Focusing on the IPC (International Patent Classification) of patent, two retrieval methods were examined. One is to use the IPC code of the query patent and the other is to exploit the codes assigned to top-N retrieved patents in a similar manner of pseudorelevance feedback. In both methods, the codes were used as constraint on retrieval results. As a result, we found the former is clearly effective and the latter's effect is positive but small. Another point of our experiments is exploitation

more »

... f synonyms, which were automatically collected from a machine-readable dictionary for each query term.