UC Irvine

This thesis evaluates 16 meeting summarization approaches, including commercial tools and large language models (LLMs), across 45 software maintenance meeting topics. Assessing performance on truthfulness, essence, length, structure, and clarity, the results reveal significant variability among approaches. LLMs demonstrated more consistent performance than meeting tools, yet no single approach excelled in all metrics, highlighting inherent trade-offs in summarization. Notably, prompt design significantly impacted LLM performance. The study provides insights for tool selection and directions for future summarization approaches.