Research map / Label-efficient and robust learning / Recognition and long-tail
Recommendation representation learning
Improves embeddings for recommender systems via multi-interest, inductive and collapse-aware modeling.
207 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Recognition and long-tail, keyword cluster “representations · transfer · unsupervised”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 1 tracked benchmark.
51 papers in 2023–24 and 68 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 4.1% to 5.2%.
Open questions the data points to
- Beauty (NDCG@10): best went from 0.0537 (up to 2024) to 0.0657 in 2025 — Intent Oriented Contrastive Learning for Sequential Recommendation (AAAI 2025 · 5 citations)
- Papers report COGNet on MIMIC-IV (F1) anywhere from 0.573 to 0.647: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback (AAAI 2016 · 1,008 citations)
- 2017 · POI2Vec: Geographical Latent Representation for Predicting Future Visitors (AAAI 2017 · 225 citations)
- 2018 · ATRank: An Attention-Based User Behavior Modeling Framework for Recommendation (AAAI 2018 · 279 citations)
- 2019 · Deep Interest Evolution Network for Click-Through Rate Prediction (AAAI 2019 · 1,056 citations)
- 2020 · Knowing What, How and Why: A Near Complete Solution for Aspect-Based Sentiment Analysis (AAAI 2020 · 447 citations)
- 2021 · Fairness-aware News Recommendation with Decomposed Adversarial Learning (AAAI 2021 · 119 citations)
- 2022 · Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems (NeurIPS 2022 · 27 citations)
- 2023 · FinalMLP: An Enhanced Two-Stream MLP Model for CTR Prediction (AAAI 2023 · 99 citations)
- 2024 · Learning Accurate and Bidirectional Transformation via Dynamic Embedding Transportation for Cross-Domain Recommendation (AAAI 2024 · 34 citations)
- 2025 · Augmenting Sequential Recommendation with Balanced Relevance and Diversity (AAAI 2025 · 20 citations)
- 2026 · Wavelet Enhanced Adaptive Frequency Filter for Sequential Recommendation (AAAI 2026 · 4 citations)
Earliest papers here
- VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback (AAAI 2016 · 1,008 citations)
- Improving Opinion Aspect Extraction Using Semantic Similarity and Aspect Associations (AAAI 2016 · 141 citations)
- Recommendation with Social Dimensions (AAAI 2016 · 112 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
Beauty · NDCG@10 (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 0.0537 | Theoretically Guaranteed Bidirectional Data Rectification for Robust Sequential Recommendation (NeurIPS 2023 · 5 citations) |
| 2025 | 0.0657 | Intent Oriented Contrastive Learning for Sequential Recommendation (AAAI 2025 · 5 citations) |
What has been tried
161 papers here report 1,582 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 5%, hurt 14%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add dynamic embedding transportation to joint preference exploration | Book->Movie · RMSE | 2.264 -> 1.206 | Helped | Learning Accurate and Bidirectional Transformation via Dynamic Embedding Transportation for Cross-Domain Recommendation (AAAI 2024 · 34 citations) |
| Add task-specific embeddings for all feature fields rather than none | TikTok · AUC | 0.8261 -> 0.8304 | Helped | STEM: Unleashing the Power of Embeddings for Multi-Task Recommendation (AAAI 2024 · 33 citations) |
| add multitask learning | TKY · Acc@1 | 0.2410 -> 0.2575 | Helped | Learning Time Slot Preferences via Mobility Tree for Next POI Recommendation (AAAI 2024 · 32 citations) |
| Use p=-2 instead of p=-1 | ML-1M, Phones, Sports, and Tools · Recall@20 | 25.17 -> 25.51; 13.02 -> 13.03; 8.42 -> 8.52; 7.27 -> 7.42 | Helped | Adaptive Hardness Negative Sampling for Collaborative Filtering (AAAI 2024 · 28 citations) |
| add relevant-visit selection | MIMIC-III · Jaccard | 0.5163 -> 0.5282 | Helped | VITA: ‘Carefully Chosen and Weighted Less’ Is Better in Medication Recommendation (AAAI 2024 · 22 citations) |
