Research map / Graph learning / Graph self-supervised learning
MotifGPL: Motif-Enhanced Graph Prototype Learning for Deciphering Urban Social Segregation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MotifGPL, a self-explainable graph prototype learning framework that links network motifs to urban social segregation in Beijing and uses motif distributions to guide urban graph reconstruction.
Problem
Urban social segregation (racial, residential, income) is complex and prior studies rely on surface-level indicators, lacking interpretable analysis of urban structure and mobility; the goal is to identify structural patterns driving segregation and guide reducing it.
Approach
Combine graph prototype learning on a spatial graph and an OD mobility graph with motif distribution discovery (random-walk local structures matched to statistically overrepresented motifs), then reconstruct the urban graphs toward motif patterns associated with lower segregation.
Key results
- MotifGPL achieves 0.7990 accuracy and 0.7976 F1 on segregation index classification, above GAT+Prototype (0.7896) and GCN (0.7872).
- Motif-guided reconstruction at alpha=0.8, beta=0.3 reduces Global Moran's I of the spatial graph from 0.4159 to 0.4043 while changing under 2% of edges.
- Ablation removing the spatial graph drops accuracy to 0.7212 versus 0.7990 for the full model.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add the OD graph to MotifGPL (ablation) | MotifGPL · Beijing segregation index classification | accuracy: 0.7776$\pm$0.0005 -> 0.7990$\pm$0.0005 | Helped |
| Add the spatial graph to MotifGPL (ablation) | MotifGPL · Beijing segregation index classification | accuracy: 0.7212$\pm$0.0002 -> 0.7990$\pm$0.0005 | Helped |
| Add street view features to MotifGPL (ablation) | MotifGPL · Beijing segregation index classification | accuracy: 0.7834$\pm$0.0008 -> 0.7990$\pm$0.0005 | Helped |
| Add flow index features to MotifGPL (ablation) | MotifGPL · Beijing segregation index classification | accuracy: 0.7752$\pm$0.0005 -> 0.7990$\pm$0.0005 | Helped |
| Add POI features to MotifGPL (ablation) | MotifGPL · Beijing segregation index classification | accuracy: 0.7560$\pm$0.0003 -> 0.7990$\pm$0.0005 | Helped |
| Use MotifGPL instead of GCN | Beijing segregation index classification · Beijing segregation index classification | accuracy: 0.7872$\pm$0.0002 -> 0.7990$\pm$0.0005 | Helped |
| Guide spatial graph reconstruction with motif distributions at α=0.8, β=0.3 | Beijing spatial graph · Beijing spatial graph reconstruction | Moran's I: 0.4159 -> 0.4043 | Helped |
| Guide spatial graph reconstruction with motif distributions at α=0.8, β=0.2 | Beijing spatial graph · Beijing spatial graph reconstruction | Moran's I: 0.4159 -> 0.3751 | Helped |
| Guide spatial graph reconstruction with motif distributions at α=0.8, β=0.1 | Beijing spatial graph · Beijing spatial graph reconstruction | Moran's I: 0.4159 -> 0.3169 | Helped |
| Guide OD graph reconstruction with motif distributions at α=0.8, β=0.3 | Beijing OD graph · Beijing OD graph reconstruction | Moran's I: 0.2410 -> 0.2341 | Helped |
| Guide OD graph reconstruction with motif distributions at α=0.8, β=0.2 | Beijing OD graph · Beijing OD graph reconstruction | Moran's I: 0.2410 -> 0.2118 | Helped |
| Guide OD graph reconstruction with motif distributions at α=0.8, β=0.1 | Beijing OD graph · Beijing OD graph reconstruction | Moran's I: 0.2410 -> 0.1805 | Helped |
Surprising findings
- ProtGNN and PGIB, the interpretable baselines, underperform plain GCN, GAT and GIN on this node-level task.
Limitations
- Evaluated on a single city (Beijing) with 842 blocks.
- Reconstruction effects are measured with Moran's I rather than validated against actual segregation change over time.
- Motif-based recommendations for urban planning are not validated with real interventions.
Experimental setup
Datasets: Beijing urban blocks spatial graph (842 nodes, 6132 edges) Beijing OD graph (842 nodes, 36334 edges) POI counts (21 categories) hourly pedestrian flow indices street view images (20 summer and 20 winter per block) socioeconomic indicators (income, education, age)
Benchmarks: Beijing urban blocks segregation index classification
Metrics: accuracy F1-score Global Moran's I added/removed/unchanged edge percentage
Baselines: GCN GAT GIN ProtGNN PGIB GAT+Prototype GIN+Prototype
Builds on: ProtGNN GCN massey1988dimensions segregation indicators moro2021mobility segregation index
Closest papers by meaning
- Towards Dynamic Spatial-Temporal Graph Learning: A Decoupled Perspective (AAAI 2024 · similarity 0.83)
- UrbanDataLayer: A Unified Data Pipeline for Urban Science (NeurIPS 2024 · similarity 0.83)
- Satellites Reveal Mobility: A Commuting Origin-destination Flow Generator for Global Cities (NeurIPS 2025 · similarity 0.83)
- UrbanVLP: Multi-Granularity Vision-Language Pretraining for Urban Socioeconomic Indicator Prediction (AAAI 2025 · similarity 0.83)
- UrbanKGent: A Unified Large Language Model Agent Framework for Urban Knowledge Graph Construction (NeurIPS 2024 · similarity 0.83)
- LIVS: A Pluralistic Alignment Dataset for Inclusive Public Spaces (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.