Chinese Text Classification Based on ERNIE-RNN

Text classification is a popular task of natural language processing. At present, text classification has been applied to multiple language like English, Chinese, Arabic et.al. However, Chinese text classification has many challenges especially in feature extraction and feature selection. This paper proposes the structure of ERNIE-RNN model. We use ERNIE to learn the complete semantic representation of a semantic unit instead of the traditional model to generate the word vector. RNN model is attached to the ERNIE model which can better process sequence information especially when preceding input is related to the following input for some certain task like text classification. In this paper, we use the data from our project which is tax consulting questions correspond to tax application scenarios. Our experimental results showed solid performance of all models on our data with a top performance of 94.63%, achieved by our model ERNIE-RNN.