RISK WARNING ALGORITHM FOR OVER-COLLECTION OF PERSONAL INFORMATION IN APP FORMAL PROGRAM

To address the issues of large quantization errors in personal information over-collection risk and low precision in alarm level classification, this study proposes a formal program personal information over-collection risk alarm algorithm for Apps. Multi-source heterogeneous data of Apps (static data, dynamic behavior data, etc.) are obtained and transformed into an initial feature space through formal parsing of policy texts, permission-information type mapping, and dynamic behavior vectorization. Multi-dimensional features of App formal programs are extracted, based on which the personal information over-collection risk factors are calculated-personal information sensitivity, personal information collection behavior abnormality, and personal information collection behavior compliance deviation. The personal inform mation over-collection risk quantization result is calculated through a generalized linear model and Sigmoid function. According to the distribution of personal information over-collection risk quantization results, adaptive thresholds are determined and a graded alarm mechanism for personal information over-collection risk is established, thereby achieving the research objectives. Experimental results show that the maximum cosine similarity between the multi-dimensional feature vectors of App formal programs extracted by the designed algorithm and the actual feature vectors reached 0.95, the minimum quantization error of personal information over-collection risk reached 1%, and the personal information over-collection risk alarm level classification results were identical to the annotated alarm level results of the experimental samples.