RecBole v1.2.1 Release Notes

After months of focused effort, we're excited to announce the completion of the RecBole upgrade and the release of a new version: RecBole v1.2.1.

In this update, we've listened to user feedback and evolving needs to make meaningful improvements to the overall experience and usability of RecBole. To better serve our growing community, we've added more benchmark models and datasets. We've also addressed various bugs reported by users and updated our documentation to make it clearer, more comprehensive, and easier to navigate.

In a word, RecBole v1.2.1 is more efficient, convenient and flexible than previous versions. More details will be introduced in the following part:

• New Features
• Bug Fixes
• Docs

New Features

• Add 3 new models:
• General recommendation (1): AsymKNN (#2122).
• Sequential recommendation (2): GRU4RecCPR and SASRecCPR (#1912).
• Add 1 new dataset: Amazon2023 (#2081).
• Modify the expensive tensor multiplication in RepeatNet (#1916).
• Support time-based split mode with ratios in evaluation (#2094).

Bug Fixes

• Model:
• Fix a bug in DiffRec: fix the running error of DiffRec when noise_scale=0 (#1915).
• Fix dropout instantiation in NGCF and GRU4RecKG (#1946).
• Fix bugs when running Random, ADMMSLIM, and SLIMElastic with negative sampling evaluations (#1997).
• Fix neg sample eval bugs for DiffRec and LDiffRec (#1999).
• Fix the endless loop training bug in FEARec (#2024).
• Fix the loss weight of S3Rec (#2028).
• Evaluator：
• Fix only use cpu in the final step in collector.py to reduce the usage of cpu (#2071).
• Typo:
• Fix typos on logger warning in general_dataloader.py (#1976).

Docs

• Add acknowledgment (#2007).
• Update README: Standardize citation format (#2012).
• Fix argument value in automatic selection of model (#2053).
• Update Sequential_recommendation.md (#2091).
• Update README_CN.md: add colab link for convenient practice (#2130).