entity can be in only one cluster and can
lar or dissimilar entities based on their features. be divided into two types:
Generally, entities are considered more similar – Hierarchical clustering [3]: For a given
if they share more common features [38]. After set of n entities, hierarchical clustering
the computation of the similarity among all algorithms are divided into two types:
pairs of software entities, a similarity matrix ∗ Agglomerative (bottom-up) algo-
can be generated for the next step. The most rithms: The algorithms begin with n
commonly used measures in the considered singletons (each of one entity) and
papers are shown in Figure 14. Clearly, "Jaccard merge them until a single cluster
distance," "Cosine distance," and "Euclidean is obtained. The two most similar
distance" are the most common, accounting for clusters are merged in each step
approximately 12% (17/143), 8% (12/143), and of the process. Agglomerative al-
6% (8/143) of the total publications, respec- gorithms provide different perspec-
tively. tives of software clustering: the ear-
lier clustering iterations present a
4.4.3 Cluster Creation (RQ8) detailed view of the software archi-
Here, a clustering algorithm must be applied tecture and the later ones present a
to similar group entities of the target system high-level view [4].
based on specific features. The selection of a ∗ Divisive (top-down) algorithms:
suitable algorithm to apply to a task is difficult. These algorithms start with one
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 17

cluster (containing all n entities) accounting for approximately 50% (71/143) fol-
and split it until n clusters are lowed by "Hierarchical Clustering", with ap-
obtained. proximately 31% (44/143), and "Fuzzy Clus-
– Partitional clustering: For a given set tering", with approximately 1% (2/143) of the
of n entities, this approach simply di- total publications.
vides the set of entities into nonover-
lapping clusters such that each entity
is in exactly one cluster.
Notably, some studies have combined
more than one clustering algorithm. For
example, hierarchical and partitional
clustering can be combined to achieve a
common goal. This type of clustering is
called cooperative clustering [84].

• Soft clustering (also called fuzzy cluster-

ing): This approach is based on fuzzy logic, Fig. 16: Number of published papers vs. clus-
where a software entity can be in one tering type.
or more clusters. Thus, a probability or a
membership degree of that entity in those In hierarchical clustering, several methods
clusters is assigned [39]. can be used to compute the distance between
A deep analysis of the selected papers re- clusters. An examination of the selected studies
veals that twenty-two clustering algorithms reveals the seven most commonly used meth-
are the most commonly used, as shown in ods. Figure 17 shows this result. Clearly, the
Figure 15. Notably, "agglomerative hierarchical top three methods are "complete linkage", with
clustering" and "k-means" are the most com- approximately 9% (13/143), "single linkage",
monly used clustering algorithms in the con- with approximately 7% (10/143), and "average
text of software module clustering, with per- linkage", with approximately 6% (9/143) of the
centages of approximately 28% (40/143) and total publications.
11% (16/143) of the total publications, respec-
tively. Figure 15 clearly shows this result.

Fig. 17: Number of published papers vs. dis-

tance computation methods in hierarchical
clustering.

Fig. 15: Number of published papers vs. clus- The analysis of the selected studies indicated
tering algorithm. the use of several different termination condi-
tions. The clustering process can be terminated
Figure 16 shows that "Partitional Clustering" based on one of the following cutoff conditions
is the most commonly used type of clustering, [116], [4]:
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 18

• A good solution has been reached (e.g.,

when satisfying results are obtained by
applying one or more clustering evaluation
metrics).
• A maximum number of iterations has been
reached.
• Improvement has not been found for a
long time.
• All software entities are combined into a
single cluster.
• A pre-specified termination condition has
been met (e.g., a specific number of clus-
ters has been formed).

4.4.4 Results Visualization of the Clustering

(RQ9)
Visualization is used after the software mod- Fig. 18: Number of published papers vs. clus-
ule clustering process to view the clustering tering evaluation metric.
results as graphs, dendrograms, or distribution
maps. Visualization of the results is also used
to enable software engineers to efficiently and
• Results Compared with Other Results: The
conveniently examine the clustering results.
results obtained from the applied cluster-
To achieve this goal, the visualization tools
ing algorithm are compared to those of
used in the literature ease and automate the
previously published studies.
visualization process. For example, graph vi-
• Modularization Quality (MQ): MQ mea-
sualization tools compile a graph description
sures the cohesion and coupling of mod-
language and generate an image file outlin-
ules and is used to ensure that the clusters
ing the subsystems as the output of software
of a system are the cohesive modules in the
clustering [77]. Analysis of the selected studies
clusters and the loose connections between
reveals that researchers have not focused on
clusters.
the visualization of the clustering results. A
• MoJo Similarity Metrics: They measure the
potential reason for this situation is related to
similarity between the partition produced
the software module clustering systems’ end-
by the software clustering process and
users, which are the developers themselves or
the partition created by an expert (also
domain experts. However, Table 11 shows the
called the expert decomposition, authorita-
visualization tools used in three of the studies.
tive decomposition, benchmark decompo-
sition, reference decomposition, gold de-
4.4.5 Results Evaluation Metrics of the Clus- composition, ground truth decomposition,
tering (RQ10) or baseline decomposition) using MoJo
The software clustering results should be eval- metric and its family based on the num-
uated in accordance with specific criteria to ber of Move and Join operations required
assess their quality [77]. Several methods in to transform one decomposition into the
the literature can be used to assess the qual- other [77]. The similarity between the two
ity of software clustering algorithms. Although partitions should be as high as possible.
crucial, validation of the results produced by • Execution Time: The clustering process
these algorithms is difficult. Figure 18 shows should not require a very long time [77].
the most commonly used evaluation methods • Extremity: Also called Nonextremity
in the considered papers. The following points cluster distribution (NED), it measures
summarize these evaluation methods: whether the clusters of the partition
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 19

TABLE 11: Results Visualization Tools.

Tool URL Supported Languages
Prefuse http://vis.stanford.edu/papers/prefuse Java
Sotograph http://www.hello2morrow.com/products/sotograph C/C++, C#, Java, PHP, and Typescript
Graphviz http://www.graphviz.org C/C++, Java, PHP, Python, Ruby, Perl, Guile, and TCL

produced by the software clustering accuracy of the clustering methods by cal-

process have extreme values or not. culating the weighted average of recall and
A good clustering process should not precision [145].
produce a partition that includes (a) • Accuracy: The percentage of software en-
Enormous clusters – clusters with too tities that are correctly classified.
many software entities would reduce • Sum of Squared Error (SSE): The sum of
the cohesion. (b) Singleton clusters (also the squared differences between entities
called isolated clusters) or small clusters in each cluster generated by a clustering
– clusters with one or too few software algorithm and the entities of each cluster
entities would increase the coupling of in the expert decomposition. Thus, SSE can
software [128]. be used as a measure of variation between
• Cohesion (intraconnectivity): It measures clusters. For example, if the SSE between
the density of connections among the soft- entities of two clusters is equal to 0, then
ware entities in a single cluster. High co- they are matched perfectly and have no
hesion indicates good clustering, as highly error. Smaller SSE is better, and obtaining
dependent modules are grouped into the clusters that minimize the SSE is always
same cluster [140]. desirable [139].
• Precision/Recall: Precision measures the • Arbitrary Decisions: An arbitrary decision
percentage of entity pairs (in the same is required when two or more entities hold
cluster) produced by the clustering algo- the same similarity. Thus, the percentage of
rithm that are also in different clusters in arbitrary decisions is employed to evaluate
the expert decomposition. High precision the clustering [83].
indicates good clustering. Recall measures • Inverted Generational Distance (IGD): It
the percentage of entity pairs (in different is calculated as the average Euclidean
clusters) in the expert decomposition that distance from each reference point (true
were also found by the clustering algo- Pareto front) to the nearest solution (Pareto
rithm. High recall indicates good cluster- front obtained by the algorithm) in the
ing [116]. solution set. Here, the set of Pareto opti-
• Stability: It measures whether the cluster- mal solutions produced by all algorithms
ing process produces similar partitions in overall runs is used as the true Pareto front
the case of small changes between suc- [30].
cessive versions of an evolving software • Meaningfulness: Generated clusters
system [81]. should resemble the subsystems of the
• Coupling (interconnectivity): It measures original system [77].
the density of connections among software • Chi-Square Test: It determines if the enti-
entities in different clusters. Low coupling ties of two different clusters are related in
indicates good clustering, as the clusters terms of some features.
are highly independent of each other [140]. • Number of Clusters: It measures the com-
• Human Experts: One or more domain ex- pactness of the clusters created during
perts manually evaluate the results ob- the software clustering process. A high
tained from the applied clustering algo- number of clusters indicates that they are
rithm. highly compact (cohesive), while a low
• F-Measure: It measures the goodness or number of clusters indicates that they are
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 20

