微博僵尸用户检测研究
发布时间：2019-02-17 18:49
【摘要】：随着在线社交网络的盛行，微博作为一种方便快捷的信息传播载体，已经成为人们交流互动的重要方式。微博服务拉近了网民之间的距离，使用户可以快速的发布、接收及传播信息。微博在国内外迅速流行的同时，粉丝数逐渐成为了衡量用户知名度及用户排名所参考的一项重要指标。随之衍生出的僵尸粉（即僵尸用户）扰乱微博正常秩序，引发微博信任危机。僵尸用户经过长期地演变，其行为表现变得越发类似真实用户，因此，如何快速准确地甄别僵尸用户已成为维护微博公信力所亟待解决的一项问题。 本文选取国内最具影响力、发展最迅速的微博平台之一——新浪微博作为数据分析对象，并使用新浪API接口获取用户数据信息，用于研究分析及模型有效性验证。通过数据分析，本文找出了僵尸用户和真实用户的粉丝关系网络是否存在聚类现象上所呈现的明显差异。此外，结合僵尸用户和真实用户在粉丝数、关注数及发微博频率等行为上的差异，提出用户可信度计算算法及用户活跃度计算方式，并构建得出基于用户粉丝聚类现象的僵尸用户检测模型。经实验验证，，此模型在检测准确性及稳定性上表现良好，但是检测效率偏低。 同时，考虑到微博用户信息量巨大，数据处理较为耗时，本研究在原有检测模型的基础上结合云计算技术，将僵尸用户检测模型中较为耗时的四个模块利用MapReduce技术做出改进，提高模型的可用性。经搭建Hadoop集群将改进前后的模型建立对比实验，实验结果表明改进后的模型在保持原有检测准确率及稳定性的基础上，检测效率有了明显的提高。并且，随着Hadoop集群节点的增多，检测效率增长趋势呈现出接近线性的加速比。
[Abstract]:With the popularity of online social networks, Weibo, as a convenient and fast carrier of information dissemination, has become an important way for people to communicate and interact. Weibo service draws the distance between Internet users, so that users can quickly publish, receive and disseminate information. With the rapid popularity of Weibo at home and abroad, the number of fans has gradually become an important index to measure the popularity and ranking of users. The resulting zombie powder (that is, zombie users) disrupts Weibo's normal order, triggering a crisis of confidence in Weibo. After a long period of evolution, the behavior of zombie users becomes more and more similar to real users. Therefore, how to identify zombie users quickly and accurately has become an urgent problem to maintain Weibo's credibility. In this paper, one of the most influential and rapidly developing Weibo platforms in China is selected as the object of data analysis, and the Sina API interface is used to obtain user data information for research and analysis and validation of model validity. Through data analysis, this paper finds out whether there are obvious differences in clustering between zombie users and real users. In addition, considering the differences between zombie users and real users in the number of followers, the number of attention and the frequency of Weibo, the calculation algorithm of user credibility and the calculation method of user activity are put forward. And construct a zombie user detection model based on the phenomenon of user fan clustering. Experimental results show that the model performs well in accuracy and stability, but the detection efficiency is low. At the same time, considering Weibo's huge amount of user information and time-consuming data processing, this study combines cloud computing technology with the original detection model, and improves the four modules of zombie user detection model using MapReduce technology. Improve model availability. The experimental results show that the improved model can improve the detection efficiency on the basis of maintaining the original detection accuracy and stability. Moreover, with the increase of Hadoop cluster nodes, the increasing trend of detection efficiency is close to linear speedup.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092
本文编号：2425478
[Abstract]:With the popularity of online social networks, Weibo, as a convenient and fast carrier of information dissemination, has become an important way for people to communicate and interact. Weibo service draws the distance between Internet users, so that users can quickly publish, receive and disseminate information. With the rapid popularity of Weibo at home and abroad, the number of fans has gradually become an important index to measure the popularity and ranking of users. The resulting zombie powder (that is, zombie users) disrupts Weibo's normal order, triggering a crisis of confidence in Weibo. After a long period of evolution, the behavior of zombie users becomes more and more similar to real users. Therefore, how to identify zombie users quickly and accurately has become an urgent problem to maintain Weibo's credibility. In this paper, one of the most influential and rapidly developing Weibo platforms in China is selected as the object of data analysis, and the Sina API interface is used to obtain user data information for research and analysis and validation of model validity. Through data analysis, this paper finds out whether there are obvious differences in clustering between zombie users and real users. In addition, considering the differences between zombie users and real users in the number of followers, the number of attention and the frequency of Weibo, the calculation algorithm of user credibility and the calculation method of user activity are put forward. And construct a zombie user detection model based on the phenomenon of user fan clustering. Experimental results show that the model performs well in accuracy and stability, but the detection efficiency is low. At the same time, considering Weibo's huge amount of user information and time-consuming data processing, this study combines cloud computing technology with the original detection model, and improves the four modules of zombie user detection model using MapReduce technology. Improve model availability. The experimental results show that the improved model can improve the detection efficiency on the basis of maintaining the original detection accuracy and stability. Moreover, with the increase of Hadoop cluster nodes, the increasing trend of detection efficiency is close to linear speedup.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092
【参考文献】
相关期刊论文 前3条
1 胡瑶迪;;微博传播对传统媒体的影响[J];新闻世界;2010年06期
2 李鸿彬;林浒;杨雪华;林荣;;一种基于社会网络的SIP垃圾即时消息的检测方法[J];小型微型计算机系统;2012年08期
3 姚永明;吕建平;;基于Android平台的用户管理软件的设计与实现[J];西安文理学院学报(自然科学版);2013年01期
相关博士学位论文 前1条
1 韩毅;社会网络分析与挖掘的若干关键问题研究[D];国防科学技术大学;2011年
本文编号：2425478
本文链接：https://www.wllwen.com/guanlilunwen/ydhl/2425478.html
