Determine engineers’ use of feature-attribution tools for debugging news recommender systems

Determine the extent to which engineers in news organizations use feature-attribution tools to debug news recommender systems.

Background

The paper reports that some engineers may use feature-attribution methods to understand recommendation models, but the actual prevalence of this practice within the interviewed organizations is unresolved. Although engineers are permitted to use such tools, the study does not establish whether, how often, or for which debugging tasks they are applied in practice. Clarifying this usage would help assess the role of algorithmic explainability in maintaining and debugging production news recommender systems.

References

When referring to engineers using feature attribution for debugging in their team, P3 said ``they probably use everything;'' usage is currently unknown, but engineers are free to use the tools as needed.

— Do We Care About Personalization and Explainability? An Interview Study with News Recommendation Engineers  (2609.21547 - Kareem et al., 18 Sep 2026) in Section “Results,” subsection “Debugging approaches”