TOPIC BOUNDARY DETERMINATION METHOD AND DEVICE, AND TOPIC BOUNDARY DETERMINATION PROGRAM

摘要

PPROBLEM TO BE SOLVED: To improve the precision of detection of both a small topic section and a large topic section. PSOLUTION: In this method, a text is morphologically analyzed and decomposed to words, and a concept base storing vectors of the words is retrieved to acquire a vector corresponding to each word. The center of gravity of word vectors corresponding to at least two or more words constituting the text, which are decomposed in every section, and the distance between the word vectors are calculated to calculate the cost value consisting of the word vectors and distance of every section. The cost value is stored in a storage means, and a combination of sections minimizing the cost value is specified from all combinations of sections. PCOPYRIGHT: (C)2004,JPO&NCIPI