Improving Query Expansion Using Wikipedia

Abstract

Query expansion is one of important technologies used to improve retrieval efficiency. Many studies focus on query expansion with relationships between terms only extracted from the single local domain corpus. In fact, because the single local domain corpus is relatively small, there exist many no-landing terms which have no candidates for query expansion resulting in low retrieval performance. Therefore, to address such problem, relationships between terms captured from Wikipedia are superimposed to the basic Markov network that pre-built using the local domain corpus. A new larger Markov network is formed with more and richer relationships for each term. A graph mining technology, clique, is implemented to measure inter-relationships in Markov network for query expansion. The proposed techniques of superimposed Markov network and clique-based query expansion are benefit to improve precision and recall of information retrieval and to reduce the risk of topic drift.