基于深度学习和特征分析的隐蔽性有害信息识别方法研究

Research on hidden harmful information recognition method based on deep learning and feature analysis

互联网上的有害信息层出不穷,而随着各项技术的发展,有害信息采用各种方式隐藏其核心内容,以躲避各类算法的识别和检索。目前,对此类隐蔽性有害信息识别的常用方法主要是基于人工智能技术,采用人工标注和训练学习的方式,但是算法较为复杂,对资源的需求较大。因此,一种新思路为:从隐蔽性有害信息的特征出发,分析出其特征规律,并基于以上结果设计一种新方法,同时降低人工参与和算法复杂度。最后,通过不同的样本库,对基于深度学习的方法和基于特征分析的方法效果进行对比分析,得到不同场景下的应用方案,为识别隐蔽性有害信息工作提供参考。

There is an endless stream of harmful information on the Internet,and with the development of various technologies,harmful information uses various ways to hide its core content,so as to avoid the recognition and retrieval of various algorithms.At present,the general method of identifying such hidden harmful information is mainly based on artificial intelligence technology,using manual annotation and training and learning,but the algorithm is more complex,and the demand for resources is large.Therefore,a new way is proposed to analyze the pattern of hidden harmful information from its characteristics.Based on the pattern,a new method is designed to reduce both human participation and algorithm complexity.Finally,through different sample libraries,the effects of the method based on deep learning and the method based on feature analysis are compared and analyzed,and application schemes in different scenarios are obtained,which provides reference for identifying hidden harmful information.