As widely acknowledged, data science has been explored and applied across numerous scenarios. However, there remains limited understanding of how knowledge in data science is integrated and diffused, particularly in terms of its interdisciplinary nature. In this study, we not only adopted a citation-based strategy to define the "scope" of data science and used bibliometric methods to analyze scientific publications in the field, creating its “bibliometric cartography,” but also employed text analysis techniques to examine the themes of data science papers and their evolutionary process. This multi-layered approach allows us to gain a more comprehensive understanding of the knowledge integration and diffusion patterns of data science across various disciplines. The findings reveal that data science publications draw from various knowledge sources and significantly influence multiple disciplines. Moreover, the diffusion of data science progresses more rapidly during the virality diffusion phase compared to the broadcasting diffusion phase within the scientific community. we emphasize the importance of technical methods in data science and their application in specific domain scenarios. Through the analysis of theme evolution maps, we discovered that the evaluation of model effectiveness serves as a crucial link between the application of technical methods and specific application scenarios. This study offers diverse perspectives that can stimulate theoretical advancements and serves as a crucial resource for policymakers and funding bodies to gain a deeper understanding of data science.