NATURE: A TOOL RESULTING FROM THE UNION OF ARTIFICIAL INTELLIGENCE AND NATURAL LANGUAGE PROCESSING FOR SEARCHING RESEARCH PROJECTS IN COLOMBIA

International Journal of Artificial Intelligence & Applications (IJAIA), Vol.12, No.4, July 2021
DOI: 10.5121/ijaia.2021.12401 1
NATURE: A TOOL RESULTING FROM THE UNION
OF ARTIFICIAL INTELLIGENCE AND NATURAL
LANGUAGE PROCESSING FOR SEARCHING
RESEARCH PROJECTS IN COLOMBIA
Felipe Cujar-Rosero, David Santiago Pinchao Ortiz,
Silvio Ricardo Timarán Pereira and Jimmy Mateo Guerrero Restrepo
Systems Department, University of Nariño, Pasto, Colombia
ABSTRACT
This paper presents the final results of the research project that aimed for the construction of a tool which
is aided by Artificial Intelligence through an Ontology with a model trained with Machine Learning, and is
aided by Natural Language Processing to support the semantic search of research projects of the Research
System of the University of Nariño. For the construction of NATURE, as this tool is called, a methodology
was used that includes the following stages: appropriation of knowledge, installation and configuration of
tools, libraries and technologies, collection, extraction and preparation of research projects, design and
development of the tool. The main results of the work were three: a) the complete construction of the
Ontology with classes, object properties (predicates), data properties (attributes) and individuals
(instances) in Protegé, SPARQL queries with Apache Jena Fuseki and the respective coding with
Owlready2 using Jupyter Notebook with Python within the virtual environment of anaconda; b) the
successful training of the model for which Machine Learning algorithms were used and specifically
Natural Language Processing algorithms such as: SpaCy, NLTK, Word2vec and Doc2vec, this was also
performed in Jupyter Notebook with Python within the virtual environment of anaconda and with
Elasticsearch; and c) the creation of NATURE by managing and unifying the queries for the Ontology and
for the Machine Learning model. The tests showed that NATURE was successful in all the searches that
were performed as its results were satisfactory.
KEYWORDS
Artificial Intelligence, Natural Language Processing, Ontology, Machine Learning, Search Engine,
Semantic Web.
1. INTRODUCTION
The Internet was conceived by Tim Berners-Lee as a project to manage and share knowledge and
information among a select group of scientists. With the pass of the time and with the advances in
the development of hardware that made possible the communication around the world, the
necessary applications were developed to meet the needs of users. The large volume of content
available online makes searching and processing difficult, the need to devise new ways to
optimize the treatment given to such content has been vital; for the information available on the
Web to be interpreted by computers without the need for human intervention, the Semantic Web
is required. It is said that in Internet computers are not only capable of presenting the information
contained in web pages, else they should also “understand” such information [1].

2
According to Berners Lee and Hendler, on the Semantic Web, information is offered with a well-
defined meaning, allowing computers and people to work cooperatively. The idea behind the
Semantic Web is to have data on the Web defined and linked so these can be used more
effectively for discovery, automatization, integration and reuse between different applications.
The challenge of the Semantic Web is to offer the language that expresses data and rules to
reason about many data and also allows the rules on any knowledge representation system to be
exported to the Web, providing a significant degree of flexibility and “freshness” to traditional
centralized knowledge representation systems, which become extremely overwhelming, and its
growing in size is unmanageable. Different web systems can use different identifiers for the same
concept; thus, a program that wants to compare or combine information between such systems
has to know which terms mean the same thing; ideally the program should have a way of
discovering the common meanings of whatever database it encounters. One solution to this
problem is to add a new element to the Semantic Web; collections of information called
Ontologies [2].
In the same way, it is known that the large amount of textual information available on the WEB
with the increase in demand by users, makes necessary to have systems that allow access to that
interest information in an efficient and effective way for saving time in the search and
consultation. Among the existing techniques to achieve this efficiency and effectiveness, and in
turn to provide access or facilitate the management of text document information are Machine
Learning techniques, using them is highly convenient, this can be evidenced in a large number of
applications in different areas [3].
This is because the factors that have generated the success of the Internet have also caused
problems such as: information overload, heterogeneity of sources and consequent problems of
interoperability. The Semantic Web helps to solve these problems by allowing users to delegate
tasks to software tools. By incorporating semantics in the Web, the software is capable of
processing content, reasoning with it, combining it and making logical deductions to solve
problems automatically. Automatic ability is the result of the application of artificial intelligence
techniques, which require the participation of intelligent agents that improve searches, adding
values for reasoning and making decisions to web services that store high content [4].
According to Kappel, it is pertinent to make use of semantics, which is reflected in the responses
that a user receives to their requests in search engines, since these go beyond the state in which
users simply asked a question and received a set sorted by web page priority. Users want targeted
answers to their questions without superfluous information. Answers should contain information
from authorized sources, terms with the same meaning as those used in the question, relevant
links, etc. So, the Semantic Web tries to provide a semantic structure to the significant contents of
the Web, creating an environment in which software agents navigate through the pages
performing complex tasks for users [1].
It is assumed that this Web has the ability to build a knowledge base on the preferences of users
and that, through a combination of its ability to understand patterns and the information available
on the Internet, it is able to meet exactly the information demands from users, for example:
restaurant reservation, flight scheduling, medical consultations, purchase of books, etc. Thus, the
user would obtain exact results on a search, without major complications because the Semantic
Web provides a way to reason on the Web as it is an infrastructure based on metadata (highly
structured data describing information), thus extending its capabilities. That is, it is not a magic
artificial intelligence that allows web servers to understand the words of the users, it is only the
construction of a skill arranged in a machine, in order to solve well-defined problems, through
similar operations well defined to be carried out on existing data [4].

