Analysing Digital Culture 101: Recommendation Systems

Foreword

Nowadays, media are so present in our everyday life that we tend not to notice their influence and power anymore. Because of this, the Analysing Digital Culture series explores the key approaches to the analysis of new-media and reflects on their social and cultural significance from different theoretical perspectives. The series touches on some of the most popular debates on digital media such as their economy, their emancipatory and limiting power, the content recommendation systems, the circulation of memes and viral content, and the selfies culture.

Analyzing Digital Culture 101 will be divided into the following chapters:

1. Digital Culture 101: The Impact of New Media

2. Digital Culture 101: Media and Democracy

3. Digital Culture 101: Online Business Models

4. Digital Culture 101: Recommendation Systems

5. Digital Culture 101: Cultural Meaning of Memes

6. Digital Culture 101: The Attention Economy

Digital Culture 101: Recommendation Systems

Recommendations are suggestions of products, such as friends, songs, or movies, that websites provide you based on your preferences (Gillespie, 2014). These recommendations are the algorithms' predictions of what information a user might need (Gillespie, 2014). Nowadays, algorithms’ role has become so central in our everyday life that they have become "the scientific instruments of a society at large" (Gitelman, 2006, p. 5). For these reasons, this article will analyse how online recommendation systems shape the availability of content and the choices for users by applying these theories to the business models of popular platforms such as Netflix and Spotify. Also, the societal and cultural relevance of recommendation systems and algorithms is explored.

Algorithms

Gillespie (2014), a Senior Principal Researcher at Microsoft Research New England, explains in his article “The Relevance of Algorithms” the influence algorithms have on users’ everyday lives. Gillespie (2014) starts his article by saying that “algorithms play an increasingly important role in selecting what information is considered most relevant to us, a crucial feature of our participation in public life” (p.167). With these words, he addresses algorithms’ recommendation features that map our preferences and highlights the content which matches them (Gillespie, 2014). Additionally, Langlois (2012), a professor in Communication and Media Studies at York University, claims that algorithms govern the flow of information on the Internet by deciding what is meaningful for every user. This is because, in addition to helping us find the information we are looking for, algorithms also tell us what we should know and how to know it (Gillespie, 2014). Algorithms’ recommendation system is based on the information they gather from each user such as activities, preferences, and expressions (Gillespie, 2014). For this reason, algorithms, also defined as public relevance algorithms, and people mutually influence each other (Gillespie, 2014).

Figure 1: A visual representation of algorithms (Deguil, n.d.).

It is crucial to analyse algorithms sociologically, so as mechanisms that shift public discourse behind which lie human and institutional choices, and not as mere abstract technical achievements like it has been done often in the past (Gillespie, 2014) [Figure 1]. Stevenson (2018), a professor of Media Studies at the University of Amsterdam, explains that the development of technologies involves different actors, such as users, producers, and investors. These people constantly decide which technologies or media to develop and use (Stevenson, 2018). However, these decisions are not taken rationally and individually, but “rather, they are shaped by culturally specific values, beliefs and practices, political and commercial interests, as well as the material constraints of available technology” (Stevenson 2018, p. 70). This is because algorithms are systems at the centre of our information ecosystem that are the cause and also the result of different cultural forms (Striphas, 2009; Andreson, 2011). Galloway (2004), author and professor of Media, Culture, and Communication at New York University, explains that by selecting the information we see and receive, algorithms became communication technologies such as broadcasting and publishing but way more computationally organized. This suggests that algorithms’ technology and culture are not independent domains as is commonly imagined; technology’s effectiveness and rationality intertwine with culture’s subjectiveness and artistic expression (Seaver, 2014). To explain this, anthropologist Nick Seaver (2014) uses the example of films which are cultural products extremely dependent on technical devices such as cameras and editing programs. Nowadays, their distribution is strictly reliant on algorithms, as in the case of Netflix (Seaver, 2014). The same occurs for the cultural product of music, whose biggest distributor is Spotify: a platform to stream music that remains unbeaten in the music streaming market because of the customer experience its recommendation system can provide (Marius, 2021).

