Titre : Netflix: A Black Box Within a Black Box 

Auteurs : Zoé Buranello
, Elsa Negre
, Juliette Rouchier,  

Revue : Open Journal in Information Systems Engineering 

Numéro : Special Issue 

Volume : 6 

Date : 2026/01/28 

DOI : 10.21494/ISTE.OP.2026.1404 

ISSN : 2634-1468 

Résumé : This article focuses on the recommendation system of the subscription video-on-demand (SVOD) platform Netflix®, specifically exploring what can be understood from an end-user perspective. To this end, we investigate how viewing choices influence the recommendations displayed during the subsequent platform visit. To conduct this analysis, we designed two experiments comparing two user profiles: one long-standing profile active for 7 years, and another newly created profile. We concentrated on elements observable directly by the user: categories, recommended titles within personalized categories, and the « top banners » displayed prominently on the homepage. Our findings revealed the following: first, recommendations for a recent profile are more quickly and strongly influenced by the content viewed on that profile, whereas an older profile shows little change. Second, the new profile receives recommendations spanning a wide variety of genres, including popular content as well as some personalized suggestions. Finally, over time, while the older profile initially received increasing numbers of documentary recommendations based on its viewing history, only a few days of inactivity were enough for these recommendations to disappear entirely. Conversely, the recent profile continued to receive documentary suggestions. These experiments also allowed us to observe the evolution of suggested categories for each profile. The significant diversity of categories and the variability in how films are ranked within them emerged as important factors. This observation raises further questions about how the recommendation system creates and uses categories to encourage user engagement. 

Éditeur : ISTE OpenScience