Transformer-Based Fine-Tuned Social Bot Detection and Understanding Enhancement

Abstract

The rapid growth of social media platforms has led to the widespread presence of automated accounts, commonly known as social bots, which can manipulate public opinion, spread misinformation, and disrupt online communities. Detecting such bots has become a critical challenge for maintaining the integrity of digital communication. This study proposes a fine-tuned framework for enhancing social bot detection using advanced machine learning and transformer-based classification techniques. The developed system integrates a web-based application using the Django framework to manage user interactions, dataset processing, and prediction services. The dataset is preprocessed by removing missing values, balancing class distribution through resampling, and normalizing features using a standard scaler. Multiple classification algorithms, including Logistic Regression, Support Vector Machine, Random Forest, XGBoost, and an Artificial Neural Network (ANN), are implemented and evaluated to identify the most effective model for bot detection. The system extracts behavioral features such as tweet frequency, follower-following ratio, account age, retweet ratio, and spam content indicators to classify accounts as human users or social bots. Model performance is assessed using accuracy metrics, and the best-performing classifier is deployed for real-time prediction within the application. Experimental results demonstrate that the proposed approach significantly improves detection accuracy and provides an efficient mechanism for identifying suspicious accounts. The framework supports scalable deployment and offers practical assistance in combating automated malicious activities on social media platforms. The implementation details of the classification pipeline and prediction workflow are demonstrated through the developed Django-based system.