Feature Selection-based Integrated Learning Model in Violation Comment Detection

With the rapid development of Internet technology and the increasing popularity of social network media, people often encounter insults, abuse, slander, and other illegal comments on the Internet, which is called “network violence”. These Social phenomena are continuously emerging, leading to vicious incidents. To maintain a healthy network environment, illegal speech must be identified and monitored. It is essential to use technology governance to regulate the network environment and curb cyber violence incidents. Based on the data set of Chinese violation comments, we proposed an ensemble learning model based on feature selection using TF-IDF to extract text features. The feature importance method of random forest was used for feature selection. Then, the grid search method was used to optimize the parameters of multiple classification models. After identifying several models with strong classification performance, they were integrated to obtain the ensemble learning model with the best performance. Compared with other traditional machine learning models, the proposed model showed superior detection ability and identified illegal comments.