A BERT-based Ensemble Model for Chinese News Topic Prediction
Jingang Liu, Chunhe Xia, Xiaojian Li, Haihua Yan, Tengteng Liu · 2020
With the rapid development of big data mining technology in the Chinese commercial field, the news topic prediction becomes increasingly important. Since the accuracy of Chinese news topic classification can directly affect the personalized recommendation effect of the Chinese news system and then affect business profits, the news category prediction performance needs to be higher as possible. With the great success of the BERT model in the past two years, using the BERT model alone has achieved extremely good performance on Chinese text classification tasks. Therefore, using the advantages of the BERT to study more effective methods for the Chinese news classification will become more meaningful. In this paper, we propose a model that combines the advantages of both BERT and the long short-term memory (LSTM) network, named BERT ensemble LSTM-BERT(BERT-LB). Our method is more effective than using BERT alone. This model uses a three-step method to calculate and integrate Chinese news text features. Besides, we use two datasets to evaluate our method and other baseline methods. We demonstrate that the proposed method has the promising ability to predict Chinese news topics and prove its generalization ability.