Study and Development of Question Answering System based on Ontology Query

Abstract

With respect to the ontology knowledge representation of refrigerator, a question answering system was designed based on ontology triple structure query. The ontology knowledge representation of refrigerator is built with representing the refrigerators' instance, attributes and values with the subject-predicate-object triples format. We analyze the natural language question, extract key words, analyze the dependencies between keywords, and build the query for ontology triples. The question answering system is implemented based on the Jena API and supports natural language querying OWL ontology. The results showed that, compared with Yang Tianqi's method, this method improve recall ratio and precision by 4.96% and 3.81% respectively. Ontology can clearly express the relationship between things and their relationship, plays an important role in the semantic analysis of the automatic question answering system. Ontology in question answering system is mainly used to calculate the distance of concepts, methods WordNet meaning analysis of keyword similarity calculation(1), to extend keywords. This approach extends the semantic information of a single keyword, but ignored the relationship between keywords, resulting in omission user statement semantic information. In recent years, with the rapid development of Linked Open Data, more and more companies will put these data in the Semantic Web with URI and RDF format, building the association with other data sources. The standard approach to query RDF is using structured queries in triple-pattern-based language like SPARQL, but only professional programmers are able to precisely build a structured query triples. as for ordinary users, the only choice to query knowledge base is by keywords or natural language sentences. Integrated the above two points, we design and implementation a method of querying RDF format data by natural language, thus implement the question answering system. We design ontology to express formula information, using Stanford Dependency analyzing the user's natural language question, build a triple query, returned to the user a precise answer. II. PREVIOUS WORK At home and abroad in recent years, many natural language query ontology question answering system were designed. Pythia(2) and ORAKEL(3) based on formal grammar, assign each lexical unit a syntactic and semantic expression, the expression of vocabulary has been consistent with the ontology of vocabulary, through calculation of the grammar portion of combination of semantic questions the semantic information of the whole. Aqualog and PowerAqua(4) system map the language structure to a compatible with ontology semantic structure, the subject of natural language matches the knowledge base of the elements in question and the predicate object. FREyA(5) name after feedback, refinement and Extended vocabulary Aggregation. FREyA and Querix(6) both interact with the user to further confirm the user's query information, carries on the deep syntactic analysis, in order to eliminate ambiguity. Yahya etal.(7) based on an integer linear program, mapping the phrase of natural language question to the class, individual, and properties of the knowledge base , constructs the SPARQL structured query, in this process to achieve entity disambiguation and predicate disambiguation jointly. In the domestic, Zhang Zongren and Yang Tianqi(8) proposed a method to translate natural language into SPARQL query, using the Stanford parser do the syntactic analysis of user's natural language question, construct three tuple SPARQL query. Xu Kun and Feng Yansong(9) put forward the concept of semantic query graph, transform the natural language into SPARQL queries.