重庆邮电大学学报（自然科学版）2026，Vol.38Issue(3)：561-573,13.DOI:10.3979/j.issn.1673-825X.202504080077
基于强化学习与多模态特征融合个性化轨迹发布方法
Personalized trajectory data publishing method based on reinforcement learning and multimodal feature fusion
摘要
Abstract
With the widespread adoption of location-based services,trajectory data privacy protection faces challenges of singularity,static mechanisms,and insufficient efficiency.To address these issues,this paper proposes a personalized trajectory publishing method driven by reinforcement learning and multimodal feature fusion.A semantic label generation approach that integrates a Transformer-based bidirectional encoder(bidirectional encoder representations from transformers,BERT)and graph neural networks(GNN)is employed to identify sensitive locations automatically.Furthermore,a deep deterministic policy gradient(DDPG)-based optimization strategy dynamically adjusts privacy weight parameters to achieve an adaptive balance between privacy protection and data utility.In addition,an individualized differential privacy algorithm based on the exponential mechanism is designed to control noise injection according to user-specific privacy levels in an adaptive manner.Comprehensive experiments conducted on real-world datasets demonstrate that,compared with existing mechanisms,the proposed approach improves data utility by 14.85%～53.94%and enhances privacy protection by 10.28%～42.72%,while maintaining high semantic label accuracy.关键词
差分隐私/强化学习/指数机制/个性化隐私保护/轨迹数据发布Key words
differential privacy/reinforcement learning/exponential mechanism/personalized privacy protection/trajectory data publishing分类
信息技术与安全科学引用本文复制引用
刘沛骞,李梦毫,王辉,申自浩..基于强化学习与多模态特征融合个性化轨迹发布方法[J].重庆邮电大学学报（自然科学版）,2026,38(3):561-573,13.基金项目
国家自然科学基金项目(61300216) （61300216）
河南省高等学校重点科研项目(23A520033) （23A520033）
河南理工大学博士基金项目(B2022-16)National Natural Science Foundation of China(61300216) （B2022-16）
Key Scientific Research Project of Colleges and Universities in Hennan Province(23A520033) （23A520033）
Ph.D.Foundation of Henan Polytechnic University(B2022-16) （B2022-16）
