The purpose of this study is to improve prediction rate with a novel model of bidirectional encoder representations for transformers (BERT) compared with random forest algorithms. A dataset of size 1100 is used to compare Novel BERT’s performance with Random Forests. With Random Forest, a framework for identifying fake news in electronic media networks is proposed. clinical calculates a sample size of 20 according to the framework. With regard to Precision rate, the Novel Bert algorithm beats the Random Forest algorithm by 8.33%. In comparison to the random forest algorithm, BERT achieves a rate of 0.002 that is significantly better than it. It is concluded that the novel BERT algorithm outperforms Random Forest in the prediction of fake news in this study.