noncohesive [55].
Figure 18 shows that "Results Compared
with Others Results" is the most commonly
used clustering evaluation method, accounting
for approximately 53% (76/143) of all publi-
cations, followed by "Modularization Quality",
30% (43/143), and "MoJo Similarity Metrics",
16% (23/143). Many studies use more than one
clustering evaluation method; thus, the results
overlap.
Some clustering evaluation methods (e.g.,
MoJo Similarity Metrics and precision/recall) Fig. 19: Publication ratio by source of expert
work only when expert decomposition is avail- decomposition.
able as a comparison standard. Examination of
the total publications reveals that a number of
approaches have been presented for obtaining software system, e.g., the folder structure,
expert decomposition. A summary of these the package structure, or the file structure.
approaches is presented as follows: The advantage of using this approach is
• Domain Expert-Based Decomposition: The
that its decompositions have good quality
decomposition is performed by software because they have been created by the
domain experts. Domain experts are per- original developers and domain experts
sonnel with experience in software design [81].
• Original Developer-Based Decomposition:
and development [55]. The experts either
evaluate the results produced by a clus- The decomposition is obtained by ap-
tering algorithm (e.g., if the results have proaching the original developers of the
a positive impact on the system’s under- target software system. The drawback of
standability) or they provide a clustering this approach is that the original develop-
benchmark that can be compared with the ers are typically not available.
• Documentation-Based Decomposition: The
results produced by a clustering algorithm
[38]. Well-known drawbacks of this ap- decomposition is obtained using the key
proach are as follows: (a) Experts may pro- functional concepts extracted from the
vide many valid ways to decompose a soft- software architecture documentation.
• Maintenance Log-Based Decomposition:
ware system into meaningful subsystems
[84]. (b) They might lead to poor decom- The decomposition is obtained by extract-
position if they did not fully understand ing information embedded in maintenance
the purpose of the clustering approach. (c) logs, which can be utilized to produce
They may lead to poor decomposition if multiple decomposition stages of the target
their experience and knowledge are not system.
sufficient [114]. In addition, finding do- Many clustering approaches use expert de-
main experts with suitable experience, es- composition to evaluate the clustering results.
pecially on open-source software systems Figure 19 shows the distribution of the expert
[37] and legacy systems [92], is difficult. (d) decomposition, where 24 approaches are based
Software systems are constantly evolving on domain experts, 15 are based on factual
and maintaining an up-to-date expert de- information, 14 are based on the original de-
composition can be a tedious, error-prone, velopers, two are based on the documentation,
and time-consuming task [93]. and one is based on maintenance logs.
• Factual Information-Based Decomposition: Many factors affect the quality and efficacy
The decomposition is obtained using the of a clustering process. The following points
current factual information of the targeted summarize those factors [65]:
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 21

• The domain, type, size, and architecture of through all the displayed results. Thus, the
the targeted software system. For example, user can better understand the results. The
a clustering algorithm that is successful for visualization tools should also automati-
a procedural program or a small software cally associate labels with the generated
system might be unsuccessful for a large clusters.
system developed in an object-oriented • CASE tool: It is essential to consider mak-
paradigm [87]. ing the whole clustering process a working
• The choice of factbase sources, preprocess- tool (e.g., third-party libraries, plugins, and
ing of data extracted from them and find- standalone applications) available for re-
ing an appropriate representation of them searchers, software engineers, and practi-
[100]. tioners to perform further experimentation
• The entities, features, and relationships be- and collect feedback for future improve-
tween them in terms of how well they have ments.
been selected. • Targeted systems: The targeted or subject
• The type of similarity measures and al- software systems used in the experiments
gorithms used for the clustering process are mostly Java-based open-source sys-
[100]. tems. It would be interesting to examine
• Arbitrary decisions during the clustering the use of non-open-source systems and
process influence the quality and perfor- systems written in other popular program-
mance of clustering [82]. ming languages. Furthermore, considering
different software application domains to
keep the results generic and widely ap-
4.5 Potential Future Directions of Research
plicable requires further study. When the
(RQ11)
targeted systems are selected from vari-
Answering RQ11 will help to identify pos- ous application domains, a specific cluster-
sible research areas that may require further ing algorithm may exhibit various perfor-
investigation. Based on the analysis of the con- mance characteristics. Therefore, the var-
sidered papers, several potential directions of ious performance features of a particu-
research were identified. The following points lar clustering algorithm, when applied to
summarize and categorize these future research multiple target systems, should be investi-
directions: gated.
• Scalability: The clustering approach should • Entity features: Some software entities
handle a growing quantity of input with- such as files and classes have a large num-
out decreasing the clustering results’ qual- ber of different types of features (e.g., lines
ity. This process can be achieved by per- of code, executable statements, number
forming clustering in parallel using mul- of functions, and number of variables or
tithreading programming techniques and objects) to be extracted. From a practical
hardware systems that have multicore pro- perspective, addressing this large number
cessors. The consistency of clustering must of various features is not preferable. There-
be ensured, i.e., performing the clustering fore, experimental studies to determine the
multiple times on the same dataset should number and type of features needed for
produce the same results. better clustering results should be con-
• Visualization: The display of the clustering ducted. Further experiments may reveal
results should be improved in cases where a clustering approach that is more suit-
large outputs are presented on the screen able for target systems that present specific
simultaneously. This improvement can be characteristics (e.g., size or implemented
achieved by applying filters to separate functionality).
the results into different abstraction lay- • Factbase sources: There are different
ers. As a result, clusters within a specific sources for factbase extraction. Each has
layer only can be viewed instead of going its own set of features and drawbacks.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 22

Thus, experimental studies that can graphs), such as utilizing the defined nam-
determine the type of factbase sources ing of modules and internal variables or
that provide better clustering results other factbase sources as part of clustering
should be performed. The impact of arrangement criteria, would be a useful
integrating different factbase sources on endeavor.
the overall clustering accuracy should be • Beyond clustering: The ultimate aim of
studied. software module clustering is to help soft-
• Cooperative clustering: Only a few studies ware engineers apply the recommended
have combined more than one clustering clustering results to their software projects.
algorithm to achieve a common goal. Cur- However, do software engineers really ad-
rently, clusters that are produced by the here to the recommended clustering re-
first clustering algorithm are subdivided sults? If yes, what will be the impact,
and reclustered by the second clustering cost, or effort of realizing the suggested
algorithm, a situation that is often unde- results? No research has comprehensively
sirable. Thus, more experiments should be addressed these issues. Thus, a thorough
performed, and tools to address the situa- investigation in this respect may be a good
tion by not reclustering all the clusters that step for further study.
are part of the initial solution set should be
developed.
• Selection of clustering algorithms: The se-
5 T HREATS TO VALIDITY
lection of appropriate software clustering Every literature mapping study has a num-
algorithms plays a significant role in pro- ber of threats that might affect its validity. In
ducing meaningful clustering results. The this study, several threats were eliminated by
authors in [163] proposed guidelines for considering well-known recommendations and
selecting or rejecting a clustering algorithm guidelines on conducting literature mapping
for a given software system. However, studies as follows:
there are no comprehensive methods for • Coverage of research questions: The threat
clustering algorithm selection. Thus, fur- here is that the research questions of this
ther research and experiments can be con- study may not cover all the aspects of the
ducted to provide formal selection meth- state-of-the-art research in software mod-
ods based on empirical evidence. ule clustering. To address this threat, all
• Clustering with aesthetic aspects of the the authors of this study used brainstorm-
software design: The computational deter- ing to define the desired set of research
mination of the optimized cluster is often questions that cover the existing research
mechanistic, ignoring the fact that software in the area.
is a creative artifact. Typically, cluster ar- • Coverage of relevant studies: It cannot be
rangement is determined via the grouping guaranteed that all the relevant studies
of nodes from dependency graphs. Having in software module clustering have been
an optimized clustering of modules versus identified. Accordingly, different literature
a meaningful set of clusters are two sides databases have been used, and a PICO
of the same coin with competing objec- method-based search string with various
tives. On the one hand, it is desirable to term synonyms (each author of the pa-
maximize cohesion and minimize coupling per suggested different terms that lead
at any cost. On the other hand, one may to desired clustering concepts) has been
also need to capture semantic as well as applied to obtain the relevant research
the essence and aesthetic aspects of the publications. However, some unidentified
software design. Integrating natural lan- papers may remain. To address this issue,
guage processing with deep learning and the snowballing method was intensively
considering on other criteria (apart from applied to reduce the possibility of missing
simply grouping nodes from dependency important related papers.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 23

