Diversity based text summarization

Abstract

Diversity of selected sentences is an important factor in automatic text summarization to control redundancy in the summarized text. In paper, we propose a method called maximal marginal importance (MMI) for text summarization based on the idea of the well-known diversity approach maximal marginal relevance (MMR) where an emphasis is on the diversity based binary tree is used to exploit the diversity among the document sentences, where the whole document is clustered into a number of clusters, and then each cluster is presented as one binary tree or more. In our method, the sentence is evaluated based on its importance and its relevance. Our experimental results shown that the proposed method outperforms the three benchmark methods used in this study.