Fake News Detection Method Based On Prompt Learning And Associated Feature Enhancemen

With the rapid development of the Internet,the number and scope of fake news have been continuously expanding,making it increasingly difficult for people to obtain genuine information.Fake news detection tasks can effectively help people filter out false information,improving the accuracy and credibility of information.However,in practical applications,this task faces challenges such as high annotation costs,limited samples,poor training results,and weak model generalization capabilities.Furthermore,with the diversification of news dissemination,news information has shifted from a single-text format to a multimodal composition comprising text and images.Therefore,to comprehensively capture news information features,multimodal fake news detection has become increasingly important.However,due to the significant differences between different modal data,there is insufficient correlation between features,which increases the difficulty of fake news detection.To address these challenges,this paper proposes a fake news detection algorithm based on prompt learning and correlated feature enhancement,which separately targets unimodal and multimodal data to effectively improve the accuracy and generalization capability of fake news detection.Experiments on relevant datasets were conducted,and the results verified the effectiveness and feasibility of the proposed method.The main contributions of this paper are:(1)A fake news detection algorithm based on knowledge-enhanced prompt learning is proposed.To address the issue of insufficient sample data in real-world scenarios,this research proposes a fake news detection algorithm based on knowledge-enhanced prompt learning.This algorithm uses the T5 model-generated prompt templates,converting the fake news detection task into a prompt learning-based fake news detection,and introducing external entity knowledge features to enhance the detection capabilities.Through well-designed prompts,the latent knowledge of the model is effectively activated,improving fake news detection performance in low-sample scenarios and enhancing the model’s generalization capabilities.In addition,experiments were conducted on the Politi Fact and Gossip Cop datasets,with the results indicating good performance on both datasets,demonstrating the effectiveness and feasibility of the algorithm.(2)A multimodal fake news detection algorithm based on correlated feature enhancement is proposed.To address the issue of insufficient feature correlation due to differences between different modal data,this research proposes a multimodal fake news detection method based on correlated feature enhancement.The algorithm uses attention mechanisms to enhance the correlation between modal features,effectively improving the fusion of multimodal data.The algorithm is divided into four modules: text feature extraction,image feature extraction,attention mechanism,and classifier.First,rich semantic text features are extracted using the BERT model,followed by the Text-CNN model further filtering noise and enhancing feature representation.Then,the attention mechanism-introduced Res Net-51 model captures and weights the most correlated image features with text features.Finally,the text and image features are combined into a high-information multimodal representation as input for the fake news detection classifier.Experiments were conducted on the Weibo and Twitter datasets,with the results demonstrating the effectiveness of the proposed method.