[Objective]The operation of modern police service relies on a large amount of alarm text data. How to intelligently classify the massive alarm text data has become a topic of concern in the field of public security.
[Methods]A BERT-DPCNN-based text classification model is proposed for the alarm text classification task. The BERT pre-training model is used to generate text word vectors, and the activation function and dynamic learning rate are also modified in the DPCNN to improve the performance of classification.
[Results]Experiments are carried out for the comparison of BERT-DPCNN with six other models, which are BERT, BERT-CNN, BERT-RCNN, BERT-RNN, BERT-LSTM, and ERNIE, respectively. The results show that BERT-DPCNN achieves the best performance in terms of accuracy, recall, and precision. In the binary classification task, the accuracy of BERT-DPCNN reached over 98%, and in the eleven classification task, the accuracy reached over 82%, verifying the effectiveness of the proposed model.
[Limitations]Due to the large number of model parameters and limited iterations, further testing is still needed to fully evaluate the performance of the proposed model.
[Conclusions]The BERT-DPCNN-based text classification model effectively improves the accuracy of alarm text classification and provides a certain data support for public security agencies to analyze and judge police situations.