A Query Specific Graph Based Approach to Multi-document Text Summarization: Simultaneous Cluster and Sentence Ranking

Abstract

Recently the focus of query independent summary is shifted to query specific document summarization. This paper presents a graph based method to find query specific multi-document summarization. Our system is divided into two stages, off-line and on-line. We construct document as graph by considering paragraph as nodes in off-line stage. Edge scores are represented node similarities. In online stage, query specific weight are calculated and assigned to node. We then perform keyword search on the document graph and search a minimum top spanning tree for finding relevant nodes that satisfy the keyword search. Resultant summary looks coherent due to simultaneous cluster and sentence ranking. Experimental results for multi-document scenarios are encouraging.