An Improved Chinese Text Classification Ensemble Model

Autor: Zehong Hong, Jun Liang, Songsen Yu, Zhaofeng Zeng
Rok vydání: 2021
Předmět:
Zdroj: ICDSP
Popis: This paper gave an improved ensemble model for Chinese text classification based on CNN and RNN structure. We first changed the bidirectional LSTM structure in the TextRCNN model to a bidirectional GRU structure and added a learning rate reduction mechanism into it. Then TextCNN model was also added into a learning rate reduction mechanism. Finally, our ensemble model was constructed through equal weight voting between the improved TextRCNN and improved TextCNN models. Experimental results showed that compared with TextCNN [7], TextRCNN [8], LSTM+attention [15], DPCNN [5] and FastText [6], the classification error rate of our model on SINA_P News dataset dropped by 8.36%, 13.2%, 20.23%, 7.33% and 29.17% respectively; dropped by 8.28%, 3.98%, 10.68% and 5.86% and rose by 3.8% on THUC_P news dataset. Hence, our ensemble model has higher classification accuracy than most other text classification models and is a competitive method.
Databáze: OpenAIRE