Ai-powered Twitter Content Moderation Using Sbert, Cnn And Bi-lstm With Severity-based Alert System

Abstract

Social Media Platforms Such As Twitter (now X) Generate Large Volumes Of Real-time Content, Including Spam, Malicious Links, And Harmful Messages. The Rapid Spread Of Such Content Poses Significant Challenges For Manual Moderation Due To High Data Velocity And Evolving Patterns Of Misuse. This Paper Proposes An AI-powered Twitter Content Moderation System That Integrates Semantic Analysis, Deep Learning, And Rule-based Validation For Effective Detection Of Harmful Content. The Proposed System Consists Of A Multi-stage Pipeline. Ini-tially, A URL Threat Detection Module Is Employed To Identify Suspicious Links Such As Malicious Domains And Shortened URLs. Cleaned Tweet Content Is Then Processed Using Sentence-BERT (SBERT) To Generate Semantic Embeddings. These Embeddings Are Passed Through A Hybrid Deep Learning Model Combining Convolutional Neural Networks (CNN) For Local Feature Extrac-tion And Bidirectional Long Short-Term Memory (Bi-LSTM) Networks For Capturing Contextual Dependencies. The Model Classifies Tweets Into Categories Such As Malicious, Spam, And Non-spam. To Enhance Reliability, Additional Decision Layers Including Confidence Threshold Checks And Rule-based Overrides Are In-corporated To Handle Uncertain Predictions And Known Spam Patterns. Furthermore, A Severity-based Alert Mechanism Is Im-plemented To Trigger Real-time Notifications For High-risk Content. Experimental Evaluation Demonstrates That The Proposed Hybrid Model Improves Classification Accuracy, Robustness, And Real-time Applicability Compared To Traditional And Standalone Approaches.

Keywords