Multi-document Summarization using Tensor Decomposition

Abstract

The problem of extractive text summarizationfor a collection of documents is defined as selecting asmall subset of sentences so the contents and meaningof the original document set are preserved in the bestpossible way. In this paper we present a new modelfor the problem of extractive summarization, where westrive to obtain a summary that preserves the informationcoverage as much as possible, when compared to theoriginal document set. We construct a new tensor-basedrepresentation that describes the given document setin terms of its topics. We then rank topics viaTensor Decomposition, and compile a summary from thesentences of the highest ranked topics.