DEEP LEARNING BASED SENSITIVE DATA DETECTION

The growing popularity of edge techniques, such as IoT, 5G, blockchain, make it increasingly challenging to protect sensitive data due to the amount of data increases and the growing volume of regulatory policies. To properly protect sensitive data, it is very important to identify sensitive data and implement data anonymization to ensure the quality and proper use of data anonymization techniques. This work focuses on proactively sensitive data identification, classification and anonymization using machine learning techniques. We first investigated the sensitive data extraction from both structured data and unstructured data, in which Bert models and Regular expressions were used to achieve the identification of sensitive data in real-time. Meanwhile, we propose a comprehensive sensitive detection framework combining the Bert model with regular expressions that can achieve high precision and good generalization capability with not so large corpus. The experimental results demonstrate the effectiveness of proposed solution.