The Data Don’t Speak for Themselves: The Humanity of VOD Recommender Systems

Abstract

There is a widespread myth and rhetoric, even in academic discourse, about data and VOD recommender systems, especially with regard to the notion of automation and the innocence of this presumed automation. Behind this rhetoric lies the de-humanization of machine computation, i.e. the removal of all the processual, decisional, ‘oriented’ aspects informing every online recommender system. This essay focuses on content-to-content video recommendations, which are based on patterns of similarity between different contents, and it intends to show that there is nothing neutral — even in the most seemingly ‘objective’ form of video recommendation. The aim is to rediscover those very processual elements of the ‘data supply chain’ — regarding how metadata are created and collected, and how algorithms are con gured — so as to make them critically observable again: the funnels, decision points, the multiple layers of human mediation and ltering, in both their relevance and sensitivity.