Reference implementation of the deconfounded recommender

This folder contains the code for Causal Inference on Recommender Systems (Wang et al., 2020):

• the empirical study on simulated datasets (Section 3.2)

• the empirical study on random test sets (Section 3.3)

Environment

python 2

tensorflow 1.5.0

edward 1.3.5

How to execute the scripts

Download the datasets

Download the Yahoo R3 dataset and the coat dataset to dat/raw/

• Yahoo R3: https://webscope.sandbox.yahoo.com/catalog.php?datatype=r

• Coat (Schnabel et al., 2016): https://www.cs.cornell.edu/~schnabts/mnar/

Preprocess the datasets

Yahoo R3

Run the script src/preproc/prep_R3_weakgen.py and src/preproc/prep_R3_stronggen.py to preprocess the dataset.

Coat (Schnabel et al., 2016)

Run the script src/preproc/prep_coat_weakgen.py and src/preproc/prep_coat_stronggen.py to preprocess the dataset.

Simulated datasets

Run the script src/simdat/simulate_generic.py to simulate the datasets.

Run the script src/preproc/prep_simulate_weakgen.py and src/preproc/prep_simulate_stronggen.py to preprocess the dataset.

Perform recommendation

Yahoo R3

Run the script src/causalrec/run_sweep_R3_fitA.sh and then src/causalrec/run_sweep_R3.sh to perform recommendation.

Coat (Schnabel et al., 2016)

Run the script src/causalrec/run_sweep_coat_fitA.sh and then src/causalrec/run_sweep_coat.sh to perform recommendation.

Simulated datasets

Run the script src/causalrec/run_sweep_simulation_fitA.sh and then src/causalrec/run_sweep_simulation.sh to perform recommendation.

Aggregate results

Run the script src/causalrec/merge_csv.py to aggregate results.

Output

The files res/*_allres.csv include output from this implementation.

References

Y. Wang, D. Liang, L. Charlin, and D.M. Blei. (2020) Causal inference on recommender systems. Proceedings of the 14th ACM Conference on Recommender Systems.

Tobias Schnabel, Adith Swaminathan, Ashudeep Singh, Navin Chandak, Thorsten Joachims (2016). Recommendations as Treatments: Debiasing Learning and Evaluation. Proceedings of The International Conference on Machine Learning (ICML).