TALASH: A SEMANTIC AND CONTEXT BASED OPTIMIZED HINDI SEARCH ENGINE

International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.3, June 2012
DOI : 10.5121/ijcseit.2012.2305 71
TALASH: A SEMANTIC AND CONTEXT BASED
OPTIMIZED HINDI SEARCH ENGINE
Nandkishor Vasnik1
, Shriya Sahu2
, Devshri Roy3
1
Department of CSE, MANIT, Bhopal, MP, India
vasnik.nd@gmail.com
2
s.shriya88@gmail.com
3
droy.iit@gmail.com
ABSTRACT
The traditional search engine have shortcoming that they retrieve irrelevant information. Query expansion
with relevant words increases the performance of search engines, but finding and using the relevant words
is an open problem. This paper presents a Hindi search engine in which we describe three models for
query enhancement. They are based on lexical variance, user context and combination of both techniques.
KEYWORDS
Information retrieval, context, lexical resources, query enhancement, HindiWordNet.
1. INTRODUCTION
The Web is growing as the fastest communication medium. This technology in combination with
latest electronic storage devices, enable us to keep track of enormous amount of information
available to the society. Information present in web are present in different language like English,
Arabic, Bengali, Hindi many more. The Web is full of unstructured information and traditional
search engines practice a keyword-oriented search scheme which leads to the problem of
retrieving numerous irrelevant information. Such searching scheme with a specific keyword may
eventuate to unsatisfactory, but with its synonym to appropriate results .Many of the documents
retrieved for general queries are irrelevant to the subject of interest and other documents are
missing because the query does not include the exact keywords. Users are not confident about the
languages used to formulate their queries, refined queries with restrictive Boolean operators may
result in a few or even no documents .This motivates the use of natural language interfaces as an
adequate way of communicating with search engines.
One solution to improve the relevancy of retrieved results is to expand the user query with more
relevant words. But achieve suitable relevant words is a challenging problem. Along with this,
sometimes the query is nebulous, its domain or context is unpredictable and as a result, selection
of enhancing keywords is really difficult [13]. Instances of these queries are words which have
various meanings in different domains.

72
To improve the quality of the search on Internet, NLP techniques approaches have typically been
adopted. In which query extensions and improving the quality of information retrieved using
NLP-based systems. Result of search engine is depend on database present and how structured is
it. In web very limited Hindi document is present, so traditional searching scheme will not work
properly. We have applied some technique to get better result from limited Hindi document.
Our proposal is to provide linguistic mechanisms that transform and extend the user query by
integrating HindiWordNet semantic database, [12] and user context. The main goal of system,
TALASH: A Hindi Search Engine is to improve the result provided by Google search engine
through the extension of user Hindi input. The rest of the paper is organized as follows. Next
section discusses related research in concerned fields. In Section 3, suggest the proposed query
expansion models. Section 4 presents Result and experiment and finally, Section 5 gives
conclusions and directions for future work.
2. RELATED WORK
According to research accomplished by Imaee and colleagues [1], query expansion has a
significant role in the Information retrieval. They showed that it can eventuate to the increase of
precision in information retrieval. Since the emergence of Semantic Web in late 1990, semantic
search has been one of the most distinguished areas. Strategies for combining information of
ontology [2] and electronic dictionary with search patterns are studied [3,4,5]. They show that
exploiting only one semantic relation, such as Hypernym or Synonym is not effective, so it is
better that a combination of semantic relations to be used. In the studies that a combination of
these relations has been used [6, 7], promising results have been reported.
Dey et al. [8] define context as any information that can be used to characterize the situation of an
entity. An entity is a person, place, or object that is considered relevant to the interaction between
a user and an application, including the user and their applications. Schilit et al. [9] were one of
the first groups to identify context-aware computing as an important aspect of mobile computing.
They identify who one is, one’s location, and the people, objects, or other resources nearby as
important aspects of context. This also aligns well with Bellotti and Edward’s classification of
necessary types of context. They believe context-aware systems must include responsiveness to
the environment, to the person, and to the person’s social group [10]. Erickson addresses issues
for Context-Aware computing in a CACM article [11]. He discusses how context aware software
development often requires the software to act autonomously without a specific request from the
user. For example, a software system may automatically adjust the volume of a speaker system in
order to compensate for the volume of a speaker’s voice. Erickson points out that it is difficult for
the software to accurately make decisions without input from the user. Google provide searching
in Hindi, Bengali, Tamil and many more but due to unstructured or limited document it retrieve
many irrelevant document. Hindi documents present in web are not so huge so use get irrelevant
information most of the time.
In Hindi there is limited work has been done on query enhancement to improve the performance
of search engine and this motivates for developing enhance Search Engine in which user will
pose a query in Hindi and get relevant information related to input.
3. THE PROPOSED QUERY EXPANSION MODEL
We have developed three methods for query enhancement for Hindi input in system. First method
use lexical resource, second method use user context [14] and last method is combination of this
two method [17].

