基于TextCNN的涉密文本识别

Confidential Text Recognition Based on TextCNN

摘要: 保密工作直接关系到社会稳定、经济增长、国家安全。新时代信息化和网络迅速普及和发展，办公信息化逐渐成为了主流，在带给办公便利的同时也导致了泄密行为的发生。人工筛选涉密文本极为浪费时间，并且可能会出现人为失误。本文利用爬虫技术构建涉密文本数据集，结合word2vec和TextCNN模型在自建数据集上进行训练。实现准确识别出包含涉密信息的文本。经过实验对比测试，相较于传统的卷积神经网络，TextCNN结合word2vec在自建数据集上达成的效果更好。

Abstract: Confidentiality is directly related to social stability, economic growth and national security. With the rapid popularization and development of informatization and network in the new era, office in-formatization has gradually become the mainstream, which brings convenience to office work and also leads to the occurrence of leakage of secrets. Manual screening of classified texts is a waste of time and may result in human error. In this paper, we use crawler technology to build a secret re-lated text dataset, and combine word2vec and TextCNN models to train on the self built dataset to accurately identify the text containing secret related information. Through experimental compari-son test, compared with the traditional convolutional neural network, TextCNN combined with word2vec achieves better results on the self built dataset.