A Multi-Document Summarization Approach Based on Extracts Classification

摘要

We present in this paper an original approach of automatic multi-document summarization based on a classification process. This process operates on a multitude of summaries generated from the source documents in order to choose the best summary. The classification process uses a set of criteria which aim to approve the quality of the summary. These criteria were determined by a learning step based on human written model summaries. Furthermore, the proposed approach allows the extract revision in order to improve its structural quality. We present in this paper the basic principles of our approach, as well as the details of its implementation and evaluation applied for the corpora disseminated in DUC'04 and DUC'07 conferences.