3
In the systematic review of the literature, a search engine is defined as an application and / or
computer resource that allows information to be located on the servers of a certain website,
resulting in a list that is consistent with the files or materials stored on the corresponding servers
and responding to the needs of the user. Search engines make easy to locate the information that
is scattered around the Web, but it is crucial to know the way in which the search is being carried
out [5]. Syntactic search engines make use of keywords, where the search result depends on an
indexing process, which is the one that will allow organizing searches with these keywords or
through the use of hierarchical trees categorized by a certain topic. Despite the power shown by
syntactic search engines, they are still far from being able to provide to the user adequate results
for the queries made, since the number of results can be too many and therefore it will be quite
tedious to find the desired result or else not getting any results, with the addition that much of the
responsibility for the search can fall into the hands of the user, who would have to filter and
categorize their search to get a clear and concise answer [6].
In this way, it can be observed that these problems can be solved with the use of semantic search
engines which, on the other hand, facilitate the user's work, are efficient in the search since they
find results based on the context, thus providing information more exact about what is sought,
offering a more biased number of results, facilitating the work of filtering the results by the user.
In this way that these search engines interpret user searches by making use of algorithms that
symbolize comprehension or understanding, offering precise results quickly and thus recognizing
the correct context for the search words or sentences. It is nothing more than a semantic search
engine, one that performs the search by looking at the meaning of the group of words that are
written [7].
ERCIM digital library [8], NDLTD [9], Wolfram Alpha [10] use semantics to find results based
on context. The last one is capable of directly answering the questions asked by the user instead
of providing a list of documents or web pages that could contain the answer, as Google does.
Once the question is asked, the tool calculates different answers by selectively choosing the
information from the Web to end up giving a precise answer. Swotti is another search engine that
uses Semantic Web technologies to extract the opinions made by users in blogs and forums about
companies or products. It is able to identify the adjectives and verbs that define what people are
looking for, and therefore allows people to deduce if the comment is positive or negative. When
people make a search in Swotti they get not only results, else a qualitative assessment [11].
Swoogle is a document search engine for the Semantic Web, a Google for the Semantic Web
although it is not aimed at the end user yet, it has been created at the University of Maryland, it is
not intended for the common user, but for the crawling of semantic web documents whose
formats are OWL, RDF or DAML. Swoogle is a search engine that detects, analyzes and indexes
the knowledge encoded as Semantic Web documents. Swoogle understands by Semantic Web
documents those that are written with some of the languages oriented to the construction of
Ontologies (RDF, OWL, DAML, N3, etc). It retrieves both documents written entirely in these
languages (which for Swoogle are strict Semantic Web documents) and documents partially
written with some of them. It also provides an algorithm also inspired by Google's Page Rank
algorithm, which for Swooglehas been called Ontology Rank. The Ontology Rank algorithm has
been adapted to the semantics and usage patterns found in the Semantic Web documents.
Swoogle currently has around 1.5M Semantic Web documents indexed. This information is
available through an internal link to statistical data related to their status [12]. Other works such
as that of Camacho Rodríguez in her undergraduate work to obtain the degree in Telematics
Engineering propose incorporating a semantic search engine in the LdShake platform for the
selection of educational patterns. This work was developed at the PompeuFabra-UPF University
of Barcelona, Spain in 2013. This work analyzes the efficiency of using Ontologies to
considerably improve the results and at the same time gain speed in the search [13]. Amaral
presents a semantic search engine for the Portuguese language where it makes use of Natural

4
Language Processing tools and a multilingual lexical corpus where the user's queries are
evaluated, for the disambiguation of polysemic words, it uses pivots shown on the screen with the
different meanings of the word where the user chooses the meaning with which he wants to make
the query [14]. Aucapiña and Plaza in their thesis for obtaining the Degree in Systems
Engineering propose a semantic search engine for the University of Cuenca in Cuenca, Ecuador
in 2018, where they describe in detail the use of SPARQL as a query language and the various
stages carried out to achieve the prototype of the semantic search engine following proven
methodologies and in certain cases those are supported by automated processes [15].Umpiérrez
Rodríguez in his final graduate project in Computer Engineering called “SPARQL Interpreter” at
the University Of Las Palmas Of Gran Canaria, developed in 2014, where he explains how
SPARQL Interpreter addresses the problem of communication between a query language and a
database of specific data [16]. Baculima and Cajamarca in their graduate thesis in Systems
Engineering developed a “Design and Implementation of an Ecuadorian Repository of Linked
Geospatial Data” at the University of Cuenca Ecuador, in 2014, they work on the solution for
generation, publication and visualization of data Geospatial Links, for which they rely on web
search engines, this since the Web focuses on the publication of this type of data, allowing them
to be structured in such a way that they can be interconnected between different sources. This
work is supported by SPARQL and GEOSPARQL to be able to carry out queries, insert
modification and elimination of data [17]. Iglesias, developed his project at the Simón Bolívar
University of Barranquilla, his objective was to build an ontological search engine that allows
semantic searches to be carried out online for master's and doctorate training works, where people
can find this kind of work or topics that can serve as a guide for new research to emerge, thus
improving searches when selecting research topics for undergraduate projects [18]. Bustos
Quiroga in the thesis in the Master's Degree in Computer and Systems Engineering develops a
“Prototype of a system for integrating scientific resources, designed to function in the space of
linked open data to improve collaboration, efficiency and promote innovation in Colombia” in
2015 at the National University of Colombia. In this work he used the Semantic Web in linked
data to improve integration in timelessness between applications and facilitate access to
information through unified models and shared data formats [19]. Moreno and Sánchez in their
undergraduate work to obtain the title of Systems and Computing Engineer propose a prototype
of semantic search engines applied to the search for books on Systems Engineering and
Computing in the Jorge RoaMartínez library of the Technological University of Pereira. This
work was developed in 2012. This prototype was developed based on the existing theoretical
foundations and the analysis that was carried out on the technologies involved, such as intelligent
software agents, Ontologies that are implemented in languages such as RDF and XML, and other
development tools [20]. Likewise, at the University of Nariño, Benavides and Guerrero
developed the undergraduate work project to obtain the title of Systems Engineer, in 2013, called
“UMAYUX: a knowledge management software model supported by a coupled-weakly dynamic
Ontology with a database manager for the University of Nariño” whose objective was to convert
the knowledge that was tacit, in the academic and administrative processes of the University of
Nariño, into explicit knowledge that allows to collect, structure, store information and transform
through the use of domain-specific Ontologies, in a way that each academic unit or administrative
unit can build and couple to the model. The UMAYUX model was implemented through the
construction of MASKANA, a knowledge management tool supported by a dynamic Ontology on
graduate works of undergraduate students of the Systems Engineering program of the Systems
department of the Faculty of Engineering, weakly coupled with the PostgreSQL DBMS (Data
Base Management System) [21].
Currently in the Research System of the University of Nariño in the VIIS, the engineer in charge,
at the time of searching for these types of research: graduate projects, student projects and
teaching projects, comments that there is a delay in the processes, he says that these processes are

