Skip to main content
Fig. 2 | Journal of Big Data

Fig. 2

From: KAGN:knowledge-powered attention and graph convolutional networks for social media rumor detection

Fig. 2

The overall framework of KAGN. (1)Post texts encoder uses Word2Vec and CNN to get the posts texts representation. (2)Knowledge distillation module extract entities and concepts sets of posts contents from knowledge graphs. (3)Knowledge attention encoder employs Bi-LSTM, self-attention and multi-head attention to obtain the representations of entities and concepts, which are then fused with posts texts representation using a gate control mechanism (4)Knowledge graphs encoder creates a graph of posts texts, entities, concepts to learn the global knowledge using GCN. (5)The results of (1), (3) and (4) are concatenated and passed to a fully connected softmax layer whose output is the probability distribution over all the categories

Back to article page