In the digital networking age, safeguarding individual privacy encounters numerous legal hurdles, including data breaches and improper utilization. Differential privacy algorithms, serving as the central privacy – safeguarding technology, achieve the goal of making query outcomes and adjacent datasets indistinguishable from one another. This is accomplished by introducing random noise into the data, thereby averting the disclosure of personal information.This paper, building upon the differential privacy algorithm, puts forward a differential privacy protection approach (DeepLIFT) founded on an adaptive Gaussian mechanism. This method first computes the correlation between the input and output. It then guarantees that the differential privacy criteria are satisfied. Subsequently, it employs the feature perturbation technique to introduce noise into the correlation. After that, it constructs subsequent hidden layers to preserve data usability.The findings indicate that the DeepLIFT algorithm can efficiently quantify the contribution of features in privacy protection applications. Moreover, it offers data – based support for assessing and optimizing privacy measures. However, its effectiveness depends on the specific protection measures and model structure, so in practical applications, it is necessary to combine multiple methods to ensure comprehensive privacy protection. To summarize, this paper suggests that the problem of preventing the leakage of personal privacy information can be solved through the joint participation of individuals, enterprises and governments.
