Enhancements to Graph based methods for Multi Document Summarization

Journal of Applied Computer Science & Mathematics

关键词: Page Rank;    Lexical Rank;    Sentence Rank;    Recommendation;    Degree;    Damping;    Threshold;    Effectiveness;    Discounting;   

DOI  :  

来源: DOAJ

【 摘 要 】

This paper focuses its attention on extractivesummarization using popular graph based approaches. Graphbased methods can be broadly classified into two categories:non- PageRank type and PageRank type methods. Of themethods already proposed - the Centrality Degree methodbelongs to the former category while LexRank and ContinuousLexRank methods belong to later category. The paper goes on tosuggest two enhancements to both PageRank type and non-PageRank type methods. The first modification is that ofrecursively discounting the selected sentences, i.e. if a sentence isselected it is removed from further consideration and the nextsentence is selected based upon the contributions of theremaining sentences only. Next the paper suggests a method ofincorporating position weight to these schemes. In all 14methods –six of non- PageRank type and eight of PageRanktype have been investigated. To clearly distinguish betweenvarious schemes, we call the methods of incorporatingdiscounting and position weight enhancements over LexicalRank schemes as Sentence Rank (SR) methods. Intrinsicevaluation of all the 14 graph based methods were done usingconventional Precision metric and metrics earlier proposed byus - Effectiveness1 (E1) and Effectiveness2 (E2). Experimentalstudy brings out that the proposed SR methods are superior toall the other methods.

【 授权许可】

Unknown