A Document Graph Based Query Focused Multi-Document Summarizer

This paper explores the research issue and methodology of a query focused multidocument summarizer. Considering its possible application area is Web, the computation is clearly divided into offline and online tasks. At initial preprocessing stage an offline document graph is constructed, where the nodes are basically paragraphs of the documents and edge scores are defined as the correlation measure between the nodes. At query time, given a set of keywords, each node is assigned a query dependent score, the initial graph is expanded and keyword search is performed over the graph to find a spanning tree identifying relevant nodes satisfying the keywords. Paragraph ordering of the output summary is taken care of so that the output looks coherent. Although all the examples, shown in this paper are based on English language, we show that our system is useful in generating query dependent summarization for nonEnglish languages also. We also present the evaluation of the system.