International Journal On Engineering Technology and Sciences – IJETS™ ISSN (P): 2349-3968, ISSN (O): 2349-3976 Volume 2 Issue 1, January 2015 ONTOLOGY BASED QUESTION ANSWERING SYSTEM Dr.C.Kumar Charliepaul 1 Principal A.S.L Pauls College of Engg & Tech, Coimbatore . [email protected] ABSTRACT Question Answering (QA) is a application field of information retrieval and natural language processing. It is concerned with building systems that automatically answer questions posed by humans in a natural language. Question answering systems are considered more complex than information retrieval (IR) systems and require extensive natural language processing techniques to provide an accurate answer to the natural language questions. Domain-specific question answering systems require pre-constructed knowledge sources, such as domain ontology. Domain ontology is used as a reliable source of knowledge in information retrieval systems such as question answering systems. Automatic ontology construction is possible by extracting concept relations from unstructured large-scale text. Here, proposed a methodology to extract concept relations from unstructured text using a syntactic and semantic probability-based Naive Bayes classifier. A set of hand-coded dependency parsing pattern rules and a binary decision tree-based rule engine were developed for this purpose. This ontology construction process is initiated through a question answering process. For each new query submitted, the required concept is dynamically constructed, and ontology is updated. 1. Introduction sources, such as domain ontology. A major challenge in knowledge-based QA system development is building a huge knowledge base An ontology is a formal framework for with objective and correct factual knowledge in representing knowledge. This framework names the preferred domain. The process of collecting and defines the types, properties, and useful knowledge from various sources and interrelationships of the entities in a domain of maintaining this information in a knowledge discourse. Ontology compartmentalizes the repository is a useful process when providing a variables needed for some set of computations, required answer on demand with greater and establishes the relationships between them. accuracy and efficiency. The fields of artificial intelligence, the Semantic Web, systems engineering, software engineering, biomedical informatics, library science, enterprise bookmarking, and information architecture all creates ontologies to limit complexity and to organize information. The ontology can then be applied to problem solving. Question answering systems in general use external knowledge sources to extract answers. Domain-specific question answering Figure 2.1- Proposed Architecture. systems require pre-constructed knowledge 13 International Journal On Engineering Technology and Sciences – IJETS™ ISSN (P): 2349-3968, ISSN (O): 2349-3976 Volume 2 Issue 1, January 2015 2. Description The pattern-matching engine considers the presence of the ‘‘Attribute lexical-pattern’’ 2.1 Sentence Extraction to identify the predicate used along with the Sentence extraction is a technique used connectors as one of the attributes of the to identify the most salient sentences of a text. concept C. The nearest VP node to target object Sentence extraction is a low-cost approach is (right most NP) only considered for compared to more knowledge-intensive deeper identifying attributes except the case ‘‘if’’ verb approaches which require additional knowledge pattern is <TO + VB>OR <VBS>, ‘‘then’’ bases such as ontologies or linguistic convert VB into VBZ and attach BY to it (VBZ knowledge. Extractive methods work by +BY) to construct the attribute (e.g., ‘‘to selecting a subset of existing words, phrases, or create’’ as ‘‘created by’’). The following three sentences in the original text. basic pattern components are used to generalize 2.2 Parse tree construction the rule: A parse tree is an ordered, rooted tree A= [NP, {PPER|NN|PRF}] that represents the syntactic structure of a string B =[VP closest to C] & [right child of according to some context-free grammar. Parse VP{S|NP|PP|ADJP|ADVP}] trees are usually constructed according to one of C= [NP, {PPER|NN}] two competing relations, either in terms of the Precedence: A<B<C constituency relation of constituency grammars The mapping C (A, B) denotes that the B is one (phrase structure grammars) or in terms of the of the attributes that describes the concept A dependency relation of dependency grammars. with the value C. Parse trees are distinct from abstract syntax 2.4 Automatic relation classification using a trees (also known simply as syntax trees), in that Naive Bayes classifier their structure and elements more concretely reflect