Multi-Document Summarization: Methodologies and Evaluations

Abstract

This paper describes a system for the summarization of multiple documents. The system produces multi-document summaries using clustering techniques to identify common themes across the set of documents. For each theme, the system identifies representative passages that are included in the final summary. We also describe a methodology for evaluation of our system which is based upon a question answering task. Results of our evaluation are also presented. 1. Introduction Multi-document summarization differs greatly from single-document summarization. In fact, single-document summarization can be considered as one of the critical sub-tasks of multi-document summarization. However, there exist several other important sub-tasks including identification of important common themes or aspects across the documents, selecting representative summaries for each of these themes, and organizing the representative summaries for the final summary. In addition to these sub-tasks, an important aspect ...