Sentence Similarity Estimation for Text Summarization Using Deep Learning

Abstract

One of the key challenges of natural language processing (NLP) is to identify the meaning of any text. Text summarization is one of the most challenging applications in the field of NLP where appropriate analysis is needed of given input text. Identifying the degree of relationship among input sentences will help to reduce the inclusion of insignificant sentences in summarized text. Result of summarized text always may not identify by optimal functions, rather a better summarized result could be found by measuring sentence similarities. The current sentence similarity measuring methods only find out the similarity between words and sentences. These methods state only syntactic information of every sentence. There are two major problems to identify similarities between sentences. These problems were never addressed by previous strategies provided the ultimate meaning of the sentence and added the word order, approximately. In this paper, the main objective was tried to measure sentence similarities, which will help to summarize text of any language, but we considered English and Bengali here. Our proposed methods were extensively tested by using several English and Bengali texts, collected from several online news portals, blogs, etc. In all cases, the proposed sentence similarity measures mentioned here was proven effective and satisfactory.