Early Detection of Fake News Based on Pre-training Representation and Broad Learning
In order to achieve early detection of fake news,a method based on pre-training representation and broad learning was proposed. Firstly,the news text was input into the RoBERTa large-scale pre-training language model to obtain the contextual semantic representation of the corresponding news text.Secondly,the obtained contextual semantic representation was fed into the feature nodes and enhanced nodes of broad learning. By leveraging these broad learning nodes,both linear and non-linear features were extracted from the news text,enabling the construction of a classifier for predicting the authenticity of the news. Finally,comparative experiments were conducted on three real datasets,and the results demonstrated that the proposed method was capable of detecting fake news within 4 h with an accuracy rate exceeding 80％,surpassing the performance of the baseline method.
early detectionfake newspre-training representationbroad learningtext classification
万方数据
维普