5
not optimal, sometimes it is difficult to find what he wants, in many occasions he has not been
able to find what he needs. This indicates that there is neither efficiency nor
Effectiveness guaranteed at the time of conducting research searches, since the search is being
performed manually. That is to say that by having a manual information search system that does
not even have the qualification of a syntactic search engine, it is deduced that the information
does not have a clear structure to be presented and that the processes are inefficient in the
searches. This leads to the conclusion that it is absolutely necessary to build the intelligent
semantic search engine, since if the problem persists, as the information increases, the searches
will be more tedious and wasteful, additionally with the intelligent semantic search engine the
amount of user population searching on a certain topic will be greater and will be satisfied by
finding the desired results.
Thus, this work provides a tool that allows teachers, students and other researchers to search and
consult about the researches that have been carried out at the Universidad de Nariño. A semantic
search engine has been built using semantics through the SPARQL query language and RDF
language with the management of Ontologies, all this integrated with a Machine Learning model
that uses algorithms and libraries such as: NLTK, SpaCy, Word2Vec, Doc2Vec, TF-IDF (Term
Frequency- Inverse Document Frequency), BOW (Bag Of Words), among others. In this way we
can facilitate the work and allow researchers and the community in general to retrieve and find
the requested information efficiently from the research projects that are digitized in the Research
System of the Universidad de Nariño. The work was developed by student researchers of the
GRIAS research group of the Systems Engineering program of the Faculty of Engineering of the
Universidad de Nariño.
2. METHODOLOGY
The methodology used for the work comprises the following stages: appropriation of knowledge;
installation and configuration of tools, libraries and technologies; collection, extraction and
preparation of research projects; design and development of NATURE.
3. RESULTS
3.1. Appropriation of knowledge
It is highlighted the result of the acquired knowledge of all the topics covered by the project, as
well as the various tools and languages used. The learning of topics such as: Semantics, Semantic
Web, Ontologies, Search Engines, Machine Learning, Natural Language Processing, Artificial
Intelligence and Methontology was obtained. In the same way, the learning in languages such as
Python, XML, RDF, OWL and SPARQL was known and reinforced.
3.2. Installation and Configuration of Tools, Libraries and Technologies
It is highlighted the result of the installation and configuration of: Jupyter notebook, Protégé,
Owlready2, Apache Jena Fuseki, Elasticsearch, Visual Studio Code, Anaconda, Gensim with
Word2Vec and Doc2Vec, Pandas, Numpy, NLTK, SpaCy, etc.

6
3.3. Collection and Extraction of Research Projects
It is highlighted the result of collecting and extracting information from the research projects of
teaching projects, student projects and graduate works that are stored in the research system of
the University of Nariño.
It is clarified that currently the difference between student projects and graduate works is that
student projects are registered from the first semesters of the university career (from first to
eighth) while graduate projects are registered from the last semesters of the university career
(seventh onwards) until the moment of appearing as a graduate (if it is the case).
3.4. Preparation of Research Projects
The result of preparing the research projects is highlighted, in such a way that this allowed for
navigating through the following stages, anticipating and avoiding inconveniences, errors or
problems with respect to the quality of the data.
In this order of ideas, the following phases (from the stage of preparation of research projects) are
highlighted:
3.4.1. Data Organization Phase
In this phase, algorithms (created by the authors of this work) were applied to the research
projects, this because the projects in the collection and extraction phase were untidy and in
conditions not suitable to be treated, managed and worked. Jupyter Notebook was used with
Python and Pandas scripts to facilitate the handling of data in series and data frames.
3.4.2. Corpus Creation Phase
In this phase, the corpus for the research projects was created, which was the most powerful input
of semantics, as can be seen in the later stages. This corpus resulted from unifying all the data
from the research projects (already organized in the previous phase), which were: title and
summary of the research; keyword 1, keyword 2, keyword 3, keyword 4, keyword 5; names,
surnames, program, faculty, department, research group and line of research for each of the
authors and advisers. In this phase, like the previous one, Jupyter Notebook, Python and Pandas
were also used to facilitate the handling of data in series and data frames.
3.4.3. Data Pre-processing Phase
In this phase, the NLTK and SpaCy libraries were used to preprocess the data obtained in the
previous phase. For this, the following subphases (from the data-preprocessing phase) were used:
Figure. 1 shows an example of a project fragment dealing with physics, to which it will be shown
how the different subphases were applied.

