Online public opinion events occur frequently. The effective governance of online public opinion is in an increasingly important position. This study creatively constructs an Opinion Lexicon to categorize and assign scores to words in major news commentaries. The model utilized in the article is based on KNN algorithm and BP neural network. The usefulness of the model is to screen and score samples of news comments from three representative media platforms, namely “People’s Daily Online”, “IFENG.com” and “Paper”. Through data processing and result comparison of the samples, the study can correctly predict the value of news opinion guidance and the scores assigned to comments. On the one hand, the study wants to provide an effective way for public opinion governance of Internet News. On the other hand, it promotes other news media to think about the key and initiatives of online opinion guidance from the perspective of social opinion orientation.