融合敏感性与关联性的隐私风险评估及实证研究OACHSSCD

[Purpose/Significance]This study aims to evaluate the privacy risks associated with unstructured text within online community user-generated content,addressing the challenge that such data poses as a high-risk vector for privacy leakage due to contextual variations in sensitivity and correlation-based vulnerabilities.[Method/Process]A pri-vacy risk quantification framework integrating sensitivity and relevance was proposed,with experimental validation con-ducted on the'Member Networking'section of the academic platform Muchong.This framework employs a BERT-BiLSTM-CRF deep learning model to achieve attribute extraction from unstructured text.Attribute sensitivity was quanti-fied using a privacy lexicon,attribute correlation was measured via Pointwise Mutual Information(PMI),and these factors were integrated with privacy principal identification metrics to compute privacy risk scores,followed by risk stratification.[Result/Conclusion]Ablation studies and manual validation demonstrate its capability to identify,assess,and stratify pri-vacy risks in unstructured textual data.These findings offer new insights for improving privacy protection policies and plat-form privacy governance.