Loading player Between the end of 2019 and 2020, the popular Netflix streaming service introduced a list of the 10 most viewed TV series and 10 movies on the platform in the main interface for presenting the contents available to users. It did not disclose the specific parameters used to compile those daily rankings, but later updated the view calculation criteria in the compilation of the statistics periodically reported to the shareholders: still today, any view of at least two minutes of a series or of a film during the first 28 days of programming.
Based on these criteria and the data provided by the company, the recent South Korean series Squid Game has been defined as the most viewed original series produced by Netflix ever: about two out of three subscribers have seen at least two minutes, for a total of 142 million accounts.
The metrics Netflix used to rank content have not always been these and may not be these in the future. Before the introduction of the 10 most viewed content lists, a single view was defined as watching at least one episode of a series or at least 70 percent of a movie. As reported in a recent letter to shareholders, Netflix plans to introduce a further update on how to compile statistics on its most viewed original content in the coming months: they will be ranked based on the amount of total hours viewed rather than the number of accounts they have seen at least two minutes of that content.
– Read also: A South Korean series
Recently, and in particular after the great success of Squid Game, several reflections shared among technology experts, sociologists, journalists and film critics have resumed an existing debate on how much the algorithmic criteria and procedures used by large streaming platforms, and in particular by Netflix, in defining the “popularity” of content, in the classification of that content and in the personalization of the user experience influence the collective choices and tastes of millions of viewers.
It is part of a very broad discussion, which concerns other large platforms and the conditioning they exert on popular culture by ordering and suggesting through complex and automated technologies what we read, watch and listen to every day. And it is a discussion made even more complex by the fact that the use of these tools by each of the companies that manage the platforms is essentially regulated according to the economic interests of the companies themselves.
The dynamics through which 142 million users end up seeing the same program, in the face of thousands of alternative possibilities present in the catalog, are only partially known. Net of the very partial and therefore not very relevant views, according to internal documents cited by Bloomberg Netflix estimates that 89 per cent of people who started watching Squid Game have seen more than one episode (at least 75 minutes) and that 66 per one hundred viewers – around 87 million people – finished the series in the first 23 days.
“Netflix differs from movie studios and television networks in that it doesn't generate sales based on specific titles, but uses its catalog and new releases at a steady pace to attract customers every week. But the company has a wealth of data about what its customers are looking at, data that the company uses to determine the value derived from individual programs, ”Bloomberg wrote.
Services like Netflix are known to use algorithms to organize content and direct users' attention within an exceptionally large overall offering. Based on the preferences expressed by subscribers through their activity on the platform and through an evaluation system (a thumbs up or down), Netflix proposes different contents from time to time, perfecting the offer according to the tastes and probable satisfaction of each of the subscribers. The processes of personalization of the user experience based on machine learning, right from the first access to the platform, are therefore aimed at keeping the user “active” and making it easier for him to discover content that he might find interesting.
– Read also: Netflix wants to avoid you spending the evening choosing what to watch
Data collected over time and on a large scale is then aggregated and analyzed to predict the success of certain content, based on estimates of the likely size of the target audience. These forecasts, for example, can serve the company to promote the marketing of certain content or prepare voiceovers and subtitles in advance in countries where content similar to the one promoted has already been widely successful in the past.
According to the English sociologist David Beer, an expert in new media and a lecturer at the University of York, many cultural practices are today significantly conditioned by the processes of automation of streaming platforms, which overlap with factors – experiences, social and cultural background, class of belonging – which previously regulated and influenced those practices. “These algorithms not only respond to our tastes, but shape and influence them,” wrote Beer in The Conversation, arguing that the ways of organizing content on streaming platforms – not just those visible to viewers – are now responsible for a profound cultural transformation.
The effect of the current and more widespread digital data classification systems – systems common to several large platforms, including musical ones – would essentially be a redefinition of traditional genres. In addition to being useful in distinguishing the various forms of artistic and cultural expression, those genres contributed to shaping widely shared identities and worldviews. Beer instead uses the expression “classification imagination” to articulate a less rigid, more contingent and transitory notion of “gender”, typical of “decentralized media”. He understands it as a sort of continuous cultural classification that reflects the elasticity of the categories used by the platforms and their ability to constantly define new genres by identifying new connections in the analyzed data.
– Read also: On Netflix there is more and more clickbait
To delineate the musical genres up to 50 years ago, for example, contributed – apart from traditional sector media, radio and criticism – the experiences of people in sharing music with friends and in searching through the different shelves in the shops of discs. The use of those genres, despite many limitations and frequent straining, was part of wider social practices. Today, from the consultation of the usual reports shared at the end of the year on the most listened to music, it is estimated that there are 5,071 distinct musical genres on Spotify, including apparently meaningless labels such as “escape room”, “taprun”, “chamber psych “Or” sound “.
