15
IEEE Network • July/August 2018
0890-8044/18/$25.00 © 2018 IEEE
AbstrAct
While mobile social networks (MSNs) enrich
people’s lives, they also bring many security issues.
Many attackers spread malicious URLs through
MSNs, which causes serious threats to users’ pri-
vacy and security. In order to provide users with a
secure social environment, many researchers make
great efforts. The majority of existing work is aimed
at deploying a detection system on the server and
classifying messages or users in MSNs through
graph-based algorithms, machine learning or other
methods. However, as a kind of instant messaging
service, MSNs continually generate a large amount
of user data. Without affecting the user experience,
with existing detection mechanisms it is difficult to
implement real-time detection in practical applica-
tions. In order to realize real-time message detection
in MSNs, we can build more powerful server clusters
or improve the utilization rate of computing resourc-
es. Assuming that computing resources of servers
are limited, we use edge computing to improve the
utilization rate of computing resources. In this arti-
cle, we propose a multistage and elastic detection
framework based on deep learning, which sets up
a detection system at the mobile terminal and the
server, respectively. Messages are first detected on
the mobile terminal, and then the detection results
are forwarded to the server along with the messag-
es. We also design a detection queue, according
to which the server can detect messages elastically
when computing resources are limited, and more
computing resources can be used for detecting
more suspicious messages. We evaluate our detec-
tion framework on a Sina Weibo dataset. The results
of the experiment show that our detection frame-
work can improve the utilization rate of computing
resources and can realize real-time detection with a
high detection rate at a low false positive rate.
IntroductIon
In recent years, MSNs such as Twitter, Facebook
and Sina Weibo have become important plat-
forms for people to obtain information, spread
information and make friends [1]. Twitter's month-
ly active users (MAU) were 200 million in 2012,
and the figure rose to 328 million in 2017, with
20 million tweets being posted every hour. While
MSNs enrich people's lives, some security issues
have emerged. Attackers spread attacks through
MSNs, such as phishing, drive-by download, mali-
cious code injection and so on. According to new
research, up to 15 percent of Twitter accounts are
in fact bots rather than people [2].
Malicious URLs are one of the most com-
mon methods used by attackers to initiate cyber
attacks [3]. Attackers trick users into clicking
malicious URLs, clicking pictures containing mali-
cious URLs, scanning QR codes with malicious
URLs, and so on by disguising themselves as well
known accounts, advertisements of discounted
merchandise, or by using mutual trust between
friends. In these ways, attackers lure victims to a
phishing website for phishing attacks, or embed
malicious software in the victim's computer to
control the target host or perform an APT attack,
which will cause huge losses to individuals, busi-
nesses, governments and organizations. Many
MSNs use blacklist techniques to filter URLs sent
by users, such as Google Safe Browsing, Phishing
Tank, URIBL, and so on. However, there is often a
delay in blacklist technology, and research shows
that 90 percent of victims click on malicious URLs
before they are blacklisted [4].
In order to provide users with a secure MSN
environment, researchers have proposed many
strategies to deal with online social network
attacks. Existing detection methods are mainly
divided into two categories. The first category
includes detection algorithms based on the rela-
tion graph of social networks. Many kinds of rela-
tions exist in social networks, so researchers use
relations in social networks to build a social graph.
By analyzing the characteristics of the user’s loca-
tion in the graph, a detection algorithm can be
designed based on the graph to identify suspi-
cious messages or users. The second category
includes detection methods based on machine
learning algorithms. Researchers extract features
from social network data such as users’ personal
information, social behaviors, relationships with
friends, message content and so on, then use
machine learning algorithms to train classifiers to
identify malicious messages or users.
Although researchers have proposed a variety
of methods, there are still some shortcomings in
them. Many graph-based detection methods are
done offline, because they often need to build a
large graph structure, and the calculation over-
head and time cost are very large, which makes it
hard to apply them online. For methods based on
machine learning, if we use some time-consuming
features, it is difficult to realize real-time detection,
such as closeness centrality, betweenness central-
ity and so on. In addition, some behavior-based
detection methods use the abnormal behavior of
accounts to detect suspicious accounts. Abnormal
behavior can be detected only after it occurs, so
Multistage and Elastic Spam Detection in Mobile Social Networks through Deep Learning
Bo Feng, Qiang Fu, Mianxiong Dong, Dong Guo, and Qiang Li
EXPLORING DEEP LEARNING FOR EFFICIENT AND RELIABLE
MOBILE SENSING
Bo Feng, Qiang Fu, Dong Guo and Qiang Li are with Jilin University; Mianxiong Dong is with the Muroran Institute of Technology.
Qiang Li is the corresponding author
Digital Object Identifier:
10.1109/MNET.2018.1700406