MORPHEUS: Modeling Role from Personalized Dialogue History by Exploring and Utilizing Latent Space

TL;DR

MORPHEUS tackles the PDG challenge of generating persona-consistent responses without relying on external role data. It introduces a three-stage training framework that builds a latent Persona Codebook to represent roles, aligns encoded persona information with dialogue history, and jointly trains code indices and codes to enable generalization to unseen roles. Empirical results on English and Chinese datasets show MORPHEUS improves personalization, coherence, and diversity, while remaining parameter-efficient for fine-tuning large language models. The approach offers privacy-preserving, scalable personalization with strong potential for deployment in multi-domain dialogue systems.

Abstract

Personalized Dialogue Generation (PDG) aims to create coherent responses according to roles or personas. Traditional PDG relies on external role data, which can be scarce and raise privacy concerns. Approaches address these issues by extracting role information from dialogue history, which often fail to generically model roles in continuous space. To overcome these limitations, we introduce a novel framework \textbf{MO}dels \textbf{R}oles from \textbf{P}ersonalized Dialogue \textbf{H}istory by \textbf{E}xploring and \textbf{U}tilizing Latent \textbf{S}pace (MORPHEUS) through a three-stage training process. Specifically, we create a persona codebook to represent roles in latent space compactly, and this codebook is used to construct a posterior distribution of role information. This method enables the model to generalize across roles, allowing the generation of personalized dialogues even for unseen roles. Experiments on both Chinese and English datasets demonstrate that MORPHEUS enhances the extraction of role information, and improves response generation without external role data. Additionally, MORPHEUS can be considered an efficient fine-tuning for large language models.

Figures (4)

• Figure 1: In the task of PDG when role data is masked, an example generated by two state-of-the-art (SOTA) models, MSP and CLV, alongside our model MORPHEUS, is presented. Roles A and B are similar, and we expect models to generalize from the data about A to learn how to generate B's response.
• Figure 2: The overview structure of the proposed model.
• Figure 3: Experiments with the different N on the ConvAI2 dataset. For ease of viewing, BLEU-1, rouge-L, P-Co are magnified by a factor of three and Dist-1 by a factor of ten.
• Figure 4: The evaluation criteria of human.