Item-Underrecommendation-Bias

Code for the SIGIR20 paper -- Measuring and Mitigating Item Under-Recommendation Bias inPersonalized Ranking Systems

Data

ml1m-2, yelp-2, and amazon-2 are the three datasets with two sensitive groups. ml1m-6, yelp-4, and amazon-4 are the three datasets with multiple sensitive groups. There is no original data files in this repo, if you want to get the original data files, please refer to the paper to see the original sources of these datasets.

Requirments

python 2
tensorflow 1.13.0
numpy
sklearn
pandas
matplotlib

Excution

Run DPR_RSP.py to run DPR-RSP model, run DPR_REO.py to run DPR-REO model, and run BPR.py to run BPR model. All the hyperparameters are described in the paper.