Early detection of fake news based on multiple information features

Early detection of fake news based on multiple information features

연구 분야: Safety

학회: DSIT 2021: 2021 4th International Conference on Data Science and Information Technology

As the popularity and influence of online social media continue to expand, automatic detection of fake information has attracted widespread attention. The detection method based on a kind of information characteristic cannot ABSTRACT sufficient characteristic information in the early stage of information dissemination, so the accuracy rate is usually low in the early detection of information dissemination. In response to the above problems, this article proposes a model composed of three modules: ABSTRACT, SCORE and CONCATENATE. ABSTRACT uses CNN and GRU neural networks to extract the propagation path and text features of the information. SCORE mines user features based on user behavior, and CONCATENATE detects the information after integrating the above features. Experimental results show that the model can detect fake information with an accuracy of 93.7% within 30 minutes of information dissemination. Compared with the benchmark method, ASC has achieved a better balance between accuracy and timeliness.