Language Modeling (LM) has been successfully applied toInformation Retrieval (IR). However, most of the existing LMapproaches only rely on term occurrences in documents, queriesand document collections. In traditional unigram based models,terms (or words) are usually considered to be independent. Insome recent studies, dependence models have been proposed toincorporate term relationships into LM, so that links can becreated between words in the same sentence, and termrelationships (e.g. synonymy) can be used to expand thedocument model. In this study, we further extend this family ofdependence models in the following two ways: (1) Termrelationships are used to expand query model instead of documentmodel, so that query expansion process can be naturallyimplemented; (2) We exploit more sophisticated inferentialrelationships extracted with Information Flow (IF). Informationflow relationships are not simply pairwise term relationships asthose used in previous studies, but are between a set of terms andanother term. They allow for context-dependent query expansion.Our experiments conducted on TREC collections show that wecan obtain large and significant improvements with our approach.This study shows that LM is an appropriate framework toimplement effective query expansion.
展开▼
