Tag
This paper presents a framework for opinion summarization using LLMs that combines multidimensional classification and stratified sampling to reduce token usage while preserving semantic diversity and balance across viewpoints.