Research on Gesar event detection method based on attention mechanism

Inhalt:
In the process of constructing the event graph of Gesar domain, it is necessary to classify the event text of Gesar domain. However, experiments have found that the traditional models of convolution neural network (CNN) and long short-term memory network (LSTM) have weak generalization ability and insufficient semantic feature extraction ability on the text of Gesar events. Therefore, Albert-ARNN model is proposed. The model first uses Albert pre-training language model to obtain the global semantic information of the text, and then uses the BiLSTM model with attention mechanism to obtain the contextual feature representation of the text. The experimental results show that on the Gesar domain text dataset, the F1 value reaches 94.21%, which is 0.81% higher than that of Albert model. Compared with Albert-BiLSTM model, the F1 value is improved by 0.78%.