We introduce Mobility-Embedded POIs (ME-POIs), a dynamic, mobility-informed framework that allows AI models to understand the temporal activity rhythms of places over time, significantly improving predictions about real-world attributes like opening hours, price levels, and busyness. While world-class language models like Gemini are proficient at processing text, their geospatial representations can be enriched by incorporating the real-world functional dynamics of the urban environment.
Rather than treating a place as a frozen set of words, ME-POIs uses a self-supervised approach to blend text descriptions with large-scale, anonymized mobility patterns from public benchmarks. A three-step pipeline — visit alignment, spatial multiscale visit propagation, and text-mobility synergy — transforms raw geographic points into a mathematical signature encoding both a place's identity and its dynamic functionality. The framework also addresses the "long tail" problem of data sparsity by statistically transferring aggregated visit patterns from busy, data-rich neighbors to nearby sparse places across multiple spatial scales.
Integrating ME-POIs with advanced text models delivered up to an 81.9% relative gain in predicting visit intent, a 75.1% improvement in price level classification, and a 24.7% increase in busyness estimation accuracy across unseen places. Extensive testing across two culturally distinct metropolitan areas confirmed that adding real-world mobility data provides a noticeable "context advantage".
Source: Google AI Blog · Summarized by HeadlinesBriefing