Papers
Topics
Authors
Recent
Search
2000 character limit reached

We're Still Doing It (All) Wrong: Recommender Systems, Fifteen Years Later

Published 11 Sep 2025 in cs.IR and cs.AI | (2509.09414v1)

Abstract: In 2011, Xavier Amatriain sounded the alarm: recommender systems research was "doing it all wrong" [1]. His critique, rooted in statistical misinterpretation and methodological shortcuts, remains as relevant today as it was then. But rather than correcting course, we added new layers of sophistication on top of the same broken foundations. This paper revisits Amatriain's diagnosis and argues that many of the conceptual, epistemological, and infrastructural failures he identified still persist, in more subtle or systemic forms. Drawing on recent work in reproducibility, evaluation methodology, environmental impact, and participatory design, we showcase how the field's accelerating complexity has outpaced its introspection. We highlight ongoing community-led initiatives that attempt to shift the paradigm, including workshops, evaluation frameworks, and calls for value-sensitive and participatory research. At the same time, we contend that meaningful change will require not only new metrics or better tooling, but a fundamental reframing of what recommender systems research is for, who it serves, and how knowledge is produced and validated. Our call is not just for technical reform, but for a recommender systems research agenda grounded in epistemic humility, human impact, and sustainable practice.

Summary

  • The paper challenges the reliance on traditional accuracy metrics, arguing that they overlook real user satisfaction.
  • It exposes methodological stagnation with opaque evaluations that compromise reproducibility and reliability.
  • The study advocates for sustainable, human-centric recommender systems that address ethical and environmental concerns.

Recommender Systems: Revisiting 15 Years of Research Challenges

Introduction

In the last decade and a half, the field of recommender systems has achieved significant algorithmic advancements and industrial adoption. However, a critique originally posited by Xavier Amatriain in 2011, which highlighted foundational flaws in the field's approach, remains pertinent. This essay discusses the enduring critical issues within recommender systems research, emphasizing methodological shortcomings, reproducibility challenges, and emerging concerns in modern practice. It further explores necessary systemic changes to realign the field with values oriented towards human welfare and sustainability.

Persistent Epistemological Failures

Central to the critique of recommender systems research is the persistent emphasis on performance improvement metrics at the expense of meaningful user experience enhancement. Despite shifts from predictive ratings to top-N recommendations, the foundational assumptions underpinning these systems have remained largely unchallenged. Recommender systems continue to prioritize algorithmic optimization, often optimizing for metrics such as RMSE or nDCG without questioning their relevance to actual user satisfaction or experiential quality. This fixation on quantitative improvements perpetuates a mechanical understanding of human preferences and behaviors.

Methodological Stagnation and Structural Defaults

The field's methodological practices exhibit a reluctance to move beyond traditional evaluation metrics. While beyond-accuracy metrics such as diversity or fairness are recognized, their application remains rare. Furthermore, reproducibility barriers persist due to variabilities in experimental setups, evaluation frameworks, and dataset preprocessing methods. Research often relies on single datasets without methodological transparency, compromising result reliability.

In addition, the community’s predilection for behavioral data as ground truth reinforces flawed assumptions. Interactions are often misinterpreted as explicit preferences, neglecting the contextual and temporal complexities inherent in user behavior. These assumptions drive the optimization effort, overshadowing the need for deeper understanding and interdisciplinary exploration.

Emerging Issues in Modern Recommender Systems

The integration of resource-intensive models, such as those based on LLMs, introduces new concerns, including environmental sustainability. Pareto-efficient architectural choices remain underexplored, with energy consumption metrics seldom reported, raising ethical concerns about the ecological footprint of advanced models. Moreover, the recent trend of employing LLMs in recommendation contexts often overlooks critical evaluation of their necessity or empirical advantage over simpler models.

Ethically, the field grapples with insufficient incorporation of fairness and user autonomy in design and evaluation processes. Evaluative focus remains predominantly retrospective and procedural, neglecting forward-looking assessment regarding societal impacts and user empowerment.

Paradigm Shift: Toward Sustainable and Human-Centric Research

Recommender systems research requires a paradigm shift toward human-centric and sustainable practices. A comprehensive evaluation should incorporate varied datasets and mixed-methods approaches aligned with human goals. Transparency in reporting experimental conditions, computational expenses, and algorithmic assumptions is crucial. Additionally, researchers must adopt epistemic humility, acknowledging data limitations and the volatile nature of user preferences.

There is a growing consensus on the field’s role in addressing societal challenges. Workshops such as AltRecSys encourage exploration of critical and interdisciplinary perspectives, fostering discussions about the normative dimensions of recommender systems. The push for participatory design models further emphasizes the inclusion of stakeholders in system conception and goal formulation.

Conclusion

Fifteen years after Amatriain’s foundational critique, recommender systems continue to confront significant epistemological and structural shortcomings. These challenges demand a reevaluation of research priorities, methodologies, and ethical frameworks. The future of recommender systems lies in cultivating an interdisciplinary, reflective, and context-aware research agenda that prioritizes human impacts and embraces broader societal goals. Such an orientation not only promises more meaningful technical advancements but also aligns the field with its broader responsibility toward human welfare and ecological sustainability.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Collections

Sign up for free to add this paper to one or more collections.

Tweets

Sign up for free to view the 2 tweets with 16 likes about this paper.