Unbiased Pairwise Learning from Biased Implicit Feedback

---

About

This repository accompanies the real-world experiments conducted in the paper "Unbiased Pairwise Learning from Biased Implicit Feedback" by Yuta Saito, which has been accepted by ICTIR'20.

Dependencies

• python>=3.7
• numpy==1.18.1
• pandas==0.25.1
• scikit-learn==0.23.1
• tensorflow==1.15.2
• pyyaml==5.1.2

About

Datasets

To run the simulation with real-world datasets, the following datasets need to be prepared as described below.

• download the Yahoo! R3 dataset and put train.txt and test.txt files into ./data/yahoo/raw/ directory.
• download the Coat dataset and put train.ascii and test.ascii files into ./data/coat/raw/ directory.

Running the code

First, to preprocess the datasets, navigate to the src/ directory and run the command

python preprocess_datasets.py -d coat yahoo

Then, run the following command in the same directory

for data in yahoo coat
  do
  for model in wmf relmf bpr ubpr upl_bpr relmf_du ubpr_nclip
  do
    python main.py -m $model -d $data -r 10
  done
done

This will run real-world experiments conducted in Section 4. After running the experimens, you can summarize the results by running the following command in the src/ directory.

python summarize_results.py -d yahoo coat

Once the code is finished executing, you can find the summarized results in ./paper_results/ directory.

Acknowledgement

We thank Minato Sato for his helpful comments, discussions, and advice.