Detecting and Characterizing Bot-Like Behavior on Twitter

Abstract

Social media is becoming a platform of choice for people to voice their opinion on topics of discussion. To evaluate these opinions, it is important to have an accurate assessment of who is saying what. Unfortunately, social media are also the home of bots which makes the assessment difficult. Bots are computer programs designed to mimic human behavior online in social networks. They are used to pursue a variety of goals, including, but not limited to, spreading information, and influencing targets.

In this paper, we describe a machine learning framework that uses content-based features extracted from Twitter to detect bot-like behavior on the platform. Unlike other machine-learning approaches to bot detection, we seek to generate explanations of why specific accounts are categorized as bots; thus allow us to modify these criteria as bots’ behaviors evolve. We have therefore developed the criteria mentioned in an article published in Medium [1] to detect bot-like behavior in our dataset then evaluate the results. We then explain the different types of bots that used as our datasets and compare the significant features for each type of bots in a logistic regression method.