Fairness and Explainability in Recommendation Systems

The field of recommendation systems is moving towards a greater emphasis on fairness and explainability. Researchers are exploring new concepts and techniques to ensure that recommendations are not only accurate but also fair and transparent. This includes the development of new fairness metrics, such as envy-freeness up to one item, and the use of explainability techniques, such as model-agnostic post-hoc explainability and database views as explanations. Noteworthy papers in this area include 'Database Views as Explanations for Relational Deep Learning', which presents a novel framework for explaining machine-learning models over relational databases, and 'We're Still Doing It (All) Wrong: Recommender Systems, Fifteen Years Later', which argues that the field needs a fundamental reframing of what recommender systems research is for and how knowledge is produced and validated. Additionally, 'Model-agnostic post-hoc explainability for recommender systems' develops a systematic application of deletion diagnostics to quantify the influence of specific users or items on the recommender system.

Sources

Envy-Free but Still Unfair: Envy-Freeness Up To One Item (EF-1) in Personalized Recommendation

Database Views as Explanations for Relational Deep Learning

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

Maximizing social welfare among EF1 allocations at the presence of two types of agents

Model-agnostic post-hoc explainability for recommender systems

Diversified recommendations of cultural activities with personalized determinantal point processes

The Entropy of Parallel Systems

What News Recommendation Research Did (But Mostly Didn't) Teach Us About Building A News Recommender

Exploring the entropic region

Between proportionnality and envy-freeness: k-proportionality

Built with on top of