Building a Generative AI Comment Review System for Content Compliance

With the rapid development of online social platforms, compliance review of comment content faces unprecedented challenges. Efficient and accurate review mechanisms are crucial for maintaining a healthy online environment. This study proposes and implements an intelligent comment review system based on generative AI technology, utilising deep learning models to achieve content comprehension and violation identification. Employing a distributed microservices architecture, the system integrates the BERT-Large pre-trained model for semantic comprehension. It further combines XGBoost and GPT-3.5 to achieve precise identification of non-compliant content and generate moderation recommendations. Experimental results demonstrate the system's stable performance when processing large-scale comment datasets, achieving an accuracy rate of 94.8%. In practical social platform applications, it can process over 10 million comments daily. Implementation outcomes reveal a 62.5% reduction in manual review costs, a 400% increase in processing efficiency, and significantly enhanced detection accuracy for spam and harmful content. This provides efficient and reliable technical support for online content governance.