EASTAR: Event-Anchored Semantic and Task-Adaptive Region Representation Learning
Organizations: College of Applied Science, Shenzhen University, Shenzhen, China · School of Artificial Intelligence, Shenzhen Technology University, Shenzhen, China
Abstract
Urban region representation learning supports a wide range of urban analytics tasks by encoding regional functions and interactions, with human mobility serving as a central signal because it directly captures inter-region interactions. However, most mobility-based methods learn from aggregated origin-destination matrices, where trips driven by different activity purposes are mixed into purpose-agnostic interaction patterns. Although static urban attributes such as POIs and land use can enrich region-level semantics, they do not explicitly ground mobility interactions in activity-purpose signals. As a result, downstream adaptation is often performed over entangled embeddings, making it difficult to select task-relevant functional semantics. To address these limitations, we propose EASTAR, an event-anchored and task-adaptive framework for urban region representation learning. EASTAR treats offline activity events as sparse semantic anchors and uses them to factorize aggregated mobility representations into multiple purpose-related semantic channels. It further introduces free channels to capture latent mobility patterns not covered by event data and a global channel to preserve overall spatial structure. To adapt the learned channel space to different downstream objectives, EASTAR designs a task-adaptive prompt router that initializes task-channel relevance from LLM-derived priors and refines the routing weights through downstream supervision. Experiments on two real-world urban datasets across four prediction tasks show that EASTAR consistently outperforms representative baselines, achieving up to 39.15% reduction in MAE. Ablation and interpretability analyses further demonstrate the effectiveness of event anchoring, semantic channel construction, and task-adaptive routing.
Figures & tables
| Semantic Anchor | Associated Activity Categories |
|---|---|
| Tech | Career/Business, Tech |
| Social | Cars/Motorcycles, Dancing, Games, Singles, Socializing |
| Arts | Fashion/Beauty, Fine Arts/Culture, Writing, Hobbies/Crafts, Movies/Film, Music, Photography, Sci-Fi/Fantasy |
| Health | Health/Wellbeing, Outdoors/Adventure, Support, Fitness, Sports/Recreation |
| Education | Book Clubs, Education/Learning |
| Community | Movements/Politics, Parents/Family, Community/Environment, Pets/Animals |
| Dataset | NYC | CHI | Source |
|---|---|---|---|
| Regions | 180 | 77 | Open Portal [ 41 , 42 ] |
| POIs | 24,496 | 57,891 | OpenStreetMap [ 38 ] |
| POI categories | 26 | 26 | OpenStreetMap [ 38 ] |
| Land-use categories | 11 | 12 | OpenStreetMap [ 38 ] |
| Taxi trips | 10,953,879 | 3,381,807 | Open Portal [ 41 , 42 ] |
| Satellite images | 180 | 77 | Google Maps [ 43 ] |
| New York City | Crime | Check-in | Service Call | Population | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | |||||
| MVURE [ 3 ] | 67.9 1.1 | 0 93.8 1.9 | 0.591 0.016 | 306.7 0 8.2 | 499.6 12.9 | 0.627 0.019 | 1428 33 | 2180 0 46 | 0.367 0.027 | 2814 67 | 3564 0 79 | 0.545 0.008 |
| MGFN [ 7 ] | 70.2 2.3 | 0 89.6 2.5 | 0.630 0.020 | 292.6 17.1 | 451.8 28.1 | 0.690 0.040 | 1554 81 | 2286 115 | 0.303 0.069 | 2822 42 | 3706 0 98 | 0.509 0.026 |
| ReCP [ 31 ] | 83.1 2.4 | 108.7 2.2 | 0.459 0.022 | 246.9 0 4.3 | 400.1 23.6 | 0.761 0.029 | 1516 66 | 2199 0 49 | 0.356 0.029 | 3305 90 | 4353 153 | 0.322 0.047 |
| HAFusion [ 5 ] | 56.1 1.3 | 0 76.1 2.2 | 0.734 0.015 | 202.8 0 7.2 | 322.8 12.6 | 0.844 0.012 | 1273 20 | 1951 0 27 | 0.493 0.014 | 2497 50 | 3277 0 82 | 0.616 0.019 |
| ComSRE [ 32 ] | 72.3 1.1 | 0 99.6 6.5 | 0.545 0.059 | 251.7 17.1 | 434.8 44.4 | 0.716 0.057 | 1459 69 | 2254 0 98 | 0.324 0.058 | 2964 95 | 3916 100 | 0.452 0.028 |
| Stage | Embedding Learning | Downstream Task | ||
|---|---|---|---|---|
| Method | NYC | CHI | NYC | CHI |
| MVURE | 12 | 5 | 0.02 (0.0005) | 0.04 (0.0010) |
| MGFN | 35 | 42 | 0.01 (0.0006) | 0.04 (0.0011) |
| HAFusion | 26 | 16 | 0.02 (0.0005) | 0.01 (0.0005) |
| ComSRE | 14 | 11 | 0.02 (0.0006) | 0.01 (0.0003) |
| UrbanCLIP | 506 | 874 | 59 (0.0028) | 64 (0.0032) |
| Variant | Crime | Check-in | Service Call | Population |
|---|---|---|---|---|
| HAFusion | 0.734 | 0.844 | 0.493 | 0.616 |
| HAFusion M | 0.744 | 0.857 | 0.505 | 0.620 |
| MGFN | 0.630 | 0.690 | 0.303 | 0.509 |
| MGFN M | 0.632 | 0.652 | 0.407 | 0.488 |
| HREP | 0.681 | 0.700 | 0.398 | 0.571 |
| HREP M | 0.660 | 0.667 | 0.368 | 0.536 |
| Variant | Crime | Check-in | Service Call | Population |
|---|---|---|---|---|
| EASTAR-POI | 0.778 | 0.874 | 0.508 | 0.584 |
| EASTAR-Random | 0.731 | 0.768 | 0.435 | 0.519 |
| EASTAR | 0.847 | 0.949 | 0.658 | 0.744 |
| Variant | Crime | Check-in | Service Call | Population |
|---|---|---|---|---|
| FlexiReg / ToPT (SOTA) | 0.751 | 0.899 | 0.497 | 0.701 |
| EASTAR-DeepSeek | 0.832 | 0.941 | 0.606 | 0.705 |
| EASTAR-Qwen | 0.821 | 0.928 | 0.661 | 0.710 |
| EASTAR-Gemini | 0.806 | 0.940 | 0.599 | 0.704 |
| EASTAR-GPT | 0.847 | 0.949 | 0.658 | 0.744 |