Extended Information Retrieval Model Based On Markov Cliques

The most important part of information retrieval is finding relevant contents that user needs. In this part, the most important process is: For a given query, Calculation the correlation between each document and query. In this process, Query expressed the user's information needs. It's the most important data in the Information retrieval process. The accuracy of queries related to search results. However, due to the query is too short or some other reasons, the search results usually very poor. A common solution is to treatment the query. This solution makes the query better express the user's intent.Current, the most commonly methods used to treatment the query are relevance feedback, query reconstruction and automatic query expansion. According to document sets, Automatic query expansion technique divided into local analysis method and global analysis method. As an early appeared query expansion method. Global analysis method has been widely used.The traditional global analysis method didn't take the indirect relationship between words into account. This paper proposes an information retrieval model based on Markov cliques, we used the extraction algorithm based on clique extraction method, the clique describes the deep-rooted relations between the terms. Moreover, we used the Markov network model to further express the indirect relationship between words.In this paper, we first build a Markov network model, which is constructed by calculating the similarity between terms, and strengthens the description of relationship between candidate terms. Then, we extracted Cliques from Markov network. Finally, we merged candidate terms and query terms in the clique for query expansion. Experiments on five data sets show that: query expansion based on Markov random walk matrix outperforms than that query expansion based on the similarity matrix; query expansion based on clique extraction method outperforms query expansion based on the general extraction method.