新浪微博恶意用户研究及检测

发布时间：2018-03-26 18:16

本文选题：新浪微博　切入点：恶意用户检测　出处：《上海交通大学》2014年硕士论文

【摘要】：近年来，随着互联网的普及，微博作为一种新的媒体形式和交流平台也越来越流行。与此同时，微博平台上不断涌现出的各种恶意行为，，开始成为破坏微博安全和内容质量的最大威胁。如何鉴别恶意用户，成为社交网络研究领域的一个重要课题。现有的针对微博的研究大多数以Twitter为平台，其成果无法直接应用于新浪微博等中文微博平台。因此，专门针对中文微博的研究也是相当有必要的。本文以新浪微博为研究对象，提出了一种基于行为特征检测的微博恶意用户鉴别方法。我们采用了多种方法收集了恶意用户样本：通过架设主动式蜜罐账号，我们吸引了114个恶意用户；利用微博爬虫程序，我们捕获了1459个恶意用户。此外，我们还从淘宝购买了微博推广服务，从而获得了8000个恶意用户样本。我们将恶意用户样本按照行为模式进行分类，归纳总结了三种不同的恶意行为模式：过度广告、重复转发以及过度关注。我们针对所得到的三种不同行为模式的恶意用户进行了深入的分析，通过与正常用户比较统计特征值的方法，归纳总结了他们的特点并且找到了能够区分恶意用户与正常用户的特征。在此基础上，我们利用机器学习技术实现了多种恶意行为分类器，并且将这些分类器整合在一起构成了自动检测系统。我们利用重新收集的测试数据集对该系统进行了测试。该系统能检测出82.06%的恶意用户，而误报率则只有5.92%。
[Abstract]:In recent years, with the popularity of the Internet, micro-blog as a new form of media and communication platform is more and more popular. At the same time, a variety of malicious behavior on the micro-blog platform continue to emerge, become the greatest threat to micro-blog security and content quality. How to identify malicious users, has become an important research topic in the field of social network the existing micro-blog. Most of the studies on Twitter platform, the results can not be directly applied to Sina micro-blog and micro-blog Chinese platform. Therefore, the research on Chinese micro-blog is quite necessary.
This paper takes Sina and micro-blog as the research object, presents a micro-blog malicious user identification method based on behavior detection. We used a variety of methods to collect the samples through the erection of a malicious user: active honeypot account, we attracted 114 malicious users; using micro-blog crawler, we captured 1459 malicious users. In addition, we also purchased from Taobao micro-blog promotion service, so as to obtain 8000 malicious users. We will sample the sample according to the malicious user behavior patterns are classified, summarized three kinds of malicious behavior in different modes: excessive advertising, repeated and excessive attention. We aim at three kinds of behavior patterns obtained by malicious users an in-depth analysis of eigenvalue method by comparison with normal user statistics, summarizes their characteristics and found to distinguish malicious The characteristics of the user and normal user. On this basis, we use machine learning techniques to achieve a variety of malicious behavior classifier, and the classifier together constitute an automatic detection system. We set the system is tested by test data to collect. This system can detect 82.06% malicious users, and false alarm rate it is only 5.92%.

【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092

【共引文献】