Markov Network Retrieval Model Based on Document Cliques

The query expansion is an effective way to improve the efficiency of information retrieval. But many of the query expansion methods to select the expansion terms did not take fully account of the correlation between the terms as well as terms and documents, which may reduce retrieval performance.Due to the information of the correlation between terms and documents is able to improve the efficiency of retrieval, this paper calculates the correlation between documents and terms, and mapping terms to documents to build a Markov network retrieval model; and then extracts term clique according to the mapping information. The mappting information is used to divide the term cliques into two categories. One is based on document and another is not based on document.The terms cliques based on document are more relevant with the query topic, so to the terms cliques are given greater weight based on document and the information of the two kinds of terms cliques is used to assist retrieval. Therefore, the method we propose in this paper can make the extension content more relevant to query. Experimental results show the proposed model can improve the retrieval efficiency.