7
Figure 1. Example of a project fragment on physics
3.4.3.1. Data Tokenization Subphase
In this subphase, algorithms from the NLTK library were executed to separate all the words and
to be able to work with them individually.
Figure 2 shows the data tokenization subphase for Figure 1.
Figure 2. Data tokenization subphase in all the words of Figure 1
3.4.3.2. Data Normalization Subphase
For this subphase, many algorithms were applied so that all the data were under the same
standard.
Figure 3 shows the data normalization subphase for Figure 2
Figure 3. Data normalization subphase in all the words of Figure 2

8
3.4.3.3. Data Cleaning Subphase
In this subphase, NLTK and SpaCy algorithms were applied together with regular expressions so
that the data is totally clean, this with the elimination of null data, punctuation marks, “non-
ascii” characters and stopwords.
Figure 4 shows the data cleaning subphase for Figure 3.
Figure 4. Data cleaning subphase in all the words of Figure 3
3.4.3.4. Data Lemmatization Subphase
Finally, in this subphase, the data resulting from the cleaning stage were lemmatized.
Figure 5 shows the data lemmatization subphase for Figure 4.
Figure 5. Data lemmatization subphase in all the words of Figure 4
3.5. Design of NATURE
Once the previous stage of preparation of the research projects was completed, NATURE was
designed. This design was carried out taking into account the conceptualization phase of the
Methontology methodology, where the following results stand out:
3.5.1. Conceptualization Phase
Within the conceptualization phase, eleven specific tasks were developed that allowed to
successfully conceptualize: classes, attributes, relationships and instances of Ontology. The most
important of these are the following:
Task 1. Build the glossary of terms:
This task listed all the important terms selected after analyzing the previous specification phase
with its knowledge acquisition process, also this task presented a brief description of each term as
shown in Table 1.

9
Table 1. Glossary of terms of Ontology
Term Description
Universidad The education-oriented entity that contains faculties.
Facultad The entity that contains academic departments.
VIIS Vice-Chancellor of Research and Social Interaction, is the entity in
charge of the research aspect throughout the University, is the one who
manages the economic resources for
research projects.
Departamento The entity that contains academic programs.
Convocatoria This term refers to the convocatory by the VIIS for researchers
to come to this convocatory and submit projects in order for them to be
financed.
Programa It is the academic program that is conformed by teachers and
students.
Grupo de investigación It is the group conformed by teachers and/or research students in order
to submit projects to the VIIS convocatory.
Docente Is a researcher who belongs to the University, who carries out projects
of teaching type.
Estudiante Is a researcher who belongs to the University, who carries out
projects of student type and/or graduate works.
Investigadorexterno Is a researcher who is external to the University but who
presents for the convocatory for VIIS.
Línea de investigación It is a branch that the research group manages, focused on aspecific area
of knowledge.
Investigador Is the one who develops research projects and submits them tothe VIIS
convocatory. This researcher may be a teacher, student or external
researcher.
Proyecto de
investigación
It is perhaps the most important entity within the researchdomain that
contains everything related to a research project.
Palabra This entity refers to each of the words that conforms the research
project, these were used for building the thesaurusand generating a big
part of the semantic.
Task 2. Build concept taxonomies:
This task defined the taxonomy or hierarchy of ontology concepts or classes that were obtained
from the glossary of terms in task 1, this taxonomy is shown in Figure. 6.

10
Figure 6. Taxonomy of ontology concepts Task 3. Build binary relation diagrams:
In this task the binary relations diagram that contains the predicates of Ontology was elaborated.
The relations of the most important class of Ontology are visualized in Figure 7, which is
“Proyecto de investigación”.
Figure 7. Binary relations diagram for class: “Proyecto de Investigación”
3.6. Development of NATURE
NATURE was developed based on three phases in which the following results are highlighted:

11
3.6.1. Development with Methontology Phase
For this phase the three subphases of Methontology were applied which are: formalization,
implementation and evaluation.
3.6.1.1. Formalization Subphase
This phase highlights the results obtained after using the Protégé tool for the construction of
Ontology in semi-computable terms.
3.6.1.2. Implementation Subphase
This phase highlights the results of using the Owlready2 library to encode a computable version
of Ontology. Scripts were created and encoding was performed for the handling of Ontology with
Python where an entire process of instantiating objects of all classes was performed:
Owlready2 “DataProperties” that correspond to ontology attributes, along with Owlready2
“ObjectProperties” that correspond to Ontology relations were also encoded within the scripts;
for each of these elements mentioned, the domain and range were determined. It should be said
that Owlready2 reverse relationships are executed in the background, so it was only enough
execute the direct relationship.
In synthesis, all classes, attributes, and relations were instantiated within Ontology.
3.6.1.3. Evaluation Subphase
This phase highlights results after having performed functional tests locally and having
successfully retrieved the data and other components of ontology with the use of SPARQL and
Apache Jena Fuseki server by handling triples of RDF (subject predicate object).
3.6.2. Development with Machine Learning Phase
This phase highlights the results of training with the Machine Learning algorithm with Natural
Language Processing such as Word2Vec, which helped to find the context that a word has, in
addition a model was trained with the Doc2Vec algorithm, which relies on Word2Vec to find
documents that relate to each other, these models make use of neural networks. In this case, the
model was trained with the algorithms previously mentioned based on the Skip-Gram model,
which attempts to predict words or documents in context given a word or set of base words to
search for.
It should be clarified that the output returned by Word2Vec was the input for the process
performed with Doc2Vec, this is possible since both algorithms work hand in hand to achieve
discover semantic relationships and retrieve information semantically effectively.
To perform the search for similarity between words or documents, of a set of given words, the
Gensim library was used, which makes use of the normalization of the vectors obtained from the
words to be searched and the calculation of the product point between the normalized vector and
each of the vectors corresponding to each word or document trained.
The model was created with data from the preparation stage of research projects, the respective
hyper parameters were assigned, the model was trained, the results were evaluated and the hyper
parameters were re-fed to satisfactory results, as it is evidenced in Table 2.

