Bilateral Personalized Dialogue Generation with Contrastive Learning

Abstract

Generating personalized responses is one of the major challenges in natural human-robot interaction. Current researches in this field mainly focus on generating responses consistent with the robot’s pre-assigned persona, while ignoring the user’s persona. Such responses may be inappropriate or even offensive, which may lead to the bad user experience. Therefore, we propose a Bilateral Personalized Dialogue Generation (BPDG) method for dyadic conversation, which integrates user and robot personas into dialogue generation via designing a dynamic persona-aware fusion method. To bridge the gap between the learning objective function and evaluation metrics, the Conditional Mutual Information Maximum (CMIM) criterion is adopted with contrastive learning to select the proper response from the generated candidates. Moreover, a bilateral persona accuracy metric is designed to measure the degree of bilateral personalization. Experimental results demonstrate that, compared with several state-of-the-art methods, the final results of the proposed method are more personalized and consistent with bilateral personas in terms of both automatic and manual evaluations.

Contrastive Learning Reduces Hallucination in Conversations

“…On the task of neural text generation, CL (Jiang et al 2022), a.k.a. unlikelihood training (Welleck et al 2020 or negative training (He and Glass 2020), alleviates undesirable properties of the generated output, e.g., repetition (Shirai et al 2020;Jiang et al 2022), maliciousness (He and Glass 2020), dullness (Li et al 2020b(Li et al , 2022, or inconsistency (Li et al 2020a). Moreover, Cao and Wang (2021) propose a sentence level contrastive learning method to reduce the hallucinations of text summarization model.…”