Talk (Work in Progress): Silvia Milano (MCMP)

Title:

Automated recommendations: What’s the problem?

Abstract:

In this talk, I start by introducing and critically examine the problem formulation that is dominant in recommender systems research, which focuses on the prediction of users’ preferences or utility. This problem formulation lies on untenable assumptions about the interpretation of data available to the system, and implausible normative assumptions regarding the objective of a recommendation task. This has several consequences, including: the main evaluation metrics for recommender systems fail to capture what’s important for a ‘good’ recommendation; and it muddles the issue of defining and operationalising fairness in recommendation. I then examine some desiderata and candidates for alternative formulations of the recommendation problem.