Research map / Architectures and efficiency / RNNs and normalization
A Meta-Learning Perspective on Cold-Start Recommendations for Items
Manasi Vartak, Arvind Thiagarajan, Conrado Miranda, Jeshua Bratman, Hugo Larochelle
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use non-linear bias adaptation instead of linear weight adaptation | meta-learning recommender · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +1.76% -> +2.65% | Improved |
| NLBA vs production model PROD-BEST | NLBA · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +2.54% -> +2.65% | Improved |
| add user CTR as input to NLBA | NLBA · Tweet engagement (Twitter production data) | AUROC gain (w/ CTR): +2.65% -> +2.76% | Improved |
| add user CTR as input to LWA (ablation) | LWA · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +1.76% -> +2.43% | Improved |
| add item-only fallback network to LWA (LWA*) (ablation) | LWA · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +1.76% -> +1.98% | Mixed |
| use meta-learning LWA instead of MF user embeddings | LWA vs MF (deep) · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +0.55% -> +1.76% | Improved |
| increase hidden layers from 1 to 5 in item representation network (ablation) | LWA, simple average · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +1.8% -> +2.20% | Improved |
| increase hidden layers from 5 to 9 (ablation) | LWA, weighted average · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +2.42% -> +1.65% | Hurt |
| increase hidden layers from 5 to 9 (ablation) | LWA, simple average · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +2.20% -> +2.09% | No clear effect |
| use weighted average instead of simple average for class representatives (ablation) | LWA, 5 layers · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +2.20% -> +2.42% | Improved |
| use weighted average instead of simple average with 1 hidden layer (ablation) | LWA · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +1.8% -> +2.31% | Improved |
| drop negative engagements from item history (ablation) | NLBA · Tweet engagement (Twitter production data) | AUROC gain over PROD-BASE: +2.54% -> +1.76% | Hurt |
Closest papers by meaning
- Cold-start Sequential Recommendation via Meta Learner (AAAI 2021 · similarity 0.88)
- A Dynamic Meta-Learning Model for Time-Sensitive Cold-Start Recommendations (AAAI 2022 · similarity 0.86)
- Temporally and Distributionally Robust Optimization for Cold-Start Recommendation (AAAI 2024 · similarity 0.86)
- Counterfactual Task-augmented Meta-learning for Cold-start Sequential Recommendation (AAAI 2025 · similarity 0.86)
- Feature-Structure Adaptive Completion Graph Neural Network for Cold-start Recommendation (AAAI 2025 · similarity 0.85)
- Addressing Cold-Start Problem in Click-Through Rate Prediction via Supervised Diffusion Modeling (AAAI 2025 · 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.