• Paper inclusion/exclusion criteria: Appli- that enabled the clustering process were briefly
cation of the criteria can suffer from single- discussed. Finally, as there are many research
author judgment and personal bias. To ad- studies on software module clustering, novice
dress this issue, each paper was included researchers are likely to experience difficulties
or excluded for this study only after the in addressing different aspects of the area.
authors reached a consensus. Therefore, we propose this analysis study as
• Accuracy of data extraction: Data extrac- a primary reference to simplify the process of
tion can suffer from the single-author expe- finding the most relevant information.
rience. Accordingly, each author individ-
ually performed the data extraction pro- 7 ACKNOWLEDGEMENT
cess, and the outcomes of all authors were This work has been partially funded by
compared in an online meeting. In the the Knowledge Foundation of Sweden (KKS)
meeting, all authors discussed differences through the Synergy Project AIDA - A Holistic
between the outcomes until a final and AI-driven Networking and Processing Frame-
agreed consensus was reached. Automatic work for Industrial IoT (Rek:20200067).
filtering provided by Microsoft Excel was
also used to ensure the accuracy of the data R EFERENCES
extraction process. [1] A. Adolfsson, M. Ackerman, and N. C. Brownstein, “To
• Reproducibility of the study: The issue cluster, or not to cluster: An analysis of clusterability
here is whether other researchers can per- methods,” Pattern Recognition, vol. 88, pp. 13–26, 2019.
[2] P. Antonellis, D. Antoniou, Y. Kanellopoulos, C. Makris,
form this study with similar results. Ac- E. Theodoridis, C. Tjortjis, and N. Tsirakis, “Clustering
cordingly, all the steps followed and per- for Monitoring Software Systems Maintainability Evolu-
formed in this study were reported in the tion,” Electronic Notes in Theoretical Computer Science, vol.
233, pp. 43–57, 2009.
research methodology (see Section 3). [3] G. Şerban and I.-G. Czibula, “Object-Oriented Software
Systems Restructuring through Clustering,” in Interna-
tional Conference on Artificial Intelligence and Soft Comput-
6 C ONCLUSION ing (ICAISC), 2008, pp. 693–704.
[4] J. Feng and H. Seok, “Applying agglomerative hierar-
This paper systematically reports the state- chical clustering algorithms to component identification
of-the-art empirical contributions in software for legacy systems,” Information and Software Technology,
vol. 53, no. 6, pp. 601–614, 2011.
module clustering. Thus, to ascertain the re- [5] R. A. Bittencourt and D. D. S. Guerrero, “Comparison of
cent clustering applications in software engi- Graph Clustering Algorithms for Recovering Software
neering, the algorithms and tools used to en- Architecture Module Views,” in the 13th European Con-
ference on Software Maintenance and Reengineering. IEEE,
able the software module clustering process 2009, pp. 251–254.
were identified. A total of 143 papers from [6] B. Joshi, P. Budhathoki, W. L. Woon, and D. Svetinovic,
popular literature databases published in the “Software Clone Detection Using Clustering Approach,”
in International Conference on Neural Information Process-
area of software module clustering from 2008- ing, 2015, pp. 520–527.
2019 were selected for this study. The pub- [7] R. Naseem, A. Ahmed, S. U. Khan, M. Saqib, and
lished papers were a combination of works M. Habib, “Program restructuring using agglomerative
from conferences, journals, symposiums, and clustering technique based on binary features,” in 2012
International Conference on Emerging Technologies. IEEE,
workshops. However, most of the published 2012, pp. 1–6.
papers were from conferences. From different [8] M. Kargar, A. Isazadeh, and H. Izadkhah, “Multi-
perspectives and based on several identified programming language software systems modulariza-
tion,” Computers and Electrical Engineering, vol. 80, pp.
RQs, the selected studies were thoroughly re- 1–22, 2019.
viewed and analyzed. The findings were in [9] Y. Yu, J. Lu, J. Fernandez-Ramil, and P. Yuan, “Com-
different categories. For instance, statistics on paring web services with other software components,”
in IEEE International Conference on Web Services (ICWS),
the published studies, their publication venues, 2007, pp. 388–397.
active authors, and countries were reported. [10] N. Arunachalam, A. Amuthan, C. Kavya, M. Sharmilla,
Then, software module clustering applications K. Ushanandhini, and M. Shanmughapriya, “A survey
on web service clustering,” in 2017 International Con-
were categorized. All the algorithms, tools, tar- ference on Computation of Power, Energy Information and
get software systems, evaluations, and metrics Commuincation (ICCPEIC), 2017, pp. 247–252.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 24

[11] M. Shtern and V. Tzerpos, “Clustering Methodologies for [27] A. Parashar and J. K. Chhabra, “Clustering Dynamic
Software Engineering,” Advances in Software Engineering, Class Coupling Data to Measure Class Reusability Pat-
vol. 2012, pp. 1–18, 2012. tern,” in International Conference on High Performance Ar-
[12] V. Singh, “Software module clustering using metaheuris- chitecture and Grid Computing, 2011, pp. 126–130.
tic search techniques: A survey,” in 2016 3rd International [28] Amarjeet and J. K. Chhabra, “An empirical study of
Conference on Computing for Sustainable Global Develop- the sensitivity of quality indicator for software module
ment (INDIACom), 2016, pp. 2764–2767. clustering,” in 2014 Seventh International Conference on
[13] F. Morsali and M. R. Keyvanpour, “Search-based soft- Contemporary Computing (IC3). IEEE, 2014, pp. 206–211.
ware module clustering techniques: A review article,” in [29] A. Prajapati and J. K. Chhabra, “A Particle Swarm
2017 IEEE 4th International Conference on Knowledge-Based Optimization-Based Heuristic for Software Module Clus-
Engineering and Innovation (KBEI), 2017, pp. 0977–0983. tering Problem,” Arabian Journal for Science and Engineer-
[14] M. Harman, “The current state and future of search ing, vol. 43, no. 12, pp. 7083–7094, 2017.
based software engineering,” in Future of Software En- [30] Amarjeet and J. K. Chhabra, “FP-ABC: Fuzzy-Pareto
gineering (FOSE ’07), 2007, pp. 342–357. dominance driven artificial bee colony algorithm for
[15] M. Harman, S. A. Mansouri, and Y. Zhang, “Search- many-objective software module clustering,” Computer
based software engineering: Trends, techniques and ap- Languages, Systems & Structures, vol. 51, pp. 1–21, 2017.
plications,” ACM Computing Surveys (CSUR), vol. 45, [31] M. Kargar, A. Isazadeh, and H. Izadkhah, “Semantic-
no. 1, pp. 11:1–11:61, 2012. based software clustering using hill climbing,” in 2017
[16] A. Ramírez, J. R. Romero, and S. Ventura, “A survey International Symposium on Computer Science and Software
of many-objective optimisation in search-based software Engineering Conference (CSSE). IEEE, 2017, pp. 55–60.
engineering,” Journal of Systems and Software, vol. 149, pp. [32] A. Rathee and J. K. Chhabra, “Software Remodulariza-
382–395, 2019. tion by Estimating Structural and Conceptual Relations
[17] K. Petersen, S. Vakkalanka, and L. Kuzniarz, “Guidelines Among Classes and Using Hierarchical Clustering,” in
for conducting systematic mapping studies in software International Conference on Advanced Informatics for Com-
engineering: An update,” Information and Software Tech- puting Research (ICAICR), 2017, pp. 94–106.
nology, vol. 64, pp. 1–18, 2015. [33] Amarjeet and J. K. Chhabra, “Improving package struc-
[18] B. Kitchenham and S. Charters, “Guidelines for Perform- ture of object-oriented software using multi-objective
ing Systematic Literature Reviews in Software Engineer- optimization and weighted class connections,” Journal of
ing,” version 2.3., EBSE Technical Report EBSE- 2007-01, King Saud University - Computer and Information Sciences,
Software Engineering Group, School of Computer Science vol. 29, no. 3, pp. 349–364, 2017.
and Mathematics, Keele University, UK and Department of [34] A. Rathee and J. Kumar, “Improving Cohesion of a
Computer Science, University of Durham, 2007. Software System by Performing Usage Pattern Based
[19] O. Pedreira, F. García, N. Brisaboa, and M. Piattini, Clustering,” Procedia Computer Science, vol. 125, pp. 740–
“Gamification in software engineering – A systematic 746, 2018.
mapping,” Information and Software Technology, vol. 57, [35] Amarjeet and J. K. Chhabra, “Many-objective artificial
pp. 157–168, 2015. bee colony algorithm for large-scale software module
[20] M. Usman, R. Britto, J. Börstler, and E. Mendes, “Tax- clustering problem,” Soft Computing, vol. 22, no. 19, pp.
onomies in software engineering: A Systematic mapping 6341–6361, 2018.
study and a revised taxonomy development method,” [36] A. Rathee and J. K. Chhabra, “A multi-objective search
Information and Software Technology, vol. 85, pp. 43–59, based approach to identify reusable software compo-
2017. nents,” Journal of Computer Languages, vol. 52, pp. 26–43,
[21] B. S. Ahmed, K. Z. Zamli, W. Afzal, and M. Bures, 2019.
“Constrained Interaction Testing: A Systematic Litera- [37] A. Corazza, S. Di Martino, and G. Scanniello, “A Proba-
ture Study,” IEEE Access, vol. 5, pp. 25 706–25 730, 2017. bilistic Based Approach towards Software System Clus-
[22] P. Brereton, B. A. Kitchenham, D. Budgen, M. Turner, tering,” in 2010 14th European Conference on Software
and M. Khalil, “Lessons from applying the systematic Maintenance and Reengineering. IEEE, 2010, pp. 88–96.
literature review process within the software engineering [38] G. Scanniello, A. D’Amico, C. D’Amico, and T. D’Amico,
domain,” Journal of Systems and Software, vol. 80, no. 4, “Using the Kleinberg Algorithm and Vector Space Model
pp. 571–583, 2007. for Software System Clustering,” in 2010 IEEE 18th In-
[23] C. Wohlin, “Guidelines for snowballing in systematic ternational Conference on Program Comprehension. IEEE,
literature studies and a replication in software engineer- 2010, pp. 180–189.
ing,” in Proceedings of the 18th International Conference on [39] S. Romano, G. Scanniello, M. Risi, and C. Gravino, “Clus-
Evaluation and Assessment in Software Engineering - EASE tering and lexical information support for the recovery
’14. ACM Press, 2014, pp. 1–10. of design pattern in source code,” in 2011 27th IEEE
[24] Google. Google trends. [Online]. Available: International Conference on Software Maintenance (ICSM).
https://trends.google.com/trends/?geo=US,Accessed: IEEE, 2011, pp. 500–503.
10/03/2020. [40] G. Scanniello and A. Marcus, “Clustering Support for
[25] O. Pedreira, F. García, N. Brisaboa, and M. Piattini, Static Concept Location in Source Code,” in 2011 IEEE
“Gamification in software engineering - a systematic 19th International Conference on Program Comprehension,
mapping,” Information and Software Technology, vol. 57, 2011, pp. 1–10.
pp. 157–168, 2015. [41] A. Corazza, S. D. Martino, V. Maggio, and G. Scanniello,
[26] R. E. Lopez-herrejon, L. Linsbauer, and A. Egyed, “A “Investigating the Use of Lexical Information for Soft-
systematic mapping study of search-based software en- ware System Clustering,” in 2011 15th European Confer-
gineering for software product lines,” Information and ence on Software Maintenance and Reengineering. IEEE,
Software Technology, vol. 61, pp. 33–51, 2015. 2011, pp. 35–44.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 25

