spatial multi-scale visit propagation

Spatial multi-scale visit propagation is a technique that fills in the activity patterns of a place with very little visit data by borrowing mobility data from nearby places instead.

Think of a small shop that just opened in a neighborhood. It has too few visitors to know when it's typically busy. In this case, you can look at data from the shop next door, the surrounding block, and the wider neighborhood as a whole, moving from the closest scale outward, to make an estimate like "this area tends to be busy at lunch and quiet at night." This step-by-step process of pulling in information from neighbors at increasingly wider scales to fill the gaps is exactly what this technique does.

The concept appears as a processing step in a place-understanding framework released by Google Research. Famous landmarks or large stores have plenty of visit records, making it easy to extract patterns from them. But small shops on neighborhood side streets have so few records that reliable patterns are hard to build — a classic "long-tail" problem. Spatial multi-scale visit propagation is used as a device to compensate for this lack of data.

The term appears in the article as a processing step within a place-understanding pipeline, in a passage like: "At the spatial multi-scale visit propagation and text-mobility synergy stage, this mobility signal is combined with text-based representations to merge them into a single embedding." A common misunderstanding is thinking this is a standalone product or a separate AI model — it is actually just one computational step inside a framework called ME-POIs.