A Small amount of Labeled Data Chinese Online Course Review Target Extraction via ALBERT-IDCNN-CRF Model. (November 2020)

• Record Type: Journal Article
• Authors: Min, Liang; Miao, Xianglin; Bi, Peng; He, Feijuan
• Is Part Of: Journal of physics. Volume 1651(2020)
• Publication Date: 2020-11
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1651/1/012049
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Aspect sentiment analysis of online course reviews is of great significance in helping users choose courses and improve course quality. Review target extraction is particularly important as the basis of aspect sentiment analysis. Because the current models mostly rely on a large amount of annotation data, there is fewer relevant research on the extraction of online course review targets with higher annotation costs. This paper proposes an ALBERT-IDCNN-CRF review target extraction model for a small amount of labeled data. First, using ALBERT pre-trained sentences obtained dynamic model Chinese word vector coding; Simultaneously, using ALBERT pre-trained model of Transformer obtain sentence abstract features. Then, abstract features are input into the dilated convolutional neural network (IDCNN) to reduce the number of neuron layers and parameters. Finally, conditional random field (CRF) is used to decode and annotate the review sentences to extract the appropriate review objectives. The experimental results on the school online real Chinese online course review data set show that our model has achieved better results than existing models.