Global and Local Term Expansion for Text Retrieval

This paper describes our work at the fourth NTCIR workshop on the subtasks of monolingual information retrieval (IR). Global and local query expansions were explored. For global query expansion, co-occurred terms accumulated across the entire collection were selected and added to the initial query. For local query expansion, a method of blind relevance feedback (BRF) was implemented. Our experiments verified that BRF is effective and can be easily implemented without much parameter tuning. If best term selection can be achieved, global query expansion based on co-occurred terms can perform similarly well and combining both local and global expansion can outperform each method alone.