Comparison of social bot detection techniques

Abstract

Online Social Networks act as a major platform for communication. The origin of social bots is one of the consequences of increasing popularity and utilization of social networks by people. A social bot is an automated application that clones the behavior of a human and creates a faux impression on real users. The Social bot can be classified as either benign and malicious based on their actions. Benign bots are used to perform tasks a lot quicker than humans, sharing vital information like weather reports, etc. Whereas, malicious bots begrime the social media with false information and may also be involved in malicious activities such as spamming, stealing private information, creating noise within the conversations, etc. This nature of bots led to the necessity of social bot detection techniques.

Various social bot detection techniques have been proposed based on different algorithms. In this research, proposed social bot detection techniques are reviewed and several of them are implemented. A comparison of these techniques based on their input requirements, approach, and accuracy is performed. The implementation of the techniques has been applied to three completely different data sets collected from the Twitter social network. Four metrics: precision, recall, accuracy, and Cohen's Kappa coefficient are calculated using the results obtained by implementing the techniques. These metrics have been used to decide the efficiency of techniques and provide a comparison of them.