In his article, Gillespie (2014) also provides an elaborate description of how algorithms work in practice. A theorist of digital culture, Manovich (1999) explains that despite data structures and algorithms are often mistakenly considered as a single working apparatus, they are two distinct halves. This is because “algorithms are inert, meaningless machines until paired with databases on which to function” (Gillespie 2014, p. 169). Before being paired with algorithms, information needs to be collected and at times omitted (Gillespie, 2014). Despite algorithms being thought to be neutral and objective, they are a technology created by humans and shaped around political and organizational principles: “the result will not appear if it is child pornography, it will not appear in China if it is dissident political speech, and it will not appear in France if it promotes Nazism” (Gillespie 2014, p. 179). After a selective collection that excludes offensive content, information is rendered into data and oriented to face the algorithm (Gillespie, 2014). Afterwards, a series of anticipation cycles start where algorithms anticipate users’ inputs by gathering information about them (Gillespie, 2014). This creates an algorithmic identity, also called shadow bodies, that does not include every kind of information, but only those necessary for the algorithm to function in an anticipated way such as geolocation, activity on the website, and friends (Balka, 2011). The last step consists of the evaluation of the relevance: the selection of content that can satisfy users (Gillespie, 2014). The evaluative criteria behind this process are obscured from the public and can vary (Gillespie, 2014).

Figure 2: Netflix's business model and recommendation system (Oome, 2019).

Recommendation Systems in the Film and Music Industry

Oomen (2019) is specialized in communication, and she explains in her article "Netflix: How a DVD rental company changed the way we spend our free time” that the company has become so influential in the film industry that the verb Neflixing is now used as a synonym of bing-watching. Oomen (2019) explains that this streaming platform presents a service-based business model. This is because it does not sell a product but a service. By paying a monthly "all-you-can-eat subscription", users are provided with a streaming service that "lets you watch your favourite shows anywhere and at any time you want.” (Oomen, 2019, p.1) [Figure 2]. Madrigal (2014) is an American journalist working for the journal The Atlantic who studied the complex recommendation system behind Netflix’s business model. In his article “How Netflix Reverse Engineered Hollywood”, Madrigal (2014) analyses how the web behemoth Netflix managed to create an appealing customer experience (Seaver, 2014). The journalist explains that the recommendations that Netflix furnishes to its users depend mainly on three factors (Madrigal, 2014). Firstly, it is used a content-based filtering process in which the characteristics of the favourite movies of a user are matched with movies that have similar features (Madrigal, 2014). Secondly, it is also used collaborative filtering which consists of suggesting products that were liked by users with similar favourites to yours, so in this case with similar viewing habits (Gillespie, 2014). In this way, different types of films are recommended to different kinds of users. Thirdly, content-based and collaborative filtering are enabled and made more efficient by the characterisation of movies into very specific genres, called personalized genres (Madrigal, 2014).

Madrigal (2014) claims that Netflix deconstructed Hollywood by creating 76,897 extremely specific and unique ways to describe film genres. Madrigal (2014) explains that the extreme specificity of these genres is what makes serendipity to encounter different content possible. Netflix’s developers and programmers managed to do this with the help of large teams of people that were trained to be able to rate films (Madrigal, 2014). After a 36 pages training document, these people got paid to watch films and rate them on dozens of different attributes such as goriness, romance levels, and plot conclusiveness (Madrigal, 2014). The result was that the creators of the platform captured many different movie features that were later combined with content-based filtering and collaborative filtering (Madrigal, 2014). By doing so, Netflix is able to make its users stick around and keep using the service because they feel understood and known (Madrigal, 2014). Moreover, with the data it collects, Netflix is also able to predict what filmmakers should create in order to be recommended and sold (Madrigal, 2014). Nevertheless, Oomen (2019) argues that, since the system that Netflix uses divides its public into severe niches, it is impossible to have unique tastes and this makes viewers nothing more than the algorithmic identity that Balka (2011) describes as shadow bodies.

Figure 3: Infographic of Netflix’s data collection types (2018).

Spotify is a digital music and podcast service that changed the world of music (Marius, 2021). The data scientist explains that the platform uses the same recommendation system as Netflix, so a combination of content-based filtering and collaborative filtering (Marius, 2021). In this case, content-based filtering relies on information about the user-song relationship so the characteristics of the most frequently listened songs of a user are matched with songs that have similar features (Marius, 2021) [Figure 3]. Differently, collaborative filtering consists of suggesting songs that were liked by users in the past and assuming that they keep liking the same music genres (Marius, 2021). Both in Netflix and Spotify's case, this type of recommendation system leaves very little space for serendipity and does not allow users to have the ability to explore other products different from their tastes (Turk, 2021). The journalist, writer, and editor Turk (2021) discusses this singularly in relation to Spotify and streaming music. Nevertheless, these ideas can apply to any recommendation system because the more information about the user’s taste the algorithm has, the more precise the recommended content will be, so the easier it will be to get stuck in a loop of very similar recommendations (Turk, 2021).

