A proximity language model for information retrieval

The proximity of query terms in a document is a very important information to enable ranking models go beyond the "bag of word" assumption in information retrieval. This paper studies the integration of term proximity information into the unigram language modeling. A new proximity language model (PLM) is proposed which views query terms' proximity centrality as the Dirichlet hyper-parameter that weights the parameters of the unigram document language model. Several forms of proximity measure are developed to be used in PLM which could compute a query term's proximate centrality in a specific document. In experiments, the proximity language model is compared with the basic language model and previous works that combine the proximity information with language model using linear score combination. The experiment results show that the proposed model performs better in both top precision and average precision.