Event Log Behavioral Privacy Protection Method Based on Relaxed Behavioral Profiles

Privacy preservation techniques for event logs prevent sensitive information leakage by injecting noise into log data, thus mitigating privacy risks in business process management. While ensuring privacy, this approach often reduces data utility and impacts subsequent analysis accuracy. Current research mainly optimizes noise injection based on data attributes to preserve utility, but overlooks how behavioral relationship changes affect data quality. To address this, this paper proposes a privacy protection method for event logs based on relaxed behavioral profiles. First, a behavioral privacy attack model is formalized to reveal how behavioral relationships can lead to privacy leakage. Building on this, data attributes are utilized to extend behavioral profiles, and a behavioral profile extension algorithm is proposed to analyze implicit behavioral relationships between activities. This allows building a multi-perspective perturbation set combining behavioral and data attributes. Finally, perturbations are injected using differential privacy methods. Experiments on both real-world and synthetic datasets demonstrate that the event logs generated by this method are behavior-unchanged at the level of the modeled relations, effectively mitigating the utility loss caused by log anonymization and attaining F1-scores of up to 0.874 on synthetic data, comparable to those of the baselines.