Research And Implementation Of A Social Network Rumor Detection System Based On Multi-Feature Fusion

With the development of society and the progress of the Internet,social networks have gradually become the main way for us to obtain information.The news of modern social networks spreads extremely fast,and through them,the situation of various events is sent to us at the first time.But at the same time many fake news are also spread rapidly,which provide the ground for reactionary activities,shake people's confidence,or increase the unnecessary work burden of staff.In social media,the content of fake news is updated and spreads quickly.Currently,Sina Weibo mainly relies on manual audit to detect rumors,which is relatively inefficient,and has lagging problems,which consumes time and labor costs.Automated fake news detection can well alleviate these pressures and is therefore of great importance.The carriers of disinformation are mainly text,and the means of propagation are mainly retweets and comments in social networks.Previous works have been based on blog post text and multimedia,and also on propagation trees or temporal features,and few works have considered event-level features associated between blog posts.Another persistent challenge in disinformation detection is how to detect whether previously unseen information is not disinformation.To address these two issues,this paper combines text features with event features and proposes a migratable fusion model,and experiments prove that the model works better than previous work in unseen event detection,and the main work in this paper is as follows.To address the problems of large data volume,repetitive texts,unclassified events described by texts and different domains to which the events belong,this paper uses the unsupervised text clustering method of single-pass clustering,which embeds a text vector for each text based on the TF-IDF method,determines the text similarity by the vector cosine similarity,and puts the similarity into the single-pass clustering algorithm.And by extracting the key phrases of determined fake information to match the unknown text to achieve the purpose of saving computational resources and improving the accuracy rate.1)To address the problems of large data volume,repetitive texts,unclassified events described by texts and different domains to which the events belong,this paper uses unsupervised text clustering method of single-pass clustering,based on TF-IDF approach to embed text vectors for each text,and the similarity of the clustering algorithm is based on the vector cosine similarity.The similarity of the clustering algorithm is based on the vector cosine similarity,and the key statements are extracted to match the unknown text to achieve the efficiency and accuracy.2)To address the problems of different textual representations of the same event and weak detection ability for unknown events,i.e.,generalization ability,this paper proposes a fusion detection model.The model uses natural language processing models to extract the textual information of blog posts,graph neural networks to mine the event-level features of the text,and generative adversarial networks to enhance the generalization performance for integrated detection of false information.3)To address the needs of fake information detection in real social networks,this paper designs a set of online false information detection,retrieval and analysis tools based on social network false information.