An Information Retrieval Model Based on Semantics

Abstract

Models of document indexing and document retrieval are mainly based on statistical NLP method. Computation of meaning is mainly based on the semantics. The former makes it possible to construct a high performance IR system easily. However, the latter is one of the significant methods which can substantially make computer well understand the language. The goal of this paper is to find the conjoined point which can combine the advantages of both schemes, and thus to propose an IR approach. Consequently, a concept-based IR model is proposed. This model is composed of two kernel schemes: the first is a domain language model, which is derived from the traditional language model. Its basic idea is to compute the conditional probability P(Q|D) . The concept extracting approach, which is the second kernel scheme of the proposed model, originates from the traditional linguistics. It can help to well extract the meaning of a term. Thus, we can take the concept (the formalized meaning), instead of the lexical term, as the processing object in the proposed model, and consequently resolve the word sense ambiguity. Experiments on the TREC6 Chinese collection show that the proposed model outperforms the traditional TF-IDF methods, especially in the average precision and the overall search time.