Automatic summarization of Japanese sentences and its application to a WWW KWIC index

This paper presents a system which creates a KWIC index of WWW texts in Japanese by automatic summarization. The system consists of three modules: a WWW spider, an extractor of important sentences, and a sentence summarizer. The most effective module is the last one which employs a robust and fairly accurate Japanese parser: KNP. It segments an input sentence into phrases or simple sentences and assembles a summary. The accuracy of the important sentence extractor was 62.8% and that of the sentence summarizer was 76.5%