Ontology-Driven Query Reformulation in Semantic Search

Abstract

Semantic search is a research area in which the goal is to understand the users intended meaning of the query. This requires disambiguation of the user query and interpreting the semantics of the query. Semantic search would thus improve the users search experience through more precise result sets. Moreover, ontologies are explicit conceptualizations of domains, defining concepts, their properties, and the relations among them. This makes ontologies semantic representations of specific domains, suitable to use as a basis for semantic search applications. In this thesis we explore how such a semantic search system based on ontologies may be constructed. The system is built as a query reformulation module that uses an underlying search engine based on Lucene. We employ text mining techniques to semantically enrich an ontology by building feature vectors for the concepts of the ontology. The feature vectors are tailored to a specific document collection and domain, reflecting the vocabulary in the document collection and the domain. We propose four query reformulation strategies for evaluation. The interpretation and expansion of the user query is based on the ontology and the feature vectors. Finally the reformulated query is fired as a weighted query into the Lucene search engine. The evaluation of the implemented prototype reveals that search is in general improved by our reformulation approaches. It is however difficult to give any definite conclusion to which query types benefit the most from our approach, and which reformulation strategy improves the search result the most. All four of the reformulation strategies seem to on average perform quite equally.