An enhanced LSA-based approach for update summarization

Abstract

Update summarization is a challenge in automatic text summarization. The task aims to distill evolved messages from a collection of new articles, under the assumption that the reader has already browsed the previous articles. In this paper, we reviewed some state-of-the-art approaches for extracting update summarization and then focused on a LSA-based one. After the analysis of LSA-based approach's framework, we improved the approach by enhancing the approach's performance in accuracy. First, we utilized TOPIC SIGNATURE algorithm to extract the terms' novel information and incorporated the information to the process of evaluating topic's novelty score, which makes the evaluation more accuracy. Second, we excluded the least novel and important topics when generating summary, which helps improving the quality of the summary. The evaluation result on the update summarization task of Text Analysis Conference (TAC) 2008 indicates the validity of our modification.