Bibliometric techniques and social network analysis are employed in this study to evaluate 14,955 papers on air pollution and health that were published from 2001 to 2021. To track the research hotspots, the principle of machine learning is applied in this study to divide 10,212 records of keywords into 96 clusters through OmniViz software. Our findings highlight strong research interests and the practical need to control air pollution to improve human health, as evidenced by an annual growth rate of over 15.8% in the related publications. The cluster analysis showed that clusters C22 (exposure, model, mortality) and C8 (health, environment, risk) are the most popular topics in this field of research. Furthermore, we develop co-occurrence networks based on the cluster analysis results in which a more specific keyword classification was obtained. These key areas include: "Air pollutant source", "Exposure-Response relationship", "Public & Occupational Health", and so on. Future research hotspots are analyzed through characteristics of the cluster groups, including the advancement of health risk assessment techniques, an interdisciplinary approach to quantifying human exposure to air pollution, and strategies in health risk assessment.
Keywords: air pollution; bibliometric; co-occurrence network; health; machine learning; social network analysis.