基于BERT文本分类模型的APP隐私政策完整性评价研究

Research on Integrity Evaluation of APP Privacy Policy Based on BERT Model

[目的/意义]隐私泄露多发的原因之一在于部分APP隐私政策不够完整合规。APP隐私政策的完整性评价有助于推动隐私政策合规性研究,并揭示运营方收集和使用用户隐私的现状。[方法/过程]研究针对主流APP的隐私政策文本,利用Bert分类模型和L2归一化等方法计算得到其完整性得分,并从完整性的角度比对分析了不同类别和典型APP的隐私政策。[结果/结论]研究发现,Bert模型分类效果极佳,同时APP间隐私政策完整性差别较大。研究提出的隐私政策自动分类评价方法有助于隐私政策评价研究向自动化和智能化方向发展,并为隐私政策合法合规性研究提供了借鉴。

[Purpose/Significance]One of the major reasons for APP’s privacy leaks is that some APP privacy policies are not fully integrated and legitimate.The integrity evaluation of APP privacy policy aims to promote the research of privacy policy legitimacy and reveal how online platforms collect and use users’ personal information.[Method/Process]Targeting mainstream mobile applications privacy policies, the research used Bert classification model and L2 normalization to calculate their integrity scores, compared and analyzed the privacy policies of typical APPs in different categories from the perspective of integrity.[Result/Conclusion]The study finds that the Bert model works great in text-classification, and the privacy policy integrity varies greatly between APPs.In conclusion, the automatic classification and evaluation method proposed in this paper promotes the privacy policy evaluation research in the direction of automation and intelligence.It also provides a reference for researching privacy policy legal compliance.