CoCoA: Collaborative Chain-of-Agents for Parametric-Retrieved Knowledge Synergy in Retrieval-Augmented Generation

ACL ARR 2025 May Submission176 Authors

Abstract:

Retrieval-Augmented Generation (RAG) has emerged as a promising framework for enhancing the capabilities of Large Language Models (LLMs), especially in knowledge-intensive tasks.
Despite its advantages, current RAG methods often struggle to *fully exploit knowledge during generation*. In particular, the synergy between the model’s internal parametric knowledge and external retrieved knowledge remains limited. Retrieved contents may sometimes mislead generation, while certain generated content can guide the model toward more accurate outputs.
In this work, we propose **Co**llaborative **C**hain-**o**f-**A**gents, a framework designed to enhance synergy over both parametric and retrieved knowledge.
Specifically, we first introduce **CoCoA-zero**, a training-free multi-agent RAG framework that first performs knowledge induction and then generates answers.
Further, we develop a long-chain training strategy for **CoCoA**,
which synthesizes long trajectories from the CoCoA-zero framework to fine-tune LLMs, improving their ability to explicitly integrate and collaboratively leverage internal and external knowledge.
Experimental results demonstrate the superiority of CoCoA in open-domain QA and multi-hop QA.
Our code will be available on GitHub.

Paper Type: Long

Research Area: Question Answering

Research Area Keywords: open-domain QA, retrieval-augmented generation, fine-tuning, reasoning

Contribution Types: NLP engineering experiment, Publicly available software and/or pre-trained models

Languages Studied: English

Submission Number: 176