Revealing the Blackmarket Retweet Game: A Hybrid Approach

Abstract

The advent of online social networks has led to a significant spread of important news and opinions. In the case of Twitter, the popularity of a tweet is measured by the number of retweets it gains. A significant number of retweets help to broadcast a tweet well and makes the topic of the tweet popular. Individuals and organizations involved in product launches, promotional events, etc. look for a broader reach in their audience and approach blackmarket services. These services artificially provide a gain in retweets of a tweet as the retweets’ natural increase is difficult and time-consuming. We refer to such tweets as collusive tweets. Users who submit their tweets to the blackmarket services gain artificial boosting to their social growth and appear credible to the end-users, leading to false promotions and campaigns. Existing methods are mostly centered around the problem of detection of fake, fraudulent, and spam activities. Thus, detecting collusive tweets is an important yet challenging problem that is not yet well understood.

In this paper, we propose a model that takes into account the textual, retweeters-centric, and source-user-centric characteristics of a tweet for an accurate and automatic prediction of tweets submitted to blackmarket services. By conducting extensive experiments on collusive tweets’ real-world data, we show how our model detects tweets submitted to blackmarket services for collusive retweet appraisals. Moreover, we extract a meaningful latent representation of collusive tweets and their corresponding users (source users and retweeters), leading to some exciting discoveries in practice. In addition to identifying collusive tweets, we also analyze different types of collusive tweets to evaluate the impact of various factors that lead to a tweet getting submitted to blackmarket services.