Persona-Aware Alignment Framework for Personalized Dialogue Generation

2025-11-13

AI Summary

Personalized dialogue generation often suffers from generic and inconsistent responses due to implicit personality modeling, neglecting explicit persona representation. To address this, we propose PAL, a two-stage training framework: (1) Persona-aware Learning jointly encodes dialogue context and persona attributes into unified representations; (2) Persona Alignment explicitly optimizes for persona feature consistency between input personas and generated responses. Furthermore, we introduce a semantic-level “Select-then-Generate” inference strategy that first retrieves persona-aligned response segments from candidates and then refines them, enhancing persona sensitivity. Evaluated on multiple benchmarks, PAL significantly outperforms state-of-the-art personalized dialogue models and large language models, achieving new SOTA performance in persona relevance, consistency, and response diversity. Notably, PAL is the first method to formulate persona alignment as an explicit, differentiable optimization objective.

Technology Category

• Natural Language Processing: Conversational AI/Dialog Systems
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Personalized dialogue generation aims to leverage persona profiles and dialogue history to generate persona-relevant and consistent responses. Mainstream models typically rely on token-level language model training with persona dialogue data, such as Next Token Prediction, to implicitly achieve personalization, making these methods tend to neglect the given personas and generate generic responses. To address this issue, we propose a novel Persona-Aware Alignment Framework (PAL), which directly treats persona alignment as the training objective of dialogue generation. Specifically, PAL employs a two-stage training method including Persona-aware Learning and Persona Alignment, equipped with an easy-to-use inference strategy Select then Generate, to improve persona sensitivity and generate more persona-relevant responses at the semantics level. Through extensive experiments, we demonstrate that our framework outperforms many state-of-the-art personalized dialogue methods and large language models.

Problem

• Addresses persona neglect in dialogue generation models
• Improves persona relevance through semantic-level alignment
• Enhances response consistency using persona-aware training objectives

Innovation

• Persona-Aware Alignment Framework treats persona alignment as training objective
• Two-stage training includes Persona-aware Learning and Persona Alignment
• Select then Generate inference strategy improves persona relevance