This repository contains the implementation of many popular sampling strategies, along with various explicit/implicit/sequential feedback recommendation algorithms. The code accompanies the paper "On Sampling Collaborative Filtering Datasets" [ACM] [Public PDF] where we compare the utility of different sampling strategies for preserving the performance of various recommendation algorithms.

We also provide code for Data-Genie which can automatically predict the performance of how good any sampling strategy will be for a given collaborative filtering dataset. We refer the reader to the full paper for more details. Kindly send me an email if you're interested in obtaining access to the pre-trained weights of Data-Genie.

If you find any module of this repository helpful for your own research, please consider citing the below WSDM'22 paper. Thanks!

@inproceedings{sampling_cf,
  author = {Noveen Sachdeva and Carole-Jean Wu and Julian McAuley},
  title = {On Sampling Collaborative Filtering Datasets},
  url = {https://doi.org/10.1145/3488560.3498439},
  booktitle = {Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining},
  series = {WSDM '22},
  year = {2022}
}

Code Author: Noveen Sachdeva (nosachde@ucsd.edu)

$ pip install -r requirements.txt

Once you've correctly setup the python environments and downloaded the dataset of your choice (Amazon: http://jmcauley.ucsd.edu/data/amazon/), the following steps need to be run:

The following command will create the required data/experiment directories as well as download & preprocess the Amazon magazine and the MovieLens-100K datasets. Feel free to download more datasets from the following web-page http://jmcauley.ucsd.edu/data/amazon/ and adjust the setup.sh and preprocess.py files accordingly.

$ ./setup.sh

• Edit the hyper_params.py file which lists all config parameters, including what type of model to run. Currently supported models:

• Finally, type the following command to run:

$ CUDA_VISIBLE_DEVICES=<SOME_GPU_ID> python main.py

• Alternatively, to train various possible recommendation algorithm on various CF datasets/subsets, please edit the configuration in grid_search.py and then run:

$ python grid_search.py

• Edit the data_genie/data_genie_config.py file which lists all config parameters, including what datasets/CF-scenarios/samplers etc. to train Data-Genie on

• Finally, use the following command to train Data-Genie:

$ python data_genie.py

MIT