A literature review of sockpuppet account detection for Facebook

Abstract

Sockpuppet accounts are widely used by people with malicious intents to commit vandalism, fraud, or spread fake information on digital networks. They can get around security precautions and violate the platform's terms of service by employing various strategies. Facebook, a social networking platform, is full of sockpuppet accounts. The main motivations behind creating sock puppet accounts on Facebook are to manipulate public opinion, deceive people, or carry out malicious actions. Facebook currently lacks a detection system for these sockpuppet accounts, as its current detection system is powered by a manual, human driven process. Due to this inefficient detection system, users are able to continue with their malicious actions through the creation of alternate accounts. The current research mainly uses two algorithms - the Random Forest (RF) and Bayesian Network (BN) algorithms - for the purpose of sockpuppet detection. The RF model is generally superior in terms of the F1 scores (RF - 99.8%, BN - 99.6%) and high accuracy being more effective when dealing with sockpuppets on Facebook. However, the BN model is strong when dealing with uncertain or probabilistic data.