1. Introduction. IJCTA Sept-Oct 2016 Available ISSN:

Transcription

1 Advanced Semantic Web Information Retrieval Technique using Ontological Data P. Nandhakumar¹ Adjunct Faculty, Faculty Centre for Computer Science Applications and Research Ramakrishna Mission Vivekananda University, Coimbatore Campus Coimbatore, Tamil Nadu, India M. Hemalatha² Dean of Sciences, Dr.SNS Rajalakshmi College of Arts and Sciences Coimbatore, Tamil Nadu, India R. Sridhar³ Head, Faculty Centre for Computer Science Applications and Research Ramakrishna Mission Vivekananda University, Coimbatore Campus Coimbatore, Tamil Nadu, India Abstract Semantic web is a kind of webs that is able to describe things to be understood by computers. Automatically answering any query without human interactions is one of the key challenges in computer science area. Semantics can help in answering such queries. Consequently, extracting information from unstructured documents and transforming them into semantic web form is an important trend. Semantic web mining is a combination of two trends; semantic web and web mining. This paper concentrates on extracting data from the web page tables which are mostly unstructured and converted into ontological format. Ontology construction and structuring system is proposed initially to know the semantic markups being used and depth first search is done using those semantic markups. The user given query is analyzed and keywords are extracted by parsing and the generated keywords are search by automatically building the SPARQL query. The use of exact semantic markup and efficient search algorithms helps to retrieve more exact links. The extracted links are ranked to find out the most appropriate link to the given query. Experimental results show that the data of interest can be located easily and most accurately. 1. Introduction In this emerging online world, web search plays a vital role and the success of this depends on the search engines being used. The web search engines help millions of users daily to find out the answers for their queries. Hence researchers focus on many ways to increase the efficiency of these search engines. Recently many new are designed in such a way that it can gather, share and reuse the linked data to produce efficient results. Semantic web based search engines makes use of linked data to provide more appropriate results, by making use of semantic relationships that exists between texts than its contents [1, 8]. Semantic search engine is used as a tool which gets formal ontology based[11] queries from the user, executes them using the ontological database and then returns the records of the ontology values that satisfy the user query [2, 3, 4, and 5]. This process uses Boolean search models which are based on the information space which consist of non-redundant, non-ambiguous, pieces of knowledge. A knowledge item can be correct or incorrect reply to a given request, but the results are considered to accurate and appropriate answer to the query is obtained is not validated with the query given. This model also faces the following drawback 1).The information must be fully represented as ontological knowledge base[10]. The huge amount of unstructured information currently available world wide has to be converted into formal ontological knowledge base and may incur cost which may result in problem. 2). A huge document may be break down into smaller information units and may be well interlinked and formalized. It is difficult to store it as a single original document. 3).This type of technique performs their search based on the semantic string values (markups). The key word based search 643

2 practiced in this model do not search beneath the long pieces of text stored as markups, hence it cannot obtain the perfect match to the query [8,15]. This search technique adopted becomes useless if the search space is very large. To over come this ranking of relevant answers has to be performed. Hence, focusing on this drawback a new technique has been proposed to retrieve documents efficiently by using ranking mechanism to return most accurate results. The proposed technique constructs the ontological database and inconstencies are removed to make the search more efficient. The constructed ontological base is used in search process. The user query is parsed and analyzed using word net and the appropriate keywords are obtained. The SPARQL query is automatically constructed and search is performed using depth first search algorithm. The retrieved links are ranked to obtain the most appropriate results for the query. 2. Semantic Information retrieval using Depth first search search. For web crawling application, the depth limit is not known a priori, in such cases DFS also lends itself to make use of heuristic methods for choosing a likely-looking branch. 3. Proposed Methodology The proposed technique works in several phases. Initially the ontological database is constructed and the inconsistencies are removed. Next the user given query is analyzed and parser to find out the meaning full keyword. Then SPARQL query is constructed using those keywords and depth first search is performed. The response to the query is ranking using relevance ranking mechanism. Final results are send to the user and response to the query request. The architecture of the proposed technique is shown in figure.1. The flow of data is depicted in the flowchart shown in figure 1. Depth-first search (DFS) is a popular algorithm which can be used for searching ontological tree based data structures[11]. The search process starts at the root node and explores as far as possible along each branch before backtracking. In general, DFS is typically used to traverse an entire ontological tree, and takes time, linear in the size of the graph. In these applications it also uses space in the worst case to store the stack of vertices on the current search path as well as the set of alreadyvisited vertices. For applications of DFS in relation to specific domains, such as searching for solutions in webcrawling, the ontological tree to be traversed is often either too large to visit in its entirety or infinite[6]. In case of too huge data, traversing is only performed to a limited depth; because of limited resources, such as memory or disk space, data structures are not used to keep track of all the previously visited set of vertices. When search is performed to a limited depth, the time is still linear in terms of the number of expanded vertices and edges but the space complexity of this variant of DFS is only proportional to the depth limit, and as a result, is much smaller than the space needed for searching to the same depth using breadth-first Figure 1: Architecture of Proposed Technique 3.1 Ontology construction and inconsistency removal In this work geopolitical owl database is constructed using neon tool. This owl file is a RDF type with classes, instances and properties. Geographical region classes contain the details about the countries available in the political world map. The inconstencies for this owl file has also been removed using in-cons handler to reduce the processing time and error rate. 644