the syntax of the input language. Parse Naive Bayes (NB) classifiers have been trees may be generated for sentences in natural proven to be very effective for solving large- languages, as well as during processing of scale text categorization problems with high computer languages, such as programming accuracy. In this research, we used an languages. expectation– maximization-based Naive Bayes In this proposed work, the Stanford classifier for classifying the relation between the dependency parser(Marie-Catherine de seed concept and predicate object through the Marneffe, 2008) are used for generating a parse predicate that exits in a sentence. Thus, the tree for each individual sentence in relevant sentence classification problem is converted to a documents concerning the seed concept. Each concept relation classification problem. individual sentence are formed with tree structure with a part-of-speech tags. 2.5 Ontology Construction 2.3 Hand-coded rules for concept triple The rough schema of the ontology extraction concept is dynamicallymodelled using the set of Then, the proposed binary decision tree-based concept relations extracted for thegiven seed rule engine applied the set of hand-coded rules concept. The ontology schema is generated with to the dependency parsing pattern. The outcome abottom-up approach in which the attributes are of the rule engine is a set of triples consisting of identifiedusing instances. An attribute is three components: candidate key word, which considered for inclusion intothe target schema represents the given seed concept; predicate and when there is an existing relationshipbetween target object, which is considered the associated the candidate concept and associated concept concept.The concept triple extraction from the keywordin the instance. A sample ontology dependency parsing pattern is performed using schema constructedusing this approach. hand-coded rules. The rules are formulated by empirical analysis. 14 International Journal On Engineering Technology and Sciences – IJETS™ ISSN (P): 2349-3968, ISSN (O): 2349-3976 Volume 2 Issue 1, January 2015 2.6 Query Analysis and Processing ontology-based question answering with collections of In this module, Natural language user queries. Inf. Process. Manage. 45, 175–188. 5) Gacitua, R., Sawyer, P., Rayson, P., 2008. A questions provided by the user are identified and flexible framework to experiment with ontology learning question patterns were generated which contains techniques. Knowl. Based Syst. 21, 192–199. question type. Questions patterns are then 6) Hearst, M., 1998. Automated discovery of converted to query string which provide WordNet relations. In: Fellbaum, Christiane. (Eds.), keywords in that natural language questions are WordNet: An Electronic LexicalDatabase and Some of its Applications, MIT Press, USA, pp. 131–151. analysed. 7) Hirschman, Lynette, Gaizauskas, Robert, 2001. After the question is analysed, query Natural language question answering: the view from here. processor will search in the constructed Nat. Lang. Eng. 7, 275–300. ontology for keywords from the query string. If 8) Horrocks, I., 2008. Ontologies and the semantic keyword is found, answer is generated in a web. Commun. ACM 51, 58–67. 9) Hou, Xin, Ong, S.K., Nee, A.Y.C., Zhang, X.T., grammatical structure from ontology otherwise Liu, W.J., 2011. GRAONTO: a graph-based approach for ontology is updated by using classifier for that automatic construction of domain ontology. Expert Syst. relevant keyword. Appl. 38, 11958–11975. 10) Jung, Yuchul, Ryu, Jihee, Kim, Kyung-Min, 3. Conclusion Myaeng, Sung-Hyon, 2010. Automatic construction of a large-scale situation ontology by mining how-to Domain-specific question answering instructions from the web. Web Semant.: Sci. Serv. Agents World Wide Web 8, 110–124. systems require pre-constructed knowledge 11) Leacock, Claudia, Chodorow, Martin, 1998. sources, such as domain ontology.Automatic Combining Local Context and WordNet Similarity for ontology construction is possible by extracting Word Sense Identification. The MIT Press, USA. concept relations from unstructured text. Here, 12) Lopez, V. et al, 2007. AquaLog: an ontology- proposed a methodology to extract concept driven question answering system for organizational semantic intranets. J. Web Semant. 5, 72–105. relations from unstructured text using a 13) Maria, R, et al, 2004. Automatic extraction of syntactic and semantic probability-based Naive semantic relationships for WordNet by means of pattern Bayes classifier. A set of hand-coded learning from Wikipedia*, Computer Science Dep., dependency parsing pattern rules and a binary Universidad Autonoma de Madrid, pp. 1-13. decision tree-based rule engine were developed for this purpose.Then constructed