Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

A challenge for rounded evaluation of recommender systems

Nature Machine Intelligence volume 5, pages 181–182 (2023)Cite this article

Subjects

The organizers of the EvalRS recommender systems competition argue that accuracy should not be the only goal and explain how they took robustness and fairness into account.

Recommender systems are embedded in most applications we use every day. From streaming services to online retailers and social networks, their performance is a key factor to the success of many products and the proper functioning of our digital society. The evaluation of these complex systems is often performed using point-wise accuracy metrics over a held-out testing dataset1,2. However, such hit-or-miss and ranking metrics provide only a partial perspective on the true generalization capabilities of recommender systems. They result in overconfidence in models, which can often be tricked by slightly out-of-distribution inputs3. Moreover, they don’t provide the desired guidance for selecting recommender systems, as two different models may have identical global mean reciprocal rank (MRR) values but different behaviour on different subsets of the target distribution4. Furthermore, as a general problem in machine learning competitions, ‘leaderboard chasing’ often comes at the expense of other important desiderata.

This is a preview of subscription content, access via your institution

Access options

Rent or buy this article

Get just this article for as long as you need it

$39.95

Learn more

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: A radar chart showcasing performance across the main metric types.

References

  1. Tagliabue, J. et al. Preprint at https://arxiv.org/abs/2104.09423 (2021).

  2. Baigorria Alonso, M. Preprint at https://arxiv.org/abs/2108.09858 (2021).

  3. Ribeiro, M. T., Wu, T., Guestrin, C. & Singh, S. in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics 4902–4912 https://doi.org/10.18653/v1/2020.acl-main.442 (2020).

  4. Chia, P. J., Tagliabue, J., Bianchi, F., He, C. & Ko, B. In Companion Proc. Web Conference 2022 99–104 https://doi.org/10.1145/3487553.3524215 (2022).

  5. Tagliabue, J. et al. Proceedings of the CIKM 2022 Workshops co-located with 31st ACM International Conference on Information and Knowledge Management Vol. 3318 CEUR-WS (2022).

  6. Schedl, M. In Proc. 2016 ACM on International Conference on Multimedia Retrieval 103–110 https://doi.org/10.1145/2911996.2912004 (2016).

  7. Ke Yang, J. S. In Proc. 29th International Conference on Scientific and Statistical Database Management 1–6 https://doi.org/10.1145/3085504.3085526 (2017).

  8. Dixon, L., Li, J., Sorensen, J., Thain, N. & Vasserman, L. In Proc. 2018 AAAI/ACM Conference on AI, Ethics, and Society 67–73 https://doi.org/10.1145/3278721.3278729 (2018).

Download references

Acknowledgements

EvalRS is based on RecList, an open source library whose development is supported by Comet, Neptune and Gantry.

Author information

Authors and Affiliations

  1. Tandon School of Engineering, New York University, Brooklyn, NY, USA

    Jacopo Tagliabue

  2. South Park Commons, New York, NY, USA

    Jacopo Tagliabue

  3. Computer Science Dept., Stanford University, Stanford, CA, USA

    Federico Bianchi & Ciro Greco

  4. Microsoft, Redmond, WA, USA

    Tobias Schnabel

  5. Department of Computing Sciences, Bocconi University, Via Roberto Sarfatti, Milano, Italy

    Giuseppe Attanasio

  6. NVIDIA, Santa Clara, CA, USA

    Gabriel de Souza Moreira

  7. Coveo, Montreal, Quebec, Canada

    Patrick John Chia

Authors
  1. Jacopo Tagliabue
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Federico Bianchi
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Tobias Schnabel
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Giuseppe Attanasio
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Ciro Greco
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Gabriel de Souza Moreira
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Patrick John Chia
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Jacopo Tagliabue.

Ethics declarations

Competing interests

The authors declare they have no conflicting interests.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Tagliabue, J., Bianchi, F., Schnabel, T. et al. A challenge for rounded evaluation of recommender systems. Nat Mach Intell 5, 181–182 (2023). https://doi.org/10.1038/s42256-022-00606-0

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s42256-022-00606-0

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing