Sentence Similarity Measurement for Bengali Abstractive Text Summarization

Abstract

Text summarization is a massive research area in natural language processing. It reduces the larger text and provided the prime meaning of a text document. Find the meaning of the larger text needed of a proper text analysis which gives a better text summarizer. Abstractive text summarizer gives a summary which can present or not present in the text document. The machine produces a text summary after learning from the human given summary. Sentence similarity is a way to judge a better text summarizer. It is exploring the similarity between sentences or words. This paper we discuss several methods of sentence similarity and proposed a method for identifying a better Bengali abstractive text summarizer. We used human given summary and machine response summary sentences for similarity measurement where both sentences contain a Bengali short text. There are several approaches to English sentences similarity measurement, and we applied some of the approaches for similarity measure for our Bengali text which give a satisfying result. For our given methods we collect data from online and social media and create a summary of those texts. After creating a summary pre-processing this text and generate a summary from our abstractive text summarization model. All summary sentence similarity measurement cases using the method provided an effective value and optimal result.