| add worst-case factor | Amazon all · Recall@20 | 0.0121 -> 0.0130 | Helped | Temporally and Distributionally Robust Optimization for Cold-Start Recommendation (AAAI 2024 · 21 citations) |
| Add Single-sequence Augmentation | Beauty, Sports, Yelp, Home · N@10 | Beauty: 0.0415 -> 0.0455; Sports: 0.0209 -> 0.0242; Yelp: 0.0152 -> 0.0164; Home: 0.0097 -> 0.0128 | Helped | Augmenting Sequential Recommendation with Balanced Relevance and Diversity (AAAI 2025 · 20 citations) |
| Add dual-perspective metapath encoding instead of shared LSTM encoding | Zhaopin, candidates · R@5 | 0.8936 -> 0.9477 | Helped | Knowledge-Aware Explainable Reciprocal Recommendation (AAAI 2024 · 20 citations) |
| add ICL to the backbone | Sports, Beauty, Toys, Yelp · sum of HR@5, HR@20, NDCG@5, and NDCG@20 | 0.1343 -> 0.1379; 0.239 -> 0.2398; 0.2664 -> 0.2675; 0.1258 -> 0.1262 | Helped | End-to-end Learnable Clustering for Intent Learning in Recommendation (NeurIPS 2024 · 18 citations) |
| Add the preference-aware contrastive loss to PAD-CLRec | Amazon Rec and Amazon Fashion, All/Warm/Cold · Recall@20 | 3.04 -> 3.22; 27.41 -> 28.11; 3.91 -> 4.18; 25.18 -> 27.39; 4.51 -> 4.89; 30.11 -> 30.41 | Helped | Preference Aware Dual Contrastive Learning for Item Cold-Start Recommendation (AAAI 2024 · 15 citations) |
| Add s1 (full-whitening sequence embedding) | Arts, Toys, Tools · R@20 | 0.1650 -> 0.1710; 0.1217 -> 0.1307; 0.0866 -> 0.0918 | Helped | Dual-View Whitening on Pre-trained Text Embeddings for Sequential Recommendation (AAAI 2024 · 12 citations) |
| Use only topics in preference pre-training | Twibot-22 · performance | No clear effect | SeGA: Preference-Aware Self-Contrastive Learning with Prompts for Anomalous User Detection on Twitter (AAAI 2024 · 11 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| COGNet | MIMIC-IV | F1 | 0.573 – 0.647 | 2 papers, e.g. NeurIPS 2023 and AAAI 2024 |
| COGNet | MIMIC-IV | PRAUC | 0.661 – 0.707 | 2 papers, e.g. NeurIPS 2023 and AAAI 2024 |
What reviewers will expect
Datasets and baselines named in the 37 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Amazon Beauty · 4 of 37 papers
- Tmall · 4 of 37 papers
- Yelp · 3 of 37 papers
- Diginetica · 3 of 37 papers
- AmazonElectronics · 2 of 37 papers
- Ali-CCP · 2 of 37 papers
- Amazon Sports · 2 of 37 papers
- Amazon Toys · 2 of 37 papers
Baselines
- SASRec · 6 of 37 papers
- GRU4Rec · 5 of 37 papers
- BERT4Rec · 5 of 37 papers
- DeepFM · 4 of 37 papers
- BPR · 3 of 37 papers
- ICLRec · 3 of 37 papers
- NARM · 3 of 37 papers
- SR-GNN · 3 of 37 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 119 papers; from OpenAlex and the papers' affiliations).
- Renmin University of China · 8 papers
- University of Science and Technology of China · 7 papers
- Tsinghua University · 6 papers
- Zhejiang University · 6 papers
- Alibaba Group · 5 papers
- Peking University · 5 papers
- Northeastern University · 4 papers
- The Hong Kong Polytechnic University · 4 papers
- Huawei · 4 papers
- University of Electronic Science and Technology of China · 4 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Contrastive representation theory (495 papers)
- Correspondence and causal representation learning (447 papers)
- Dense features and foundation priors (482 papers)
- Data augmentation and generalization bias (409 papers)
- Multilingual language representation pretraining (507 papers)
Elsewhere in the topic
- Partial and multi-label learning (345 papers)
- Loss design and classifier theory (273 papers)
- Group-robust classification (251 papers)
- Learning from unlabeled and positive-unlabeled data (240 papers)
- Confidence calibration objectives (129 papers)
- Margin-based face recognition losses (124 papers)
- Supervised person re-identification (108 papers)
- Self-explaining and visual explanation networks (108 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).