P-Tailor: Customizing Personality Traits for Language Models via Mixture of Specialized LoRA Experts

2024-06-18

arXiv.org

AI Summary

Existing research predominantly focuses on explicit role modeling, neglecting the profound influence of deep personality traits on behavior and decision-making—thereby hindering the development of human-like, psychologically grounded AI. This paper introduces a Theory of Mind–driven paradigm for LLM personalization, pioneering the modeling of the Big Five personality traits as customizable, composable implicit behavioral representations tailored for high-stakes, sensitive applications such as psychotherapy. To this end, we propose P-Tailor: a LoRA-based Mixture-of-Experts architecture, a personality-specialized loss function, and the first high-quality Personality Choreography Dataset (PCD), augmented by prompt-enhanced data construction and fine-tuning strategies. Experiments demonstrate substantial improvements in fine-grained controllability across all five personality dimensions and enhanced cross-topic consistency, while achieving 37% higher parameter efficiency compared to baseline approaches.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Personalized large language models (LLMs) have attracted great attention in many applications, such as intelligent education and emotional support. Most work focuses on controlling the character settings based on the profile (e.g., age, skill, experience, and so on). Conversely, the psychological theory-based personality traits with implicit expression and behavior are not well modeled, limiting their potential application in more specialized fields such as the psychological counseling agents. In this paper, we propose a mixture of experts (MoE)-based personalized LLMs, named P-tailor, to model the Big Five Personality Traits. Particularly, we learn specialized LoRA experts to represent various traits, such as openness, conscientiousness, extraversion, agreeableness and neuroticism. Then, we integrate P-Tailor with a personality specialization loss, promoting experts to specialize in distinct personality traits, thereby enhancing the efficiency of model parameter utilization. Due to the lack of datasets, we also curate a high-quality personality crafting dataset (PCD) to learn and develop the ability to exhibit different personality traits across various topics. We conduct extensive experiments to verify the great performance and effectiveness of P-Tailor in manipulation of the fine-grained personality traits of LLMs.

Problem

• Modeling Big Five personality traits in LLMs
• Capturing nuanced individual trait expressions
• Generating topic-adaptive personality-based reactions

Innovation

• Mixture of specialized LoRA experts
• Personality Specialization Loss (PSL)
• OCEAN-Chat human-verified dataset