How mobility gives language models a deeper understanding of place
AI models excel at understanding text but need physical-world awareness. A new framework, ME-POIs, blends language models with anonymized mobility data—arrival times, stay durations, and movement patterns—to capture places’ dynamic rhythms, not just static metadata like categories. Using self-supervised learning, it creates richer embeddings combining identity and functionality. This method boosts performance significantly, improving visit-intent prediction by 81.9%, price-level classification by 75.1%, and busyness estimation by 24.7%, marking a leap toward context-aware urban AI.
