Frequency Adaptive Hypergraph Constraint for Collaborative Filtering(FAHC) devises a Frequency Adaptive graph convolution neural network and hypergraph constraint loss. To solve the issue of difficulty in capturing signals with different frequency for conventional graph neural networks, and a new loss is proposed to improve recommendation performance in collaborative filtering.
The codes of FAHC are implemented and tested under the following development environment:
conda install pytorch torchvision torchaudio pytorch-cuda=11.6 -c pytorch -c nvidia
conda install numpy=1.23.4
conda install scipy==1.9.3
We utilized three datasets to evaluate HCCF: Yelp, MovieLens, and Amazon. Following the common settings of implicit feedback, if user has rated item , then the element is set as 1, otherwise 0. We filtered out users and items with too few interactions. The datasets are divided into training set, validation set and testing set by 7:1:2.
For more information, please refer to https://github.com/akaxlh/HCCF .
Please unzip the datasets first.Here is an example of a yelp dataset
python main.py --data=yelp
More detail hyperparameter in Params.py
The log file is generated in /Log
and the model saved file name is in Saved
.
xxxx