[42] M. Risi, G. Scanniello, and G. Tortora, “Using fold-in [57] I. Czibula and G. Czibula, “Hierarchical clustering for
and fold-out in the architecture recovery of software adaptive refactorings identification,” in 2010 IEEE In-
systems,” Formal Aspects of Computing, vol. 24, no. 3, pp. ternational Conference on Automation, Quality and Testing,
307–330, 2012. Robotics (AQTR). IEEE, 2010, pp. 1–6.
[43] G. Scanniello, C. Gravino, A. Marcus, and T. Menzies, [58] Z. Marian, I. G. Czibula, and G. Czibula, “A hierarchical
“Class level fault prediction using software clustering,” clustering-based approach for software restructuring at
in 2013 28th IEEE/ACM International Conference on Auto- the package level,” in Proceedings of 19th International
mated Software Engineering (ASE). IEEE, 2013, pp. 640– Symposium on Symbolic and Numeric Algorithms for Sci-
645. entific Computing (SYNASC). IEEE, 2017, pp. 239–246.
[44] A. Corazza, S. Di Martino, V. Maggio, and G. Scanniello, [59] A. Alkhalid, C.-H. Lung, and S. Ajila, “Software architec-
“Weighing lexical information for software clustering in ture decomposition using adaptive K-nearest neighbor
the context of architecture recovery,” Empirical Software algorithm,” in 2013 26th IEEE Canadian Conference on
Engineering, vol. 21, no. 1, pp. 72–103, 2016. Electrical and Computer Engineering (CCECE). IEEE, 2013,
[45] S. Mohammadi and H. Izadkhah, “A new algorithm for pp. 1–4.
software clustering considering the knowledge of depen- [60] A. Alkhalid, C.-h. Lung, D. Liu, and S. Ajila, “Software
dency between artifacts in the source code,” Information Architecture Decomposition Using Clustering Tech-
and Software Technology, vol. 105, pp. 252–256, 2019. niques,” in 2013 IEEE 37th Annual Computer Software and
[46] H. Masoud and S. Jalili, “A clustering-based model Applications Conference. IEEE, 2013, pp. 806–811.
for class responsibility assignment problem in object- [61] D. Liu, C.-h. Lung, and S. A. Ajila, “Adaptive Clustering
oriented analysis,” Journal of Systems and Software, vol. 93, Techniques for Software Components and Architecture,”
pp. 110–131, 2014. in 2015 IEEE 39th Annual Computer Software and Applica-
[47] M. Akbari and H. Izadkhah, “Hybrid of genetic algo- tions Conference. IEEE, 2015, pp. 460–465.
rithm and krill herd for software clustering problem,” [62] J. Huang and J. Liu, “A similarity-based modularization
2019 IEEE 5th Conference on Knowledge Based Engineering quality measure for software module clustering prob-
and Innovation (KBEI), pp. 565–570, 2019. lems,” Information Sciences, vol. 342, pp. 96–110, 2016.
[48] A. H. Farajpour Tabrizi and H. Izadkhah, “Software [63] J. Huang, J. Liu, and X. Yao, “A multi-agent evolutionary
Modularization by Combining Genetic and Hierarchical algorithm for software module clustering problems,” Soft
Algorithms,” in 5th Conference on Knowledge Based Engi- Computing, vol. 21, no. 12, pp. 3415–3428, 2017.
neering and Innovation (KBEI). IEEE, 2019, pp. 454–459. [64] W. Pan, B. Li, J. Liu, Y. Ma, and B. Hu, “Analyzing
[49] M. Kargar, H. Izadkhah, and A. Isazadeh, “Tarimliq: A the structure of Java software systems by weighted k-
new internal metric for software clustering analysis,” in core decomposition,” Future Generation Computer Sys-
27th Iranian Conference on Electrical Engineering (ICEE). tems, vol. 83, pp. 431–444, 2018.
IEEE, 2019, pp. 1879–1883. [65] M. Risi, G. Scanniello, and G. Tortora, “Architecture
[50] R. Naseem, O. Maqbool, and S. Muhammad, “Cooper- Recovery Using Latent Semantic Indexing and K-Means:
ative clustering for software modularization,” Journal of An Empirical Evaluation,” in 2010 8th IEEE International
Systems and Software, vol. 86, no. 8, pp. 2045–2062, 2013. Conference on Software Engineering and Formal Methods.
[51] R. Naseem and M. M. Deris, “A New Binary Similar- IEEE, 2010, pp. 103–112.
ity Measure Based on Integration of the Strengths of [66] A. Alkhalid, M. Alshayeb, and S. Mahmoud, “Software
Existing Measures: Application to Software Clustering,” refactoring at the function level using new Adaptive K-
in International Conference on Soft Computing and Data Nearest Neighbor algorithm,” Advances in Engineering
Mining (SCDM), 2016, pp. 304–315. Software, vol. 41, no. 10-11, pp. 1160–1178, 2010.
[52] R. Naseem, M. M. Deris, O. Maqbool, and S. Shahzad, [67] A. Alkhalid, M. Alshayeb, and S. Mahmoud, “Software
“Euclidean space based hierarchical clusterers combi- refactoring at the package level using clustering tech-
nations: an application to software clustering,” Cluster niques,” IET Software, vol. 5, no. 3, pp. 274–286, 2011.
Computing, vol. 22, pp. 7287–7311, 2019. [68] M. D. O. Barros, “An analysis of the effects of composite
[53] Z. Shah, R. Naseem, M. A. Orgun, A. Mahmood, and objectives in multiobjective software module clustering,”
S. Shahzad, “Software Clustering Using Automated Fea- in Proceedings of the 14th international conference on Genetic
ture Subset Selection,” in Advanced Data Mining and and evolutionary computation conference (GECCO). ACM
Applications: 9th International Conference, ADMA Proceed- Press, 2012, pp. 1205–1212.
ings, ser. Lecture Notes in Computer Science. Berlin, [69] M. C. Monçores, A. C. Alvim, and M. O. Barros, “Large
Heidelberg: Springer Berlin Heidelberg, 2013, pp. 47–58. Neighborhood Search applied to the Software Module
[54] S. Muhammad, O. Maqbool, and A. Abbasi, “Evaluat- Clustering problem,” Computers & Operations Research,
ing relationship categories for clustering object-oriented vol. 91, pp. 92–111, 2017.
software systems,” IET Software, vol. 6, no. 3, pp. 260– [70] P. Antonellis, D. Antoniou, Y. Kanellopoulos, C. Makris,
274, 2012. E. Theodoridis, C. Tjortjis, and N. Tsirakis, “Employing
[55] R. Naseem, M. B. M. Deris, O. Maqbool, J. peng Li, Clustering for Assisting Source Code Maintainability
S. Shahzad, and H. Shah, “Improved binary similarity Evaluation according to ISO / IEC- 9126,” in Artificial
measures for software modularization,” Frontiers of In- Intelligence Techniques in Software Engineering Workshop
formation Technology and Electronic Engineering, vol. 18, (AISEW), 2008, pp. 1–5.
pp. 1082–1107, 2017. [71] S. Arshad and C. Tjortjis, “Clustering Software Metric
[56] I. G. Czibula and G. Czibula, “Clustering based auto- Values Extracted from C# Code for Maintainability As-
matic refactorings identification,” in Proceedings of the sessment,” in Proceedings of the 9th Hellenic Conference on
2008 10th International Symposium on Symbolic and Nu- Artificial Intelligence - SETN ’16. New York, New York,
meric Algorithms for Scientific Computing, SYNASC 2008, USA: ACM Press, 2016, pp. 1–4.
2008, pp. 253–256. [72] D. Papas and C. Tjortjis, “Combining Clustering and
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 26

