Curation or Control: Analyzing the Socioeconomic Context of Recommendation Engines for Cultural Content

Abstract

Even though recommendation engines are unmissable when it comes to recommending either online news or cultural content, there seems to be a different discourse for both spheres. The use of recommendation engines for selecting and recommending news and information and the potential problems that come with it are elaborately discussed and studied in public and academic spheres, but when it comes to the possible negative consequences of recommendation engines for cultural content both the public and academic debate seem to be largely absent. This thesis scrutinizes the socioeconomic factors that could potentially play a role in this difference in debate. Through the framework of José van Dijck’s analysis of platforms as distinct microsystems I have studied recommendation engines in their appropriate context. I have scrutinized the socioeconomic elements of music streaming service Spotify and uncovered to which extent they create a different situation than is the case for recommendation engines for news and information.

In doing so, I have found that Spotify is currently in the unique position where their “key user metrics are very strongly associated with diverse listening” since users that are interested in a more diverse musical diet are more likely to pay for a subscription to the platform (Anderson et. al 2156). This unique position could be explained by means of their subscription-based model. This subscription based model guarantees a certain kind of quality: if the user is paying for a service, in this case a personalized music experience that is diverse in its recommendations, a platform would ultimately profit from being as accurate with this service as possible. This is in contrast to platforms for news and information: as most of these platforms are freely accessible, they mostly have to rely on user engagement in order to sell more advertisements. The longer users stay on a platform and interact with it in various ways, the more advertisements can be sold. Therefore, the platform’s recommendation engines are designed with the goal of engagement in mind. The guarantee of accurate recommendations disappears when a user is not paying for a specific service, and recommendation engines that are implemented on platforms that are dependent on advertising revenues are therefore mostly built to work for the platform itself rather than the user. This combination of engagement rather than accuracy, the disappearance of a guaranteed quality of recommendations and the general black-boxed nature of algorithms is a recipe for popular imaginaries such as the filter bubble, echo chamber, popularity bias or confirmation bias - imaginaries that heavily shape the critical debate surrounding these recommendation engines for news and information platforms.

As is a general consensus in the debate surrounding the potential negative consequences of recommendation engines for news and information, recommendation engines and the algorithms that they consist of are heavily intertwined with a platform’s socioeconomic context, and more often than not they are built to work for the platform itself rather than for its users. I have found that the socioeconomic elements of my case study Spotify are indeed different as compared to those of news and information. Not only could my findings offer a possible explanation for Spotify’s unique position in which the company would profit from more diverse recommendations, but they have also highlighted the potential of scrutinizing socioeconomic elements as a way of interpreting an increasingly important debate.