Dehallucinating Parallel Context Extension for Retrieval-Augmented Generation

TL;DR

This paper introduces DePaC, a method to reduce hallucinations in retrieval-augmented generation by using context-aware negative training and information-calibrated aggregation, improving factual accuracy.

Contribution

DePaC is a novel approach that effectively mitigates hallucinations in RAG by addressing fact fabrication and omission through targeted training and aggregation strategies.

Findings

DePaC significantly reduces hallucinations in RAG tasks.

DePaC improves performance across nine RAG benchmarks.

The method effectively guides LLMs to avoid unsupported claims.

Abstract

Large language models (LLMs) are susceptible to generating hallucinated information, despite the integration of retrieval-augmented generation (RAG). Parallel context extension (PCE) is a line of research attempting to effectively integrating parallel (unordered) contexts, while it still suffers from hallucinations when adapted to RAG scenarios. In this paper, we propose DePaC (Dehallucinating Parallel Context Extension), which alleviates the hallucination problem with context-aware negative training and information-calibrated aggregation. DePaC is designed to alleviate two types of in-context hallucination: fact fabrication (i.e., LLMs present claims that are not supported by the contexts) and fact omission (i.e., LLMs fail to present claims that can be supported by the contexts). Specifically, (1) for fact fabrication, we apply the context-aware negative training that fine-tunes the LLMs with negative supervisions, thus explicitly guiding the LLMs to refuse to answer when contexts are not related to questions; (2) for fact omission, we propose the information-calibrated aggregation which prioritizes context windows with higher information increment from their contexts. The experimental results on nine RAG tasks demonstrate that DePaC significantly alleviates the two types of hallucination and consistently achieves better performances on these tasks.