IRACST – Engineering Science and Technology: An International Journal (ESTIJ), ISSN: 2250-3498,

Vol.2, No. 2, April 2012

Folksonomy based Personalized Search using

ConceptNet

Rohith Krishna V Sujatha R

School of Information Technology and Engineering School of Information Technology and Engineering
VIT UNIVERSITY VIT UNIVERSITY
Vellore, India Vellore, India

Abstract— The amount of information on the web is increasing considered and what to follow for the designing a system for
day by day at an exponential rate. There is a need for retrieving personalized search.
the necessary information required for the user. The requirement The rest of the paper is organized as follows. The
of the information varies from user to user (example: variation in section II will discuss the past and ongoing research related to
the required information between a geek and geek’s mother). So
there is a need for the personalized search where the results
this field and the section III will propose a system for better
should vary based on the user interests. This paper will discuss designing of personalized search with explanations along with
about the research going on in this particular area and also serve the architecture. The section IV will contain the conclusion for
as a guide in designing a personalized search engine efficiently this paper. The section V will discuss on the future work
and effectively using the ConceptNet with a detailed architecture. related to this area.

Keywords- ConceptNet; Folksonomy; Tagging; WordNet; II. RELATED WORK

A lot of work is being carried out for the personalization of
I. INTRODUCTION search based on the user interests and comparatively a small
amount of work considering folksonomy for personalization.
There is a lot of research going on in order to personalize the
search based on search history of the user, user profile etc...
The number of digital resources is increasing in everyday
The problem with the search history is that the user may be
lives; especially the Worldwide Web has become the largest
using different systems and search history varies from
source of the digital resources with the resources growing at
machine to machine and when a profile [4] is used then if the
exponential rate. In this situation there is a need for the user to
user wanted resources in different domains it becomes
identify the required resources from the vast available data.
difficult. These follows a Taxonomic approach.
The search engines like Google [1], Yahoo [2] and a lot more
We have already discussed about the problems with the
are serving this purpose. But a problem with these search
taxonomic approach in the introduction. This session mainly
engines is that each search is considered independent which
concentrates on the folksonomy based personalized search.
means that the same results will be provided for different users
This section is further organized as follows:
irrespective of the their level of interest. Consider for example
a set of user’s type java as a keyword for searching, of which
some may require java tutorials and some may require java A. Overcoming Folksonomy Disadvantages of ambiguity
software and some requires java code but the results will be and inconsistency
same for the users if they use Google or Yahoo. Thus the need
for the personalized results which is based on their interests is Thus in order to overcome the disadvantages of folksonomy
becoming more important for the users. Udell (2005) [5] provided a method for vocabulary
stabilization around the common choices in tags but there is a
The common vocabulary extracted from social loss of some valuable information which does not serve our
tagging is called as Folksonomy. But considering folksonomy purpose. Kipp and cambell (2006) [6] used co-word analysis
there are many problems such as ambiguity, inconsistency and to derive patterns from tags used in social bookmarking
lack of precession which results in tags that do not best delicious. This showed good results but there is a problem
describe certain web resources and the results will vary from with the same word having different meanings. (For example
what is expected thus violating our goal. “apple” may represent “fruit apple” or the “company apple”).
Using WordNet [8] we identify similarity of various tags by
This paper will discuss on the various methods for measuring the relatedness of their concepts.
personalization and also solutions to the above discussed
problems thus providing an idea on what issues to be

281
 IRACST – Engineering Science and Technology: An International Journal (ESTIJ), ISSN: 2250-3498,
Vol.2, No. 2, April 2012
B. Personalizing Search based on Folksonomy gun”. Then the concept that is identified from the concept net
will be “Robbery” (based on the data it seems like that). Thus
The delicious website[3] provide an interface for tagging the search will be done by searching for the keyword robbery
where personalized results will be returned based on the which will yield accurate Results. More details about the
search tag but the problem with this is, the search tag must ConceptNet will be found in [11]. Let us now take an example
exactly match with a tag in the folksonomy (Dataset) i.e. it from the fig.1. The query if contains “bank, wallet, ATM”
does not consider the sematic aspect of searching. Peter and then the concept that is identified from the ConceptNet is
stock (2007) [7] proposed natural language processing “money”. This can be seen in the figure which is the root node
methods to identify the entities referenced by the tags that of the three concepts.
could be useful for retrieval by processing the sentences
provided as a query for the search. Heung-nam Kim et al.
(2011) [9] Proposes a method which uses user-tag matrix,
item-tag matrix and user-item matrices. Where the tag
similarities and the item similarities are identified and based
on the cosine similarity calculation of various tags and items.
The results are personalized based on the tag used by the user
and their similarity with the other tags is used for identifying
the items.

