Research map / Probabilistic and scientific modeling / Molecules and proteins
SPOT-Trip: Dual-Preference Driven Out-of-Town Trip Recommendation
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 SPOT-Trip, a static-dynamic preference-aware framework that recommends intermediate POI sequences for out-of-town trips, combining a POI attribute knowledge graph (relation-aware attention), neural ODE with a temporal point process for dynamic preferences, and a fusion head.
Problem
Out-of-town trip recommendation: generating a sequence of intermediate POIs for travelers from a hometown to an unvisited region given origin, destination and number of stops, under sparse out-of-town check-ins and entangled static/dynamic preferences.
Approach
Disentangle static preferences (knowledge-graph-enriched hometown preferences, aligned to out-of-town via an MLP) from dynamic preferences (latent neural ODE with an NHPP intensity over irregular check-in times), then fuse both with the query representation to score POIs per trip position.
Key results
- On Yelp, SPOT-Trip reaches F1 0.0399 and PairsF1 0.0190, improving over the best baseline by up to 17.01% in F1.
- On Foursquare, SPOT-Trip reaches F1 0.0400 and PairsF1 0.0109, a 6.67% F1 and 15.96% PairsF1 gain over the best baseline.
- Removing the knowledge-enhanced static preference module (w/o KS) drops Yelp F1 by 10.02% and PairsF1 by 21.79%.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SPOT-Trip instead of the strongest baseline | Out-of-town trip recommendation · Foursquare | F1: +6.67% | Helped |
| Use SPOT-Trip instead of the strongest baseline | Out-of-town trip recommendation · Foursquare | PairsF1: +15.96% | Helped |
| Use SPOT-Trip instead of the strongest baseline | Out-of-town trip recommendation · Yelp | F1: +17.01% | Helped |
| Use SPOT-Trip instead of the strongest baseline | Out-of-town trip recommendation · Yelp | PairsF1: +13.90% | Helped |
| Add knowledge-enhanced static preference learning (KSPL) (ablation) | SPOT-Trip · Yelp | F1: 10.02% drop when removed | Helped |
| Add knowledge-enhanced static preference learning (KSPL) (ablation) | SPOT-Trip · Yelp | PairsF1: 21.79% drop when removed | Helped |
| Add dynamic preference learning (ODPL) (ablation) | SPOT-Trip · Foursquare and Yelp | F1 and PairsF1 | Helped |
| Add spatial information to ODPL (ablation) | SPOT-Trip · Foursquare and Yelp | F1 and PairsF1 | Helped |
| Replace KSPL with KDDC (ablation) | SPOT-Trip · Foursquare and Yelp | F1 and PairsF1: 0.0400 -> 0.0385; 0.0109 -> 0.0089; 0.0399 -> 0.0361; 0.0190 -> 0.0174 | Hurt |
| Replace ODPL with CNN-ODE (ablation) | SPOT-Trip · Foursquare and Yelp | F1 and PairsF1: 0.0400 -> 0.0390; 0.0109 -> 0.0107; 0.0399 -> 0.0351; 0.0190 -> 0.0186 | Hurt |
| Replace TransE alternative training with TransR (ablation) | SPOT-Trip · Foursquare and Yelp | F1 and PairsF1: 0.0400 -> 0.0378; 0.0109 -> 0.0091; 0.0399 -> 0.0366; 0.0190 -> 0.0182 | Hurt |
| Replace parallel KSPL and ODPL design with ODPL followed by KSPL (ablation) | SPOT-Trip · Foursquare and Yelp | F1 and PairsF1: 0.0400 -> 0.0351; 0.0109 -> 0.0078; 0.0399 -> 0.0344; 0.0190 -> 0.0153 | Hurt |
Surprising findings
- Using only 80% of hometown data gives better F1 on Foursquare and PairsF1 on Yelp than using full data, which the authors attribute to removal of noisy check-ins.
Limitations
- Spatial coordinates are unavailable at recommendation time, so the normalized query position is used as a surrogate time grid.
- Evaluation is limited to two check-in datasets (Foursquare and Yelp).
Experimental setup
Datasets: Foursquare Yelp
Benchmarks: Foursquare Yelp
Metrics: F1 PairsF1 Full-F1 Full-PairsF1
Baselines: Popularity POIRank GraphTrip MatTrip AR-Trip Base + KDDC Base + CNN-ODE Base + PPROC
Builds on: neural ODE TransE AR-Trip KDDC PPROC
Closest papers by meaning
- Learning Time Slot Preferences via Mobility Tree for Next POI Recommendation (AAAI 2024 · similarity 0.86)
- GeoMamba: Towards Multi-granular POI Recommendation with Geographical State Space Model (AAAI 2025 · similarity 0.85)
- Enhancing Long-and Short-Term Representations for Next POI Recommendations via Frequency and Hierarchical Contrastive Learning (AAAI 2025 · similarity 0.85)
- Spatial-Temporal Knowledge Distillation for Takeaway Recommendation (AAAI 2025 · similarity 0.85)
- Integrating Personalized Spatio-Temporal Clustering for Next POI Recommendation (AAAI 2025 · similarity 0.84)
- TRACE: Tourism Recommendation with Accountable Citation Evidence (NeurIPS 2026 · similarity 0.84)
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.