Cantonese rumor detection method based on graph convolutional network and external knowledge embedding

The invention discloses a Cantonese rumor detection method based on a graph convolutional network and external knowledge embedding, which comprises the following steps of: firstly, collecting Cantonese rumor data containing original tweet, forward tweet and comments on a social network platform, strictly performing manual annotation, and constructing a relatively perfect structured Cantonese rumor data set; secondly, constructing a directed heterogeneous knowledge graph, and obtaining external knowledge embedding by using a heterogeneous graph convolutional neural network; then, using BERT to extract text features of the tweet, and obtaining correlation vectors of the tweet text and external knowledge through a comparison network; in addition, a bidirectional graph convolutional neural network is used for obtaining structural features in the rumor spreading process; and finally, constructing a Cantonese rumor detection model BGK based on a bidirectional graph convolutional neural network embedded with externa