C. Page Ranking after Personalization

ching-chieh kiu et al. (2010) [10] Proposed a method

for ranking of pages for personalized search. The rank of a
page resulted from a search engine is re-ranked based on the
pre-defined user profile which includes user interests and top-
level domains of webpages which considers only few domains
and will not produce the results from other domains. Heung-
nam Kim et al.(2011) [9] Proposed a method called
Figure 1. A Sample Example of ConceptNet
Folksonomy Boosted Ranking for ranking of the resources
according to the user interests which uses the normalized user-
tag matrix and tag-item matrix and displays the results which
more suits user interests. B. Personalizing the search based on user profile

The next important issue is searching for the

identified concepts. The issue here is that the search should be
III. FOLKSONOMY BASED PERSONALIZED SEARCH personalized i.e. it has to be based on the user profile. Here the
We have seen the various solutions to the problems faced by user profile is based on the tags used by the user for the
the Folksonomy based personalized system. This section will resources. Now first the identified concepts should be
propose a method which provides accurate results of searched in the folksonomy dataset. If a concept is novel i.e. if
personalization. The major issues that we have to consider are: a concept which has been identified is not present in the
dataset then it can be identified by making analogies with the
known concepts.
A. Identifying concepts from the query Another problem here is to search a word with
different meanings. For example consider “Eat, chips” here the
The query given by the user should be processed before ConceptNet[11] will return the keyword as Potato chips than
it is used for searching. The initial step that has to be micro-chips which has more relevance to situation so this is
performed is to identify the parts of speech of the given solved by using ConceptNet.
sentence and extract the subject, object, verb and adjective (if Now considering user based personalization it is
any present) of the given sentence and the concepts relating all better to incorporate two methods [9][10] i.e. identifying the
these should be extracted using the ConceptNet [11]. This top level domain of interests of the users based on/her his
method of using ConceptNet for extracting the concepts serves dataset and also using his/her tag similarities used for various
better as the matching will be done based on the concepts resources(items). For example consider apple. The top-level
identified and not the word matching so the results will be domain of the user will mention what type of domain he is
accurate. interested in whether “fruit apple” or “company apple”.
Let us see this with an example. Suppose the
keywords that are extracted from the query are “money, Theft,

282
 IRACST – Engineering Science and Technology: An International Journal (ESTIJ), ISSN: 2250-3498,
Vol.2, No. 2, April 2012

C. Page Ranking
E. Architecture
After identifying the personalized results it is also important
that the relevant interested results should come first as
everyone will be looking for the interested resources in the top
10 results. Thus in order to provide the interested results page
ranking using Folksonomy Boosted Rank [9] will serve the
purpose better as it will consider the user profile from the tags
tagged by the user and returns the returns based on his profile
i.e. based on the users interests the ranking of resources(items)
will be done.

D. Maintaining stable and consistent Folksonomy

Vocabulary

The vocabulary should be stable and consistent. In

order to serve the purpose using WordNet or ConceptNet [11]
will be better. The advantage of ConceptNet over WordNet is
its integrated Natural language Processing Engine [11]. Thus
using ConceptNet for storing the concepts along with the tags
will maintain a unambiguous and stable vocabulary, as the
concepts are stable and are well structured i.e. linked with the
other concepts using the relations. The ConceptNet has 20
relations based on which it is related (or linked) with other
concepts. The figure 2 will show the relations between the Figure 3. Architecture Diagram for
concepts eliminating the three k –line relations omitted. Folksonomy based Search

The fig.3 represents the architecture of the system designed.

