Novel Hybrid Methods for Journal Article Summarization Combining Graph Method and Rough Set TFIDF Method with Pegasus Model

Journal article summarization shortens an article, including all the relevant and significant topics. As the number of journal articles published every year is increasing rapidly, the significance of research on journal article summarization also increases. This work generates a summary of journal articles from the ScisummNet dataset using extractive, abstractive and novel hybrid methods. The summary generated by these methods is then analyzed using word overlap and semantic similarity scores to find the best summary. The precision, recall and f-measure scores of rouge-1, rouge-2, rouge-L and rouge-Lsum scores are analyzed to find the word overlap score and precision, recall and f-measure of the BERT score to find the semantic similarity score. The summary generated by the novel hybrid methods combining the extractive methods like the graph with the sum of weights, the graph with bushy path and the rough set TFIDF methods with the abstractive Pegasus fine-tuned model gives the best rouge and BERT scores among the other hybrid methods.