BERT-POS: Sentiment Analysis of MOOC Reviews Based on BERT with Part-of-Speech Information

Wenxiao Liu, Shuyuan Lin, Boyu Gao, Kai Huang, Weilin Liu, Zhongcai Huang, Junjie Feng, Xinhong Chen, Feiran Huang

Abstract

Sentiment analysis and part-of-speech tagging are two main tasks in the field of Natural Language Processing (NLP). Bidirectional Encoder Representation from Transformers (BERT) model is a famous NLP model proposed by Google. However, the accuracy of sentiment analysis of the BERT model is not significantly improved compared with traditional machine learning. In order to improve the accuracy of sentiment analysis of the BERT model, we propose Bidirectional Encoder Representation from Transformers with Part-of-Speech Information (BERT-POS). It is a sentiment analysis model combined with part-of-speech tagging for iCourse (launched in 2014, one of the largest MOOC platforms in China). Specifically, this model consists of the part-of-speech information on the embedding layer of the BERT model. To evaluate this model, we construct large data sets which are derived from iCourse. We compare the BERT-POS with several classical NLP models on the constructed data sets. The experimental results show that the accuracy of sentiment analysis of the BERT-POS is better than other baselines.
BibTeX
@inproceedings{liu2022bert,
  author = {Wenxiao Liu and Shuyuan Lin and Boyu Gao and Kai Huang and Weilin Liu and Zhongcai Huang and Junjie Feng and Xinhong Chen and Feiran Huang},
  title = {{BERT-POS: Sentiment Analysis of MOOC Reviews Based on BERT with Part-of-Speech Information}},
  booktitle = {AIED},
  year = {2022},
  doi = {10.1007/978-3-031-11647-6\_72},
}

Loading related content...

Figure