Abstract: Naive Bayes is a kind of classifier based on probability and used for the prediction of individual categories with the prior probability and conditional probability of each category. Targeting at the current “negative information all over the Internet” phenomenon, the paper offers an early warning model based on Naive Bayes method. Different from general text classification, emotion recognition aimed at large-scale network information mainly focuses on words with subjective emotiosn and requires high execution efficiency. To solve these problems, we conducted the corresponding optimization, such as extracting Emotional Tendency Stop Words List, detailing the management of negative words, and taking 20000 Twitters as sample to test the effectiveness of the model on text emotion recognition. Experiments showed these strategies have ideal execution efficiency and accuracy.

Key words: Negative information, Sentiment analysis, Machine learning, Naive Bayes, Public opinion monitoring, Early warning