Analysis of multi-document viewpoint summarization using multi-dimensional genres

Abstract

An interactive information retrieval system that pro-vides different types of summaries of retrieved docu-ments according to each user’s information needs can be effective for understanding the contents. The pur-pose of this study is to build a multi-document sum-marizer to produce summaries according to such view-points. As an exploratory stage of investigation, we ex-amined the effectiveness of genre for source documents to produce different types of summaries. Once a set of documents on a topic is provided to our summarization system, a list of topics discussed in the given document set is presented, so that the user can select a topic of in-terest from the list as well as the summary type, such as opinion-oriented, fact-reporting or knowledge-focused, according to their requirements. We assume a relation-ship between a summary type and human recognition of information types included in the source: a docu-ment genre. We also analyzed the results of the multi-document summarization using automatic genre classi-fication to reveal the association between genre dimen-sions and the summary types.