DML4Rec

Code for the paper:
Metric Learning for Session-based Recommendations
Bartłomiej Twardowski, Paweł Zawistowski, Szymon Zaborowski
European Conference on Information Retrieval 2021
arxiv, ECIR2021

@inproceedings{twardowski2021metric,
  title={Metric Learning for Session-Based Recommendations},
  author={Twardowski, Bart{\l}omiej and Zawistowski, Pawe{\l} and Zaborowski, Szymon},
  booktitle={European Conference on Information Retrieval},
  pages={650--665},
  year={2021},
  organization={Springer}
}

Environment set-up

Environment is based on Conda distribution. All dependencies are in environment.yml file. For Docker version check out docker directory.

Create env

To create new environment check out repo and type:

conda env create --file environment.yml --name dml4rec

Notice: set appropriate version of your CUDA driver for cudatoolkit in environment.yml.

Environment activation/deactivation

conda activate dml4rec
conda deactivate

Datasets preparation

See README in data/ directory.

Running single experiment

Available recommendation algorithms to run:

alg	Class
RND	RandomRecommender
POP	MostPopularRecommender
SPOP	PopularityInSessionRecommender
SKNN	SessionKnnRecommender
VSKNN	VMSessionKnnRecommender
MARKOV	MarkovRecommender
DML	DMLSessionRecommender

All hyper-parmas are in the init() method of the recommender class (like in sklearn). Simples file will have alg key with selected algorithm and optional parameters.

Sample experiments files are in examples/ directory.

To run single experiment type:

./run.sh experiments/experiment.py examples/dml-RSC15_64-MaxPool-SDMLAllLoss.json

To select GPU device use it with CUDA_VISIBLE_DEVICES=X prefix.

Full results will be stored in json file in results/ directory and only some evaluation metrics will be presented in logs as below:

2021-01-06 16:54:45,811 - rec.eval - Test prediction time: 25.048 sec.
2021-01-06 16:54:45,811 - rec.eval - Eveluation - test sessions num: 31352
2021-01-06 16:54:45,815 - rec.eval - Evaluation - ground truth items num: 93510
2021-01-06 16:54:45,817 - rec.eval - Evaluation - predictions items num: 627040
2021-01-06 16:54:45,873 - rec.eval - Unique items predicted: 8605
2021-01-06 16:54:46,023 - rec.eval - REC@20: 0.519
2021-01-06 16:54:52,709 - rec.eval - HR@20: 0.6478
2021-01-06 16:54:53,789 - rec.eval - Evaluation time: 7.915 sec.

Running baselines

All simple baselines can be run at once using script:

./run.sh experiments/baseline.py

Unittests

To run all unit tests type:

pytest tests/