73
3.1. Method-I: Query enhancement using lexical resource
Extending web queries using lexical resources focusing on the Query Generator component, it
deals with lexical variation of significant words of user queries in order to enhance document
searching. Lexical variation is accomplished by using HindiWordNet(HWN) [12], a lexical
database that is structured as a top concept ontology that reflects different explicit relationships.
Words are organized in synonym sets, called synsets (each synset represents a concept). The
synsets are related by hyponym and hypernym (IS-A) relationships.
In Query Generator module the natural language query is tokenized, and keywords that have to be
extended by using HWN lexical resources. HWN is used for extracting semantically related
terms, in this approach synonym and hypernym words are used[16]. For instance, given the noun
“झोपड़ी” (Hut), HindiWordNet provides output shown in figure 1.
Then, nominal and adjectival keywords are then expanded with gender and number variations
and verbal keywords with their corresponding infinitive lemma .This step is necessary, as the
search engine does not perform any kind of stemming/ morphological inflection for Hindi
language. In short, the strategy for Boolean query generation is this: semantically related terms
are added to each relevant term connected by ORs [13].
Synset [0] 9718 - NOUN - [झोपड़ी, झोपड़ा, झ पड़ा, झ पड़ी, मड़ई, मड़ैया, मड़ाई, मढ़ई, मढ़ा, मढ़ , आ शयाना,
आ शयाँ]
HYPERNYM : 1901 - NOUN - [घर, गृह, मकान, सदन, शाला, आलय, धाम, नके तन, नलय, के तन, पण,
गेह, सराय, अमा, नषदन, अवसथ, अव थान, आगार, आगर, आयतन, आ य]
HYPONYM : 18476 - NOUN - [टपर , टपरा]
ONTO_NODES : भौ तक थान (Physical Place)
Figure.1: Output of HindiWordNet
For instance, the term “सूरज” (Sun) is expanded as: (सूरज OR दवाकर OR भा कर OR दनकर OR
र व). Finally, all the sets of expanded terms are connected by the AND operator. For example
user query is “सूय नम कार" (Sun Salutation) is translated into: “( सूय ORसूरज OR दवाकर OR
भा कर OR दनकर OR र व) AND (नम कार OR नमन OR अ भवादन OR अ भवंदन OR नम ते)”.
This query is ready to be used as input for the web search engine

74
Figure.2: Architecture of Method-I
Architecture for Method-I is shown in figure 2.User impose their Hindi query in system then it
pass to HWN. HWN extract semantically related terms and send to Formal Query Generator
module. Where Boolean query is builds with the help of semantic related terms and send Google
search engine. Search engine retrieve information from web according to Boolean query then
show top ten results on interface.
3.2. Method-II: Query Enhancement using user Context
Users are required to register to the system and fill out their profile at the first time they are
logging in to the system. In the subsequent times, they are identified by the system. A user can
change his profile information in any time he enters the system. The profile information consists
of name, username, password, age, profession and sexuality.
3.2.1. Context identification
At first the context information that describe the interaction environment between the user and the
system are acquired, which consists of [14]:
(a)User role: User role determines main activity of a user in its current environment. For example
a woman is a software engineer then if she searches the query “जावा” (Java) in system, she
probably aims “जावा ो ा मंग भाषा” (programing language) instead of other meanings of “जावा”
(java) such as coffee or an island. If same query is fired by a house wife, she expected result on
coffee or island not programming language .Hence, user role can be exploited to disambiguate
multi-meaning queries.
(b)User location: Location of user will be retrieved from the user’s Google calendar. Alternately,
the system may collect this information from a local Outlook calendar on the user’s local
machine. If any user living in particular area then result should be related to that area. If user
situated in India and fire query “रा ट य फू ल” (National flower) then result should be related to
"कमल” (lotus) national flower of India.

