Bilateral Personalized Dialogue Generation with Contrastive Learning

TL;DR

This work tackles the problem of generating responses that respect both user and robot personas in dyadic dialogue. It introduces Bilateral Personalized Dialogue Generation (BPDG), a GPT-2 based framework that uses a dynamic persona-aware fusion module to integrate bilateral personas with dialogue context, guided by a CMIM-based candidate selection augmented with contrastive learning. The approach employs multi-task learning (language modeling, persona presence prediction, and dialogue generation) and leverages explicit bilateral profiles to achieve bilateral persona-consistency. Experimental results on a bilateral PersonalDialog-derived dataset show improvements in automatic metrics (BPAcc, BLEU, F1) and human judgments, with ablations confirming the contributions of dynamic fusion, pre-training, and CMIM. This work advances open-domain dialogue by enabling responses that are more personalized and aligned with both participants, offering potential gains in trust and engagement for human-robot interaction.

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.

Figures (6)

• Figure 1: Exemplar dialogues with/without bilateral persona-consistent in dyadic conversation. The general GPT2 model with unilateral persona can generate a response that only meets the robot's persona, but ignores the persona of the other party. The proposed method can incorporate bilateral personas and generate a response that matches the personas of both parties.
• Figure 2: The overview of the proposed BPDG method.
• Figure 3: The structure of personalized history embeddings.
• Figure 4: The structure of the dynamic persona-aware fusion module.
• Figure 5: Illustration of conditional mutual information. The circles represent the information entropies of the different variables. The dashed circle represents the information entropy of the generated responses.