Identification of source and sink points of population flow based on POI-KG embedding

The source and sink points of urban population flow represent key areas of population aggregation and dispersion, crucial for optimizing resource allocation and urban development planning. Identifying these points heavily relies on high-quality population flow data, which presents challenges due to data acquisition difficulties and limited coverage. This study proposes a method using knowledge graph embedding that incorporates the semantic and spatial information of points of interest (POI) to represent factors influencing population flow, allowing for source and sink point identification with minimal reliance on population flow data. First, a knowledge graph is constructed to represent factors like socio-economic location, semantic aggregation, semantic similarity, and spatial aggregation of POIs influencing population flow. Then, POI semantic and spatial information is aligned in a unified space to construct the integrated semantics and spatial information-knowledge graph embedding (ISS-KGE). By integrating total population flow and net inflow/outflow rates, the urban grid is divided into strong sources, strong sinks, and stable points. Weakly supervised learning is applied using ISS-KGE to identify these types within the urban grid. Experimental results, in Wuchang District, Wuhan, show that ISS-KGE achieves an F1-score of 0.69 in identifying population flow source and sink points, outperforming baseline models Word2Vec and GloVe.