75
(c)User interests: User interests are entered by users to the system and system augments this by
inserting hypernyms of the words [15]. System categorized the user interest in following category
and sub categories shown in table 1.Interest of a user is saved for their subsequent search [19].
Users can modify their interest for every new search.
Table 1. Category and subcategory of user interest.
Figure.3: Architecture of Method-II
Category Subcategory
श ा (Education) इ तहास(History),भूगोल(Geography),राज न त(Politics),मनो व ान
(Psychology),
सं कृ त(Culture),खगो लय(Astronomy),दशनशा (Philosophy),समाजशा
(Sociology),
सा ह य(Literature).
व ान (Science) जीव व ान(Zoology), रसायन व ान
(Chemistry),भौ तक(Physics),वन प त
(Botany), ो ा मंग भाषा(Programming language), क यूटर हाडवेयर
(Computer hardware),सॉ टवेयर(Software), अ भयां क (Engineering).
मनोरंजन
(Entertainment)
फ म(Movies),गीत(Songs),छ व(Images), खेल(Sports) ,समाचार(News)
अ य (Other)

76
Architecture for Method-II is shown in figure 3. User login in system through interface and
choose their interest. With help of user information and interest formal query generator module
form Boolean query and send to Google search engine. Search engine retrieve relevant
information from internet according to Boolean query and show top ten results on interface.
3.3. Method-III: Query Enhancement using hybrid technique
Here system uses hybrid technique which is combination of both query enhancement using
lexical resource and user Context technique for input hindi query [15]. System builds formal
query by submerging query get from both technique.
Figure.4: Architecture of Method-III.
Architecture of Method-III is shown in figure 4. User have login in system and impose their
query in hindi along with choose one of the category or subcategory of interest. Both semantic
related term and user information along with user interest send to formal query generator. It form
Boolean Query through received information and send to Google search engine. Search engine
retrieve information from web database and top ten results.
4. RESULT AND EXPERIMENT
In order to evaluate how the linguistic knowledge affects the retrieval results, we study of thirty
queries in four types of experiments, all of them executed in the Google Search mode. These
experiments are: (1) Simple Google search (2) Method-I (query enhance with HWN), (3)
Method-II(query enhance with user context) and (4)Method-III (expansion with HWN and user
context). In this experiment thirty queries are given one by one to each propose model and
analyzed top ten results. Some links are useful and most of results are irrelevant due to
insufficient database for Hindi document. Some query give better result as compare to other
queries because more Hindi document present in web related to that query. Precision values are
measured considering uppermost retrieved documents by experiment [18]. Precision values
obtained in the four experiments are: (1) 0.57, (2) 0.67, (3) 0.69 and (4) 0.79. These results show
that a combination of Method-I and Method-II could enhance information retrieval.

77
Table 2.Experiment Results
The recall on the other hand is the ability of a retrieval system to obtain all or most of the relevant
documents in the collection. Thus it requires knowledge not just of the relevant and retrieved but
also those not retrieved. There is no proper method of calculating absolute recall of search
engines as it is impossible to know the total number of relevant in huge databases. So we have
adapted the traditional recall measurement for use in the Web environment by giving it a relative
flavour. The relative recall value is thus defined as below, where A is total number of document
retrieve by search engine and B is sum of document retrieve by all four experiments.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Google Method-
I
Method-
II
Method-
III
Precision
relative recall
F-measure
Figure.5: Experimental graph
5. CONCLUSIONS AND FUTURE WORK
In this paper three methods are discuss for query expansion by utilizing HWN, user context and
combination of both has been proposed . In first method use semantically related terms of query
for this approach synonym and hypernyms words are used. Second method deal with user
context, for context, we have investigated the use of personal calendar/PIM information to help
determine what a user is currently doing, their physical location and personal interest. Last
method implements combination of both methods. The experimental results show that by
combining the principles of search with HWN and context awareness, more relevant search
results may be produced for a user and, the precision of information retrieval increases. F-

