Real-Time Detection of Fake Social Media Accounts Using CNN-BERT And XGBOOST Ensemble

Authors:

V.Anvith, S.Sathvika, S.Yashwanth, E.Ramu

Department of Computer Science & Engineering Kakatiya Institute of Technology and Science Warangal, Telangana, India

Dr. T. Ranjith Kumar

Associate Professor,Department of CSE Kakatiya Institute of Technology and Science Warangal, Telangana, India

Abstract—There is an increasing challenge of fake accounts on the social media platforms, which propagate misinformation, facilitating scams, dismantling the trust of the user. These are automated bots to advanced impersonators, and unless a complex system of rules is used, they are hard to detect. The issue is compounded by the fact that the strategies of bad actors to penetrate an organization are constantly changing, as bad actors keep evolving to avoid detection systems. It provokes a sharp necessity to find a more intelligent, adaptive solution which is able to examine various dimensions of account behavior and content to make the correct decisions.

The proposed project creates an artificial intelligence-based fake accounts detection system that uses machine learning to detect potential fraudulent social media accounts. The system is an amalgamation of natural language processing power and conventional machine learning method to process both textual and behavioral data. The system will be able to distinguish between real and false accounts with a high score by analyzing profile descriptions, follower ratios, post frequency, and account age. Its implementation employs a CNN-BERT-based deep text analyzer and XGBoost to process numerical features, and it creates a strong ensemble to be able to adapt to different kinds of fake accounts, such as spam bots, impersonators, and inauthentic accounts.The solution is implemented as a RESTful API with FastAPI and could be easily integrated with the social media systems or moderation systems already in place. Having the capacity to process real-time accounts and give explainable decisions, this system can be a useful instrument in keeping the platforms intact, safeguarding users against frauds, and contributing to the fact that fake news do not spread in the social networks.

Index Terms—Bidirectional Encoder Represen-tations from Transformers, Convolutional Neural Network, Extreme Gradient Boosting, Application Programming Interface.