Detecting and characterizing anomalous followers on social media

Official URL: https://risc01.sabanciuniv.edu/record=b2865358_(Table of contents)

This paper aims to detect anomalies in target social media accounts. Previous work on behalf of this topic includes bot detection applications with different types of methods. However, Anomaly detection is a more general framework that encapsulates social bots. Capturing anomalies starts with specifying possible anomaly types that can be seen in social media. In this study, we covered collective and point anomaly detection types targeting Twitter. Our algorithm includes features extracted from Twitter and synthetically created features which is the ratio of two or more other features. We create experiments to build an anomaly detection approach that can detect real-world examples. Our study focused on both supervised and unsupervised machine learning models that capture the above anomaly types in an experimental environment. These models contain classifying algorithms. We used several scenarios to understand whether our model will be useful to use in a non-experimental environment, Twitter. We tested these scenarios under an anomaly ratio between 1%-10%.In conclusion, the experiments in this study have the purpose to demonstrate the outcomes of supervised & unsupervised anomaly detection techniques can capture these anomalous accounts.