Implementation of various collaborative filtering methods for recommender systems with implicit feedback.
The methods are:
- SVD
- Denoising AutoEncoder (DAE): https://arxiv.org/abs/1802.05814
- Variational AutoEncoder (VAE): https://arxiv.org/abs/1802.05814
- Bayesian Personalized Ranking (BPR): https://arxiv.org/abs/1205.2618
- Light Graph Convolutional Network (LightGCN): https://arxiv.org/abs/2002.02126 and https://arxiv.org/abs/2108.08735 for reading about signBPR loss
- Embarrassingly Shallow AutoEncoders (EASE): https://arxiv.org/abs/1905.03375
- Neural Network version of EASE (NeuEASE)
Note: the models are not tuned and the accuracies are just for demonstration.
Note: im not sure about my negative sampling procedure but the methods implementations are correct.