Revealing the research differences of AI between China and the U.S using semantic deviation

China and the United States are recognized as leading forces in Artificial Intelligence (AI) research, with distinct research inclinations within their communities. Understanding the research differences between these two nations is crucial for grasping the global AI landscape, especially for revealing its collaborative division of labor and competitive situation. This paper moves beyond traditional methods reliant on frequency statistics and topic analysis by introducing an innovative approach that highlights the semantic deviation, which can help differentiate the details of research preference of a given research concept in two countries. We construct a matrix that includes two dimensions: research scale and semantic deviation, positioning each research concept into four areas including Discrepant Research, Interest-Vary Research, Consensus Research and Scale-Gap Research. Based on which, we conducted co-word network analysis to explore the research differences of China and U.S. on macro level, and utilized semantic field analysis to further explore its details in the case of “Face Recognition” at micro level. We found that in AI research between China and the U.S., the research scale difference is not significant for over 90 % of all domain entities, but 37.5 % of entities show a clear semantic deviation. The high-frequency entities that represent popular research issues also show the same results. Our findings indicate that AI researchers from both countries have a relatively consistent level of attention to the vast majority of domain concepts, yet there is still a significant difference in the content preferences between the two nations in terms of research being conducted. Our framework enables a thorough examination of research differences with various types, providing valuable insights into the distinctive research profiles and competition advantages in AI between China and U.S.