Searching Messages Based on Semantic Context

Clients' queries upon keywords or other informed description do not usually provide complete and unambiguous retrieval of information. Expansion of the queries based on semantic relation and phrase patterns is an effective approach to improve the retrieval. In this paper, a novel approach to queries expansion is presented. In the first step, keywords and phrases in a query are extracted, and the query is classified using Bayesian classifier. The classification defines the domain of user interest which serves as the context around the query. The phrase is then matched against fixed patterns which are automatically extracted from the set of domain documents and serve as context within the query. This is followed by expansion of the remaining keywords that are not in the phrases. This expansion is based on synonyms and hyponyms in the domain ontology, and is controlled by measuring information gain. Finally, the similarity between the expanded query and the document of the domain is computed by combining the weighted phrases and keywords. Experimental results indicate that the proposed approach improves the precision and recall of information retrieval.