Store agglomeration in a commercial district is considered to follow synergistic and complementary relationships between stores. The micro-interactions, that is, co-occurrence relationships, between stores are thought to create commercial districts’ universality and characteristics. This study aims to empirically identify latent attributes that can explain the mechanism of store agglomeration from the co-occurrence relationships between stores in commercial districts, and to identify universal characteristics in store agglomeration. We represented store agglomeration as a store co-occurrence network with stores as nodes and co-occurrence relationships between stores as links in 10 major commercial districts in Japan. We estimated latent attributes of stores that generated the store co-occurrence network and empirically clarified the quantitative and qualitative nature of store agglomeration. The co-occurrence networks of stores with estimated latent attributes were compared across those of districts, and the co-occurrence patterns common to latent attributes were clarified. The estimated latent attributes were empirically shown to have more explanatory power for the store agglomeration than the conventional business category classification, suggesting their usefulness as a new classification axis for stores. In addition, by comparing 10 districts, common co-occurrence relationships were extracted, and the universal spatial structure of store agglomeration was empirically clarified.