Evaluation of Recommender Systems with Impressions

This repository contains the source code of the paper:

• Towards the Evaluation of Recommender Systems with Impressions. Fernando B. Pérez Maurera, Maurizio Ferrari Dacrema, and Paolo Cremonesi. RecSys 2022. DOI: 10.1145/3523227.3551483.
  • Paper Published Version (ACM DL),
  • Github
  • If you use our work, please cite it. You can click on the Cite this repository button or use the following bibtex:

    @inproceedings{conf/recsys/PerezFC22/towards-the-evaluation-of-recommender-systems-with-impressions,
      author    = {P{\'{e}}rez Maurera, {Fernando B.} and {Ferrari Dacrema}, Maurizio and Cremonesi, Paolo},
      title     = {Towards the Evaluation of Recommender Systems with Impressions},
      booktitle = {Sixteenth ACM Conference on Recommender Systems (RecSys '22), September 18--23, 2022, Seattle, WA, USA},
      url       = {https://doi.org/10.1145/3523227.3551483},
      doi       = {10.1145/3523227.3551483},
    }

See our website for more information on our research group. We are actively pursuing this research direction in evaluation and reproducibility, we are open to collaboration with other researchers. Follow our project on ResearchGate

This repo is divided in three folders:

• impressions-evaluation: Contains the implementation of the impressions recommenders (time-aware, re-ranking, impressions as user profiles), scripts to create the dataset splits and run the experiments.
• recsys-framework: Contains the implementation of baselines, hyper-parameter tuning loop, and evaluation framework.
• recsys-framework-extensions: Contains utility classes and functions that extend the functionalities provided by the recsys-framework. Functions for common data-splitting are defined here.

You'll find instructions to install this project and run the experiments in the
README inside impressions-evaluation.

NOTE: all commands must be run from a terminal using poetry and inside the impressions-evaluation folder.

Contact us

The best place to ask questions or raise issues is in the issues' page of the repo. For further inquiries, you may contact Fernando Pérez via email.