Improving Multi-document Text Summarization Performance using Local and Global Trimming

Abstract

Multi-document summarization can produce a condensed representation of the contents of multiple related text documents. With this summarization facility, web users can judge rapidly the relevance of a group of documents returned by the search engines and decide whether those should be discarded. This reduces the total search cost for the users. This paper presents a multi-document summarization system, which has two components: (1) the sentence extraction component that produces draft summaries by sentence extraction and (2) the sentence-trimming component that eliminates the low content and redundant elements from the sentences in the draft summaries for improving the summarization performance. In this paper, we also introduced several new local and global sentence-trimming rules. Our experiment on DUC 2004 data set shows that the local and global trimming can improve the extractive multi-document summarization performance in many cases.