Research map / Image generation / Autoregressive and latent generation
Faithful Trip Recommender Using Diffusion Guidance (Student Abstract)
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 a DDPM-based incremental knowledge injection module that gradually injects known source and destination embeddings into the diffusion denoising process to generate faithful trip recommendations.
Problem
Deep-learning trip recommenders generate sub-trajectories and merge them with user-specified sources and destinations, causing trajectory distortion and mismatch with user intent.
Approach
Instead of post-hoc clipping-merging, inject ground-truth source and destination embeddings at every DDPM denoising step with a decaying weight F(t)=exp(-lambda t), so the generated full trajectory is steered toward the user query gradually.
Key results
- Ours achieves F1 of 0.774 on Edinburgh vs 0.728 for the best baseline (CTLTR/SelfTrip) and 0.838 on Glasgow vs 0.834 for the best baseline (DeepTrip).
- Ours reaches 100% Sratio and Eratio on both datasets, versus at most 97.5% and 75.1% for CTLTR on Edinburgh.
- Dmed drops to 0.411 on Edinburgh (best baseline 0.442) and 0.386 on Glasgow (best baseline 0.782).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add DDPM-based incremental knowledge injection | Trip recommender · Edinburgh | F1: 0.728 -> 0.774 | Helped |
| Add DDPM-based incremental knowledge injection | Trip recommender · Glasgow | F1: 0.838 -> 0.830 | Hurt |
| Add DDPM-based incremental knowledge injection | Trip recommender · Edinburgh | Dmed: 0.442 -> 0.411 | Helped |
| Add DDPM-based incremental knowledge injection | Trip recommender · Glasgow | Dmed: 0.782 -> 0.386 | Helped |
| Add DDPM-based incremental knowledge injection | Trip recommender · Edinburgh | Sratio: 97.5% -> 100% | Helped |
| Add DDPM-based incremental knowledge injection | Trip recommender · Edinburgh | Eratio: 75.1% -> 100% | Helped |
| Add DDPM-based incremental knowledge injection | Trip recommender · Glasgow | Sratio: 82.1% -> 100% | Helped |
| Add DDPM-based incremental knowledge injection | Trip recommender · Glasgow | Eratio: 54.5% -> 100% | Helped |
Limitations
- Evaluation limited to two datasets (Edinburgh and Glasgow) and a short student abstract.
- Hyperparameters such as lambda and alpha are pre-defined without reported sensitivity analysis in the text.
- Only F1 and distortion metrics reported; no details on compute or training setup.
Experimental setup
Datasets: Edinburgh POI trajectories Glasgow POI trajectories
Benchmarks: Edinburgh Glasgow
Metrics: F1 score Dmed Sratio Eratio
Baselines: Popularity PoiRank DeepTrip CTLTR SelfTrip
Builds on: DDPM DeepTrip CTLTR SelfTrip
Closest papers by meaning
- SPOT-Trip: Dual-Preference Driven Out-of-Town Trip Recommendation (NeurIPS 2025 · similarity 0.83)
- On Efficiency-Effectiveness Trade-off of Diffusion-based Recommenders (NeurIPS 2025 · similarity 0.82)
- Enhancing Long-and Short-Term Representations for Next POI Recommendations via Frequency and Hierarchical Contrastive Learning (AAAI 2025 · similarity 0.81)
- TAU: Trajectory Data Augmentation with Uncertainty for Next POI Recommendation (AAAI 2024 · similarity 0.81)
- Learning Multiple User Distributions for Recommendation via Guided Conditional Diffusion (AAAI 2025 · similarity 0.80)
- Enhancing Diffusion Model with Auxiliary Information Mining-Exploration and Efficient Sampling Mechanism for Sequential Recommendation (AAAI 2025 · similarity 0.80)
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.