This paper addresses the impact of Large Language Model (LLM)-based pseudonymization on cross-platform identifier similarity and re-identification risk. Cross-platform profile linking enables organizations to analyze user behavior across multiple social networking platforms. However, this practice raises significant privacy and security concerns. Identifiers such as usernames and real names can be used to reconstruct user identities when aggregated across different online platforms. Therefore, addressing these challenges is highly relevant to current research on privacy-preserving data analytics, digital identity protection, and secure information management. Using Linktree profile connection datasets from Facebook, Instagram, LinkedIn, and Twitter/X, we analyze how pseudonymization affects identifier similarity and the potential recovery of original identifiers. Our findings demonstrate that pseudonymization significantly reduces identifier similarity, resulting in low re-identification success rates. The results highlight the potential of LLM-driven privacy-enhancing techniques for mitigating identity disclosure risks while preserving limited analytical utility. Consequently, this work contributes to ongoing research in privacy-aware data processing, secure cross-platform analytics, and trustworthy LLMs for online social networks.     «

This paper addresses the impact of Large Language Model (LLM)-based pseudonymization on cross-platform identifier similarity and re-identification risk. Cross-platform profile linking enables organizations to analyze user behavior across multiple social networking platforms. However, this practice raises significant privacy and security concerns. Identifiers such as usernames and real names can be used to reconstruct user identities when aggregated across different online platforms. Therefore, a...     »