基于评论数据的恶意移动应用检测方法  

Malicious Mobile Application Identification based on End-User Comment

作　　者：朱璋颖 马永 燕锦华[3] 吴振宇[3] 徐文博 ZHU Zhang-ying;MA Yong;YAN Jin-hua;WU Zhen-yu;XU Wen-bo(Pwnzen InfoTech Co.,LTD.,Shanghai 201199,China;Shanghai Dianji University,Shanghai 200240,China;East China Institute of Computing Technology,Shanghai 201808,China)

机构地区：[1]上海犇众信息技术有限公司,上海201199 [2]上海电机学院,上海200240 [3]华东计算技术研究所,上海201808

出　　处：《通信技术》2019年第2期449-454,共6页Communications Technology

基　　金：中国电科联合基金(No.20166141B08020101)~~

摘　　要：恶意移动应用通过动态代码加载等手段绕过移动应用市场安全审核,对终端用户造成威胁。为了实现对这些应用进行事后审计,提出一种基于自然语言处理(NLP)的恶意应用检测模型。通过搜集、处理移动应用市场中用户对应用的评论数据,建立恶意分类检测模型。通过对评论数据的处理分类,判断应用是否存在恶意行为,以此对移动应用进行事后安全检查。实验结果表明,建立的恶意应用检测模型准确率达到81%,可以有效识别恶意移动应用。Malicious mobile applications bypass the mobile application market security audit by means of dynamic code loading and pose a threat to end users. In order to implement post-auditing of these applications, a malicious language detection model based on NLP(natural language processing) is proposed. A malicious classification detection model is established by collecting and processing the user’s comment data on the application in the mobile application market. By performing processing and classifying of review data, whether or not the application has malicious behavior is determined, and in this way, afterthe-fact security checks are carried out for mobile application. The experimental results show that the established malicious application detection model has an accuracy rate of 81%, which can effectively identify malicious mobile applications.

关 键 词：恶意移动应用 自然语言处理 机器学习 用户评论 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]