README.md

metadata

pretty_name: KuaiRec
license: cc-by-sa-4.0
language:
  - zh
size_categories:
  - 10M<n<100M
tags:
  - tabular
  - recommendation
  - arxiv:2202.10842
configs:
  - config_name: interactions_big
    data_files:
      - split: train
        path: interactions_big/train/*.parquet
  - config_name: interactions_small
    data_files:
      - split: train
        path: interactions_small/train/*.parquet
  - config_name: users
    data_files:
      - split: train
        path: users/train/*.parquet
  - config_name: users_raw
    data_files:
      - split: train
        path: users_raw/train/*.parquet
  - config_name: item_daily_features
    data_files:
      - split: train
        path: item_daily_features/train/*.parquet
  - config_name: item_categories
    data_files:
      - split: train
        path: item_categories/train/*.parquet
  - config_name: item_caption_category
    data_files:
      - split: train
        path: item_caption_category/train/*.parquet
  - config_name: item_categories_raw
    data_files:
      - split: train
        path: item_categories_raw/train/*.parquet
  - config_name: social_network
    data_files:
      - split: train
        path: social_network/train/*.parquet

KuaiRec

Parquet republication of KuaiRec for the Lizard benchmark. This is not the official repository. The files were converted from the Zenodo release without adding a train/test cut.

The authors use the big interaction matrix for training and the fully observed small matrix for evaluation. Those are separate released tables, so they are separate configs here. Each config has one split, train, containing every row of that table.

Original sources

• Homepage: https://kuairec.com/
• GitHub: https://github.com/chongminggao/KuaiRec
• Zenodo: https://zenodo.org/records/18164998
• Paper: https://arxiv.org/abs/2202.10842
• DOI: https://doi.org/10.1145/3511808.3557220

Field definitions are on the homepage. CSV text is preserved as VARCHAR. Typed ids and timestamps, source_row, dataset_source, and synthetic are added. source_row is the 1-based source line number, assigned before sorting. No rolling history or label was derived.

License

CC BY-SA 4.0, as stated by the official homepage and GitHub repository.

Citation

@inproceedings{gao2022kuairec,
  author = {Gao, Chongming and Li, Shijun and Lei, Wenqiang and Chen, Jiawei and Li, Biao and Jiang, Peng and He, Xiangnan and Mao, Jiaxin and Chua, Tat-Seng},
  title = {KuaiRec: A Fully-Observed Dataset and Insights for Evaluating Recommender Systems},
  booktitle = {Proceedings of the 31st ACM International Conference on Information and Knowledge Management},
  series = {CIKM '22},
  year = {2022},
  pages = {540--550},
  doi = {10.1145/3511808.3557220},
  url = {https://doi.org/10.1145/3511808.3557220}
}