Comparative Analysis of Hindi Text Summarization for Multiple Documents by Padding of Ancillary Features

Abstract

There is an enormous amount of textual material, and it is only growing every single day. The data available on Internet comprised of Web pages, news articles, status updates, blogs which are unstructured. There is a great need to reduce much of these text data to shorter, focused summaries that capture the salient details so that the user can navigate it more effectively as well as check whether the larger documents contain the information that we are looking for. Text summary is generating a shorter version of the original text. The need of summarization arises because every time it is not possible to read the detailed document due to lack of time. Automatic text summarization methods are greatly needed to address the ever-growing amount of text data available online both to better help discover relevant information and to consume relevant information faster. To address the issue of time constraint, an extractive text summarization technique has been proposed in this research work which selects important sentences from a text document to get a gist of information contained in it. A fuzzy technique has been used to generate extractive summary from multiple documents by using eight and eleven feature sets. The eleven feature set combines the existing eight features (term frequency-inverse sentence, length of sentence in the document, location of sentence in document, similarity between sentences, numerical data, title overlap, subject object verb (SOV) qualifier, lexical similarity) and three ancillary features (proper nouns, hindi cue phrase, thematic words). It was seen that applying fuzzy technique with eleven features gave better results for summarization than the same using eight features. The precision increases in the range of 3–5% for different datasets. Datasets used were Hindi news articles from online sources.