Deciphering Email Threads: A Comparative Study of Large Language Models

Email overload is a major challenge in modern communication, preventing efficient and effective information processing. This paper aims to harness state-of-the-art NLP techniques to improve the understandability of lengthy email conversations by delivering concise summaries. This paper aims to comprehensively compare and analyze the performance of different LLMs in the context of email-trail summarization. We compared human-generated summaries with those generated by GPT-3, Google’s Bard, and LLaMA2, employing evaluation metrics like Rouge score, BERTScore, to assess their strengths and weaknesses for Email-trail summarization. Our results show that summaries generated by GPT-3 are closer to human-generated summaries, followed by those from Bard and LLaMA2.