StyEmp: Stylizing Empathetic Response Generation via Multi-Grained Prefix Encoder and Personality Reinforcement

TL;DR

StyEmp tackles the lack of system personality in empathetic response generation by introducing a multi-grained prefix encoder that ties personality traits to empathetic signals and a personality reinforcement module that calibrates outputs via contrastive learning. The approach uses past listener responses to implicitly encode personality and employs a diverse-beam candidate set with a personality-margin-based ranking and a pairwise margin loss to align generated personality with expectations. Empirical results on EMPATHETICDIALOGUES show improvements in personality expression and empathy, aided by objective metrics and human judgments, though personality predictor accuracy remains a bottleneck. The work advances conversational AI toward more believable, personality-consistent empathy, with future directions including annotated personality data to further refine calibration and evaluation.

Abstract

Recent approaches for empathetic response generation mainly focus on emotional resonance and user understanding, without considering the system's personality. Consistent personality is evident in real human expression and is important for creating trustworthy systems. To address this problem, we propose StyEmp, which aims to stylize the empathetic response generation with a consistent personality. Specifically, it incorporates a multi-grained prefix mechanism designed to capture the intricate relationship between a system's personality and its empathetic expressions. Furthermore, we introduce a personality reinforcement module that leverages contrastive learning to calibrate the generation model, ensuring that responses are both empathetic and reflective of a distinct personality. Automatic and human evaluations on the EMPATHETICDIALOGUES benchmark show that StyEmp outperforms competitive baselines in terms of both empathy and personality expressions.

Figures (4)

• Figure 1: Different personalities exhibit distinct preferences for empathetic intents in responses richendoller1994exploringmairesse2010towards. In a given context, the user shows varying feelings to the system's responses, where the system encompasses empathetic expression and consistent personality traits, resulting in a more human-like interaction.
• Figure 2: The architecture of our proposed method that contains a multi-grained prefix encoder and personality reinforcement module.
• Figure 3: Template for human evaluation on empathy in generated responses.
• Figure 4: Template for human evaluation on personality consistency in generated responses.