| 摘要 |
PROBLEM TO BE SOLVED: To optimize sentence extracting rules in consideration of the correlation between the rules and a training text. SOLUTION: Regulation candidates in an information extracting rule candidate storing part 21 are applied to a text in an adapting text storing part 20 to extract a sentence in an information extracting part 22. Based on the extracted information, the performances of the rules are determined. The performance is indicated as the number of correct and wrong sentences and the like. A clustering part 27 processes the rules extracting one or more correct sentences in the ascending order of the number of the correct sentences, arranges, as nodes, other rules including all the correct sentences as its own correct sentences to construct a cluster tree. The clustering part 27 inspects the cluster tree in favor of depth to select the rules. COPYRIGHT: (C)2004,JPO&NCIPI
|
