Loading...

Detecting Telegram Channels with Fake Members

Mohammadi, Behnam | 2019

652 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 52109 (44)
  4. University: Sharif University of Technology
  5. Department: Management and Economics
  6. Advisor(s): Aslani, Shirin; Talebian, Masoud
  7. Abstract:
  8. Nowadays, social media and messenger applications have found widespread usage in peoples’ lives, making them spend numerous hours on these services. As a consequence, it’s imperative for businesses to maintain presence and carry out advertising campaigns in these social networks. Telegram, which started out as a messaging application, has now grown into a prevalent social media platform in many countries. As a result, Telegram channels demonstrate high potential in publishing advertisements, furthering the promotion of products, and even asserting influence on political, societal, and economic situations. The cost of advertising on channels depends on the number the channel’s subscribers (i.e. members). As a result, some channel admins (i.e. moderators) are known to artificially increase the number of subscribers by adding fake members to their channels. Since only admins have access to the list of the subscribers of their channels, detecting channels that have employed such a scheme may not be directly possible. To solve this problem, we propose an indirect approach based on Cluster Analysis. By engineering and choosing a promising set of variables, selecting the right clustering algorithm with apt parameters, and by training our model on a random sample of channels which are labeled as fake or not fake, we were able to detect three main clusters of channels, one of which included fake channels. Performing the same method on all channels in our dataset yielded similar clusters and a precision of 84% based on the expert’s judgment. Since our method relies on the historical data of only three variables (subscribers, views, posts), it can very well be used in practical applications to detect fake Telegram channels, improving the ROI of marketing campaigns in Telegram
  9. Keywords:
  10. Anomaly Detection ; Abnormal Behavior ; Fake Users ; Cluster Analysis ; Agglomerative Clustering ; Hierarchical Clustering

 Digital Object List

 Bookmark

No TOC