Recommendation Algorithms

a study on how Netflix’s recommender system works, and it is described

Abstract

The audiovisual productions watched by each user on Netflix – a streaming platform - are based, in part, on data collection, computing, and archiving about how and what was previously consumed by each user and by others. Suggestions for new content are made by recommendation systems and operated by a set of algorithms, which are often kept in commercial secret. Netflix, on its website, proposes a “high-level description” of its recommendation system “in layman's language.” This article analyzes how this text explains the functioning of these tools, articulating it with authors who were already part of the group of platform’s developers, other critics, and specialists in algorithms. The analysis demonstrated that from the collection of little user data, especially if compared with the quantity of data usually extracted from social networking sites, it is possible to implement the recommendation system in an elaborated and customized way. The collected data behave as an inclusion pattern and constitute the raw material of a database that feeds the system, creating a complex personalized profile for each user. This profile is what recommends new titles in the search system and mainly guides the item’s position in the ranks in the initial interface. Finally, the position of the title in the interface and the row in which it is displayed significantly influence the choice of production. This, in turn, has consequences in the user’s contact with the diversity of audiovisual productions, the maintenance of the subscription, and the consumption experience on the platform.