3 3.2 Query parser and Text analyzer The parsing is done to analyze the query syntactically which determines the part of speech for each and every word in the query. In this way the given query is analyzed grammatically. Here in this method Stanford parser is to analyze the query grammatically and syntactically. The output obtained from the parser is sent to the word net to get the related synsets of various words contained in the query. So here semantically related words are obtained from the output of the word net. 3.3 Relation ship extraction and query construction The relationship between the keywords is extracted since a single keyword generated may have semantically many relationships. Hence the relationship for each and every keyword is extracted and most suitable relationship is selected for traversing during query execution. The figure 2 shows the relationship extracted for the class territory which has parent relationship with the class group and has successor of relationship with the inherited class area. stored and using this template the keywords extracted are replaced in the appropriate position and thus query is constructed automatically. The automatic construction of query helps the naïve user to access the web by using the natural language itself. 3.4 Depth first search algorithm The query being constructed is executed to find the relevant results. The search makes use of depth first search algorithm as it has several benefits. The procedure how this search technique is used in semantic web is given in section 2. Using depth first search has several benefits like it requires less memory, it can be easily implemented using a queue of links (URLs) to be visited next. The new links can be added to the queue to determine the search strategy. FIFO (first in first out, append to end of the queue) gives breadth-first search. LIFO (last in first out, add to front of queue) gives depth-first search. The algorithm of depth first search algorithm is given in figure 3. Figure 2: Relation ship extracted for the class territory The automatic query construction mechanism is being used to generate the SPARQL queries. The SPARQL querying language allows users to write globally unambiguous queries [7, 9, 12]. Hence a predefined template for the query generation is Input: A tree graph G and a vertex v of G Output: All vertices reachable from v labeled as discovered Algorithm : Step 1 procedure DFS(G,v): Step 2 label v as discovered Step 3 for all edges from v to w in G.adjacentEdges(v) do Step 4 if vertex w is not labeled as discovered then Step 5 recursively call DFS(G,w) // To add a new vertex to the queue and perform iteratively: Step 6 let Q be a queue Step 7 Q.add(v) Step 8 while Q is not empty Step 9 v Q.remove() Step 10 if v is not labeled as discovered: Step 11 label v as discovered Step 12 for all edges from v to w in G.adjacentEdges (v) do Step 13 Q.add(w)Figure 3: Depth first search algorithm 645

4 This DFS algorithm visit the neighbors of each vertex in the opposite order from each other: the first neighbor of v visited by the recursive variation is the first one in the list of adjacent edges, while in the iterative variation the first visited neighbor is the last one in the list of adjacent edges. The result of the algorithm will contain the set of vertex which is most appropriate to the keywords being generated. This process is performed iteratively until all relevant links have been extracted. 3.5 Query execution and Ranking mechanism The result of query execution provides set of links which are suitable to the keywords mentioned in the query. The possible links are again refined to find the most appropriate links for the query. The relevance ranking mechanism is used to rank the links that has higher preferences[14]. The link which has the highest rank will be at the top and it provides the most accurate results to the user given query. Following charts shows the results for precision rate, recall rate for this proposed approach as compared to existing search engines[13]. 4.2 Implementation results The proposed system empowers users to create simple expressions that allow them to create complicated conjunctive queries over a knowledge base. Query answering mechanism operates under the open world assumption and the lack of unique name assumption according to the Semantic Web semantics. The evaluation result for a single query is given in figure 4. Similarly 100 queries are executed and the overall precision and recall is calculated and displayed in table Results and discussion 4.1 Evaluation Criteria Accuracy The accuracy of the proposed technique has been evaluated against the result set generated by running the query. From the Web pages returned, it can be observed how there exist possibly out-of-scope pages that have been ranked as very relevant, while potentially interesting pages are positioned at the end of the list. Precision Rate: Precision is the fraction of retrieved documents that are relevant to the search. Figure 4: Performance evaluation screen shot for a single query Recall Rate: Recall in information retrieval is the fraction of the documents that are relevant to the query that are successfully retrieved. 646

