A Hybrid Framework for Detecting Automated Spammers on Twitter: Integrating Machine Learning and Heuristic Approaches

Abstract

Twitter's open platform has become a hotspot for automated spammers who exploit its vast user base to spread malicious and misleading content. This paper proposes a hybrid approach to detect automated spammers, integrating machine learning models with heuristic rules to achieve a robust and adaptive detection framework. The methodology leverages a rich set of features, including behavioral attributes such as account age and retweet frequency, and content-based metrics like hashtag density and sentiment polarity. The hybrid model combines the adaptability of a Gradient Boosting Classifier with manually defined heuristic rules, enabling it to address the dynamic nature of spam tactics effectively. The proposed system was evaluated using a dataset of 50,000 Twitter accounts, evenly split between spam and legitimate users. Experimental results demonstrate that the hybrid approach outperforms traditional models, achieving a Precision of 91.2%, Recall of 88.9%, F1-Score of 90.0%, and Accuracy of 91.5%. In comparison, standalone models such as Logistic Regression and Support Vector Machines achieved significantly lower performance metrics. These findings highlight the hybrid approach's superior ability to accurately classify spam accounts while minimizing false positives. Despite its effectiveness, the framework's scalability for real-time applications and generalization across platforms remain areas for future work. Additionally, integrating graph-based features and exploring unsupervised techniques are promising directions to enhance detection of previously unseen spam patterns. Overall, this research provides a robust solution for mitigating the growing threat of automated spammers on social media platforms