In recent years, network fraud as a new form of fraud, to the social and economic development and the safety of citizens’ property has caused a greater threat. This paper proposes a computerized detection technique for network text fraud by design, so as to reduce the success rate of network fraud. This paper classifies the text information of network fraud, and improves the text recognition method through the method of knowledge distillation, and constructs a lightweight fraud text recognition model based on just distillation. Through performance test experiments, detect the utility of this paper’s method to detect fraudulent text. The F0.5 mean values of this paper’s lightweight fraudulent text recognition based on knowledge distillation are 0.72, 0.67, 0.73 on the online fraudulent text training set, validation set and test set, respectively, which are significantly better than other detection models. The accuracy of this paper’s method on fraudulent text classification is greater than 0.8, which clearly outperforms other text classification models. All in all, the method in this paper comes out on top in both classification and detection of online fraudulent text, with better results.
