The ubiquity of smartphones in contemporary society has cemented Location-Based Social Network (LBSN) as an integral component of modern life. Yet, the rapid evolution of LBSN services has brought to the fore pressing concerns surrounding user privacy. While prior research has yielded a plethora of privacy preservation strategies focusing on personalization and spatiotemporal dimensions, the semantic dimension, particularly semantic transitions, remains inadequately addressed. This paper introduces the Dynamic Semantic Probability Distribution Attack to identify user locations through semantic transition analysis. To counter this, we propose Semantic Transfer Entropy, quantifying protection resilience. We present the Trajectory Group Semantic Anonymization (TGSA) model to address various attack vectors, extending anonymous positioning with Group Deep Semantic Trees. Leveraging the Dummy method, we consider temporal, spatial, and semantic factors to produce K-anonymity sets balancing data utility and privacy. Rigorous experiments on the Geolife dataset demonstrate TGSA’s efficacy in addressing privacy concerns. This research contributes to advancing privacy protection mechanisms in the realm of LBSN, providing a novel solution to privacy protection issues incorporating semantic elements.