GSIMC

The official implementation for "Graph Signal Sampling for Inductive One-Bit Matrix Completion: a Closed-form Solution".

What's news

[2023.04.26] We release the PySpark version of our codes for collaborative filtering.

Results on Netflix

Dataset

We use the Netflix benchmark to evaluate model performance, where the Dataframe scheme is as follows:

Feature Name	Content
uid	user identity
mid	movie identity
time	feedback time

CSV Download: Google, 夸克

Results with Rank=50

Below we report the HR@50, HR@100 and NDCG@100 results on the above provided dataset.

Graph Regularization	Lapalacian	HR@50	HR@100	NDCG@100
Bandlimited Norm	Hypergraph	0.19623	0.29322	0.08761
Diffusion Process	Hypergraph	0.18990	0.28547	0.08682
Random Walk	Hypergraph	0.19512	0.29250	0.08766
Inverse Cosine	Hypergraph	0.19353	0.29130	0.08757
Bandlimited Norm	Covariance	0.20030	0.29517	0.08814
Diffusion Process	Covariance	0.19228	0.28798	0.08675
Random Walk	Covariance	0.19922	0.29500	0.08817
Inverse Cosine	Covariance	0.19830	0.29540	0.08831

Folder Specification

• conf/: configurations for logging
• src/: codes for model definition
• runme.sh: train or evaluate EasyDGL and baseline models

Run the Code

Download our data to $DATA_HOME directory, then Reproduce above results on Netflix benchmark:

bash runme.sh $DATA_HOME

Citation

If you find our codes useful, please consider citing our work

@inproceedings{chen2023graph,
  title={Graph Signal Sampling for Inductive One-Bit Matrix Completion: a Closed-form Solution},
  author={Chen, Chao and Geng, Haoyu and Zeng, Gang and Han, Zhaobing and Chai, Hua and Yang, Xiaokang and Yan, Junchi},
  booktitle={Proceedings of the International Conference on Learning Representations (ICLR'23)},
  year={2023},
}