Classification for Software Quality Evaluation,” in Hel- in 2011 18th Asia-Pacific Software Engineering Conference.
lenic Conference on Artificial Intelligence, 2014, pp. 273–286. IEEE, 2011, pp. 315–321.
[73] Linhui Zhong, Liangbo Xue, Nengwei Zhang, Jing Xia, [88] K. Praditwong, M. Harman, and X. Yao, “Software Mod-
and Jun Chen, “A tool to support software clustering ule Clustering as a Multi-Objective Search Problem,”
using the software evolution information,” in 2016 7th IEEE Transactions on Software Engineering, vol. 37, no. 2,
IEEE International Conference on Software Engineering and pp. 264–282, 2011.
Service Science (ICSESS). IEEE, 2016, pp. 304–307. [89] K. Praditwong, “Solving software module clustering
[74] L. Zhong, J. He, N. Zhang, P. Zhang, and J. Xia, “Soft- problem by evolutionary algorithms,” in 2011 Eighth
ware Evolution Information Driven Service-Oriented International Joint Conference on Computer Science and Soft-
Software Clustering,” in EEE International Congress on Big ware Engineering (JCSSE). IEEE, 2011, pp. 154–159.
Data. IEEE, 2016, pp. 493–500. [90] K. Kobayashi, M. Kamimura, K. Kato, K. Yano, and
[75] C. Y. Chong, S. P. Lee, and T. C. Ling, “Efficient software A. Matsuo, “Feature-gathering dependency-based soft-
clustering technique using an adaptive and preventive ware clustering using Dedication and Modularity,” in
dendrogram cutting approach,” Information and Software 2012 28th IEEE International Conference on Software Main-
Technology, vol. 55, no. 11, pp. 1994–2012, 2013. tenance (ICSM). IEEE, 2012, pp. 462–471.
[76] C. Y. Chong and S. P. Lee, “Constrained agglomerative [91] J. Misra, K. Annervaz, V. Kaulgud, S. Sengupta, and
hierarchical software clustering with hard and soft con- G. Titus, “Software Clustering: Unifying Syntactic and
straints,” Proceedings of the 10th International Conference Semantic Features,” in 2012 19th Working Conference on
on Evaluation of Novel Approaches to Software Engineering Reverse Engineering. IEEE, 2012, pp. 113–122.
(ENASE), pp. 177–188, 2015. [92] G. E. Boussaidi, A. B. Belle, S. Vaucher, and H. Mili, “Re-
[77] M. N. Adnan, M. R. Islam, and S. Hossain, “Clustering constructing Architectural Views from Legacy Systems,”
software systems to identify subsystem structures using in the 19th Working Conference on Reverse Engineering.
knowledgebase,” in Malaysian Conference in Software En- IEEE, 2012, pp. 345–354.
gineering, 2011, pp. 445–450. [93] A. Mahmoud and Nan Niu, “Evaluating software clus-
[78] K. Jeet and R. Dhir, “Software Architecture Recovery tering algorithms in the context of program comprehen-
using Genetic Black Hole Algorithm,” ACM SIGSOFT sion,” in 2013 21st International Conference on Program
Software Engineering Notes, vol. 40, no. 1, pp. 1–5, 2015. Comprehension (ICPC). IEEE, 2013, pp. 162–171.
[79] A. S. Mamaghani and M. R. Meybodi, “Clustering of [94] V. Köhler, M. Fampa, and O. Araújo, “Mixed-Integer Lin-
Software Systems Using New Hybrid Algorithms,” in ear Programming Formulations for the Software Clus-
The 9th IEEE International Conference on Computer and tering Problem,” Computational Optimization and Applica-
Information Technology. IEEE, 2009, pp. 20–25. tions, vol. 55, no. 1, pp. 113–135, 2013.
[80] Z. Han, L. Wang, L. Yu, X. Chen, J. Zhao, and X. Li, “De- [95] A. C. Kumari, K. Srinivas, and M. P. Gupta, “Software
sign pattern directed clustering for understanding open module clustering using a hyper-heuristic based multi-
source code,” in 2009 IEEE 17th International Conference objective genetic algorithm,” in 2013 3rd IEEE Interna-
on Program Comprehension. IEEE, 2009, pp. 295–296. tional Advance Computing Conference (IACC). IEEE, 2013,
[81] F. Beck and S. Diehl, “Evaluating the Impact of Software pp. 813–818.
Evolution on Software Clustering,” in 2010 17th Working [96] C. Deiters, A. Rausch, and M. Schindler, “Using spectral
Conference on Reverse Engineering. IEEE, 2010, pp. 99– clustering to automate identification and optimization
108. of component structures,” in 2013 2nd International Work-
[82] Y. Wang, P. Liu, H. Guo, H. Li, and X. Chen, “Im- shop on Realizing Artificial Intelligence Synergies in Software
proved Hierarchical Clustering Algorithm for Software Engineering (RAISE). IEEE, 2013, pp. 14–20.
Architecture Recovery,” in 2010 International Conference [97] A. Ibrahim, D. Rayside, and R. Kashef, “Cooperative
on Intelligent Computing and Cognitive Informatics. IEEE, based software clustering on dependency graphs,” in
2010, pp. 247–250. 2014 IEEE 27th Canadian Conference on Electrical and Com-
[83] Q. Zhang, D. Qiu, Q. Tian, and L. Sun, “Object-oriented puter Engineering (CCECE). IEEE, 2014, pp. 1–6.
software architecture recovery using a new hybrid clus- [98] A. M. Saeidi, J. Hage, R. Khadka, and S. Jansen, “A
tering algorithm,” in 2010 Seventh International Conference search-based approach to multi-view clustering of soft-
on Fuzzy Systems and Knowledge Discovery. IEEE, 2010, ware systems,” in 2015 IEEE 22nd International Confer-
pp. 2546–2550. ence on Software Analysis, Evolution, and Reengineering
[84] M. Shtern and V. Tzerpos, “On the Comparability of (SANER). IEEE, 2015, pp. 429–438.
Software Clustering Algorithms,” in 2010 IEEE 18th In- [99] M. Schindler, O. Fox, and A. Rausch, “Clustering
ternational Conference on Program Comprehension. IEEE, Source Code Elements by Semantic Similarity Using
2010, pp. 64–67. Wikipedia,” in 2015 IEEE/ACM 4th International Workshop
[85] I. Sora, G. Glodean, and M. Gligor, “Software architec- on Realizing Artificial Intelligence Synergies in Software
ture reconstruction: An approach based on combining Engineering. IEEE, 2015, pp. 13–18.
graph clustering and partitioning,” in 2010 International [100] J. Kaur and P. Tomar, “A software component selection
Joint Conference on Computational Cybernetics and Technical technique based on fuzzy clustering,” in 2016 1st India
Informatics. IEEE, 2010, pp. 259–264. International Conference on Information Processing (IICIP).
[86] M. Faunes, M. Kessentini, and H. Sahraoui, “Deriving IEEE, 2016, pp. 1–5.
High-level Abstractions from Legacy Software Using [101] D. Jensen and A. Lundkvist, “On the significance of re-
Example-driven Clustering,” in Proceedings of the 2011 lationship directions in clustering algorithms for reverse
Conference of the Center for Advanced Studies on Collabora- engineering,” in Proceedings of the Symposium on Applied
tive Research, 2011, pp. 188–199. Computing - SAC ’17. New York, New York, USA: ACM
[87] U. Erdemir, U. Tekin, and F. Buzluca, “Object Oriented Press, 2017, pp. 1239–1244.
Software Clustering Based on Community Structure,” [102] X. Li, L. Zhang, and N. Ge, “Framework Informa-
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 27

