Content Selection Operators for Multidocument Summarization Based on Cross-Document Structure Theory

Abstract

This paper aims at presenting an analysis of content selection techniques for multidocument summarization based on the multidocument discourse theory CST (Cross-document Structure Theory). We approach the task of content selection by using CST-based operators and focus specifically on redundancy treatment, which is an important and pervasive problem in multidocument summarization. Our experiments with Brazilian Portuguese news texts show that CST improves summaries quality by exploring relations among texts. Particularly, redundancy is reduced by identifying common information among texts, especially when compression rate is low.