The strategies using transit-oriented development (TOD) to optimize transportation sustainability have been implemented in many metropolitan areas and extended beyond the role of exclusively offering transit services. Research findings from existing literature have largely shown that metro station catchment areas can attract a substantial number of urban functions and human activities that lead metro stations to be vital and vibrant places of urban daily life. In this work, we propose a data-driven semantic framework to characterize metro stations through points of interest (POIs) in Hong Kong. The analytical results reveal four thematic topics of urban functions that are closely related to commercial, residential, tourism, and industrial activities. Given the implementation of a hierarchical clustering approach on these thematic topics, the similarities among different stations are investigated. In particular, metro stations in the same thematic group tend to be spatially concentrated, suggesting an evident geographical proximity relating to similar urban functions. Plus, results from the Multinomial Logit Model (MNLM) confirm that the surrounding built environment of metro stations has close relationships with the heterogeneity of urban functions. Ultimately, this study introduces alternative insights into the urban functional heterogeneity exhibited by metro station areas, and the practical implications for more targeted TOD strategies are discussed. •Latent Dirichlet allocation (LDA) is used to identify the thematic topics of urban functions for each station catchment area based on their POI types.•Four thematic topics are uncovered that mainly relate to commercial, residential, tourism, and industrial functions and activities.•Stations containing similar urban functions and relevant activities tend to be spatially concentrated.•Urban functional patterns within station catchment areas indicate strong connections with surrounding built environments.•The analytical findings can be used for formulating more targeted TOD strategies