Structured Document Retrieval Based On Bayesian Network

The traditional information retrieval method based on the key word matching uses the single and surfaced model, and only uses the single term to retrieval. Therefore the retrieval results are often unable to satisfy the information need of users in the recall and the precision. At the same time many researches indicate that using the term relationships can improve the performance of the Information Retrieval system.Bayesian network is a kind of model of representing the uncertainty knowledge and inference, which is the dominant approach for managing uncertainty. Bayesian network have been applied to Information Retrieval (IR) in different ways to solve a wide range of problems where uncertainty is an important feature. Especially, since Bayesian network can accurately represent the structure of the document, it is fit for applying to structured document retrieval, which is the new field of IR.This paper adopts the method based on co-occurrence analysis to learn the term relationships, applies these relationships to the structured document retrieval, and presents a structured document retrieval model based on Bayesian network, gives the topology, probability estimation and the inference process of this model. Co-occurrence analysis is the hard core of setting up the concept space in traditional IR. In order to use co-occurrence analysis in structured document retrieval, some computation formulas have been modified correspondingly. Finally the experiment results show that this model can improve the retrieval performance effectively.