Trajectory-Based Market Structure in Local Services Markets: A Heterogeneous Graph Transformer Approach

Market structure in local services markets — segmentation, pairwise competition, and the network and spatial structure of the market — is shaped by consumer routines, not by firm attributes alone. We extend the heterogeneous graph transformer (HGT) with temporal edges to integrate offline trajectory data with online reviews, restaurant attributes, and neighborhood demographics, and embed consumers and restaurants jointly in a continuous latent space. Applied to Pittsburgh dining in 2019 (9,723 consumers, 2,504 restaurants, 272,119 Yelp reviews), the framework recovers a coupled demand-supply segmentation, a pairwise competition matrix that diverges substantively from cuisine-, distance-, co-visitation-, attribute-, and collaborative-filtering-based alternatives, and the network and spatial structure of the urban market. A quasi-experimental case study of one verified market entry validates predicted demand reallocation. The framework supports a hypothesis-generation tool for operators, platforms, and consultants, and is portable to other frequent-occasion service categories.