Twitter Bot Detection Using Supervised Machine Learning Algorithms

Abstract

Human beings engage in a new way thanks to online social networks. People are becoming interested in ways to attack and exploit them through malicious activities like using them to deceive people. To do something like this, one approach is creating Twitter bots. A considerable amount of social media content is generated by automated accounts known as social bots. Content can be developed which is normal, and even helpful. But it could be designed to have a different purpose, such attracting a specific group to influence beliefs or raise someone's profile with their success even though it's not accurate. However, recent machine learning algorithms that have proven to be an improvement in accurately identifying between real users and bot accounts. The aim of this paper is to build a model using supervised machine learning algorithms to detect bot tweets. The tweets we are going to work with is only written in Arabic and specially bots that try to sell products. These tweets are collected by us. We manually labelled 2969 tweets into two classes, human “real users” and bot. Then, a pre-processing step is done by cleaning the dataset from several noises such as punctuation marks, Arabic diacritics, and repeated letters. We used three machine learning algorithms: Support vector machine, Naïve Bayes, and K nearest neighbours. All models yield an overall accuracy above 95 percent, which is relatively high compared to similar studies in the field. Support vector machine achieve the highest accuracy of 98.20 percent, followed by Naïve Bayes with accuracy of 97.86, and K-Nearest Neighbours with accuracy equal to 96.18 percent. We additionally tested a small new dataset consist of 5977 randomly selected tweets. We used the Support vector machine model to test the data. The model predicted 3671 tweets as human and 2306 tweets as bot.