Research map / Label-efficient and robust learning / Recognition and long-tail
Bites of Tomorrow: Personalized Recommendations for a Healthier and Greener Plate
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 GRAPE, a sequential food recommender with self/cross-attention over items and sustainability indicators plus two "Green Loss" functions (non-prioritized and prioritized) that push recommendations toward greener foods while preserving accuracy.
Problem
Standard recommender systems optimize accuracy and ignore the sustainability of recommended food, so they do not nudge users toward greener choices.
Approach
A sustainability-integrated attention module models user preferences over items and multiple sustainability indicators (EIS, NIS, HMI), with a user-specific attention weight vector initialized from the mean and variance of each user's past indicator values, trained with a BPR loss combined with pairwise Green Losses that favor more sustainable items among user-liked or user-disliked sets, optional
Key results
- GRAPE leads the best baseline MSSR by 8.01% in HR@20 and 4.89% in NDCG@20 (Top-20).
- At Top-10, GRAPE (L_np variant) reports HR 0.0285 and NDCG 0.0210, with EIS 83.08 versus 83.79 for MSSR, and the best-in-class NDCG among compared methods.
- Prioritized Green Loss with EIS highest priority reduces EIS to 79.47 (Table with priority orders), showing the highest-priority indicator improves most.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use GRAPE instead of MSSR | Food recommendation · Top-20 | HR: 0.0437 -> 0.0472 | Helped |
| Use GRAPE instead of MSSR | Food recommendation · Top-20 | NDCG: 0.0266 -> 0.0279 | Helped |
| Use GRAPE instead of CAFE | Food recommendation · Top-10 | EIS: 79.43 -> 83.08 | Hurt |
| Use GRAPE instead of NOVA | Food recommendation · Top-10 | NIS: 35.85 -> 31.28 | Hurt |
| Use GRAPE instead of CAFE | Food recommendation · Top-10 | HMI: 44.11 -> 44.26 | Helped |
| Increase the Normal Loss weight α (reduce Green Loss) (ablation) | GRAPE with non-prioritized Green Loss · Top-10 food recommendation | sustainability indicator scores | Hurt |
| Increase the Normal Loss weight α (ablation) | GRAPE with non-prioritized Green Loss · Top-10 food recommendation | recommendation accuracy | Mixed |
| Replace personalized initialized P with frozen all-ones P_one (ablation) | GRAPE · Food recommendation | performance | Hurt |
| Use personalized rather than shared non-personalized attention preferences (ablation) | GRAPE · Food recommendation | performance | Helped |
| Initialize personalized P from users' historical sustainability means and variances rather than random initialization (ablation) | GRAPE · Food recommendation | performance | Helped |
| Prioritize EIS over NIS and HMI (ablation) | GRAPE prioritized Green Loss · Top-10 food recommendation | EIS: 81.41 -> 83.08 | Helped |
| Prioritize NIS over EIS and HMI (ablation) | GRAPE prioritized Green Loss · Top-10 food recommendation | NIS: 30.92 -> 33.40 | Helped |
Surprising findings
- Making recommendations greener does not hurt accuracy and can improve it, according to the authors.
- Fixed all-ones user attention matrix performs significantly worse than the other variants, indicating user-specific sustainability preferences matter.
Limitations
- Evaluated on a single food dataset with three sustainability indicators.
- Green Loss priority thresholds and alpha require hyperparameter tuning and trade off accuracy against sustainability.
- Accuracy-sustainability gains are mixed across metrics (e.g., NIS lower than some baselines at Top-5).
Experimental setup
Datasets: Green food recommendation dataset (Zhang et al.) with EIS, NIS, HMI labels
Benchmarks: green food recommendation dataset (Zhang et al.)
Metrics: HR@5 HR@10 HR@20 NDCG@5 NDCG@10 NDCG@20 EIS NIS HMI
Baselines: BPR KNN SHT STOSA ICLRec NOVA CAFE FDSA-CL FHFRS MSSR
Builds on: BPR STOSA MSSR CAFE FDSA-CL
Closest papers by meaning
- Knowledge-Powered Recommendation for an Improved Diet Water Footprint (AAAI 2024 · similarity 0.86)
- Recommender Ecosystems: A Mechanism Design Perspective on Holistic Modeling and Optimization (AAAI 2024 · similarity 0.82)
- What can large language models do for sustainable food? (ICML 2025 · similarity 0.81)
- Tree of Preferences for Diversified Recommendation (NeurIPS 2025 · similarity 0.79)
- ORBIT - Open Recommendation Benchmark for Reproducible Research with Hidden Tests (NeurIPS 2025 · similarity 0.79)
- Promoting Sustainable Web Agents: Benchmarking and Estimating Energy Consumption Through Empirical and Theoretical Analysis (AAAI 2026 · similarity 0.78)
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.