Detecting Sensitive Information of Unstructured Text Using Convolutional Neural Network
Guosheng Xu, Chunhao Qi, Hai Bo Yu, Shengwei Xu, Chunlu Zhao, Jing Yuan · 2019
With the use of a large number of electronic text documents, the disclosure of sensitive information from unstructured text documents is a costly issue for individuals, businesses, and governments. How to detect sensitive information to prevent data leakage is a topic in the field of information security. At present, the practical detection methods are roughly divided into two types, sensitive word matching and traditional machine learning methods. Both methods rely on the frequency with which keywords are co-occurring with sensitive seed words. However, in practice, this may not accurately detect more complex patterns of sensitive information. In recent years, some scientists have proposed using recurrent neural networks for sensitive information detection, using the context of documents to more accurately predict the sensitivity of documents, but RNN improves the accuracy, the rate of training construction is not high. Therefore, this paper proposes to use Text-CNN instead of a recurrent neural network. While ensuring the accuracy of detection, it can also improve the training construction time of the detection model, improve the detection efficiency as a whole, and achieve efficient and accurate detection.