A Cascading Framework for Uncovering
Spammers in Social Networks
Zejia Chen, Jiahai Yang, Jessie Hui Wang
Tsinghua National Laboratory for Information Science and Technology
Dept. of Computer Science and Technology, Tsinghua University
Beijing, 100084, China
zejiachen@gmail.com, {yang,hwang}@cernet.edu.cn
Abstract—With tremendous popularity, OSNs have become
the most important platform for marketing and advertising
during the past years. Meanwhile, spamming has already become
a very serious problem in OSNs, drawing the attention of both
academic and industry communities. In this paper, we investigate
the problem of spammer detection from the perspective of user
behaviors, including relation creation, user activeness, user
interaction and tweet content. We quantitatively explore their
correlations with spammer detection and find that tweet content
is the most important factor for spammer detection, followed by
relation creation. Based on these behavior factors, we propose a
novel cascading framework CWB-SPAM for spammer detection
in OSNs. Experiments on dataset crawled from Sina Microblog
show that the proposed algorithm outperforms over all classical
algorithms we investigated in terms of F-score
1
. Experiments also
demonstrate that as a probabilistic classification model, the
proposed CWB-SPAM has a good ranking quality. It enables the
OSN operators to make tradeoff between precision and recall
easily so that the proposed algorithm can be used in different
scenarios. Besides, we also note that the proposed framework can
be used in other probabilistic binary classification models and
thus applied in more scenarios.
Keywords—social network; user behavior; spammer detection;
cascading framework
I. INTRODUCTION
As social networks gain tremendous popularity in the past
years, spammers start to utilize OSNs as a new platform to
conduct their malicious behaviors. According to a survey of
Harris Interactive, 80% of OSN users have received unwanted
friend request, messages and postings in their OSN accounts in
the past years, which means spammers are now very popular in
social networks [1]. Many users have complained about their
terrible experiences in OSNs where they have to face with too
much spam. Moreover, spamming in OSNs has harmful effects
on both OSN users and operators. Virus, phishing or malwares
contained in spam can always lead to users' treasure loss or
privacy disclosure. Users may then reluctantly choose to leave
and therefore the population of OSNs suffers a decrease.
Considering the great impact spammers in OSNs bring,
both industry and academia have been making efforts to detect
them. In industry, in order not to mistake normal users for
spammers, OSN operators choose to apply simple strategies
1
F-score refers to the harmonic mean between precision and recall.
with high precision like URL blacklist and single-feature-
threshold classifiers [2], although they have a much lower
recall. Since these approaches exploit few aspects of behaviors,
spammers can adapt their behaviors to evade detection easily.
In academia, researchers notice that spammers will form tightly
connected communities in OSNs [3,4]. They then propose
detection approaches exploiting this social-graph characteristic.
However, all these approaches rely on the assumption that the
network is fast-mixing, which has been validated to be untrue
in large scale OSNs like Facebook, Youtube, etc[5]. Recently,
researchers start to detect spammers by training classifiers on
behavior features using traditional classification algorithms like
Random Forest. However, experiments in this paper reveal that
these traditional algorithms don’t work well for spammer
detection in OSNs. In summary, although much work has
already been done by academic and industry communities,
spammer detection remains an unsolved problem. Actually,
according to research report published by NexGate, during the
first half of 2013, there has been a growth of 355% in spam in
OSNs including Facebook, Twitter, Google+ and Youtube [6].
In this paper, we propose and evaluate a novel algorithm
CWB-SPAM for spammer detection in OSNs. We build a
classifier based on features extracted from various aspects of
user behaviors so as to make the detector more solid, as it
makes it much tougher for spammers to adapt their behaviors.
To improve the classifier's precision and recall, we propose a
cascading framework, in which we train a classifier using all
data and a second classifier on instances difficult to classify.
We then use these two classifiers to detect spammers in a
cascading framework. Besides, we design CWB-SPAM as a
probabilistic model so that OSN operators can easily make
tradeoff between precision and recall in different scenarios.
Experiments show that the proposed algorithm achieves a
precision and recall of about 90%. In addition, by changing the
parameters in CWB-SPAM, we can achieve a precision of 95%
with a recall of 81% or a higher recall with a lower precision.
In summary, we frame our contributions as follows:
After investigating various aspects of user behaviors,
we find that spammers distinguish from normal users
in behavior patterns of relation creation, user
interaction, tweet content and user activeness. We
find that they all have predictive power but cannot be
used independently to detect spammers effectively.
ISBN 978-3-901882-58-6 © 2014 IFIP
下载后可阅读完整内容，剩余8页未读，立即下载
weixin_38700240
- 粉丝: 2
- 资源: 976
我的内容管理
展开
最新资源
- WebLogic集群配置与管理实战指南
- AIX5.3上安装Weblogic 9.2详细步骤
- 面向对象编程模拟试题详解与解析
- Flex+FMS2.0中文教程：开发流媒体应用的实践指南
- PID调节深入解析：从入门到精通
- 数字水印技术：保护版权的新防线
- 8位数码管显示24小时制数字电子钟程序设计
- Mhdd免费版详细使用教程：硬盘检测与坏道屏蔽
- 操作系统期末复习指南：进程、线程与系统调用详解
- Cognos8性能优化指南：软件参数与报表设计调优
- Cognos8开发入门：从Transformer到ReportStudio
- Cisco 6509交换机配置全面指南
- C#入门：XML基础教程与实例解析
- Matlab振动分析详解：从单自由度到6自由度模型
- Eclipse JDT中的ASTParser详解与核心类介绍
- Java程序员必备资源网站大全
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