Figure 4: Spotify's recommendation system (Spotify, 2020).

Conclusion

To conclude, algorithms play an essential role in today’s society because they deeply influence users’ experiences and the availability of data (Gillespie, 2014). Algorithms need to be paired with databases in order to function (Gillespie, 2014). However, these databases do not include certain information that is considered offensive by the ruling government or company (Gillespie, 2014). After collecting data, the algorithms also collect information about users in order to recommend content that they find useful and interesting (Gillespie, 2014). Very importantly, algorithms are not mere technology but an instrument that shapes and is shaped by their users and developers (Gillespie 2014; Stevenson 2018). For this reason, they have cultural relevance (Seaver, 2014). Their function is influenced by culture and at the same time, they alter and are responsible for the creation of cultural products such as films and music (Seaver, 2014). An example of this is the great success that the two service providers Netflix and Spotify had in the past years. Their recommendation systems are very similar and able to create a great customer experience by recommending very specific content that is close to users’ tastes (Madrigal 2014; Marius 2021; Seaver 2014). These apparently perfectly working mechanisms are questioned by Oomen (2019) and Turk (2021) who point out the difficulty of exploring content that does not match what the algorithm thinks is “our taste” (Turk, 2021, n.p.).

Bibliographical References

Anderson, C. W. (2011) “Deliberative, agonistic, and algorithmic audiences: Journalism's vision of its public in an age of audience.” Journal of Communication 5: 529-547.

Balka, E. (2011) “Mapping the body across diverse information systems: Shadow bodies and they make us human.” Paper presented at the annual meeting for the Society for Social Studies of Science, Cleveland, Ohio.

Galloway, A. (2004) Protocol: How Control Exists after Decentralization. Cambridge, MA: MIT Press.

Gillespie, T. (2014) “The Relevance of Algorithms.” In Media Technologies: Essays on Communication, Materiality, and Society, edited by MIT Press Scholarship Online, 87-113.

Gitelman, Lisa. 2006. Always Already New: Media, History, and the Data of Culture. Cambridge, MA: MIT Press.

Langlois, G. (2012) "Participatory culture and the new governance of communication: The paradox of participatory media." Television and New Media.

Madrigal, A. C. (2014) “How Netflix Reverse Engineered Hollywood.” The Atlantic, January 2, 2014.

Manovich, L. (1999) "Database as symbolic form." Convergence: The International Journal of Research into New Media Technologies 5 (2): 80-99.

Marius, H. (2021, December 10). Uncovering How the Spotify Algorithm Works. Medium. https://towardsdatascience.com/uncovering-how-the-spotify-algorithm-works-4d3c021ebc0

Oomen, M. (2019) “Netflix: How a DVD rental company changed the way we spend our free time.” BMI, November 29, 2019. https://www.businessmodelsinc.com/exponential-business-model/netflix/.

Seaver, N. (2014) On Reverse Engineering: Looking for the Cultural Work of Engineers. Medium.

Striphas, T. (2009) “The Late Age of Print: Everyday Book Culture from Consumerism to Control.” New York: Columbia University Press.

Turk, V. (2021). How to break out of your Spotify feedback loop and find new music. Wired UK. https://www.wired.co.uk/article/spotify-feedback-loop-new-music

Visual Sources

Cover Image: Martin Tognola (2019), “How Google Interferes With Its Search Algorithms and Changes Your Results”, Wall Street Journal https://s.wsj.net/public/resources/images/EQ-AC665_ALGO_G_20191113162940.jpg

Figure 1: Deguil, A. (n.d.) https://www.pewresearch.org/fact-tank/2019/02/13/7-things-weve-learned-about-computer-algorithms/

Figure 2: Oome, M. (2019) "Business Model Netflix" https://www.businessmodelsinc.com/en/inspiration/blogs/netflix-how-a-dvd-rental-company-changed-the-way-we-spend-our-free-time

Figure 3: Deciphering the unstoppable Netflix and the role of Big Data (2018). https://www.muvi.com/blogs/deciphering-the-unstoppable-netflix-and-the-role-of-big-data.html

Figure 3: Spotify (2020) "What Goes Into Spotify's Personalized Recommendations" https://storage.googleapis.com/pr-newsroom-wp/1/2020/10/Spotify_Streams_102720_v4.png