12
Table 2. Hyper parameters for word2vec and doc2vec models
Name Value Description
vector_size 300 Dimension of the vector of each of the words in the
corpus.
window 5 Refers to the context where the distance
between predicted words is chosen.
min_count 1 Minimum words to look for.
dm 0 0 indicates that Doc2Vec PV-DBOW is used which
is analogous to the Skip-Gram model used in
Word2vec.
1 indicates that Doc2Vec PV-DM is used
which is analogous to the CBOW model used in
Word2Vec.
dbow_words 1 0 indicates that it will train with Doc2Vec.
1 indicates that it will train with Doc2Vec taking
Word2Vec input.
hs 0 It is the value with which the neuron will be
punished in case the task done is not correct.
negative 20 Number of irrelevant words for negative
sampling.
ns_exponent -0.5 Indicates that frequencies will be sampled
equally.
alpha 0.015 Neural network learning rate
min_alpha 0.0001 Rate to be reduced during training.
seed 25 Seed to generate hash for words.
sample 5 Reduction number for high frequency words
epochs 150 Epochs, number of iterations for training.
In Figure 8, Figure 9 and Figure 10 are presented the results of executing the order to find 10
more similar and related words (according to the cosine similarity of the algorithm ordered in
percentage terms from highest to lowest) to another word that is specified within of the entire
research corpus with a method of the Word2Vec algorithm.
Figure 8 indicates the 10 words most similar and related to the word “cultivos”.
Figure 8. Result of method with Word2vec for word “cultivos”

13
Figure 9 indicates the 10 words most similar and related to the word “fresa”.
Figure 9. Result of method with Word2vec for word “fresa”
Figure 10 indicates the 10 words most similar and related to the word “historia”.
Figure 10. Result of method with Word2vec for word “historia”
Figure 11 shows possible semantic relationships between the words:
 entidades - instituciones - academicas - ministerio - promover
 impacto - condiciones
 empresas - comercio - turismo – emprendimiento
 adolescentes - escolares - escolar - bullying
 nivel - mundial
 aportar - contribuir
 relaciones - económicos
 social – sociales
 lado - humano
 programa - diseño - industrial - artes - tecnologicos

14
 investigacion - curriculo
 desarrollo - proyecto - estudios - docente - grado
 universidad – Nariño
 desarrollar – propuesta
Figure 11. First plot for words semantically related
Figure 12. Second plot for words semantically related

15
Figure 13. Third plot for words semantically related
3.6.3. Integration of Ontology and Machine Learning Phase
For this phase, Ontology and Machine learning are integrated, providing potency, effectiveness
and semantic power to optimize times, resources, and to have greater chances of finding
successful and satisfactory results to certain searches in NATURE, the results are observed in:
Figure 14, Figure 15, Figure 16 and Figure 17.
This was achieved by bringing the vectors that Doc2Vec generated to Elasticsearch; Elastic
helped in the ranking stage by having speed, scalability and being a distributed analysis engine
that favors the search and indexing of research projects.
Afterwards, scripts were created to manage the queries of the research projects for the Ontology
with SPARQL, which relies on the trained Word2Vec model to add additional words to the
search that are related to those requested and thus find research related to a certain query. In the
same way, with Doc2Vec it was possible to infer vectors from a set of supplied words, then as a
partial result, the investigations that are related to inferred vectors are presented. Finally, the
results obtained in the SPARQL query and the Doc2Vec algorithm are joined, so the final
ranking of a search will show consistent, coherent, successful and satisfactory results as requested
with the additional ability to recommend documents that may be useful and interest to the user.

16
Figure 14. NATURE tool interface
Figure 15. First results for the search: “investigaciones de psicología”

17
Figure 16. First results for the search: “investigacionessobrequímica”
Figure 17. First results for the search: “investigacionesacerca de minería de datos”
4. DISCUSSION
 NATURE provides a degree of optimization, originality and innovation compared to other
search engines and knowledge databases such as WordNet, Freebase, DLBP (Digital
Bibliography and Library Project), ERCIM digital library, Swoogle, NDLTD, Wolfram
Alpha (and others mentioned above) because ontologies are being integrated with Machine
Learning with the aforementioned scripts where the ontology is well set up and the
algorithms are well trained. In addition to this, the vectors of the words are being managed
with Elasticsearch, which save significant memory consumption. Searches are also done
with Elasticsearch which is another reason because the search engine is so fast and
accurate.

