Knowledge-Graph NLP Assistant for Mitigate Norm Violations: Live Chat Moderation Study

The rise of real-time communication through live chat and social media platforms has heightened the need for effective content moderation tools to detect and mitigate norm violations, such as hate speech, abusive language, and offensive slang, thereby ensuring user safety. This study develops and evaluates a knowledge-graph-based natural language processing (NLP) framework aimed at improving the accuracy and efficiency of content moderation in live chats. Several machine learning models were explored, including K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Neural Network (NN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM), along with their graphenhanced variants. The models were trained and evaluated on a comprehensive dataset including various norm violations. Results show that the Graph-Bi-LSTM model outperforms the others, achieving an F1-score of 0.84, precision of 0.84, and accuracy of 0.81, indicating its potential to improve harmful content detection. Despite these promising outcomes, the study underscores the need for human oversight to prevent false positives and ensure context-aware decision-making. The findings suggest that combining graph-based NLP models with human moderation tools can enhance the transparency and reliability of content moderation in real-time. Further research should optimize the inference time of models and continuously adapt the system to handle evolving language trends and diverse cultural contexts.