Assessing BigBirdPegasus and BART Performance in Text Summarization: Identifying Right Methods

In the past few years, automatic text summarization has risen as a key instrument for shrinking voluminous documents. This research study scrutinized the efficacy of the BART Large and BigBirdPegasus language models in generating abstractive summaries for biomedical scientific articles sourced from the PLOS and eLife databases. This study has employed three summarization methods and evaluated the ensuing summaries utilizing ROUGE, BLEU, and perplexity metrics. The findings demonstrated that direct summaries from articles secured the highest ROUGE and BLEU scores, while summaries derived from titles and keywords proved to be more succinct. BigBirdPegasus produced more intelligible summaries, reflected in lower perplexity scores, whereas BART Large offered summaries with superior ROUGE and BLEU scores, closely mirroring human summaries. This study emphasizes the importance of choosing an appropriate summarization technique, taking into account the particular needs and goals of a given application.