Ecosystem of spamming on Twitter: Analysis of spam reporters and spam reportees

Abstract

Lately, there has been a growing trend in the Internet space particularly among the Online Social Media (OSMs) platforms like Twitter, Facebook etc which are becoming huge repositories of information. This information, by design, is posted by users of these websites and consequently, this information is vast, un-organized, unreliable and dynamic. It is commonly observed that along with genuine users, a lot of activity is seen from spammers or users with the intent of spreading malicious or irrelevant content. In our work, we focus on spamming activity on Twitter. Spamming activity in Twitter is can typically be reported by its users, who we refer as reporters and those who indulge in spamming activities are referred as reportees. We collected data of suspected spammers, i.e. reportees as well as of the users who reported them, i.e. reporters. Thereafter, we classified them into various categories and tried to study the ecosystem of these reportees and reporters. We used three data mining techniques i.e., decision tree , K-nearest neighbors and random forest classifier for the classification tasks. Finally, we have compared these three algorithms on the basis of their accuracy.

A Semi-Supervised Learning Approach for Tackling Twitter Spam Drift

“…Advertisers or promoters are types of spammers who use Twitter to publicize themselves. These advertisers could be a company, an organization, or an individual (Sinha et al 2016). Also, some of the individual advertisers use Twitter for selling and buying followers.…”

Detecting spams in social networks using ML algorithms - a review

“…From his work, SVM showed a high performance of F-measure about 72%. (Ribeiro et al, 2015) 90% for spam - (Yang et al, 2013) 77% for twitter and yelp - (Alsudais et al, 2014) 90% for reporters and 84% for reportees - (Sinha et al, 2016) 96.07% - (Meda et al, 2014) 93.6% F-measure - (Chen et al, 2015c) Decision tree High 96.51% - (Ribeiro et al, 2015) 99% for IDS - (Jalil et al, 2010) 87.6% for spam - (Yang et al, 2013) 92% for spam reporters and 90% for spam reportees - (Sinha et al, 2016) 92% F-measure for C4.5 - (Chen et al, 2015c) Naïve Bayes High 86.63% - (Ribeiro et al, 2015) 70.9% F-measure - (Chen et al, 2015c) K-NN Average 84% for reporters and 89% for reportees - (Sinha et al, 2016) 90.5% F-measure - (Chen et al, 2015c) SVM Average 88.75% - (Du and Fang, 2004) 57% around for IDS - (Jalil et al, 2010) 79.9% F-measure - (Chen et al, 2015c) Bayes network Average 83.3% for spam - (Yang et al, 2013) 81.9% F-measure - (Chen et al, 2015c) Xu et al (2016) introduced a new point of view to efficiently detect spam in social networks. He collected two types of datasets from the Twitter and Facebook using the application programming interface, which contains both spam and non-spam contents.…”

“…By the comparison with above result, the author showed that all classifiers are shown a high performance with only TSD and FSD than the mixed spam and he also concluded that random forest classifier gives a better outcome and high performance than other classifiers. Sinha et al (2016) proposed a study on spamming activity on Twitter using K-nearest neighbour and also he compared the results using the performance obtained by decision tree C5.0 and random forest. The author used the names known reporters and reportees, who reports spamming activity is a reporter and who involved the cause for spamming is reportee.…”

Research of Distinct Algorithm of Short Text Based on Simhash

“…Simhash algorithm [4,8] has been proposed by Charikar in Google. It converts two text records into two n-bit fingerprint.…”

“…The analysis of online crime network ecosystem has been researched by Sinha [4] et al They tried to use social relationships and semantic information to find more junk accounts based on some seeds account of the computer crime. Maclean [5] believed that the opening transmission was the main reason for more and more garbage data.…”