The browser will provide the interface for the user for tagging
as well as for searching a resource on the web. When users tag
a resource on the web the data will be stored in a repository
called folksonomy dataset. Push technology is used for the fast
production of results i.e. whenever the repository is updated
the filtered dataset will be notified about the update and hence
it will update the data from the folksonomy dataset by filtering
the tags using ConceptNet or WordNet. Thus stable and
consistent vocabulary is maintained in the filtered dataset. If
pull technology is used instead of push technology the data
will be extracted by the filtered dataset every time it is called
which is unnecessary. Now considering the query the user
enters in the search box of the interface provided by the
browser, the concepts are extracted from them using the
ConceptNet. The user interests are identified based on the user
profile. Thus based on the user interests, categories and the
folksonomy vocabulary the searcher performs the job of
searching for resources on the web and after that finally the
page rank algorithm will rank the resources based on the user
interests and will return to the browser which will display the
results to the user.
IV. CONCLUSION
The system designed for the personalized search tries to
suppress some problems faced by the folksonomy such as
ambiguity and inconsistency using the ConceptNet. Because
of the reduced problems the accuracy of the results produced
Figure 2. A Treemap of ConceptNet
Relational Ontology
(with the three K-line relations omitted)
283
 IRACST – Engineering Science and Technology: An International Journal (ESTIJ), ISSN: 2250-3498,
Vol.2, No. 2, April 2012
by the system will be improved, by identifying the concepts MIT University (references).
using the ConceptNet and then searching based on the user
profile. Thus ConceptNet best serves the purpose of semantic AUTHORS PROFILE
search supporting Personalization.

V. FUTURE WORK
The system designed now needs an explicit user intervention Rohith Krishna V, is currently
i.e. a user should tag resources based on which his profile is pursuing his final year of
created and personalization is done. Instead of this MS(Software Engineering) at
personalizing based on his/her historic data over the web will VIT University and is interested
serve better as there is no need for the user to understand the in the area of Data Mining
concepts of tagging and Folksonomy as the personalization is mainly personalizing the search.
done implicitly. Another work is to integrate the folksonomy
structure in the taxonomy which will overcome some of the
disadvantages of taxonomy such as updating the taxonomy
and adding extra tags to the taxonomy, also some of the
disadvantages of folksonomy such as unstructured and
unambiguous tags. Thus the integration may help in
maintaining a common vocabulary which is consistent.

REFERENCES

[1] www.google.com (Google)

[2] Search.yahoo.com (Yahoo)
[3] www.delicious.com (Delicious)
[4] Mehmet S Aktas, Mehmet A. Nacar, “An application of
Personalized Page Rank Vectors: Personalized search Engine”
[5] J Trant, “Studying Social Tagging and Folksonomy: A review and
Framework,” Journal of Digital Information, Vol 10, No 1 (2009)
[6] Kipp, M.E., & Campbell, D.G., “Patterns and Inconsistencies in
Collaborative Tagging Systems: An Examination of Tagging
Practices” In American Society of Information Science &
Technology Annual Conference 2006,Austin, Texas (US),3-8
November 2006, unpublished
[7] Stock, W.G., “Folksonomies and science communication. A
mash-up of professional science databases and Web 2.0 services,”
Information Services & Use, 27, 97-103 (2007)
[8] Ted Petersen,Siddharth Patwardhan & Jason Michelizzi
‘WordNet::Similarity – Measuring the Relatedness of Concepts,”
2004 (references)
[9] Heung-Nam Kim, Majdi Rawashdeh, Abdullah Alghamdi,
Abdulmotaleb El-Saddik, “ Folksonomy-based personalized
search and ranking in social media services,” Inf. Syst. 37(1): 61-
76 (2012).
[10] Mehmet S. Aktas, Mehmet A. Nacar, and Filippo Menczer,
“An Application of Personalized PageRank Vectors:
Personalized Search Engine,” (references).
[11] Hugo Liu and Push singh : A Practical commonsense Reasoning
Tool Kit,
Media R.Sujatha received her B.E. degree
Laboratory, in computer science from Madras
Uninversity, in 2001, the M.E. degree
in computer science from Anna
University in 2009 and currently
pursuing the Ph.D. degree in Vellore
Institute of Technology, Vellore. She
was a lecturer and currently assistant
professor, with School of Information
Technology and Engineering in
Vellore Institute of Technology,
Vellore. Her area of research interest 284
include Data mining, Image
Processing and Management of
Information systems.