The RecSys Challenge 2022 was organized by Dressipi, Bruce Ferwerda (Jönköping University, Sweden), Saikishore Kalloori (ETH Zürich, Switzerland), and Abhishek Srivastava (IIM Visakhapatnam, India).
The challenge focused on fashion recommendations. Given user sessions, purchase data and content data about items, the task was to accurately predict which fashion item will be bought at the end of the session.
The Dataset is available at the following link.
We participated in the challenge as Boston Team Party, a team of 7 MSc students from Politecnico di Milano:
We worked under the supervision of:
The repository is divided in the following parts:
-
Notebooks, selection of notebooks used to explore the dataset and generate custom attributes
-
RecSys_Course_AT_PoliMi, fork of the course repository enriched with the GRU4Rec implementation by Theano, other custom models, utilities to handle the dataset, train models and perform inference
-
optimizer_files, scripts based on Opuna or Bayesian Optimization used to perform hyperparameter tuning of the models involved in the candidate generation
-
booster, scripts related to LightGBM, involving the creation of its dataset, the hyperparameter tuning and the inference
It's available a paper based on our experience in the challenge, describing our discoveries and implementation choices.
The paper was presented at the ACM RecSys2022 Conference and here are the slides used for the presentation.
Our model achieved a score of 0.1880 and 0.1845 in the public and private leaderboard respectively, granting us the 29th place after the first round of evaluation