Single Language Information Retrieval at NTCIR-5

We participated in the Single Language (Japanese) Information Retrieval (IR), or SLIR, subtask of the Cross-Lingual Information Retrieval (CLIR) task at NTCIR-5 to verify the practical effectiveness of a two-phase IR system that we proposed for visualizing IR while at the same time improving its precision. Al-though the proposed system performed well in rela-tively small computer experiments, the results of the NTCIR-5 task were not as good as we expected, re-vealing the need for more work in adaptating our sys-tem for large-scale practical IR tasks. This paper describes the details of our systems submitted to the NTCIR-5 task and their experimental results, and pro-poses three solutions to the scaling problem that al-low the proposed system to be effectively adapted for practical IR tasks. Very beginning additional experi-ments based on the solutions showed that our system has a comparable performance when using automatic method, and possibly a much higher precision when using a semi-automatic method, compared to the con-ventional TFIDF-based method.