Bayesian Network based Information Retrieval Model

Information Retrieval Models (IRM) that integrate term dependencies are based on the assumption that the retrieval performance of an Information Retrieval System (IRS) usually increases when the relationships among the terms, contained in a given document collection, is used. These models have to deal with two problems. The first is how to obtain a set of relevant dependence relationships efficiently form a document collection. The second problem is how best to use the set of the obtained dependencies to retrieve relevant documents, given a user query. In this work, a new information retrieval model based on Bayesian networks is proposed. Its aim is to achieve a good retrieval performance by restricting the set of dependencies between terms to most relevant ones. In order to achieve this objective, this model searches for dependence relationships within each document in the collection. Then, it creates a final list of dependencies by merging the set of lists obtained localy form each document. Experiments carried out on four standard document collections have proven the efficiency of the proposed model.