Optimization Study of a Pornographic Content Detection Model for Live Streaming Platforms Based on Improved TF-IDF and SVM: A Case Study of Bullet Screens

Along with the rapid development of the webcasting industry, the amount of information interaction on live platforms is growing explosively, and the problem of pornography-related issues on live platforms is becoming more and more serious, and the reports of pornography-related pop-up texts account for more than 60% of the total, which is difficult to be recruited by the traditional detection techniques, with limitations of similar keyword matching and high cost and poor interpretability of the deep learning model. Combining the improved TF-IDF algorithm and SVM model to develop a pornographic content detection model can effectively deal with these problems. The model reached an accuracy level of 85.3%, and the recall rate reached 82.4 percentage points, which significantly reduced the cost of manual review and the number of reports of pornographic pop-ups. The results of this achievement put forward an innovative solution for the live broadcasting platform to detect pornographic content and effectively promote the healthy development and cleaner network environment of the live broadcasting industry.