Detecting Spam on Twitter via Message-Passing Based on Retweet-Relation - Technologies and Applications of Artificial Intelligence - page 57

a phenomenon that spammers can inveigle victims easily through social media
platforms.
In 2009, the first revolutionary malware called KOOBFACE successfully and
continuously propagated through social networks [1]. This malware using fraud-
ulent messages with harmful link that taking victims to a site which uploads
malware to victims' computers. In the past few years, many researchers fo-
cused on detecting spam accounts individually by using effective account fea-
tures. Stringhini et al. [16] proposed some features based on the result of data
analysis, including the following and follower ratio, URL ratio, and content sim-
ilarity between tweets. These features can be evaded. For example, the following
and follower ratio can be easily manipulated by follower Yang et al. [24] pro-
posed Criminal account Inference Algorithm (CIA) based on the closeness of
social relationships. However, friend relations can be easily forged through the
follower markets.
A spam post on social media can potentially reach thousands or millions of
users. According to the “2013 State of Social Media Spam” research report from
Nexgate, they got the following key findings [12]:
- Only 15% of all social spams have a link that can be detected as spam, so it
is harder to detect social spam.
- During the first half of 2013, there has been a 355% growth of social spam.
Therefore finding spam tweets with the spam collusion mask becomes an im-
portant issue.
2 Background and Related Work
2.1 Types of Social Spam
Social media contain many kinds of social spams. The most common ones are
link spam and text spam.
Link spam : This type of spam may be just a link with no surrounding text.
Curious and unsuspecting users click on this link, then the users will be redirected
to a malicious website. This website may contain advertisements or malicious
software which spammers can profit from it.
Text spam : Spam content is usually very charming. Some text spam request
the recipient to respond to the spammer via a private message in order to obtain
detailed information [12].
2.2 Spammers on Twitter
In social network sites, following and follower counts may influence the ranking
of tweets by search engines [7]. Thus, spammer began to acquire more following
and follower counts. Yang et al. [24] conducted in-depth analysis of spam ac-
counts community on Twitter. They found out that spam accounts community
was made up of spam accounts and spam collusion accounts community. Spam
collusion accounts are those users who have close relationship with spammers.