基于三层心理学模型的人格智能系统
Personality Intelligence System Based on a Three-Layer Psychological Model
摘要: 随着以ChatGPT、Claude、Gemini为代表的大语言模型(Large Language Models, LLM)技术的成熟，人机交互正从“操作式”向“理解式”范式演化。传统对话系统侧重任务执行，而新一代智能体开始展现出拟人化思维与情感表达的潜能。然而，现有系统仍在人格一致性、长期记忆以及低时延交互与深度优化的平衡方面存在关键缺口。本文提出人格智能系统(Personality Intelligence System, PIS)，以三层心理学模型(大五人格、三我理论、MBTI)为核心，构建人格的定义–决策–表达一体化机制。系统引入动静分离人格建模与双时域优化机制，在时间维度上实现短期自适应与长期演化的统一；并通过人格记忆网络(Personality Memory Network, PMN)整合短期上下文、长期事实与知识图谱，实现人格状态的可解释更新。输出层采用一致性锚定(Consistency Anchoring)机制，在人格、语义与情感维度上保持输出稳定性。本文进一步给出人格状态更新的数学形式、收敛条件及学习机制分析，证明系统在收缩映射假设下的人格演化具有稳定性。研究结果表明，PIS能在认知与情感层面实现人格的连续表达，为虚拟伴侣、教育辅导与心理健康支持等领域提供理论支撑。原型系统与相关数据集将于后续公开，以促进人格智能的开放研究。
Abstract: With the maturity of Large Language Models (LLMs) such as ChatGPT, Claude, and Gemini, human-computer interaction is evolving from an “operational” paradigm to an “interpretive” one. Traditional dialogue systems focus on task execution, while the new generation of intelligent agents begins to exhibit potentials for human-like cognition and emotional expression. However, existing systems still face critical gaps in personality consistency, long-term memory, and the balance between low-latency interaction and deep optimization. This paper proposes the Personality Intelligence System (PIS), which centers on a three-layer psychological model (Big Five Personality Traits, Structural Theory of the Psyche, and MBTI) to build an integrated framework for personality definition, decision making, and expression. The system introduces dynamic-static decoupled personality modeling and a dual time-domain optimization mechanism, achieving the unity of short-term adaptation and long-term evolution over time. Through the Personality Memory Network (PMN), it integrates short-term context, long-term facts, and knowledge graphs to enable interpretable personality-state updates. The output layer employs a consistency anchoring mechanism to maintain stability across personality, semantic, and emotional dimensions. Furthermore, this paper formalizes the mathematical representation, convergence conditions, and learning mechanism of personality-state updates, proving that personality evolution under the contraction mapping assumption is stable. The proposed framework demon states the potential to enable continuous personality expression at both cognitive and affective levels, providing theoretical foundations for applications such as virtual companionship, educational tutoring, and mental health support. The prototype system and related datasets will be released subsequently to promote open research on personality intelligence.
文章引用：万学靖, 刘伟, 曹婷, 冯吭雨, 解运洲, 任付标, 过弋, 宋广奎. 基于三层心理学模型的人格智能系统[J]. 嵌入式技术与智能系统, 2025, 2(5): 305-318. https://doi.org/10.12677/etis.2025.25030

参考文献

[1] Celli, F., Kartelj, A., Đorđević, M., et al. (2025) Twenty Years of Personality Computing: Threats, Challenges and Future Directions. arxiv:2503.02082.
[2] Fang, Q., Giachanou, A., Bagheri, A., et al. (2022) On Text-Based Personality Computing: Challenges and Future Directions. arxiv:2212.06711.
[3] Besta, M., Chandran, S., Gerstenberger, R., et al. (2025) Psychologically Enhanced AI Agents. arxiv:2509.04343.
[4] Kruijssen, J.M.D. and Emmons, N. (2025) Deterministic AI Agent Personality Expression through Standard Psychological Diagnostics. arxiv:2503.17085.
[5] Klinkert, L.J., Buongiorno, S. and Clark, C. (2024) Driving Generative Agents with Their Personality. arxiv:2402.14879.
[6] Zeng, W., Wang, B., Zhao, D. and Qu, Z. (2025) Dynamic Personality in LLM Agents. arxiv:2505.01234.
[7] Osou, V. and Zhao, H. (2017) A Review of Dialogue Systems: From Eliza to Neural Models. arxiv:1706.05100.
[8] Goldberg, L.R. (1990) An Alternative “Description of Personality”: The Big-Five Factor Structure. Journal of Personality and Social Psychology, 59, 1216-1229. [Google Scholar] [CrossRef] [PubMed]
[9] Mairesse, F., Walker, M.A., Mehl, M.R. and Moore, R.K. (2007) Using Linguistic Cues for the Automatic Recognition of Personality in Conversation and Text. Journal of Artificial Intelligence Research, 30, 457-500. [Google Scholar] [CrossRef] 
[10] Zhang, S., Dinan, E., Urbanek, J., Szlam, A., Kiela, D. and Weston, J. (2018) Personalizing Dialogue Agents: I Have a Dog, Do You Have Pets Too?
https://arxiv.org/abs/1801.07243
[11] Park, J.S., O’Brien, J.C., Cai, C.J., Morris, M.R., Liang, P. and Bernstein, M.S. (2023) Generative Agents: Interactive Simulacra of Human Behavior. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, San Francisco, 29 October 2023-1 November 2023, 1-22. [Google Scholar] [CrossRef] 
[12] Kwon, J., Kim, T. and Lee, S. (2024) Dynamic Personalities in LLM-Based Negotiation Agents. arxiv:2403.18901.
[13] Smith, J., Rivera, E. and Zhao, K. (2025) Psychometrics for AI: Measuring and Validating Personality in Intelligent Agents. arxiv:2501.05678.
[14] Liu, Y., Wang, H. and Qian, L. (2024) Affective Computing in the LLM Era: Personality, Empathy, and Human-AI Emotion Alignment. arxiv:2409.10213.
[15] Chen, X., Zhang, R. and Chen, W. (2024) MemGPT: Towards Long-Term Memory-Enhanced LLM Agents. arxiv:2402.11103.
[16] Yuan, T., Zhang, W. and Xu, L. (2025) Mem0: Memory-Enhanced AI Agents for Scalable Long-Term Interaction. arxiv:2502.09401.
[17] Yao, S., Zhao, J., Yu, D., et al. (2023) ReAct: Synergizing Reasoning and Acting in Language Models. arxiv:2210.03629.
[18] Shinn, N. and Bisk, Y. (2023) Reflexion: Language Agents with Verbal Reinforcement Learning. arxiv:2303.11366.
[19] Li, G., Hammoud, H.A.A.K., Itani, H., Khizbullin, D. and Ghanem, B. (2023) CAMEL: Communicative Agents for “Mind” Exploration of Large Scale Language Models. arxiv:2303.17760.
[20] Sharma, G., Xu, P., Wang, J., Zhang, Y. and Zhou, D. (2023) Autogen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation Framework. arxiv:2308.08155.
[21] Chen, Y., Xu, J. and Zhao, F. (2023) Langgraph: Graph-Based Reasoning for Language Models. arxiv:2312.00752.
[22] Bai, Y., Kadavath, S., Kundu, S., et al. (2022) Constitutional AI: Harmlessness from AI Feedback. arxiv:2212.08073.