Efficient Dynamic Clustering-Based Document Compression for Retrieval-Augmented Generation
Published in Findings of EMNLP 2025, 2025
This paper presents EDC²-RAG, a retrieval-augmented generation framework that improves LLM inference by dynamically clustering retrieved documents to remove noise, redundancy, and irrelevant content. Built on GPT-3.5, EDC²-RAG demonstrates robust performance across knowledge-QA and hallucination detection tasks.
Recommended citation: Weitao Li, Xiangyu Zhang, Kaiming Liu, Xuanyu Lei, Weizhi Ma, Yang Liu, "Efficient Dynamic Clustering-Based Document Compression for Retrieval-Augmented Generation", Findings of EMNLP 2025
Download Paper