tion Based Java Software Architecture Recovery,” in [117] R. Adnan, B. Graaf, A. van Deursen, and J. Zonn-
24th Asia-Pacific Software Engineering Conference Work- eveld, “Using Cluster Analysis to Improve the Design of
shops (APSECW). IEEE, 2017, pp. 114–120. Component Interfaces,” in The 23rd IEEE/ACM Interna-
[103] S. M. Naim, K. Damevski, and M. S. Hossain, “Recon- tional Conference on Automated Software Engineering (ASE).
structing and evolving software architectures using a IEEE, sep 2008, pp. 383–386.
coordinated clustering framework,” Automated Software [118] K. Cassell, P. Andreae, L. Groves, and J. Noble, “Towards
Engineering, vol. 24, pp. 543–572, 2017. automating class-splitting using betweenness cluster-
[104] M. Hall, N. Walkinshaw, and P. McMinn, “Effectively ing,” in ASE2009 - 24th IEEE/ACM International Confer-
incorporating expert knowledge in automated software ence on Automated Software Engineering, 2009, pp. 595–599.
remodularisation,” IEEE Transactions on Software Engi- [119] M. Fokaefs, N. Tsantalis, E. Stroulia, and A. Chatzi-
neering, vol. 44, no. 7, pp. 613–630, 2018. georgiou, “JDeodorant: Identification and Application
[105] J. Sun and B. Ling, “Software Module Clustering Al- of Extract Class Refactorings,” in Proceeding of the 33rd
gorithm Using Probability Selection,” Wuhan University international conference on Software engineering - ICSE ’11.
Journal of Natural Sciences, vol. 23, no. 2, pp. 93–102, 2018. New York, New York, USA: ACM Press, 2011, pp. 1037–
[106] K. Z. Zamli, F. Din, N. Ramli, and B. S. Ahmed, “Soft- 1039.
ware Module Clustering Based on the Fuzzy Adaptive [120] Y. Zhang, G. Huang, W. Zhang, X. Liu, and H. Mei,
Teaching Learning Based Optimization Algorithm,” in “Towards module-based automatic partitioning of Java
Proceedings of Intelligent and Interactive Computing (IIC), applications,” Frontiers of Computer Science, vol. 6, no. 6,
2018, pp. 167–177. pp. 725–740, 2012.
[107] C. Cho, K.-S. Lee, M. Lee, and C.-G. Lee, “Software [121] A. Hussain and M. S. Rahman, “A new hierarchical
Architecture Module-View Recovery Using Cluster En- clustering technique for restructuring software at the
sembles,” IEEE Access, vol. 7, pp. 72 872–72 884, 2019. function level,” in Proceedings of the 6th India Software
[108] M. Papachristou, “Software clusterings with vector se- Engineering Conference on - ISEC ’13. New York, New
mantics and the call graph,” in Proceedings of the 27th York, USA: ACM Press, 2013, pp. 45–54.
ACM Joint Meeting on European Software Engineering Con- [122] G. Santos, M. T. Valente, and N. Anquetil, “Remodular-
ference and Symposium on the Foundations of Software En- ization analysis using semantic clustering,” in 2014 Soft-
gineering (ESEC/FSE), 2019, pp. 1184–1186. ware Evolution Week - IEEE Conference on Software Main-
[109] H. Sözer, “Evaluating the effectiveness of multi-level tenance, Reengineering, and Reverse Engineering (CSMR-
greedy modularity clustering for software architecture WCRE). IEEE, 2014, pp. 224–233.
recovery,” in European Conference on Software Architecture [123] A. S. Mamaghani and M. Hajizadeh, “Software modu-
(ECSA), 2019, pp. 71–87. larization using the modified firefly algorithm,” in 2014
[110] L. Lövei, C. Hoch, H. Köllö, T. Nagy, A. Nagyné Víg, 8th. Malaysian Software Engineering Conference (MySEC).
D. Horpácsi, R. Kitlei, and R. Király, “Refactoring mod- IEEE, 2014, pp. 321–324.
ule structure,” in Proceedings of the 7th ACM SIGPLAN
[124] L. L. Silva, M. T. Valente, and M. d. A. Maia, “Assessing
workshop on ERLANG. ACM Press, 2008, pp. 83–89.
modularity using co-change clusters,” in Proceedings of
[111] L.-H. Zhong, L. Xu, M.-s. Ye, Y. Zheng, and B. Xie, the 13th international conference on Modularity. ACM
“An Approach for Software Architecture Refactoring Press, 2014, pp. 49–60.
Based on Clustering of Extended Component Depen-
dency Graph,” in International Conference on Computa- [125] M. W. Mkaouer, M. Kessentini, S. Bechikh, K. Deb, and
tional Intelligence and Software Engineering. IEEE, dec M. Ó Cinnéide, “High dimensional search-based soft-
2009, pp. 1–4. ware engineering: Finding Tradeoffs Among 15 Objec-
tives for Automating Software Refactoring Using NSGA-
[112] M. Fokaefs, N. Tsantalis, A. Chatzigeorgiou, and
III,” in Proceedings of the 2014 conference on Genetic and
J. Sander, “Decomposing object-oriented class modules
evolutionary computation (GECCO). ACM Press, 2014,
using an agglomerative clustering technique,” in 2009
pp. 1263–1270.
IEEE International Conference on Software Maintenance.
IEEE, 2009, pp. 93–101. [126] M. Paixao, M. Harman, and Y. Zhang, “Multi-objective
[113] M. Glorie, A. Zaidman, A. van Deursen, and L. Hofland, Module Clustering for Kate,” in International Symposium
“Splitting a large software repository for easing fu- on Search Based Software Engineering SSBSE 2015: Search-
ture software evolution-an industrial experience report,” Based Software Engineering, 2015, pp. 282–288.
Journal of Software Maintenance and Evolution: Research and [127] I. Šūpulniece, I. Pol, aka, S. Běrziša, E. Ozolin, š, E. Palacis,
Practice, vol. 21, no. 2, pp. 113–141, 2009. E. Meiers, and J. Grabis, “Source Code Driven Enter-
[114] Y. Liu, G. Guo, and J. Qi, “An algorithm of system de- prise Application Decomposition: Preliminary Evalua-
composition based on laplace spectral graph partitioning tion,” Procedia Computer Science, vol. 77, pp. 167–175,
technology,” in Proceedings - International Conference on 2015.
Computer Science and Software Engineering, CSSE 2008, [128] M. Bishnoi and P. Singh, “Modularizing Software Sys-
2008, pp. 85–89. tems using PSO optimized hierarchical clustering,” in
[115] J. Dietrich, V. Yakovlev, C. McCartin, G. Jenson, and 2016 International Conference on Computational Techniques
M. Duchrow, “Cluster analysis of Java dependency in Information and Communication Technologies (ICCTICT).
graphs,” in Proceedings of the 4th ACM symposium on IEEE, 2016, pp. 659–664.
Software visuallization (SoftVis). New York, New York, [129] J. Lee, D.-K. Kim, J. Park, and S. Park, “Class Modu-
USA: ACM Press, 2008, pp. 91–94. larization Using Indirect Relationships,” in the 22nd In-
[116] B. Khan, S. Sohail, and M. Y. Javed, “Evolution Strategy ternational Conference on Engineering of Complex Computer
Based Automated Software Clustering Approach,” in Systems (ICECCS). IEEE, 2017, pp. 110–119.
2008 Advanced Software Engineering and Its Applications [130] Y. Wang, H. Yu, Z. Zhu, W. Zhang, and Y. Zhao, “Auto-
(ASEA). IEEE, dec 2008, pp. 27–34. matic Software Refactoring via Weighted Clustering in
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 28

Method-Level Networks,” IEEE Transactions on Software based technique for software quality prediction through
Engineering, vol. 44, no. 3, pp. 202–236, 2018. clustering and chi-square test,” in 2015 International Con-
[131] B. G. Varghese R, K. Raimond, and J. Lovesum, “A ference on Applied and Theoretical Computing and Commu-
novel approach for automatic remodularization of soft- nication Technology (iCATccT). IEEE, 2015, pp. 238–245.
ware systems using extended ant colony optimization [146] P. S. Sandhu, M. Kaur, and A. Kaur, “A Density Based
algorithm,” Information and Software Technology, vol. 114, Clustering approach for early detection of fault prone
pp. 107–120, 2019. modules,” in 2010 International Conference on Electronics
[132] E. Hatami and B. Arasteh, “An efficient and stable and Information Engineering. IEEE, 2010, pp. 525–530.
method to cluster software modules using ant colony [147] X. Tan, X. Peng, S. Pan, and W. Zhao, “Assessing
optimization algorithm,” The Journal of Supercomputing, Software Quality by Program Clustering and Defect
pp. 1–23, 2019. Prediction,” in 2011 18th Working Conference on Reverse
[133] Y.-S. Seo and J.-H. Huh, “GUI-based software modular- Engineering. IEEE, 2011, pp. 244–248.
ization through module clustering in edge computing [148] C. M. Rosenberg and L. Moonen, “Improving problem
based IoT environments,” Journal of Ambient Intelligence identification via automated log clustering using dimen-
and Humanized Computing, vol. 22, pp. 7287–7311, 2019. sionality reduction,” in Proceedings of the 12th ACM/IEEE
[134] S. Kebir, A.-D. Seriai, A. Chaoui, and S. Chardigny, International Symposium on Empirical Software Engineering
“Comparing and Combining Genetic and Clustering and Measurement - ESEM ’18. New York, New York,
Algorithms for Software Component Identification from USA: ACM Press, 2018, pp. 1–10.
Object-Oriented Code,” in Proceedings of the 5th Interna- [149] M. Bailey, K.-i. K.-I. K.-i. K.-I. Lin, and L. Sherrell,
tional Conference on Computer Science and Software Engi- “Clustering Source Code Files to Predict Change Prop-
neering, 2012, pp. 1–8. agation during Software Maintenance,” in Proceedings of
[135] C. Srinivas, V. Radhakrishna, and C. V. G. Rao, “Clus- the Annual Southeast Conference, 2012, pp. 106–111.
tering Software Components for Component Reuse and [150] R. Benkoczi, D. Gaur, S. Hossain, and M. A. Khan,
Program Restructuring,” in Proceedings of the Second In- “A design structure matrix approach for measuring co-
ternational Conference on Innovative Computing and Cloud change-modularity of software products,” in Proceedings
Computing - ICCC ’13. New York, New York, USA: ACM of the 15th International Conference on Mining Software
Press, 2013, pp. 261–266. Repositories - MSR ’18, 2018, pp. 331–335.
[136] V. Radhakrishna, C. Srinivas, and C. Rao, “Document [151] A. K. Malviya and V. K. Yadav, “Maintenance activities in
Clustering Using Hybrid XOR Similarity Function for object oriented software systems using K-means cluster-
Efficient Software Component Reuse,” Procedia Computer ing technique: A review,” in 2012 CSI Sixth International
Science, vol. 17, pp. 121–128, 2013. Conference on Software Engineering (CONSEG). IEEE,
[137] C. Patel, A. Hamou-Lhadj, and J. Rilling, “Software 2012, pp. 1–5.
Clustering Using Dynamic Analysis and Static Depen- [152] B. Mathur and M. Kaushik, “In Object-Oriented Software
dencies,” in 2009 13th European Conference on Software Framework Improving Maintenance Exercises Through
Maintenance and Reengineering. IEEE, 2009, pp. 27–36. K-Means Clustering Approach,” in 2018 3rd International
[138] S. Vodithala and S. Pabboju, “A clustering technique Conference On Internet of Things: Smart Innovation and
based on the specifications of software components,” in Usages (IoT-SIU). IEEE, 2018, pp. 1–7.
2015 International Conference on Advanced Computing and [153] S. H. Hamad, T. Fergany, R. A. Ammar, and A. A. Abd
Communication Systems. IEEE, 2015, pp. 1–6. El-Raouf, “A Double K-Clustering Approach for restruc-
[139] S. Hasheminejad and S. Jalili, “CCIC: Clustering analysis turing Distributed Object-Oriented software,” in IEEE
classes to identify software components,” Information and Symposium on Computers and Communications (ISCC).
Software Technology, vol. 57, pp. 329–351, 2015. IEEE, jul 2008, pp. 169–174.
[140] A. C. Kumari and K. Srinivas, “Hyper-heuristic approach [154] A. A. El-raouf, “Restructuring Distributed Object-
for multi-objective software module clustering,” Journal Oriented Software Using Hierarchical Clustering,” in
of Systems and Software, vol. 117, pp. 384–401, 2016. Proceeding ICCOMP’09 Proceedings of the WSEAES 13th
[141] V. Karande, S. Chandra, Z. Lin, J. Caballero, L. Khan, international conference on Computers, 2009, pp. 412–416.
and K. Hamlen, “BCD: Decomposing Binary Code Into [155] A. Ashish, “Clones clustering using K-means,” Proceed-
Components Using Graph-Based Clustering,” in Pro- ings of the 10th International Conference on Intelligent Sys-
ceedings of the 2018 on Asia Conference on Computer and tems and Control (ISCO), pp. 1–6, 2016.
Communications Security - ASIACCS ’18. New York, New [156] M. Sudhamani and L. Rangarajan, “Code similarity
York, USA: ACM Press, 2018, pp. 393–398. detection through control statement and program fea-
[142] M. A. Saied, A. Ouni, H. Sahraoui, R. G. Kula, K. Inoue, tures,” Expert Systems with Applications, vol. 132, pp. 63–
and D. Lo, “Improving reusability of software libraries 75, 2019.
through usage pattern mining,” Journal of Systems and [157] P. Kreutzer, G. Dotzler, M. Ring, B. M. Eskofier, and
Software, vol. 145, pp. 164–179, 2018. M. Philippsen, “Automatic clustering of code changes,”
[143] C. Psarras, T. Diamantopoulos, and A. Symeonidis, “A in Proceedings of the 13th International Workshop on Mining
Mechanism for Automatically Summarizing Software Software Repositories - MSR ’16. New York, New York,
Functionality from Source Code,” in IEEE 19th Inter- USA: ACM Press, 2016, pp. 61–72.
national Conference on Software Quality, Reliability and [158] Q. Khan, U. Akram, W. H. Butt, and S. Rehman, “Im-
Security (QRS). IEEE, 2019, pp. 121–130. plementation and evaluation of optimized algorithm for
[144] R. Islam and K. Sakib, “A Package Based Clustering for software architectures analysis through unsupervised
enhancing software defect prediction accuracy,” in 2014 learning (clustering),” in 2016 17th International Confer-
17th International Conference on Computer and Information ence on Sciences and Techniques of Automatic Control and
Technology (ICCIT). IEEE, 2014, pp. 81–86. Computer Engineering (STA). IEEE, 2016, pp. 266–276.
[145] A. Ali, K. Choudhary, and A. Sharma, “Object oriented [159] B. T. Bennett, “Using hierarchical agglomerative cluster-
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 29

