CoRank: A clustering cum graph ranking approach for extractive summarization

TL;DR

CoRank introduces a two-stage extractive summarization method combining novel clustering and graph ranking to improve content coverage and diversity, validated on standard datasets.

Contribution

The paper presents a new two-stage summarization approach that integrates clustering with graph ranking for better content coverage and diversity.

Findings

Achieves improved coverage and diversity in summaries.

Validated on DUC2001 and DUC2002 datasets.

Outperforms some existing summarization techniques.

Abstract

Online information has increased tremendously in today's age of Internet. As a result, the need has arose to extract relevant content from the plethora of available information. Researchers are widely using automatic text summarization techniques for extracting useful and relevant information from voluminous available information, it also enables users to obtain valuable knowledge in a limited period of time with minimal effort. The summary obtained from the automatic text summarization often faces the issues of diversity and information coverage. Promising results are obtained for automatic text summarization by the introduction of new techniques based on graph ranking of sentences, clustering, and optimization. This research work proposes CoRank, a two-stage sentence selection model involving clustering and then ranking of sentences. The initial stage involves clustering of sentences using a novel clustering algorithm, and later selection of salient sentences using CoRank algorithm. The approach aims to cover two objectives: maximum coverage and diversity, which is achieved by the extraction of main topics and sub-topics from the original text. The performance of the CoRank is validated on DUC2001 and DUC 2002 data sets.