Chinese News Title Classification Model Based on ERNIE-TextRCNN
Qi Wang, Xu Li · 2022
With the rapid development of the Internet, the number of online news are increasing significantly. The efficient identification and classification of news through news headlines is of great significance for news regulation and information gathering. This paper proposes a Chinese news title classification model based on the combination of ERNIE and TextRCNN. It uses the ERNIE model to generate news text title vectors. Through the convolutional layer of TextRCNN, the generated vector is extracted with a bidirectional long short-term memory network. Then, the important parts of the text are obtained through the pooling layer, and finally classified by softmax. The experimental results show that the improved ERNIE-TextRCNN model has an accuracy value of 95.23% and an F1 value of 95.21% on the public THUCNews data set, which is significantly better than other models in evaluation indicators such as accuracy, precision, recall, and F1 value.