18
 It is recommended to carry out tests with more data to see how NATURE behaves in the
face of an expandable size in the information. This is because the data used were all the
research projects that were in the VIIS Research System, but all the projects at the
University level are not within that system, but 10%.
 It is proposed to do the coupling of NATURE in other universities and in various non-
academic environments, determining the structure of the Ontology and Machine Learning
models with their possible variants.
 It is suggested to carry out an analysis of what users are looking for, analyzing the records
of searches, downloads, storing everything in the database, then applying data mining with
all the information to possibly determine aspects such as: “What semesters do belong
people who make queries about astronomy ?” or “What ages do belong people who make
queries about psychology?”. Machine Learning could be used for this future work
perfectly.
 It is also proposed to incorporate in the search engine page a view with its respective
database that allows to rate and comment on the search engine in order to observe and
analyze how users are rating NATURE, as well as to realize their opinions and whether
they are satisfied or not, thus determining the usability of NATURE.
5. CONCLUSIONS
 With the culmination of this research work, NATURE is obtained: A tool resulting from
the union of Artificial Intelligence and Natural Language Processing for searching research
projects in Colombia. Through the successful development of the project stages, the
formulated problem is solved, the objectives set are fullfiled and satisfactory results are
obtained. In this way, this tool facilitates the successful search for research projects for
teaching projects, student projects and graduate projects at the University of Nariño.
 In the stages of appropriation of knowledge and installation and configuration of the tools,
a domain of the various topics was acquired and this contributed to the development of the
work and led to the personal training of the researchers as well as made outstanding
contributions to the group of GRIAS research (Grupo de InvestigaciónAplicadoenSistemas)
and for the University of Nariño in general.
 The stages of collection, extraction and preparation of research projects were extremely
important stages that acted as preliminary and prelude stages as input for NATURE. In this
vein, it is correct to affirm that without these stages a good development of NATURE
could not have been achieved.
 Methontology was a methodology that was perfectly coupled to the project and allowed to
build the Ontology following specific phases and tasks with an order, comprehension and
accuracy in the processes.
 The Ontology integrated with Machine Learning demonstrated great potency, semantic
power and effectiveness in the processes to obtain concrete results according to the
searches carried out. This is because Machine Learning algorithms, specifically Natural
Language Processing algorithms such as Word2vec and Doc2vec work with neural
networks, which were trained with the words from the research project corpus, adapting
them to the context and finding the various semantic relationships between them. Likewise,

19
Ontology acted as a great semantic network whose instances, hand in hand with classes,
relations and attributes, interacted under the triple scheme handled by RDF and consulted
by SPARQL to extract all the knowledge from the domain of the research projects.
ACKNOWLEDGMENT
To the University of Nariño, to the VIIS (Vicerrectoría de Investigación e Interacción Social) for
financing this project and to the Research Community in general for supporting the successful
completion of this work.
REFERENCES
[1] VELÁSQUEZ, Torcoroma, PUENTES, Andrés & GUZMÁN, Jaime. Ontologías: unatécnica de
representación de conocimiento. En: AvancesenSistemas e Informática. Vol. 8. No. 2. (Julio, 2011),
p. 211-216. [Enlínea]. Disponibleen:
https://revistas.unal.edu.co/index.php/avances/article/view/26750
[2] GARCÍA, Francisco. Web Semántica y Ontologías. [Enlínea]. Disponibleen:
https://www.researchgate.net/publication/267222548_Web_Semantica_y_Ontologias
[3] MOURIÑO, M. Clasificaciónmultilingüe de documentosutilizando machine learning y la wikipedia.
[Enlínea]. Disponibleen: https://dialnet.unirioja.es/servlet/tesis?codigo=150295
[4] EFIGENIA, Ana & CANTOR, Sandoval. USO DE ONTOLOGÍAS Y WEB SEMÁNTICA PARA
APOYAR LA GESTIÓN DEL CONOCIMIENTO. En: Ciencia e IngenieríaNeogranadina. Vol. 17
No.2. (Diciembre, 2007), p.111-129. [En línea]. Disponible en:
https://dialnet.unirioja.es/descarga/articulo/2512191.pdf
[5] GALLO, Manuel, FABRE, Ernesto & GALLO, Manuel. ¿Quéesunbuscador? [Enlínea].
Disponibleen: http://media.axon.es/pdf/98234_1.pdf
[6] FAZZINGA, Bettina, GIANFORME, Giorgio, GOTTLOB, Georg & LUKASIEWICZ, Thomas.
Semantic Web Search Based On Ontological Conjunctive Queries. En: SSRN Electronic Journal.
[Enlínea]. Disponible en:
https://www.researchgate.net/publication/326473981_Semantic_Web_Search_Based_on_Ontological
_Co njunctive_Queries
[7] DE PEDRO, A. BuscadoresSemánticos, para quésirven. Usosen la AAPP. [Enlínea]. Disponibleen:
http://www.alejandrodepedro.es/buscadores-semanticos-el-paso-al-30
[8] ANDREONI, Antonella, BALDACCI Maria, BIAGONI, Stefania, CARLESI, Carlo, CASTELLI,
Donatella, PAGANO, Pasquale, PETERS, Carol & PISANI, Serena. The ERCIM Technical
Reference Digital Library. En: D-Lib Magazine. Vol. 5. No. 12. (Diciembre, 1999). [Enlínea].
Disponibleen: http://www.dlib.org/dlib/december99/peters/12peters.html
[9] NDLTD. Networked Digital Library of Theses and Dissertations. [Enlínea]. Disponibleen:
http://www.ndltd.org
[10] WolframAlpha Computational Intelligence. [Enlínea]. Disponibleen: https://www.wolframalpha.com
[11] MARTÍN, Javier. Swottibuscador de opiniones. [Enlínea]. Disponibleen: https://loogic.com/swotti-
buscador-de-opiniones
[12] BARBERÁ, Consuelo, MILLET, Mercé& TORRES, Emiliano. Estudio del
buscadorsemánticoSwoogle. [Enlínea]. Disponibleen:
https://www.uv.es/etomar/trabajos/swoogle/swoogle.pdf
[13] CAMACHO, María. Incorporación de unbuscadorsemánticoen la plataformaLdShake para la
selección de patroneseducativos. Barcelona, 2013, 76p. Trabajo de grado. Universidad PompeuFabra.
Escuela Superior Politécnica UPF. Ingeniería de Telemática. [Enlínea]. Disponibleen:
https://repositori.upf.edu/handle/10230/22172
[14] AMARAL, Carlos, LAURENT, Dominique, MARTINS, André, MENDES, Alfonso & PINTO,
Cláudia. Design and Implementation of a Semantic Search Engine for Portuguese. [Enlínea].
Disponibleen:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.129.4090&rep=rep1&type=pdf

