Intelligent Text Summarization Using a Fuzzy-Driven BART Architecture

Abstract

Every day there is huge amount of data on the internet on a specific subject, leading to the availability of massive amounts of data which includes relevant and irrelevant search. Therefore, it is necessary to summarize the text data in order to understand its key idea in limited time. This research proposes a new hybrid text summarization framework utilizing fuzzy logic and BART model. In order to improve the quality, the proposed framework is used in conjunction with extractive and abstractive summarization methods. In extractive method fuzzy logic utilizes fuzzy rules to create sentence embedding and for abstractive summarization method BART utilizes sequence-to-sequence method in which the encoder is bidirectional BERT and the decoder is autoregressive (GPT-like). To generate an intermediate summary each sentence is scored based on relevance and novelty metric. The sentences with the highest scores are selected to be the part of summary which is used as input to abstractive method. In order to evaluate the efficiency of the proposed model, extensive experiments on benchmark datasets are performed- for abstractive summarization CNN/Daily Mail, DUC-2004 and extractive summarization DUC-2006 and DUC-2007 are used for conducting the experiments and recall oriented understudy for Gisting Evaluation (ROUGE) is used for evaluation. The results indicates that the proposed model is capable of performing efficient text summarization and generate better results than other state-of-art models.