A Comprehensive Review of Large Language Models in Abstractive Summarization of News Articles

Recent years have seen significant advancements in Large Language Models (LLMs), particularly in their application to natural language processing tasks. One prominent application is the abstractive summarization of news articles, a process that distills extensive textual information into a succinct yet comprehensive representation. The significance of this task lies in its capacity to efficiently distill key information. This paper focuses on evaluating the performance of various LLMs—Llama-2-7b, T5, Bart, and Pegasus—specifically fine-tuned on the CNN Daily dataset of news articles. QLoRA and Prompt-Based Fine-Tuning Approaches are employed systematically to compare the efficacy of these models in generating summaries that capture the essence of news articles. Evaluation metrics, including ROUGE, BLEU, and cosine similarity, are utilized to assess summarization quality. The findings reveal distinctive results among the selected LLMs such that Llama-2 excels in ROUGE-1 and ROUGE-L with scoreS of 0.45 and 0.427, Pegasus demonstrates remarkable performance in ROUGE-2 and cosine similarity with scores 0.19 and 0.6 respectively.