20
[15] AUCAPIÑA, Yolanda & PLAZA, C. Buscadorsemánticouniversitario: Caso de estudio Universidad
de Cuenca. Cuenca, 2018, 200p. Trabajo de grado (Tesisprevia a la obtencióndelTítulo de
IngenieroenSistemas). Universidad de Cuenca. Facultad de Ingeniería. Ingeniería de Sistemas.
[Enlínea]. Disponibleen: http://dspace.ucuenca.edu.ec/handle/123456789/30291
[16] UMPIÉRREZ, Francisco. SPARQL Interpreter. Las Palmas de Gran Canaria, 2014, 65p. Trabajo de
grado (Trabajo Final de GradoenIngenieríaInformática). Universidad de Las Palmas de Gran Canaria.
EscuelaIngenieríaInformática. IngenieríaInformática. [Enlínea]. Disponibleen:
https://nanopdf.com/download/0701044000000000pdf_pdf
[17] BACULIMA, Jhon& CAJAMARCA, Marcelo. Diseño e Implementación de
unRepositorioEcuatoriano de DatosEnlazadosGeoespaciales. Cuenca, 2014, 131p. Trabajo de grado
(Tesis de Gradoprevia a la obtencióndelTítulo: Ingeniero de Sistemas). Universidad de Cuenca.
Facultad de Ingeniería. Ingeniería de sistemas. [Enlínea]. Disponibleen:
http://dspace.ucuenca.edu.ec/handle/123456789/19876
[18] IGLESIAS, Daniela, MEJÍA, Omar, NIETO, Julio, SÁNCHEZ, Steven & MORENO, Silvia.
Construcción de unbuscadorontológico para búsquedassemánticas de proyectos de maestría y
doctorado. En: Investigación y Desarrolloen TIC. Vol. 7. No. 1. (Mayo, 2017), p. 7-13.
[Enlínea].Disponibleen: https://revistas.unisimon.edu.co/index.php/identic/article/view/2501
[19] BUSTOS, Gabriel. Prototipo de unsistema de integración de recursoscientíficos, diseñado para
sufuncionamientoen el espacio de losdatosabiertosenlazados para mejorar la colaboración, la
eficiencia y promover la innovaciónen Colombia. Bogotá, 2018. Tesis de Maestría. Universidad
Nacional de Colombia. Facultad de Ingeniería. Ingeniería de Sistemas e Industrial. [Enlínea].
Disponibleen: https://repositorio.unal.edu.co/handle/unal/55245
[20] MORENO, Carlos & SÁNCHEZ, Yakeline. Prototipo de buscadorsemánticoaplicado a la búsqueda
de libros de Ingeniería de Sistemas y Computaciónen la biblioteca Jorge RoaMartínez de la
Universidad Tecnológica de Pereira. Pereira, 2012, 66p. Trabajo de grado. Universidad Tecnológica
de Pereira. Facultad de Ingenierías: Eléctrica, Electrónica, Física y Ciencias de la Computación.
Ingeniería De Sistemas y Computación. [En línea]. Disponible
en:http://repositorio.utp.edu.co/dspace/bitstream/11059/2671/1/0057565M843.pdf
[21] BENAVIDES, Mauricio & GUERRERO, Jimmy. UMAYUX: unmodelo de software de gestión de
conocimientosoportadoenunaontologíadinámicadébilmenteacoplado con un gestor de base de datos.
San Juan de Pasto, 2014, 145p. Trabajo de grado (Trabajo de gradopresentadocomorequisitoparcial
para optar al título de Ingeniero de Sistemas). Universidad de Nariño. Facultad de Ingeniería.
Ingeniería de Sistemas. [Enlínea]. Disponibleen: http://sired.udenar.edu.co/2030
[22] Apache Software Foundation. Apache Jena Fuseki. [Enlínea]. Disponibleen:
https://jena.apache.org/documentation/fuseki2
[23] ARAUJO, Joaquín. ¿Quées Docker? ¿Qué son loscontenedores? y ¿Porqué no usar VMs? [Enlínea].
Disponibleen: https://platzi.com/tutoriales/1432-docker/1484-guia-del-curso-de-docker
[24] BUDHIRAJA, Amar. A simple explanation of document embeddings generated using Doc2Vec.
[Enlínea]. Disponibleen: https://medium.com/@amarbudhiraja/understanding-document-embeddings-
of- doc2vec-bfe7237a26da
[25] CHALLENGER, Ivet, DÍAZ, Yanet& BECERRA, Roberto. El lenguaje de programación Python. En:
Ciencias Holguín. Vol. XX. No. 2. (Junio, 2014), p. 1-13. [Enlínea]. Disponibleen:
www.redalyc.org/articulo.oa?id=181531232001
[26] CHECA, Diego & ROJAS, Oscar. ONTOLOGÍA PARA LOS SISTEMAS HOLÓNICOS DE
MANUFACTURA BASADOS EN LA UNIDAD DE PRODUCCIÓN. En: RevistaColombiana de
Tecnologías de Avanzada. Vol. 1. No. 23. (Noviembre, 2013), p. 134-141. [Enlínea]. Disponibleen:
http://revistas.unipamplona.edu.co/ojs_viceinves/index.php/RCTA/article/view/2334
[27] CLASSORA. Sacandoprovecho a la Web Semántica: SPARQL. [Enlínea]. Disponibleen:
http://blog.classora.com/2012/11/05/sacando-provecho-a-la-web-semantica-sparql
[28] CODINA, Lluís& CRISTÒFOL, Rovira. La Web Semántica. En: JesúsTramullas (coord.).
Tendenciasendocumentación digital. Guijón: Trea, 2006. p. 9-54. [Enlínea]. Disponibleen:
http://eprints.rclis.org/8899
[29] EDWARDS, Gavin. Machine Learning An Introduction. [Enlínea]. Disponibleen:
https://towardsdatascience.com/machine-learning-an-introduction-
23b84d51e6d0ElasticsearchB.V.Elasticsearch. [En línea]. Disponible
en:https://www.elastic.co/es/what- is/elasticsearch