5 Table.1 Performance comparison of proposed method Evaluation measures Proposed Method Common web Precision Recall References [1] Berners-Lee, T., Hendler, J. and Lassila, O., 2001 The Semantic Web, Scientific American. [2] Castells, P., Foncillas, B., Lara, R., Rico, M., Alonso, J. L.: Semantic Web Technologies for Economic and Financial Information Management. 1st European Semantic Web Symposium (ESWS 2004). LNCS Vol (2004) Accuracy Time taken ms ms The results obtained in table 1 clearly shows that using depth first search algorithm in the search process has decreased the time taken and it has also high precision and recall rate than existing common web. This proves that proposed method is highly efficient than other methods. Any naïve user can make use of the proposed method as it automatically generates the query by extracting the essential keyword from the user given query. Moreover the query is analyzed to find out the exact semantics by using word net. This phase s adopted in the proposed technique enables to produce higher precision and recall. 5. Conclusion and Future Direction In this paper, the ontology-based semantic web search model has been proposed to enhance efficiency and accuracy of information retrieval for unstructured and semi-structured documents. The modular architecture and basic functionality adopted in the proposed system makes an easy to use webbased application for the semantic web. Initial functionality includes automatic semantic query generation by analyzing the text using parser and word net. The answering and browsing the results over an OWL knowledge base is performed using depth first search algorithm. The experimental result obtained shows a high precision and recall rate than the existing web based search engines. The time taken to extract the links from the ontological base is very less than the existing methods. This proves the efficiency of the proposed technique. Further research is planned for the improvement of the search method on the RDF semantic metadata. [3] Contreras, J., Benjamins, V. R., et al: A Semantic Portal for the International Affairs Sector. 14th International Conference on Knowledge Engineering and Knowledge Management (EKAW 2004). LNCS Vol (2004) [4] Castells, P., Perdrix, F., Pulido, E., Rico, M., Benjamins, V. R., Contreras, J., Lorés, J.:Neptuno: Semantic Web Technologies for a Digital Newspaper Archive. 1st European SemanticWeb Symposium (ESWS 2004). LNCS Vol (2004) [5] Maedche, A., Staab, S., Stojanovic, N., Studer, R., Sure, Y.: SEmantic portal: The SEAL Approach. In: Fensel, D., Hendler, J. A., Lieberman, H., Wahlster, W. (eds.): Spinning the Semantic Web. MIT Press, Cambridge London (2003) [6] D. Tümer, M. A. Shah, and Y. Bitirim, An Empirical Evaluation on Semantic Search Performance of Keyword-Based and Semantic Search Engines: Google, Yahoo, Msn and Hakia, 2009 [7] Google. [8] Web Pages for Testing. [9] CSE 995 Semantic Web. [10] Urvi Shah, James Mayfield, Information Retrieval on the Semantic Web, ACM CIKM International Conference on Information Management, Nov [11] David Vallet, M.Fernandes, An Ontology- Based Information Retrieval Model, European Semantic Web Symposium (ESWS),

6 [12] Yahoo Search Engine. [13] Xiajiong Shen Yan Xu Junyang Yu Ke Zhang,2007 Intelligent Search Engine Based on Formal Concept Analysis IEEE International Conference on Granular Computing, 2(4): 669. [14] Kandogan E., R. Krishnamurthy, S. Raghavan, S. Vaithyanathan, and H. Zhu, 2006"Avatar"Avatar semantic search: a database approach to information retrieval," in Proceedings of SIGMOD '06 Chicago, PP: [15] O. Corby, R. Dieng-Kuntz, and C. Faron- Zucker,2004. Querying the Semantic web with Corese Search Engine. In Proceedings of 15th ECAI/PAIS, Valencia (ES) Author s Profile Dr. M. Hemalatha (csresearchhema@gamil.com) completed M.Sc., M.C.A., M. Phil., Ph.D (Ph.D, Mother Terasa women's University, Kodaikanal). She is Professor & Head and guiding Ph.D Scholars in Department of Computer Science at Karpagam University, Coimbatore. Twelve years of experience in teaching and published more than hundred papers in International Journals and also presented more than eighty papers in various national and international conferences. She received best researcher award in the year 2012 from Karpagam University. Her research areas include Data Mining, Image Processing, Computer Networks, Cloud Computing, Software Engineering, Bioinformatics and Neural Network. She is a reviewer in several National and International Journals. Dr. P. Nandhakumar (nandhap@gmail.com) completed M.C.A., M.Phil., Ph.D (Ph.D, Karpagam University, Coimbatore). He is working as Adjunct Faculty, Faculty Centre for Computer Science Applications and Research, Ramakrishna Mission Vivekananda University, Coimbatore Campus. He is Member of Board of Studies of various autonomous colleges and has given many guest lecturers of different subjects. His research areas include Data Mining, Semantic Web Mining, Cloud Computing and Software Engineering. Dr. R. Sridhar (rmsridhar@rediffmail.com) completed M.Sc., M.C.A., M. Phil., Ph.D (Ph.D, Bharathiyar University, Coimbatore). He is the Professor & Head of Faculty Centre for Computer Science Applications and Research, Ramakrishna Mission Vivekananda University, Coimbatore Campus. Twenty five years of experience in teaching. His research areas include Environmental Management and Bio Technology Interdisciplinary Computer Applications. 648