Place2Vec: Visualizing and Reasoning About Place Type Similarity and Relatedness by Learning Context Embeddings

Abstract

Understanding, representing, and reasoning about points of interest (POI) types is a key aspect of geographic information retrieval, location-based services, and knowledge graphs. POI-type similarity and relatedness are important for query expansion if there is no direct-matched candidate for a query. Current place-hierarchy representation is mainly derived from a top-down expert design perspective which may not capture holistic geospatial semantics from real-world POI datasets. In this demo, we illustrate how to learn the POI-type embedding representations from spatial contexts and a data-driven perspective, and to visualize the corresponding POI-type similarity and relatedness from such embeddings.