RankCoT: Refining Knowledge for Retrieval-Augmented Generation through Ranking Chain-of-Thoughts

TL;DR

RankCoT tackles the challenge of misused knowledge in retrieval augmented generation by unifying ranking with chain-of-thought based knowledge refinement and a self-reflection training loop. It prompts per-document CoT outputs, uses ground-truth based signals to rank refinements with Direct Preference Optimization, and refines CoTs through a self-reflection step to produce high-quality training data. Empirically, RankCoT improves over vanilla RAG and other refinement baselines across multiple datasets and model scales, while generating shorter yet more effective refinements. This approach enhances the LLM's ability to leverage external knowledge with reduced noise, enabling more accurate answers in knowledge-intensive tasks.

Abstract

Retrieval-Augmented Generation (RAG) enhances the performance of Large Language Models (LLMs) by incorporating external knowledge. However, LLMs still encounter challenges in effectively utilizing the knowledge from retrieved documents, often being misled by irrelevant or noisy information. To address this issue, we introduce RankCoT, a knowledge refinement method that incorporates reranking signals in generating CoT-based summarization for knowledge refinement based on given query and all retrieval documents. During training, RankCoT prompts the LLM to generate Chain-of-Thought (CoT) candidates based on the query and individual documents. It then fine-tunes the LLM to directly reproduce the best CoT from these candidate outputs based on all retrieved documents, which requires LLM to filter out irrelevant documents during generating CoT-style summarization. Additionally, RankCoT incorporates a self-reflection mechanism that further refines the CoT outputs, resulting in higher-quality training data. Our experiments demonstrate the effectiveness of RankCoT, showing its superior performance over other knowledge refinement models. Further analysis reveals that RankCoT can provide shorter but effective refinement results, enabling the generator to produce more accurate answers. All code and data are available at https://github.com/NEUIR/RankCoT.

Figures (8)

• Figure 1: Illustration of RankCoT. We present how knowledge refinement can be achieved by incorporating reranking into CoT-based summarization.
• Figure 2: Illustration of RankCoT.
• Figure 3: Quality of Refined Knowledge Generated by Different Models. In Figure \ref{['fig:textsimquery']}, we first estimate the text similarity between the query and the knowledge refinement results using the BGE model chen2024bge. Then, we calculate the hit rate of these knowledge refinement results in Figure \ref{['fig:groundtruthin']}, which evaluates whether the ground truth answers are included in the knowledge refinement results.
• Figure 4: The Length of Knowledge Refinement Results Produced by Different Models. We first present the average length of the refinement results in Figure \ref{['fig:inputlengthlog']}. Then, the length change ratio relative to vanilla LLMs is illustrated in Figure \ref{['fig:inputlengthratio']}.
• Figure 5: QA Consistency of the RAG Model Using Different Knowledge Refinement Models.