基于分期BERTopic与非平衡最优传输的我国人工智能安全政策主题演化研究
Evolution of AI Safety Policy Topics in China: A Period-Specific BERTopic and Unbalanced Optimal Transport Approach
摘要: 目的/意义：随着人工智能技术的快速发展，其安全风险成为贯穿模型研发、部署与应用全流程的综合性问题。系统梳理我国人工智能安全治理的内容演变，对进一步加强安全治理具有重要意义。方法/过程：本文选取2018~2025年的人工智能安全政策文本和行业报告作为研究语料，按两年为窗口划分为四个时间片，采用分期BERTopic模型来识别主题内容；结合切片Wasserstein距离和非平衡最优传输构建跨期主题流量关系，据此识别主题演化路径。结果/结论：研究发现，我国人工智能安全主题呈现明显的阶段性特征，政策关注点由早期的训练数据与算法公平逐步转移至个人信息处理、生成内容治理、模型攻击防护及综合应急处置。这些主题表现为延续、分裂与重组并存的多层次结构特征。本文所识别的多种主题结构关系，为把握治理重心的迁移方向、完善系统性治理提供了方法支撑。
Abstract: Purpose/Significance: The rapid development of artificial intelligence (AI) has made safety risks a comprehensive issue spanning model development, deployment, and application. A systematic examination of the evolution of AI safety governance in China is of great significance for further strengthening safety governance. Method/Process: AI safety policy documents and industry reports published between 2018 and 2025 were collected and divided into four two-year periods. BERTopic was applied separately to each period to identify topics. Sliced Wasserstein distance and unbalanced optimal transport were then combined to construct cross-period topic flows and trace topic evolution paths. Result/Conclusion: AI safety topics in China exhibit distinct characteristics across periods. Governance attention has progressively shifted from early concerns about training data and algorithmic fairness to personal information processing, generated-content governance, protection against attacks on AI models, and comprehensive emergency response. These topics exhibit a multilayered structure characterized by the coexistence of continuation, splitting, and recombination. The identified topic-structure relationships provide methodological support for understanding shifts in governance priorities and improving systematic governance.
文章引用：谢桂平, 郑静. 基于分期BERTopic与非平衡最优传输的我国人工智能安全政策主题演化研究[J]. 统计学与应用, 2026, 15(10): 133-143. https://doi.org/10.12677/sa.2026.1510234

参考文献

[1] Bommasani, R., Hudson, D.A., Adeli, E., et al. (2021) On the Opportunities and Risks of Foundation Models.
https://arxiv.org/abs/2108.07258
[2] Weidinger, L., Uesato, J., Rauh, M., Griffin, C., Huang, P., Mellor, J., et al. (2022) Taxonomy of Risks Posed by Language Models. 2022 ACM Conference on Fairness Accountability and Transparency, Seoul, 21-24 June 2022, 214-229.
https://doi.org/10.1145/3531146.3533088
[3] Bender, E.M., Gebru, T., McMillan-Major, A. and Shmitchell, S. (2021) On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 3-10 March 2021, 610-623.
https://doi.org/10.1145/3442188.3445922
[4] Raji, I.D., Smart, A., White, R.N., Mitchell, M., Gebru, T., Hutchinson, B., et al. (2020) Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, Barcelona, 27-30 January 2020, 33-44.
https://doi.org/10.1145/3351095.3372873
[5] 魏弋, 李刚. 人工智能治理范式变革——基于政策分析与主题建模的研究[J]. 图书与情报, 2024, 44(5): 62-76.
[6] 李亮, 徐晓林, 明承瀚. 政务数据安全政策体系的优化路径研究——基于治理政策文本的扎根理论分析[J]. 图书馆, 2024(12): 36-45.
[7] 张涛, 张博雅, 马海群. 我国央地政府数据安全政策扩散特征及主题转移研究[J]. 情报理论与实践, 2025, 48(5): 118-127, 137.
[8] 周文泓, 熊小芳, 叶雅寒. 面向人工智能的数据治理框架研究——基于政策文本的构建与展望[J]. 信息资源管理学报, 2025, 15(4): 87-98.
[9] Blei, D.M. and Lafferty, J.D. (2006). Dynamic Topic Models. Proceedings of the 23rd International Conference on Machine Learning. 25-29 June 2006, 113-120.
https://doi.org/10.1145/1143844.1143859

为你推荐