Sound engineer and data analyst Glenn McDonald is known for collaborating on the creation of a large map developed for Echo Nest, a platform owned by Spotify, which brings together hundreds of musical micro-genres. In an interview in 2018, he explained that the machine learning used to create the map evaluates music based on numerous “subjective psychoacoustic attributes of songs,” which are also very transversal attributes such as duration or beats per minute (bpm). The algorithm then traces similarities between different musicians based on those attributes and defines new genres which, despite having “invented names”, actually correspond to the tastes of real groups of people.
According to that Spotify end of year round-up thing, my 3rd most listened to genre is 'chamber psych' and I didn't even know that was a genre until today. And I can confirm that it is indeed one of my favorite genres.
– Bob Hardy (@ B0bHardy) December 2, 2020
According to my year's recap on Spotify, my third most listened to genre is “chamber psych”, and I didn't even know it was a genre until today. I can confirm that it is indeed one of my favorite genres.
Sometimes Spotify users themselves mint new genres when they create playlists. “We are constantly stocked with new labels and categories as we consume music, film and television,” says Beer, focusing on the continuity of the cataloging processes of streaming services. Through these detailed and detailed categories, he continues, our tastes can be more specific and more eclectic, and at the same time moldable.
More and more specific categories are created and stored in the metadata – attributes not directly visible – of the contents present on the platforms, and are therefore used to personalize the viewing recommendations. The way Netflix is organized through metadata, in essence, “decides what is discovered within it,” writes Beer. And, as on Spotify, on Netflix there are thousands of categories that far exceed traditional genres in quantity. In some cases they are more detailed formulations of known genres (“cult horror”) but in many others they define a long series of different attributes (“unrealistic foreign films of the 1970s” or “sci-fi horror and suspense of the 1950s”).
– Read also: The 76 thousand hidden categories of Netflix
Although Squid Game is categorized as a “Korean, Thriller TV” series, this series is associated through metadata with many other more specific categories not necessarily known to users. And Squid Game is also a very illustrative example of how algorithms can reinforce content that is already popular on a large scale. “Like on social media, once a trend starts to take hold, algorithms can direct even more attention to it,” writes Beer. And Netflix's categories do that too, he continues, telling us what content is trending or popular in our area.
“In music, art and many other areas, our preferences are largely the result of the culture in which we grew up: a multitude of people around us who tell us what is beautiful,” wrote the journalist on Insider. Brooklyn-based and tech expert Drew Austin. According to Austin, it's as if streaming platforms are making an effort instead of pretending that tastes aren't a deeply social phenomenon. Their algorithmic recommendations appear to have a “flattening and standardizing” effect on tastes, an effect even “more pronounced when the deeply social nature of tastes is minimized and content is decontextualized”.
One of the risks associated with using these algorithms to suggest content is that we still have a very limited understanding of how these tools work and the consequences of applying them on a large scale. Furthermore, the regulation of these systems generally takes place in a context of poor transparency, in which companies are not inclined to share the data collected through various metrics, which they track internally, for fear of offering competitive advantages to the competition.
– Read also: We don't know how many people watch movies anymore
«Netflix is a closed book. We don't know anything about its viewer numbers unless they tell us, and even then, those numbers are vaguely suspicious. We also know little about how they use audience data to promote their shows, or even how they decide what lives and dies on their service – reasons that have always seemed more specific than just “no one was watching”, ”he wrote in March. 2020 journalist Adam White on the Independent.
According to journalist Rob Horning, social media expert and editor of the Marginal Utility column on The New Inquiry, the concrete objective of the algorithms used by streaming platforms is not so much to better understand the tastes of users as to “reshape” those tastes so that users “experience desire on schedule and in pre-formatted ways”.
Nothing excludes, finally, Beer puts as an example, that the cultural classification of streaming platforms can keep us away from certain categories or voices, effectively limiting our experience and shaping popular culture according to particular needs. And although it is also valid for other platforms and intermediaries that in the past effectively regulated the cultural offer according to user response, streaming platforms can now use artificial intelligence tools on very large scales. And they can predict success, Canadian journalist Jon Fingas wrote in Engadget, “in ways that conventional box office numbers and ratings reports from Nielsen probably couldn't match.”
The positive aspects of these cataloging are evident, according to Beer, and that is that in most cases they allow us to orient ourselves in the face of a very high number of possibilities and to find what we like more quickly. The less positive aspects concern the fact that at the moment it is difficult to give an exhaustive answer to the question of what the evaluations underlying the decision of companies to update certain criteria, or generate new categories, or promote certain contents instead of others, meet and to what needs. in one audience rather than another. And what significant consequences all these choices have on the cultural practices of our time.
