[Objective] Improve the accuracy and timeliness of Weibo rumor event detection .

[Methods] A method of Weibo rumor event detection based on hierarchical semantic feature learning model is proposed. Firstly, the semantic features of a single text in an event are extracted based on the BERT pre-training model; Secondly, the event propagation data is dynamically divided based on the time domain, and the convolution neural network is used to learn the semantic correlation features of the text sets in each time domain; Then, the semantic correlation features in each time domain are input into deep bidirectional gated recurrent neural network to learn the deep semantic temporal features in the process of event propagation; Finally, the attention mechanism is integrated to make the model pay more attention to the part with rumor feature in semantic temporal features.

[Results] Experimental results on Weibo public data sets show that the detection accuracy of the model reaches 95.39%, and the detection delay is within 12 hours.

[Limitations] The model requires a certain amount of forwarding and commenting information, so the detection effect is not outstanding when the event is not hot.

[Conclusions] The hierarchical semantic feature learning model realizes the learning process from local semantics to global semantics, thus improving the performance of Weibo rumor event detection.