An Extractive and Abstractive Approach of Text Summarization

Text summarization, a difficult problem in natural language processing, tries to reduce lengthy materials into summaries that are clear and instructive. The “Extract-then-Abstract” approach for unsupervised abstractive summarization is a novel method introduced in this paper. This approach aims to provide more cohesive and human-like summaries as opposed to previous methods, which have trouble with seamless transitions between phrases. The method was designed to mimic how people generate summaries by first isolating the key lines, then fluently reassembling them. The proposed approach consists of two stages. In the extraction stage, TextRank, an unsupervised graph-based algorithm for extractive summarization, is utilized to generate a summary by selecting important sentences from the document. In order to provide a structured and fluid abstractive summary, a BART-based sentence-writing paradigm is introduced. When compared to unsupervised extractive summary techniques, the Extract-then-Abstract method produces summaries with greater coherence and resemblance to human comprehension while still maintaining competitive ROUGE scores.