Feature based Extractive Text Summarization for News Documents

Abstract

Abstract newlineNow days, the rate of growth of data is expanding exponentially and World- newlineWide Web has become one of the largest data and information repository newlinein the world. Even though the large amount of data is available, to access newlinethe required data in appropriate form at right time is a big challenge. Thus newlineResearch and Development in automatic text summarization is become more newlinesignificant. The aim of text summarization process is to take one or more newlinesource documents as input and present most salient information in condensed newlineform in a manner which can meet the user s interest and requirement. newlineThe objective of this thesis is to develop extractive text summarization newlinesystem for single and multi-document. The proposed approach uses shallow newlinesentence features and fuzzy logic system for the aforementioned work. In newlineliterature various methods have been tried to use different features to carry newlineout the summarization task. Based on the feature score of sentences, high newlinescoring sentences are included in summary. However, it is not possible to newlineclaim that any of features used in summarization yields the best coverage for newlineall genres. There are no strict rules, but there are clues that could be exploited newlineto identify important topics and ideas. Initially our research work investigates newlinedifferent clues by implementing and assessing various features and identifies newlinebest features which can yield the best result for news domain. To sum-up, the newlineobjective of this assessment is to identify the best feature set which can obtain newlinegood content coverage with minimum time. newlineNext two phases of research work, utilizes the fuzzy model to deal with newlinethe imprecise and uncertainty of feature weight for single and multi-document newlinesummarization. After pre-processing and feature extraction, feature vector newlinematrix is given as input to the fuzzy logic system. The rule based decision newlinemodule is used to provide the balance weight to the extracted features and newlineupdate the feature vector matrix. We used cosine similarity measure to calculate newlinethe similarity between senten