Mining Temporal Activity Patterns On Social Media
Nikan Chavoshi · UNM’s Digital Repository (University of New Mexico) · 2018
Social media provide communication networks for their users to easily create and share content. Automated accounts, called bots, abuse these platforms by engaging in suspicious and/or illegal activities. Bots push spam content and participate in sponsored activities to expand their audience. The prevalence of bot accounts in social media can harm the usability of these platforms, and decrease the level of trustworthiness in them. The main goal of this dissertation is to show that temporal analysis facilitates detecting bots in social media. I introduce new bot detection techniques which exploit temporal information. Since automated accounts are controlled by computer programs, the existence of patterns among their temporal behavior is highly predictable. On the other hand, patterns emerge in human temporal behavior as well since humans follow cyclic schedule. Therefore, we need a solution that can differentiate between these two classes by learning patterns of each. For my Ph.D. dissertation, I focus on the temporal behavior of social media users for the following purposes: 1. to show that high temporal correlation among users is common with automated accounts, 2. to design a system, called DeBot, which detects highly correlated accounts, 3. to improve the time complexity of calculating correlation for real-time applications, and 4. to deploy deep learning techniques on temporal information to classify social media users.