An Improved Deep Belief Network for Chinese Emergency Recognition
Haoran Yin, Jinxuan Cao, Luzhe Cao, Guodong Wang · Journal of Physics Conference Series · 2020
Abstract Aiming at the defects that the RBM module in DBN can only re-represent information but cannot extract information features, and can only handle one-dimensional data, the DBN network is improved, and a Conv-DBN model is proposed to recognize emergencies. First, the text corpus is preprocessed, and the word vector matrix generated by Word2Vec is used as input, and then the word vector features are extracted through the visible layer integrated into the convolution operation. Word vector features are used as the input of the next layer. Finally, every layers are fine-tuned through back-propagation at the top layer. The softmax function is used to activate, and the recognition result is output. Simulation results show that the method proposed in this paper has improved accuracy and recall, and F value is better than other methods.