ing to locate potential aspect interference,” in Southeast-

Con 2017. IEEE, 2017, pp. 1–8.
[160] C. Ieva, A. Gotlieb, S. Kaci, and N. Lazaar, “Discovering
Program Topoi via Hierarchical Agglomerative Cluster-
ing,” IEEE Transactions on Reliability, vol. 67, no. 3, pp.
758–770, 2018.
[161] W. Jin, T. Liu, Q. Zheng, D. Cui, and Y. Cai,
“Functionality-Oriented Microservice Extraction Based
on Execution Trace Clustering,” in Proceedings - 2018
IEEE International Conference on Web Services, ICWS 2018
- Part of the 2018 IEEE World Congress on Services. IEEE,
2018, pp. 211–218.
[162] A. Vanya, L. Holland, S. Klusener, P. van de Laar, and
H. van Vliet, “Assessing software archives with evolu-
tionary clusters,” in The 16th IEEE International Conference
on Program Comprehension (ICPC). IEEE, jun 2008, pp.
192–201.
[163] M. Shtern and V. Tzerpos, “Methods for selecting and
improving software clustering algorithms,” in IEEE 17th
International Conference on Program Comprehension. IEEE,
2009, pp. 248–252.
[164] TIOBE. The tiobe programming community index. [On-
line]. Available: https://www.tiobe.com/tiobe-index,
Accessed:01/03/2020.
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 30

TABLE 12: List of all papers included in the study.

ID# Ref. Paper Title Year
1 [153] A Double K-Clustering Approach for restructuring Distributed Object-Oriented software 2008
2 [114] An algorithm of system decomposition based on laplace spectral graph partitioning technology 2008
3 [162] Assessing software archives with evolutionary clusters 2008
4 [115] Cluster analysis of Java dependency graphs 2008
5 [56] Clustering based automatic refactorings identification 2008
6 [70] Employing Clustering for Assisting Source Code Maintainability Evaluation according to ISO / IEC- 9126 2008
7 [116] Evolution Strategy Based Automated Software Clustering Approach 2008
8 [3] Object-Oriented Software Systems Restructuring through Clustering 2008
9 [110] Refactoring module structure 2008
10 [117] Using Cluster Analysis to Improve the Design of Component Interfaces 2008
11 [111] An Approach for Software Architecture Refactoring Based on Clustering of Extended Component Dependency Graph 2009
12 [2] Clustering for Monitoring Software Systems Maintainability Evolution 2009
13 [79] Clustering of Software Systems Using New Hybrid Algorithms 2009
14 [5] Comparison of Graph Clustering Algorithms for Recovering Software Architecture Module Views 2009
15 [112] Decomposing object-oriented class modules using an agglomerative clustering technique 2009
16 [80] Design pattern directed clustering for understanding open source code 2009
17 [154] Restructuring Distributed Object-Oriented Software Using Hierarchical Clustering 2009
18 [137] Software Clustering Using Dynamic Analysis and Static Dependencies 2009
19 [113] Splitting a large software repository for easing future software evolution-an industrial experience report 2009
20 [118] Towards automating class-splitting using betweenness clustering 2009
21 [146] A Density Based Clustering approach for early detection of fault prone modules 2010
22 [37] A Probabilistic Based Approach towards Software System Clustering 2010
23 [65] Architecture Recovery Using Latent Semantic Indexing and K-Means: An Empirical Evaluation 2010
24 [81] Evaluating the Impact of Software Evolution on Software Clustering 2010
25 [57] Hierarchical clustering for adaptive refactorings identification 2010
26 [82] Improved Hierarchical Clustering Algorithm for Software Architecture Recovery 2010
27 [83] Object-oriented software architecture recovery using a new hybrid clustering algorithm 2010
28 [84] On the Comparability of Software Clustering Algorithms 2010
29 [85] Software architecture reconstruction: An approach based on combining graph clustering and partitioning 2010
30 [66] Software refactoring at the function level using new Adaptive K-Nearest Neighbor algorithm 2010
31 [38] Using the Kleinberg Algorithm and Vector Space Model for Software System Clustering 2010
32 [4] Applying agglomerative hierarchical clustering algorithms to component identification for legacy systems 2011
33 [147] Assessing Software Quality by Program Clustering and Defect Prediction 2011
34 [39] Clustering and lexical information support for the recovery of design pattern in source code 2011
35 [27] Clustering Dynamic Class Coupling Data to Measure Class Reusability Pattern 2011
36 [77] Clustering software systems to identify subsystem structures using knowledgebase 2011
37 [40] Clustering Support for Static Concept Location in Source Code 2011
38 [86] Deriving High-level Abstractions from Legacy Software Using Example-driven Clustering 2011
39 [41] Investigating the Use of Lexical Information for Software System Clustering 2011
40 [119] JDeodorant: Identification and Application of Extract Class Refactorings 2011
41 [87] Object oriented software clustering based on community structure 2011
42 [88] Software Module Clustering as a Multi-Objective Search Problem 2011
43 [67] Software refactoring at the package level using clustering techniques 2011
44 [89] Solving software module clustering problem by evolutionary algorithms 2011
45 [68] An analysis of the effects of composite objectives in multiobjective software module clustering 2012
46 [149] Clustering Source Code Files to Predict Change Propagation during Software Maintenance 2012
47 [134] Comparing and Combining Genetic and Clustering Algorithms for Software Component Identification from Object- 2012
Oriented Code
48 [54] Evaluating relationship categories for clustering object-oriented software systems 2012
49 [90] Feature-gathering dependency-based software clustering using Dedication and Modularity 2012
50 [151] Maintenance activities in object oriented software systems using K-means clustering technique: A review 2012
51 [7] Program restructuring using agglomerative clustering technique based on binary features 2012
52 [92] Reconstructing Architectural Views from Legacy Systems 2012
53 [91] Software Clustering: Unifying Syntactic and Semantic Features 2012
54 [120] Towards module-based automatic partitioning of Java applications 2012
55 [42] Using fold-in and fold-out in the architecture recovery of software systems 2012
56 [121] A new hierarchical clustering technique for restructuring software at the function level 2013
57 [43] Class level fault prediction using software clustering 2013
58 [135] Clustering Software Components for Component Reuse and Program Restructuring 2013
59 [50] Cooperative clustering for software modularization 2013
60 [136] Document Clustering Using Hybrid XOR Similarity Function for Efficient Software Component Reuse 2013
61 [75] Efficient software clustering technique using an adaptive and preventive dendrogram cutting approach 2013
62 [93] Evaluating software clustering algorithms in the context of program comprehension 2013
63 [94] Mixed-Integer Linear Programming Formulations for the Software Clustering Problem 2013
64 [59] Software architecture decomposition using adaptive K-nearest neighbor algorithm 2013
65 [60] Software Architecture Decomposition Using Clustering Techniques 2013
66 [53] Software Clustering Using Automated Feature Subset Selection 2013
67 [95] Software module clustering using a hyper-heuristic based multi-objective genetic algorithm 2013
68 [96] Using spectral clustering to automate identification and optimization of component structures 2013
69 [46] A clustering-based model for class responsibility assignment problem in object-oriented analysis 2014
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 31

