Detection Methods for Sockpuppet Accounts on Reddit – A Systematic Literature Review

Abstract

Reddit, one of the largest and most diverse social media platforms, has witnessed a significant rise in the presence of malicious accounts, including trolls, spammers, and sockpuppets. The timely and accurate detection of these accounts is essential to maintain the platform's integrity and user experience. A large range of techniques is used to detect sockpuppet accounts, including machine learning algorithms such as Support Vector Machines (SVM), Random Forest, and Neural Networks, as well as rule-based approaches and natural language processing techniques. Optimized Feed Forward Neural Networks obtained best precision (92%) and recall. SVM's had the ability to handle imbalanced datasets, a common issue in troll detection. The generalizability of SVM also contributes to its impact, as it can be trained on diverse datasets from various online platforms and still yield consistent and high-performing results, making it the most reliable detection method. By understanding the details and effectiveness of various detection methods, this study contributes to the development of more reliable approaches to tackle the growing challenge of malicious accounts on Reddit.