Extracting Region Function Representations through Compressed Trajectory Embeddings

In the context of smart cities, accurately dividing urban areas into functional zones based on residents’ mobility patterns and the distribution of Points of Interest (POIs) is essential. These zones exhibit dynamic changes over time, influenced by varying crowd dynamics and purposes [1] . To address this, we’ve introduced a novel approach extracting region function representations through compressed trajectory embedding (CTE). The CTE model extracts representations of urban functional zones from crowd trajectories, effectively compressing region features. Figure 1 provides an overview of our region representation through compressed trajectory embedding method. We firstly create two networks based on the functional RotatE [2] : a General POI network and zone networks, based on POIs and region-traveling trajectories. The General POI network captures global POI distribution to generate a general function representation for each category. Region networks focus on static and dynamic function representations of region, considering temporal aspects. Then we compare them with general POI function representations to analyze the effects between compressed trajectory embeddings and region functions. Finally, we apply our CTE method to two tasks to validate the effectiveness and applicability of region representation through compressed trajectory embedding.