Research Of Key Technology And Prototype Realization Of News Authenticity

With social media becoming a mainstream communication channel,the amount of information far exceeds the ability of human review.At the same time,ensuring the effectiveness of information is the key to safeguarding social public order and good customs,so curbing the spread of fake news has become an urgent problem to be solved.However,the large number and miscellaneous forms of fake news make it difficult to regulate effectively.The traditional method is based on probability model,which is difficult to establish,high cost of technology landing and unsatisfactory effect.Focusing on news authenticity identification,this thesis takes multi-feature fusion method as the main research method and builds news identification model and image identification model based on deep learning method.(1)Multi-feature fusion news authenticity identification technology,"multi-feature" refers to news text features,news image features and news behavior features."Fusion" refers to early fusion after features are extracted.Specifically include: first,the news entity is divided into news content and news behavior,and news content can be divided according to pictures and texts.News image features,news text features and news behavior features are extracted separately through different feature extraction methods,and then classified after feature fusion.Second,for news image information,a feature extraction network is built.See the authenticity identification technology of news image mentioned below.Thirdly,in the feature fusion stage,attention mechanism and multiscale channel mechanism are introduced to ensure efficient adaptation to different size features.Fourthly,in the construction of news behavior network,Pearson correlation coefficient is introduced to measure user similarity based on the network representation learning method.(2)Authenticity identification technology of news images is a model of feature extraction in multi-feature fusion news authenticity identification technology.It mainly includes the following aspects: First,the structure is built based on the deep separable network and residual jump structure to ensure that the effective features of news images can still be extracted from the shallow feature extraction network.Secondly,regularization feature selection is carried out for news images,which reduces the attention of the model to some features,and finally improves the overall attention efficiency of the model.Third,for the news image data set,slight random conversions are performed,including scaling,rotation,horizontal flipping,brightness and hue changes,to improve generalization and robustness.(3)Based on the lightweight framework of Python Flask,the multi-feature fusion news authentication network proposed in Chapter 3 and Chapter 4 and the image authenticity authentication network based on separable convolutional network are designed and implemented in a systematic way.This system fully verifies the two algorithms proposed in the first two chapters through three modules: user management module,model and data set management module and news data management module.