README.md

RCF

This is the implementation of paper (Codes may be delay, because now I'm taking an internship at Barcelona and the codes are at my university office. I can only process through remote desktop and it's slow. I will upload it as fast as I can.):

Xin Xin, Xiangnan He, Yongfeng Zhang, Yongdong Zhang and Joemon Jose (2019). Relational Collaborative Filtering: Modeling Multiple Item Relations for Recommendation.

Please note that this code may be slow, but it' not the problem of the algorithm. At this moment, the code spends much time to generate training batch.

Citation

If you want to use our codes and datasets in your research, please cite:

@inproceedings{RCF,
  author    = {Xin Xin and
               Xiangnan He and
               Yongfeng Zhang and
               Yongdong Zhang and
               Joemon Jose},
  title     = {Relational Collaborative Filtering: Modeling Multiple Item Relations for Recommendation},
  booktitle = {{SIGIR}},
  year      = {2019}
}

Environemnt

Tensorfow with python 2.7

Dataset

We provide two processed datasets: ML100K and KKBOX

• train.txt

  • Train file.
  • Each line is a user with one of her/his interaced items: (userID and itemID).

• test.txt

  • Test file (positive instances).
  • Same format with train.txt

• test_negative.txt

  • Test file (for KKBOX).
  • For KKBOX, the ranking is performed between 1 postive instance vs 999 negative instances
  • Download from this link.

• auxiliary-mapping.txt

  • For ML100K, itemID|genreIDs|directorIDs|actorsIDs|.
  • For KKBOX, itemID|genreIDs|singerIDs|composerIDs|lyricistIDs