PersLLM: A Personified Training Approach for Large Language Models

Large language models (LLMs) exhibit humanlike intelligence, enabling them to simulate human behavior and support various applications that require both humanized communication and extensive knowledge reserves. Efforts are made to personify LLMs with special training data or hand-crafted prompts, while correspondingly faced with challenges such as insufficient data usage or rigid behavior patterns. Consequently, personified LLMs fail to capture personified knowledge or express persistent opinion. To fully unlock the potential of LLM personification, we propose PersLLM, a framework for better data construction and model tuning. For insufficient data usage, we incorporate strategies such as Chain-of-Thought prompting and anti-induction, improving the quality of data construction and capturing the personality experiences, knowledge, and thoughts more comprehensively. For rigid behavior patterns, we design the tuning process and introduce automated DPO to enhance the specificity and dynamism of the models’ personalities, which leads to a more natural opinion communication. Both automated metrics and expert human evaluations demonstrate the effectiveness of our approach.

Introduction. Large language models (LLMs) have demonstrated human-level intelligence in multiple domains due to extensive parameters and data (Brown et al., 2020; Achiam et al., 2023). This has driven research into using LLMs as human-like agents in social simulations, human-machine interactions, and multi-agent systems (Bail, 2024; Gao et al., 2024; Grossmann et al., 2023; Yang, 2024). Aligning agents with specific personalities enhances user comfort, improves knowledge mastery and facilitates collaboration (Pelau et al., 2021; Güver and Motschnig, 2017), making personification crucial for applications such as online education, consultation and public opinion analysis. Efforts to integrate personalities into LLMs typically follow two approaches. Training-based methods embed personality traits into model parameters using targeted data (Zhou et al., 2023; Wang et al., 2023c). Unfortunately, the lack of comprehensive and theoretical analysis in data construction has led to insufficient use of raw materials, focusing only on modeling isolated features such as language style or anecdotes.

Discussion / Conclusion. LLM personification is a vital research area that allows for more humanized, personalized, and knowledgeable communication. This approach facilitates deeper simulations to understand social issues. It may also enhance psychological acceptance across applications such as intelligent psychotherapy. In this article, we have identified the limitations of prompt-based personification methods, such as nonhuman communication patterns and characteristic tendencies, and proposed improved data construction and model tuning strategies for training-based methods. Our experiments demonstrate that PersLLM successfully captures the core traits of target personalities, leading to consistent opinion interactions and effective knowledge generation. We highlight three key limitations of the current approach: 1.