According to index redundancy and high complexity of online public opinions which are not conducive to supervision, a comprehensive evaluation model based on factor analysis and SVM is proposed, in this method, 14 indicators for online public opinions are reduced into three common factors by factor analysis, then 5-fold cross of genetic algorithm is used to optimize SVM parameters in the simplified index system, the early warning model for online public opinion crisis by using genetic algorithm to optimize SVM is set up, finally, two kinds of SVM are improved into one-to-many algorithm to classify four cases, as a result, the early warning on online public opinions is obtained. The empirical analysis of 10 online public opinion events in 2019 shows that the early warning error is lower than 0.51 percent, which reveal that the model is feasible and which strengthen the supervision on online public opinions. Factor analysis reduces the complexity of index system, 5-fold cross of the genetic algorithm improves the learning ability of SVM classifier, thus, the model can more accurately predict training set, one-to-many algorithm makes classification speed more quickly, which provide the help for the supervision on online public opinions.