78
measure also shows that Method-III is better than other methods. Due inadequate Hindi database
present in web affect performance of system.
Future direction of the current research consists of designing and implementing agents to extract
context automatically by search history of user and implement morphological analyzer like Hindi
Shallow Parser with Method-III.
REFERENCE
[1] H. Imai, C. Nigel, and T. Jun’ichi, “A combined query expansion approach for information retrieval” ,
Genome Informatics, Tokyo, Japan: Universal Academy Press Inc. 1999 [2] J.Davies, R. Studer and
P. Warren, Semantic Web Technologies: Trends and Research in Ontology-based Systems, John
Wiley & Sons, 2006.
[3] Z. Gong, C. Wa Cheang and U. L. Hou, “Web Query Expansion by WordNet”, In Proceedings of the
16th International Conference on Database and Expert Systems Applications ,Copenhagen, Demark,
2005,pp. 166-175.
[4] Z. Gong, C. Wa Cheang and U. L., Hou, “Multi-term Web Query Expansion Using WordNet”, the
17th International Conference on Database and Expert Systems Applications,2006.
[5] Y. Liu, C. Li, P. Zhang and Z. Xiong, “A Query Expansion Algorithm based on Phrases Semantic
Similarity”, International Symposiums on Information Processing, IEEE, 2008.
[6] E. M. Voorhees, “Using WordNet to disambiguate word senses for text retrieval”, the 16th Annual
international ACM SIGIR Conference on Research and Development in information Retrieval, ACM
Press., 1993.
[7] D. Parapar, A. Barreiro, and D. Losada, “Query expansion using Wordnet with a logical model of
information retrieval”, IADIS International Conference , Portugal, 2005
[8] D. A. Gregory, K. D. Anind, J. B. Peter, D. Nigel, S. Mark, and S. Pete, "Towards a better
understanding of context and context-awareness," in Proc. of the 1st international symposium on
Handheld and Ubiquitous Computing, Karlsruhe, Germany: Springer-Verlag, 1999.
[9] B. Schilit and M. Theimer, "Disseminating active map information to mobile hosts," IEEE Network,
vol. 8, pp. 22-32, 1994.
[10] V. Bellotti and K. Edwards, "Intelligibility and Accountability: Human Considerations in Context-
Aware Systems," Journal of Human-Computer Interaction 16, pp. 193-212, 2001.
[11] E. Thomas, "Some problems with the notion of context-aware computing," Communications of the
ACM, vol. 45, pp. 102-104, 2002.
[12] http://www.cfilt.iitb.ac.in/wordnet/webhwn/
[13] Ana Garcia-Serrano, Paloma Martine and Albert0 Ruiz “linguistic engineering approach to the
enhancement of web-searching” 2001 IEEE.
[14] Najmeh Ahmadian, M .A Nematbakhsh and Hamed Vahdat-Nejad,” A Context Aware Approach to
Semantic Query Expansion” 2011 International Conference on Innovations in Information
Technology,2011 IEEE.
[15] Nicole Anderson,” Putting Search in Context: Using Dynamically-Weighted Information Fusion to
Improve Search Results”, 2011 Eighth International Conference on Information Technology: New
Generations,2011 IEEE.
[16] Sírius Thadeu Ferreira da Silva, Samuel de Oliveira Apolonio, Adriana S. Vivacqua, Jonice Oliveira,
Geraldo B. Xexéo and Maria Luiza M. Campos,” Ontoogle: Enhancing Retrieval with Ontologies and
Facets”, Proceedings of the 2011 15th International Conference on Computer Supported Cooperative
Work in Design,2011 IEEE.
[17] Jingjing Liu, Xiao Li, Alex Acero2 and Ye-Yi Wang,” lexicon modeling for query
understanding”,2011 IEEE.
[18] C. H. C. Leungn and Yuanxi Li,” CYC Based Query Expansion Framework for Effective Image
Retrieval”, 2011 4th International Congress on Image and Signal Processing,2011 IEEE.
[19] Farag Ahmed and Andreas N¨urnberger,” A Web Statistics based Conflation Approach to Improve
Arabic Text Retrieval”, Proceedings of the Federated Conference on Computer Science and
Information Systems pp. 3–9,2011 IEEE.

79
Authors
Nandkishor Vasnik received his B.E. degree in Computer Science & Engineering from
Rajeev Gandhi Technical University, Bhopal, India, in 2010. Now he is an MTech student
at Computer Science and Engineering Department in Maulana Azad National Institute of
Technology, Bhopal, India. His interests involve Natural Language Processing (NLP) and
Ontology.
Shriya Sahu received her B.E. degree in Computer Science & Engineering from
Chhattisgarh Swami Vivekanand Technical University, Bhilai, India, in 2009. Now she is
an MTech student at Computer Science and Engineering Department in Maulana Azad
National Institute of Technology, Bhopal, India. Her interests involve Natural Language
Processing (NLP) and Ontology.
Dr. Devshri Roy is a University Distinguished Scholar Professor of Computer Science and
Engineering at Maulana Azad National Institute of Technology, Bhopal, India . She has
done her PhD from Indian Institute of Technology, Kharagpur, India. She is specialized in
Application of Computer and Communication Technologies in e-learning , Personalized
Information Retrieval,and Natural Language Processing. She published many research
papers including writing of books.

TALASH: A SEMANTIC AND CONTEXT BASED OPTIMIZED HINDI SEARCH ENGINE

More Related Content

What's hot (20)

Similar to TALASH: A SEMANTIC AND CONTEXT BASED OPTIMIZED HINDI SEARCH ENGINE (20)

More from IJCSEIT Journal (20)

Recently uploaded (20)

TALASH: A SEMANTIC AND CONTEXT BASED OPTIMIZED HINDI SEARCH ENGINE