Research map / Label-efficient and robust learning / Recognition and long-tail
NineRec: A Benchmark Dataset Suite for Evaluating Transferable Recommendation
Jiaqi Zhang, Yu Cheng, Yongxin Ni, Yunzhu Pan, Zheng Yuan, Junchen Fu, Youhua Li, Jie Wang, Fajie Yuan
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 |
|---|---|---|---|
| pre-train on the Bili_500K source dataset before fine-tuning (HasPT vs NoPT) | SASRec TransRec · Bili_Food (text) | HR@10 (%): 18.03 -> 19.59 | Improved |
| pre-train on source dataset (HasPT vs NoPT) | BERT4Rec TransRec · Bili_Food (text) | HR@10 (%): 16.70 -> 20.00 | Improved |
| train modality encoder end-to-end instead of two-stage frozen features (ablation) | TransRec with BERT · Bili_Movie (text) | HR@10 (%): 5.73 -> 12.63 | Improved |
| train modality encoder end-to-end instead of two-stage frozen features (ablation) | TransRec with BERT · Bili_500K (source) | HR@10 (%): 0.57 -> 4.69 | Improved |
| train modality encoder end-to-end instead of two-stage frozen features (ablation) | TransRec with Swin-B · Bili_Food (image) | HR@10 (%): 14.20 -> 18.72 | Improved |
| replace item IDs with a text encoder (TransRec NoPT) vs IDRec | SASRec · KU (text) | HR@10 (%): 28.36 -> 30.77 | Improved |
| replace item IDs with a pre-trained image encoder vs IDRec | SASRec TransRec HasPT · Bili_Dance (image) | HR@10 (%): 23.46 -> 22.16 | Hurt |
| jointly learn BERT4Rec with Swin Transformer image encoder | BERT4Rec TransRec NoPT · Bili_Food (image) | HR@10 (%): 17.07 -> 2.15 | Hurt |
| pre-train ResNet50 TransRec on Bili_500K (HasPT) instead of directly training (NoPT) (ablation) | ResNet50 TransRec · image recommendation | HR@10 (%) | No clear effect |
| remove ImageNet pre-training of ResNet50 (TFS) (ablation) | ResNet50 TransRec · image recommendation | HR@10 (%) | Hurt |
| use multimodal (text and image) features instead of a single modality | SASRec TransRec · target datasets | HR@10 (%) | No clear effect |
| zero-shot transfer of pre-trained TransRec instead of random ranking | ZeroRec · Bili_Music (text) | HR@10 (%): 0.16 -> 11.30 | Improved |
Closest papers by meaning
- Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations (ICML 2024 · similarity 0.90)
- RecFlow: An Industrial Full Flow Recommendation Dataset (ICLR 2025 · similarity 0.88)
- Can LLMs Outshine Conventional Recommenders? A Comparative Evaluation (NeurIPS 2025 · similarity 0.87)
- Wukong: Towards a Scaling Law for Large-Scale Recommendation (ICML 2024 · similarity 0.87)
- ORBIT - Open Recommendation Benchmark for Reproducible Research with Hidden Tests (NeurIPS 2025 · similarity 0.87)
- CURE4Rec: A Benchmark for Recommendation Unlearning with Deeper Influence (NeurIPS 2024 · similarity 0.87)
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.