70 [144] A Package Based Clustering for enhancing software defect prediction accuracy 2014
71 [28] An empirical study of the sensitivity of quality indicator for software module clustering 2014
72 [124] Assessing modularity using co-change clusters 2014
73 [72] Combining Clustering and Classification for Software Quality Evaluation 2014
74 [97] Cooperative based software clustering on dependency graphs 2014
75 [125] High dimensional search-based software engineering: Finding Tradeoffs Among 15 Objectives for Automating 2014
Software Refactoring Using NSGA-III
76 [122] Remodularization analysis using semantic clustering 2014
77 [123] Software modularization using the modified firefly algorithm 2014
78 [138] A clustering technique based on the specifications of software components 2015
79 [98] A search-based approach to multi-view clustering of software systems 2015
80 [61] Adaptive Clustering Techniques for Software Components and Architecture 2015
81 [139] CCIC: Clustering analysis classes to identify software components 2015
82 [99] Clustering Source Code Elements by Semantic Similarity Using Wikipedia 2015
83 [76] Constrained agglomerative hierarchical software clustering with hard and soft constraints 2015
84 [126] Multi-objective Module Clustering for Kate 2015
85 [145] Object oriented based technique for software quality prediction through clustering and chi-square test 2015
86 [78] Software Architecture Recovery using Genetic Black Hole Algorithm 2015
87 [6] Software Clone Detection Using Clustering Approach 2015
88 [127] Source Code Driven Enterprise Application Decomposition: Preliminary Evaluation 2015
89 [51] A New Binary Similarity Measure Based on Integration of the Strengths of Existing Measures: Application to Software 2016
Clustering
90 [62] A similarity-based modularization quality measure for software module clustering problems 2016
91 [100] A software component selection technique based on fuzzy clustering 2016
92 [73] A tool to support software clustering using the software evolution information 2016
93 [157] Automatic clustering of code changes 2016
94 [155] Clones clustering using K-means 2016
95 [71] Clustering Software Metric Values Extracted from C# Code for Maintainability Assessment 2016
96 [140] Hyper-heuristic Approach for Multi-Objective Software Module Clustering 2016
97 [158] Implementation and evaluation of optimized algorithm for software architectures analysis through unsupervised 2016
learning (clustering)
98 [128] Modularizing Software Systems using PSO optimized hierarchical clustering 2016
99 [74] Software Evolution Information Driven Service-Oriented Software Clustering 2016
100 [44] Weighing lexical information for software clustering in the context of architecture recovery 2016
101 [58] A hierarchical clustering-based approach for software restructuring at the package level 2017
102 [63] A multi-agent evolutionary algorithm for software module clustering problems 2017
103 [29] A Particle Swarm Optimization-Based Heuristic for Software Module Clustering Problem 2017
104 [129] Class Modularization Using Indirect Relationships 2017
105 [30] FP-ABC: Fuzzy-Pareto dominance driven artificial bee colony algorithm for many-objective software module 2017
clustering
106 [102] Framework Information Based Java Software Architecture Recovery 2017
107 [55] Improved binary similarity measures for software modularization 2017
108 [33] Improving package structure of object-oriented software using multi-objective optimization and weighted class 2017
connections
109 [69] Large Neighborhood Search applied to the Software Module Clustering problem 2017
110 [101] On the significance of relationship directions in clustering algorithms for reverse engineering 2017
111 [103] Reconstructing and evolving software architectures using a coordinated clustering framework 2017
112 [31] Semantic-based software clustering using hill climbing 2017
113 [32] Software Remodularization by Estimating Structural and Conceptual Relations Among Classes and Using Hierarchical 2017
Clustering
114 [159] Using hierarchical agglomerative clustering to locate potential aspect interference 2017
115 [150] A design structure matrix approach for measuring co-change-modularity of software products 2018
116 [64] Analyzing the structure of Java software systems by weighted K-core decomposition 2018
117 [130] Automatic Software Refactoring via Weighted Clustering in Method-Level Networks 2018
118 [141] BCD: Decomposing Binary Code Into Components Using Graph-Based Clustering 2018
119 [160] Discovering Program Topoi via Hierarchical Agglomerative Clustering 2018
120 [104] Effectively incorporating expert knowledge in automated software remodularisation 2018
121 [161] Functionality-Oriented Microservice Extraction Based on Execution Trace Clustering 2018
122 [34] Improving Cohesion of a Software System by Performing Usage Pattern Based Clustering 2018
123 [148] Improving Problem Identification via Automated Log Clustering using Dimensionality Reduction 2018
124 [142] Improving reusability of software libraries through usage pattern mining 2018
125 [152] In Object-Oriented Software Framework Improving Maintenance Exercises Through K-Means Clustering Approach 2018
126 [35] Many-objective artificial bee colony algorithm for large-scale software module clustering problem 2018
127 [105] Software Module Clustering Algorithm Using Probability Selection 2018
128 [106] Software Module Clustering Based on the Fuzzy Adaptive Teaching Learning Based Optimization Algorithm 2018
129 [143] A Mechanism for Automatically Summarizing Software Functionality from Source Code 2019
130 [36] A multi-objective search based approach to identify reusable software components 2019
131 [45] A new algorithm for software clustering considering the knowledge of dependency between artifacts in the source 2019
code
132 [131] A novel approach for automatic remodularization of software systems using extended ant colony optimization 2019
algorithm
133 [132] An efficient and stable method to cluster software modules using ant colony optimization algorithm 2019
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 32

134 [156] Code similarity detection through control statement and program features 2019
135 [52] Euclidean space based hierarchical clusterers combinations: an application to software clustering 2019
136 [109] Evaluating the effectiveness of multi-level greedy modularity clustering for software architecture recovery 2019
137 [133] GUI-based software modularization through module clustering in edge computing based IoT environments 2019
138 [47] Hybrid of genetic algorithm and krill herd for software clustering problem 2019
139 [8] Multi-programming language software systems modularization 2019
140 [107] Software Architecture Module-View Recovery Using Cluster Ensembles 2019
141 [108] Software clusterings with vector semantics and the call graph 2019
142 [48] Software Modularization by Combining Genetic and Hierarchical Algorithms 2019
143 [49] Tarimliq: A new internal metric for software clustering analysis 2019
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 33

TABLE 13: List of active journals with abbreviations.

Acronym Journal Full Name
Inf. Softw. Technol. Information and Software Technology
J. Syst. Softw. Journal of Systems and Software
IEEE Trans. Softw. Eng. IEEE Transactions on Software Engineering
Procedia Comput. Sci. Procedia Computer Science
Soft Comput. Soft Computing
IET Softw. IET Software
J. Software Maint. Evol. Res. Pract. Journal of Software Maintenance and Evolution: Research and Practice
Inf. Sci. Information Sciences
Wuhan Univ. J. Nat. Sci. Wuhan University Journal of Natural Sciences
IEEE Trans. Reliab. IEEE Transactions on Reliability
Future Gener. Comput. Syst. Future Generation Computer Systems
Empir. Softw. Eng. Empirical Software Engineering
Comput. Oper. Res. Computers and Operations Research
Expert Syst. Appl. Expert Systems with Applications
J. Supercomput. Journal of Supercomputing
Comput. Lang. Syst. Struct. Computer Languages, Systems, and Structures
Comput. Optim. Appl. Computational Optimization and Applications
Arabian J. Sci. Eng. Arabian Journal for Science and Engineering
Adv. Eng. Softw. Advances in Engineering Software
J. King Saud Univ. Comp. Info. Sci. Journal of King Saud University: Computer and information sciences
Front. Comput. Sci. Frontiers of Computer Science
Form. Asp. Comput. Formal Aspects of Computing
Cluster Comput. Cluster Computing
J. Ambient Intell. Humaniz. Comput. Journal of Ambient Intelligence and Humanized Computing
Comput. Electr. Eng. Computers and Electrical Engineering
J. Comput. Lang. Journal of Computer Languages
IEEE Access IEEE Access
Front. Inf. Technol. Electron. Eng. Frontiers of Information Technology and Electronic Engineering
Autom. Softw. Eng. Automated Software Engineering
ACM SIGSOFT Software Eng. Notes ACM SIGSOFT Software Engineering Notes
JOURNAL OF LATEX CLASS FILES, VOL. 11, NO. 4, DECEMBER 2012 34

TABLE 14: List of active conferences with abbreviations.

Acronym Conference Full Name
ICPC International Conference on Program Comprehension (ICPC)
CSMR European Conference on Software Maintenance and Reengineering (CSMR)
WCRE Working Conference on Reverse Engineering (WCRE)
ASE International Conference on Automated Software Engineering (ASE)
ICSM International Conference on Software Maintenance (ICSM)
CSSE International Conference on Computer Science and Software Engineering (CSSE)
MySEC Malaysian Software Engineering Conference (MySEC)
CCECE Canadian Conference on Electrical and Computer Engineering (CCECE)
COMPSAC Annual Computer Software and Applications Conference (COMPSAC)
SETN Hellenic Conference on Artificial Intelligence (SETN)
GECCO International Conference on Genetic and Evolutionary Computation (GECCO)
KBEI International Conference on Knowledge Based Engineering and Innovation (KBEI)