Smart Bot Detection for Twitter/X: A Systematic Analysis of Machine and Deep Learning based Methods
Rekha Jangra, Abhishek Kajal · International Journal of Electronics and Communication Engineering · 2025
A significant rise in online social networks has been witnessed over the past decade, particularly on social media network platforms such as X (previously known as Twitter), Instagram, and Facebook accounts, which recorded a steep hike. This further leads to the rise of automated accounts known as bots, devised to replicate the behavior of organic user accounts automatically to disseminate false information or spread spam. For identifying these social bot accounts, many machine and deep learning methods have been proposed on X accounts' features, and based on tweets’ text contents through sentiment analysis for diverse datasets; the leading works have been reviewed in this research article. However, comparing the efficacy of diverse research works in terms of problem statement, methodologies, and various evaluation parameters was extremely difficult. However, the authors put effort into a systematic literature review. It has been observed that SVM followed by RF are the most used ML algorithms for Twitter bot detection, where RF achieved the highest accuracy of 94.87% on account profile features. LSTM is also observed as the most employed DL algorithm for Twitter bot detection. While RoBERTa achieved the highest accuracy of about 98% on the COVID-19 dataset, followed by CNN on the Arabic Spam dataset. The article also summarizes the potential of methods to enhance bot detection performance and scalability over machine and deep learning methods. In addition, authors presented a demonstration with the execution of leading ML and DL methods with the combination of ReLU activation function and Adam optimizer on diverse X datasets. They presented the respective results in tabular form. During simulation with leading ML-based techniques, accuracy yielded for SVM, Naïve Bayes, and Random Forest was 81.38%, 79.47%, and 85.77%, respectively. At the same time, accuracy in the case of DL-based LSTM and Bi-LSTM approaches was 88% and 89%, respectively. Overall, this review article will provide a significant blueprint for future research on enhancing the performance of bot detection models for different online social networks.