Text classification of electricity policy information based on BERT-optimized TextRNN
Zhiyong Liu · 2022
In order to improve the effective query of power policy information, we design and implement text classification of power policy information. Aiming at the problem of sparse news headline features and strong context dependence in power policy news text classification, a semantic enhancement-based power policy news text classification method is studied. In this paper, based on the traditional TextRNN, we use the BERT language model instead of the traditional Word2Vec model for semantic representation; at the same time, for the LSTM network used by the traditional TextRNN, the Bi-LSTM network is used to extract feature vectors, and then connect the Attention layer to get fusion. The feature vector is finally input into Softmax to get the model classification result. The final experimental results show that the TextRNN network optimized based on BERT has a better classification effect.