This repository provides the source code of "Consensus Learning from Heterogeneous Objectives for One-Class Collaborative Filtering" accepted in TheWebConf (WWW2022) as a research paper.

We present ConCF Framework that exploits the complementarity from heterogeneous learning objectives throughout the training process, generating a more generalizable model.

In this work, we use five learning objectives for OCCF that have been widely adopted in recent work.

• CF-A: Bayesian Personalized Ranking (BPR) Loss
• CF-B: Collaborative Metric Learning (CML) Loss
• CF-C: Binary Cross-Entropy (BCE) Loss
• CF-D: Mean Squared Error (MSE) Loss
• CF-E: Multinomial Likelihood Loss

• The performance of the head trained by CF-A is significantly improved in ConCF.

• The consensus collaboratively evolves with the heads based on their complementarity, providing accurate supervision..

  

• ConCF (w/o CL) cannot effectively improve the performance of each head.

  

• Python version: 3.6.10
• Pytorch version: 1.5.0

Please refer to 'Guide to using ConCF.ipynb' file.