Position Debiasing Fine-Tuning for Causal Perception in Long-Term Dialogue

TL;DR

This paper introduces a novel fine-tuning framework called CPD that reduces position bias in long-term dialogue generation by enhancing causal perception, leading to more relevant and context-aware responses.

Contribution

The paper proposes a perturbation-based causal variable discovery method and a local-position awareness technique to improve causal perception during fine-tuning of large language models.

Findings

Effective in alleviating position bias in multiple LLMs

Significant improvements over baseline models

Enhanced relevance and causality in generated responses

Abstract

The core of the dialogue system is to generate relevant, informative, and human-like responses based on extensive dialogue history. Recently, dialogue generation domain has seen mainstream adoption of large language models (LLMs), due to its powerful capability in generating utterances. However, there is a natural deficiency for such models, that is, inherent position bias, which may lead them to pay more attention to the nearby utterances instead of causally relevant ones, resulting in generating irrelevant and generic responses in long-term dialogue. To alleviate such problem, in this paper, we propose a novel method, named Causal Perception long-term Dialogue framework (CPD), which employs perturbation-based causal variable discovery method to extract casually relevant utterances from the dialogue history and enhances model causal perception during fine-tuning. Specifically, a local-position awareness method is proposed in CPD for inter-sentence position correlation elimination, which helps models extract causally relevant utterances based on perturbations. Then, a casual-perception fine-tuning strategy is also proposed, to enhance the capability of discovering the causal invariant factors, by differently perturbing causally relevant and non-casually relevant ones for response generation. Experimental results on two datasets prove that our proposed method can effectively alleviate the position bias for multiple LLMs and achieve significant progress compared with existing baselines.