Skip to main content
Digitial Humanities @ uOttawa

Cinema Case Study

Netflix Logo.jpeg

Netflix is an American streaming service that offers access to a catalog full of movies, TV shows and original content, all on a subscription basis. The company which started by renting out DVDs and sending them by mail, is now responsible for 9.39% of the global internet bandwidth traffic (2022 Global Internet Phenomena Report, 2022). Of all the features Netflix offers, their recommender system is surely one that stands out. It influences about 80% of all hours streamed on the platform (Gomez-Uribe & Hunt, 2016), which demonstrate the efficiency of the service it provides to clients.

It is interesting to explore the technology behind their recommender system. In fact, the system uses a collection of various algorithmic approaches such as reinforcement learning, neural networks, causal modelling, probabilistic graphical models, matrix factorization, ensembles and bandits (Springboard India, 2019). Different algorithms that serve different use cases come together to create the Netflix user experience, including Personalized Video Ranker, Top-N Video Ranker, Trending Now, Continue Watching, Video-Video Similarity, Page Generation (Row Selection and Ranking), Evidence Together, and Search and Related Work (Gomez-Uribe & Hunt, 2016). The purpose of each tool is explained in Table 1 below. The results generated by the algorithms are based on numerous inputs, for example viewer ratings, viewing history, time duration of a viewer watching a show, viewing time of the day, information about the categories (year of release, title, genres), other viewers with similar watching preferences, and more (Springboard India, 2019).

Despite receiving a lot of praise for their recommender system, Netflix has also faced criticism. In December of 2017, Netflix announced that their recommender system had a new component named artwork personalization, which consists of personalizing the artwork or thumbnail images that represents a specific title for each user (Chandrashekar et al., 2017). In other words, the image representing a title available on Netflix will differ for each member, to highlight the aspects of a title that is relevant to them. The image could differ either by showcasing different cast members or themes present in the title. However, after the algorithm was implemented, many African-American users in the United States discovered that the thumbnail photos they were viewing were racially and ethnically biased, and were frequently inaccurate representations of the movie's actual cast (Iqbal, 2018). As the company provides its service to users from marginalized populations around the world, it must ensure that they are not misrepresented on the platform.

Table 1: Based on the recommender system's description provided in Gomez-Uribe and Hunt (2016).
Table.png