Research Of Using Term-relationships To Extend Bayesian Network Retrieval Models

Bayesian networks, or belief networks, is a directed acyclic graph to describe probability, applied to manage uncertainty and probability. Nowadays it constitutes the dominant approach of processing uncertain information. Because information retrieval process also includes a certain amount of uncertainty, therefore, since the end of the 1980s, when Bayesian network for the first time was applied to information retrieval, the research of Bayesian network based information retrieval has been developed rapidly within the last 10 years, and offered some IR models.Synonyms refer to the words, those who can be interchanged or express the same or similar idea, while relatives are the ones, despite different meanings, who always appear together and have some correlation between them. Because synonyms and correlatives of one term include some searching information of users, the important contents in information retrieval fields are the recognition of synonyms and relatives and the quantification of their relationships, and how to use them to expand the inquiry.To solve the problems that BN models don't use the relationships among index terms, this paper mines theses relationships, and extends some BN models by using them. Experiments show that new models have better performance. The significant achievements of this dissertation are as follows:By using Synonyms of Query terms and their similarity, this dissertation presents an extended belief network model for IR. The performance of proposed model is tested by experiment.Presents an improved co-occurrences frequency method to mine relationships among document index terms of belief network model, gives an expanded model by adding one term layer, and tests new model's performance by experiment.By combining synonyms evidence of query terms, presents an extended belief network model. Its performance is tested.By using terms relationships obtained from Synonyms, presents a new retrieval model based on simple Bayesian network and tests is performance.Studies term relationships by using co-occurrence analysis method, and then presents an expanded model of SID by using this relation, gives out this model's topology and retrieval process.