Enhanced web-based summary generation for search.

Automatic summarization;Search engines;Graph theory;Social media

Brent Wenerstrom, 1980-

University:University of Louisville

Department:Computer Engineering and Computer Science

关键词: Automatic summarization;    Search engines;    Graph theory;    Social media;   

Others  :  https://ir.library.louisville.edu/cgi/viewcontent.cgi?article=2546&context=etd

美国|英语

来源: The Universite of Louisville's Institutional Repository

【 摘 要 】

Afterauser types ina search query on amajor search engine, they are presented with a number of search results.Each search result is made up of a title,brief text summary and aURL. It isthen the user's job to selectdocuments for further review.Our researchaims to improve the accuracy of users selecting relevantdocuments by improving the waythese webpages are summarized.Improvements inaccuracy willlead to time improvements and user experience improvements. Wepropose ReClose,a system forgenerating webdocument summaries.ReClose generates summary content through combining summarization techniques fromquery-biasedand query-independent summary generation.Query-biased summaries generallyprovide queryterms in context.Query-independent summaries focuson summarizing documents as awhole.Combining these summary techniques ledto a10% improvement inuser decision making over Google generated summaries. Color-coded ReClose summaries providekeywordusage depth at aglance and also alert usersto topic departures.Color-coding further enhanced ReClose resultsand led to a20% improvement inuser decision making over Google generated summaries. Manyonline documents include structure and multimedia of various forms suchas tables, lists, formsand images.Wepropose to include this structure in webpage summaries.We found that the expertuser was insignificantly slowedin decision making whilethe majority of average users made decisions more quicklyusing summaries including structure withoutany decrease in decisionaccuracy. Weadditionally extended ReClose foruse in summarizing large numbers of tweets in tracking fluoutbreaks in social media.The resulting summaries have variable length and are effective at summarizing flu related trends.Users of the system obtained an accuracy of 0.86 labeling multi-tweet summaries.This showed that thebasis of ReCloseiseffectiveoutside of web documents and that variable length summaries canbe more effectivethan fixedlength. Overall the ReClose system provides unique summaries that contain more informative contentthan current search enginesproduce, highlightthe results ina more meaningful way,and add structure when meaningful.The applications of ReCloseextend far beyond searchand have been demonstrated in summarizing pools of tweets.