icodein / onlinelearningtorank

Unbiased Differentiable Gradient Descent

This repository contains the code used for the experiments in "Differentiable Unbiased Online Learning to Rank" published at CIKM 2018.

Citation

If you use this code to produce results for your scientific publication, or if you share a copy or fork, please refer to our CIKM 2018 paper:

@inproceedings{Oosterhuis2018Unbiased,
  title={Differentiable Unbiased Online Learning to Rank},
  author={Oosterhuis, Harrie and de Rijke, Maarten},
  booktitle={Proceedings of the 2018 ACM on Conference on Information and Knowledge Management},
  year={2018},
  organization={ACM}
}

License

The contents of this repository are licensed under the MIT license. If you modify its contents in any way, please link back to this repository.

Reproducing Experiments

Recreating the results in the paper can be done with the following command:

python scripts/CIKM2018.py --data_sets cikm2018 --click_models per nav inf --log_folder testoutput/logs/ --average_folder testoutput/average --output_folder testoutput/fullruns/ --n_runs 125 --n_proc 1 --n_impr 10000

This runs all experiments included in the results section of our paper. It is up to the user to download the datasets and link to them in the dataset collections file. The output folders including the folder where the data will be stored (in this case testoutput/fullruns/) has to exist before running the code, if folders are missing an error message will indicate this. Speeding up the simulations can be done by allocating more processes using the n_proc flag.