21
[30] FLORES, Pedro & PORTILLO, Julio. ELABORACIÓN DE PROPUESTA DE GUÍA DE
IMPLEMENTACIÓN DE SCRUM PARA EMPRESA SALVADOREÑA, UN CASO DE
ESTUDIO. Antiguo Cuscatlán, 2017, 117p. Trabajo de grado (MAESTRO EN ARQUITECTURA
DE SOFTWARE). Universidad Don Bosco. Arquitectura de Software. [Enlínea]. Disponibleen:
http://rd.udb.edu.sv:8080/jspui/bitstream/11715/1264/1/documento.pdf
[31] FLÓREZ, Héctor. Construcción de ontologías OWL. En: VÍNCULOS. Vol. 4. No. 1. (Diciembre,
2007), p. 19-34. [En línea]. Disponible en:
https://revistas.udistrital.edu.co/index.php/vinculos/article/view/4112
[32] Kit de herramientas de lenguaje natural. [Enlínea]. Disponibleen: https://www.nltk.org
[33] LAMY, Jean. Owlready: Ontology-oriented programming in Python with automatic classification and
high level constructs for biomedical ontologies. En: Artificial Intelligence in Medicine. Vol. 80.
(Agosto, 2017), p. 11-28. [En línea]. Disponible
en:https://www.sciencedirect.com/science/article/pii/S0933365717300271
[34] LINCOLN, Matthew. Uso de SPARQL para acceder a datosabiertosenlazados.
[Enlínea].Disponibleen: https://programminghistorian.org/es/lecciones/sparql-datos-abiertos-
enlazados
[35] LOZANO, Adolfo. Ontologíasen la Web Semántica. [Enlínea]. Disponibleen:
http://eolo.cps.unizar.es/docencia/MasterUPV/Articulos/Ontologias%20en%20la%20Web%20Seman
tica. pdf
[36] MUÑOZ, José. Introducción a flask. [Enlínea]. Disponibleen:
https://plataforma.josedomingo.org/pledin/cursos/flask/curso/u05/
[37] PEDRAZA, Rafael, CODINA, Lluís& CRISTÒFOL, Rovira. Web semántica y ontologíasen el
procesamiento de la información documental. En: El profesional de la información. Vol. 16. No. 6.
(Noviembre, 2007), p. 569-579. [Enlínea]. Disponibleen:
https://repositori.upf.edu/handle/10230/13141
AUTHORS
FELIPE CUJAR ROSERO: Research student of the GRIAS group of the University of
Nariño with publication of papers, presentations, poster exhibition and certifications in the
areas of database knowledge, artificial intelligence and web development.
Link:https://www.linkedin.com/in/felipe-
cujar/.https://scienti.minciencias.gov.co/cvlac/visualizador/generarCurriculoCv.do?cod_rh
=0001853544.https://scholar.google.com/citations?user=dX12cEAAAAAJ.
https://github.com/fcujar
DAVID SANTIAGO PINCHAO ORTIZ: Research student of the GRIAS group of the
University of Nariño with publication of papers, presentations, poster exhibition and
certifications in the areas of database knowledge, artificial intelligence and web
development. Link: https://co.linkedin.com/in/sangeeky. https://github.com/SanGeeky
SILVIO RICARDO TIMARÁN PEREIRA: Doctor of Engineering. Director of Research
Group GRIAS. Professor in the Systems Department of the University of Nariño. Link:
Researcher:http://scienti.colciencias.gov.co:8081/cvlac/visualizador/generarCurriculoCv.do?
cod_rh=0000250988
MATEO GUERRERO RESTREPO: Master in Engineering. GRIAS Group Researcher.
Professor Hora chair of Systems Department of the University of Nariño. Link: Researcher:
http://scienti.minciencias.gov.co:8081/cvlac/visualizador/generarCurriculoCv.do?cod_rh=00
01489230

NATURE: A TOOL RESULTING FROM THE UNION OF ARTIFICIAL INTELLIGENCE AND NATURAL LANGUAGE PROCESSING FOR SEARCHING RESEARCH PROJECTS IN COLOMBIA

Recommended

More Related Content

What's hot (18)

Similar to NATURE: A TOOL RESULTING FROM THE UNION OF ARTIFICIAL INTELLIGENCE AND NATURAL LANGUAGE PROCESSING FOR SEARCHING RESEARCH PROJECTS IN COLOMBIA (20)

Recently uploaded (20)

NATURE: A TOOL RESULTING FROM THE UNION OF ARTIFICIAL INTELLIGENCE AND NATURAL LANGUAGE PROCESSING FOR SEARCHING RESEARCH PROJECTS IN COLOMBIA