| News is an important way for people to obtain information,but news is often interspersed with fake news to spread.With the development of the Internet in recent years,the spread of fake news has become more rampant,so fake news detection has become an urgent research task.So far,the research on fake news has been divided into four directions,namely,knowledge base-based,writing style-based,position-based and dissemination-based.On social networks,fake news often uses distinct writing style features in order to attract traffic and guide public opinion.Based on these differences,certain research results have been produced.The traditional method based on writing style features is Take news text as a text classification problem and split it into two parts:feature engineering and classifier.However,the existing methods have the following two problems in the process of feature engineering construction: First,the selection of writing style features is incomplete.In the existing practice,some textual style features,such as keywords and symbols,are generally selected,while other textual style features are ignored.There are many types of writing style features,and each feature is of different importance to fake news detection.However,related research methods do not consider such issues,but directly use statistical methods to conduct statistics and then model.Although this type of model integrates the characteristics of writing style for research,it cannot exert the greatest advantages of various characteristics for fake news detection.So far,fake news detection methods have achieved certain research results,but few detection tools have been implemented.Aiming at the problem of incomplete selection of writing style features,this research uses statistical methods to further mine other writing style features that are effective for fake news detection.Aiming at the problem that the diversity of writing style features is not considered,this research studies the attributes of various writing style features,and classifies the writing style features according to their respective attributes according to the word dimension,sentence dimension and article dimension,and then combines the text semantic information to This research studies the relationship between writing style features and fake news,and proposes a fake news detection method based on multi-dimensional writing style features.In the model construction,the Text-CNN model is used to extract semantic information,and the Attention mechanism is added to weight the writing style features of each dimension,so as to indicate the importance of each writing style feature to fake news,making it more reasonable to deal with fake news.Make predictions.This method achieves better experimental results on the weibo dataset,with an F1 value of 86.95%.Aiming at the problems of single detection model in the pre-order method and the context information of the sentence is not considered in the sentence dimension,based on the fake news detection method based on multi-dimensional writing style features,this research presents a multi-channel fake news detection method based on mixed features.News detection methods.This method draws on the idea of multi-granularity image feature extraction,divides the fake news text into three granularities of words,words and sentences,and then uses a multi-channel neural network to extract the semantic information features of each granularity respectively.In sentence granularity feature extraction,Bi-LSTM is added to extract the context information of sentences.Then use the fusion method based on DCA algorithm to fuse each granular semantic information feature.This method achieves an F1 value of 89.89% on the weibo dataset.Aiming at the problem of insufficient implementation of fake news detection tools,this research designs and implements an automatic detection tool for fake news based on a multi-channel fake news detection method based on mixed features.The tool is built using the Python-based Flash lightweight web framework and uses a My SQL database for data storage.After demand analysis,four main functions are designed,namely fake news detection,rumor clue submission,hot news display and other rumorrefuting platform entrances. |
