摘要：

Abstract:

【Purpose/significance】With the continuous evolution of public health emergencies, people's understanding of them has a

process from vagueness to accuracy. The author quantified and graded the credibility of online information of public health emergen⁃

cies to provide data support for more detailed rumor identification.【Method/process】Five static features of the information text,

namely keywords, emotion, comment, information source and media, and two dynamic features, namely time and the number of newly

confirmed cases on the same day, were selected, the rumor index RI was quantified with the entropy method. Based on this, the "toler⁃

ance interval" was introduced, and the boundary was determined by the naive Bayes classifier. The reliability of rumor identification

results was divided into three categories: low, medium and high.【Result/conclusion】The model performed well in the training set and

the verification set, with the accuracy of 95% and 90.20%, respectively. Compared with the decision tree and SVM baseline models,

the model's performance indexes were significantly improved.【Innovation/limitation】In this study, a rumor identification model based

on information credibility evaluation was established. RI index was established and dynamic indicators related to the epidemic situa⁃

tion were innovatively incorporated to improve the accuracy of rumor identification. The credibility classification of rumor is intro⁃

duced to break through the limitation of the traditional binary classification detection method of rumor identification.