This study aims to find the appropriate form of summarization for RAPTOR method, which is a RAG (Retrieval Augmented Generation) based framework that uses document summarization in the information retrieval process. We categorized summaries into three...
This study aims to find the appropriate form of summarization for RAPTOR method, which is a RAG (Retrieval Augmented Generation) based framework that uses document summarization in the information retrieval process. We categorized summaries into three types (KeyDetail, Short, and Headline) based on length and level of detail. And we tested the performance changes when applying each type of summary, showing that the current summarizing method of “including as much detail as possible” is not optimal.