With the rapid development of communication and computer technology, Internet applications continue to penetrate into all aspects of society, and text, as a carrier for people to directly express their emotions and opinions, occupies a large proportion of network data. In this paper, we choose the comment text as the experimental data and sentiment analysis as the research task, and design a text sentiment analysis model based on graph neural network and representation learning. English is taken as the source language, and six languages, Chinese, French, German, Japanese, Korean and Thai, are respectively taken as the target languages for analysis. The experimental results show that the sentiment analysis model proposed in this paper improves the F1 value of sentiment classification in cross-language environment more, and then compares with the machine translation method and the cross-language sentiment analysis method without sentiment feature representation, which improves the prediction accuracy by about 10.21% and 9.7%, respectively, and has a better performance in text sentiment.