Smart Detection: A System for Identifying BOT Activities on Social Media

Abstract

The increasing number of automated bot accounts on social networking platforms such as Twitter creates a daunting challenge as these bots help in the spread of fake news, influence public perception and shift the online narrative. Banishment of frauds is crucial within the realm of social networks and involves the recognition and differentiation between human and bot accounts. The research seeks to propose an advanced machine learning framework for bot detection on Twitter that integrates heuristics approaches, sentiment analysis and ensemble learning approaches. The procedure includes the extraction of user profiling, content and metadata of the tweet such as sentiment polarity and subjectivity required for the determination of the unique features that separate bots from humans. The class imbalance issue was rectified during data normalization using SMOTE (Synthetic Minority Over-sampling Technique) so as to create a more just environment for training. The ensemble model consisted of Random Forest, XGBoost, Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Gradient Boosting which were enrolled as classifier in the process so as to increase the quality of the classification process. The model accuracy, precision, recall, F1-score and confusion matrix evaluation were all included in assessing the models performance. The approach we used was able to classify whether an account was a bot or user with accuracy of 95.36%,03 precision of 93.47%, recall of 94.65%, and F1 of 94.06, thus proving effectiveness of the method used for the classification. To address the arising issue of bot attacks, it can be seen that the combination of sentiment analysis and supplementing it with advanced ensemble learning techniques can significantly improve the detection and mitigation of such social media attacks.