Information Retrieval Model Based On Markov Network

With the development of computer science and internet, people now can get abundant information rapidly and expediently, but they find it is still a problem to get information they really need. Information retrieval is to solve it, and has been a focus of researchers these years.As the process of information retrieval contains many uncertainties, a user's interest is hard to describe and measure and influenced by many factors, such as user's background and circumstance, documents is difficult to represent, it takes some difficulties for information retrieval.Adding useful information into information retrieval by using learning mechanism, analyzing and studying of corpus or user's relevance feedback, has been proved to be an effective way to improve the performance of information retrieval,Graphical model is a good way to induct learning mechanism, especially Bayesian Network Model. Bayesian Network is a directed network, which makes it difficult and complicated to construct. Markov Network is an undirected graphical network that is capable of efficiently representing relevance in knowledge and can be easily gotten from training data. In this thesis, we propose and implement an information retrieval model based on Markov network. We construct the Markov network to represent the relationships between terms and the relationships between documents learned from training corpus, do query expansion and re-compute the similarity between documents and queries. Experiments show the result of our model is promising.The creatives of this thesis are:1. Propose an information retrieval model based on Markov Network for the first time; the model can effectively represent the relationship between documents and the relationship between terms, and can be used to represent any of the classic models in IR(Information Retrieval).2. Analysis and verify the performance of Markov Network model, compare the result with other information retrieval model.