Fine-Grained Sentiment Analysis of Cross-Domain Chinese E-Commerce Texts Based on SKEP_Gram-CDNN

This study aims to use pre-trained models and an improved DPCNN model to extract useful information for sentiment analysis in an e-commerce dataset by combining a general domain text dataset. However, owing to feature distribution differences between text data from different domains, the feature information obtained from a general domain text dataset may contain ambiguities and lead to a scarcity of target domain data, thereby increasing the training error and decreasing the model performance. To address these issues, this study proposes the “SKEP_Gram-CDNN” model for fine-grained sentiment analysis of cross-domain Chinese e-commerce comments. The model introduces the ERNIE_Gram+DPCNN_att model as a generator and the capsule network as a discriminator to construct a cross-domain Chinese e-commerce sentiment analysis model. The performance of the discriminator model was validated on the ASAP_ASPECT dataset released by Meituan to demonstrate its superiority. Moreover, experiments were conducted on the SE-ABSA16_PHNS and SE-ABSA16-CAME target datasets to compare the proposed model with the ERNIE_Gram-CDNN model, which not only proved the effectiveness of the proposed model but also provided favorable methods for cross-domain fine-grained sentiment analysis.