Abstract: This paper integrates the text mining technology of sentiment dictionary and machine learning, and uses the linguistic appraisal system to compare and analyze the discourse on emotions of English version of Sino-US Enterprises' CEO's letters.The research finds that:1) the accuracy of combining sentiment dictionary with machine learning algorithms are better than that of using sentiment dictionary alone.2) Chinese companies have a more positive sentiment, while American companies have a more neutral sentiment.In the letters, American companies not only show their own social responsibility performances, but also focus on the situation of their external suppliers and communities.3) different economic systems, cultural features and thinking patterns of Chinese and American enterprises lead to obvious differences in sentence structure, discourse theme and discourse distribution in the letters.This study can provide reference for Chinese enterprises to write CEO's letters in accordance with the characteristics of the other party's native language and culture, thus promoting the norms of enterprise information disclosure, and shaping a good image of Chinese enterprises, and also provide guidance for business English teaching.