Automatic Generation of Term Descriptions by Web-based Multi-Document Summarization

We developed a Web search system called " Cyclone " , which extracts high quality term descriptions and helps users to obtain encyclopedic knowledge efficiently. In the current implementation , multiple paragraph-style descriptions extracted from different Web pages are presented in response to a user keyword. However, to obtain sufficient information for a single keyword, a user usually has to read the similar or redundant content included in multiple paragraphs. To solve this problem, in this paper we propose a multi-document summarization method that uses multiple paragraphs to generate a single description for a keyword. Our method analyzes source description paragraphs with respect to viewpoints of explanation (e.g., definitions and purposes) and generates a summary by integrating the representative sentences for each viewpoint. We also show the effectiveness of our method by means of experiments.