Skip to content

RecBole-CDR v0.1.0

Latest
Compare
Choose a tag to compare
@Wicknight Wicknight released this 30 May 01:14
· 26 commits to main since this release
d05f558

RecBole-CDR v0.1.0 Release Notes

Bingo! After a long period of effort, we finally develop RecBole-CDR, a recommendation library built upon RecBole for reproducing and developing cross-domain-recommendation algorithms.

In this initial release, we partially refactored RecBole for cross-domain data and implemented several typical cross-domain-recommendation algorithms. In addition, we also published a leaderboard for reference. More details will be introduced in the following part:

  • Highlights
  • Implemented Model
  • Leaderboard

RecBole-CDR is still in its rapid development period, we warmly welcome any type of PRs, including new models, bug reports, and suggestions.

Highlights

  • Automatic and compatible data processing for cross-domain recommendation: Our library designs a unified data structure for cross-domain recommendation, which inherits all the data pre-processing strategies in RecBole. The overlapped data in different domains can be matched automatically.
  • Flexible and customized model training strategies: Our library provides four basic training modes for cross-domain recommendation, which can be combined arbitrarily by users. It is also easy to customize training strategy in original way.
  • Extensive cross-domain recommendation algorithms: Based on unified data structure and flexible training strategies, several cross-domain recommendation algorithms are implemented and compared with others fairly.

Implemented Model

Our library currently supports the following models: CMF (#2), DTCDR (#5), CoNet (#7), BiTGCF (#6), CLFM (#5), DeepAPF (#8), NATR (#26), EMCDR (#14), SSCDR (#28), DCDCSR (#27)

Dataset and Hyper-parameters setting

We collected and organized three pairs of source-target domain datasets which are commonly used in cross domain recommendation. Here we provide these datasets for reference:

We carefully tune the hyper-parameters of the implemented models on these datasets and provide these hyper-parameters for reference:

  • The leaderboard of cross-domain-recommendation on Amazon datasets;
  • The leaderboard of cross-domain-recommendation on Book-Crossing datasets;
  • The leaderboard of cross-domain-recommendation on Douban datasets;

Acknowledgement

Many thanks to the great efforts contributed by Zihan(@linzihan-backforward), Gaowei(@Wicknight), and Shanlei(@ShanleiMu). The team members come from RUC AI Box, which is supported by Prof. Wayne Xin Zhao. It's hoped that Recbole-CDR serves as an important step towards RecBole Community.

Contributors
Zihan Lin (@linzihan-backforward)
Gaowei Zhang (@Wicknight)
Shanlei Mu (@ShanleiMu)