Published Paper Details:

Text summarization is one of the basic tasks in natural language processing, which has been vastly developed over the past couple of years, with huge growth influenced by large language models. This paper focuses on a comparative analysis of five state-of-the-art LLM models for text summarization: GPT-3, T5, MPT-7B-instruct, FALCON-7B-instruct, and OpenAI ChatGPT. The models evaluate on CNN/Daily Mail and XSum datasets utilizing ROUGE-1, ROUGE-2, ROUGE-L, BLEU, and F1 score to measure the competence of generated coherent and meaningful summaries and the preservation of the meaning of the origin text. Results demonstrate that GPT-3 scores the best on most of the metrics, while having strong language understanding and generation capabilities. T5 variants: The results show that the T5-large variant performs better than the T5-base variant. This confirms the benefits of scaling up model size. Instruction-tuned models MPT-7B-instruct and FALCON-7B-instruct have slightly inferior performance compared to other approaches, which implies that general pre-training of GPT-3 and T5 could be more effective than task-specific fine-tuning. Interestingly, the recently developed ChatGPT also delivers competitive results, underscoring the continuous progress in text summarization. The paper ends with suggesting the possible research avenues; first, development of the hybrid models by integrating benefits from different approaches, study on multi-lingual summarization, domain-agnostic summarization, and then extensive evaluation framework. Such contribution is essential to take one step ahead in terms of text simplification technology advancement and application for facilitating accessible information across various diverse users.