Comparison of Global Term Expansion Methods for Text Retrieval

This paper describes our work at the fifth NTCIR workshop on the subtasks of single language information retrieval (SLIR). Several automatic global query expansion strategies were explored based on a machine-derive thesaurus. These term selection strategies were compared with manual selection and local expansion. Experiments show that all the global expansion strategies perform worse than the simple local expansion. Furthermore, even with the help of a human in selecting the global terms, the performance may not be better than an automatic local feedback method, if the human does not fully understand the information need of the search topic. However, the machine-derived thesaurus does extract relevant global terms for expansion. Proper term selection can yield great improvement in performance than local feedback alone. Keywords: Chinese IR, term association, global expansion