Evaluating recommender systems with and for children: towards a multi-perspective framework

Abstract

Children are common users of recommender systems (RSs) when watching videos on streaming services, accessing information on the web or playing games, being tablets or phones their favourite devices. Some concerns have been raised by parents and educators on the risks that these systems pose to children and the need to develop products and services that empower children by design and support children’s rights. The RSs literature shows that children scenarios are difficult for evaluation, which makes it a clear example of the need to integrate perspectives from multiple stakeholders. Motivated by the need for practical methodologies for children-centric trustworthy artificial intelligence, this paper provides a comprehensive view of the different perspectives involved in the evaluation of RSs for children. We first carry out a literature review, with a focus on the RSs literature, on children-related research, which integrates knowledge from disciplines such as engineering, cognitive science and humancomputer interaction. From this review, we identify the main opportunities, challenges and risks related to children-centred RSs and their evaluation. Finally, we propose a multi-perspective framework for the evaluation of RSs for children.