3M:Multi-document Summarization Considering Main and Minor Relationship

Anonymous

Abstract: The multi-document summary task is an important branch of the information aggregation task. Compared with the single-document summary, the input of multi-document summary is much longer and the logic is more complicated. This article proposes a hypothesis: taking the content of a document as the main body and the content of other documents as auxiliary information, a summary that combines all the information in the document collection can be generated. Based on this assumption, the multi-document summarization task can select one main document, and then combine the information of other documents for summary generation. This paper combines CopyTransformer and the Maximal Marginal Relevance (MMR) to design Multi-document summarization considering Main and Minor relationship model(3M). Empirical results on the Multi-News and DUC 2004 dataset show that the 3M brings substantial improvements over several strong baselines, manual evaluation shows that the generated abstract is fluent and can better express the content of the main document. In addition, by selecting different main documents, 3M can generate multiple abstracts with different styles for a set of documents.