Cross-Domain Sentiment Analysis of Fusion Generative Adversarial Networks and Transfer Learning

Sentiment analysis models are often difficult to work well in cross-domain scenarios due to the differences in language expressions in different domains or platforms. To address this problem, this study proposes a cross-domain sentiment analysis and transfer learning method that incorporates generative adversarial networks (GANs). The method first aligns the text in the source and target domains using a GAN generator to eliminate the expression differences between the two domains. Then, based on the idea of transfer learning, the trained sentiment analysis model in the source domain is fine-tuned in the target domain to realize cross-domain sentiment analysis. The experimental results show that the classification accuracy of the GAN algorithm is between 95% and 99% in the product test of Taobao software, and it can classify evaluation quality very accurately. The cross-domain sentiment analysis method integrating GAN and transfer learning can effectively mitigate the negative impact of language expression differences between different domains or platforms.