Social media feeds as curatorial assemblages: A conceptual framework

Abstract

Social media feeds are a central object of analysis and a frequently recurring topic in the study of digital cultures, but their complex and ephemeral nature resists conceptualization. This paper addresses this challenge by approaching feeds as “assemblages.” Using Instagram and TikTok as primary examples, we employ the concept of the “feed-as-assemblage” to understand how a feed’s constitutive elements (e.g., algorithms, data, devices, interfaces, human actors, practices) form relations both at the micro-level of individual feeds and at the macro-level between all the feeds of a given platform. These relations make up intricate sociotechnical ensembles driven by curation, which is to say, the feed-as-assemblage constantly selects specific elements and rearranges their relations in order to generate value. On the basis of this conceptual framework, we focus on three central elements in the feed-as-assemblage – recommender systems, interfaces, and user practices. These elements serve in turn as heuristic access points into the production of (and exchange between) different forms of value. The general aim of this paper is to offer a conceptual approach that brings the strengths of multiple disciplines to bear in analyzing social media feeds.

Keywords

assemblage curation interfaces recommender systems social media feeds user practices

Introduction

What the millions of people who use social media encounter on platforms such as Instagram or TikTok is commonly referred to as “feeds.” Social media feeds and their principles of algorithmic selection are frequently discussed as part of the infrastructure of platforms such as Facebook (Helmond, 2015), Instagram (Leaver et al., 2020), and TikTok (Kaye et al., 2022); in studies of user experiences (Schellewald, 2023; Sülflow et al., 2019; Ytre-Arne and Moe, 2021); or as part of social media marketing (Carah, 2014; Simatzkin-Ohana and Frosh, 2022). Some studies have focused on feed-related phenomena such as scrolling, swiping, browsing, and clicking (Anikina, 2021; Triệu et al., 2021), or on the “algorithmic imaginaries” connected to feeds (Bucher, 2017; Schulz, 2023). Others have addressed social media feeds in the context of broader platform cultures (Gillespie, 2016), discussed the politics and power relations embedded in algorithmic sorting (Bandy and Diakopoulos, 2023; Bucher, 2012, 2018) and (in)visibility (Are, 2024; Duffy and Meisner, 2023; Savolainen, 2022), investigated the archival and temporal logics of social media feeds (Geismar, 2017; Kaun and Stiernstedt, 2014), or conceptualized the relation between feeds and their interfaces (Schulz and Matzner, 2020).

Although feeds are a central topic in this growing interdisciplinary literature on social media, researchers have yet to adequately address the conceptual challenge they pose. What is a social media feed? Is it a personalized selection of images, videos, and texts? Is it a stream of digital representations on interfaces? Is it a complex combination of algorithms – a “thinking” AI-based system? Does a feed exist only through the practices and experiences of its users? Or is it perhaps all of these things at once?

In this paper, we argue that this conceptual challenge can be addressed with the help of assemblage thinking. Specifically, we propose that social media feeds be thought of as curatorial assemblages. Rather than trying to conceptualize feeds as fixed and clearly defined entities, our framework describes the ongoing process of their becoming. Assemblage thinking helps us understand how algorithms, data, devices, interfaces, human actors, practices, etc. and the relations between them participate in the ongoing curation of social media feeds. In this approach, the strengths of different disciplines contribute to a comprehensive understanding of an ephemeral phenomenon.

Our approach is intended to enrich studies on social media in general, but we focus on two particularly prominent platforms featuring a mix of visual, textual, and auditive content: Instagram and TikTok. We begin by providing a brief overview of what is commonly referred to as “assemblage thinking.” We then introduce the concept of the “feed-as-assemblage” before discussing in the third section how the assemblage follows the principles of curation. In the fourth section, we consider how three specific elements within the feed-as-assemblage – recommender systems, interfaces, user practices – create relations to other elements and in the process produce multiple kinds of value. Finally, we identify several heuristic access points enabled by our framework for the analytical investigation of social media feeds.

Assemblage thinking

The concept of the “assemblage” goes back to the philosophy of Deleuze and Guattari (1987) and was further developed by theorists such as Bennett (2005), Latour (2005), DeLanda (2006, 2016), Nail (2017), and Buchanan (2021). It has been applied in a variety of disciplines, including cultural anthropology (Hansen and Koch, 2022) and media studies (Macgregor Wise, 2017). Strictly speaking, there is no single assemblage theory, but rather a multitude of texts associated with the concept of the assemblage. Accordingly, this approach is now often referred to as “assemblage thinking” more generally (Anderson et al., 2012; Hansen and Koch, 2022: 10; Müller and Schurr, 2016; Tseng, 2023). In our work, we too adopt this term.

The idea of applying assemblage thinking to the study of digital technologies in general, and social media in particular, is not new. For example, the concept has been used to study algorithms and data in general (Bucher, 2018; Carlson et al., 2021; Kitchin and Lauriault, 2018; Rosenbaum, 2020), and artificial intelligence in particular (Bareither, 2023b, 2024; Kim et al., 2023; Lindgren, 2023: 22–42; Tseng, 2023; Vepřek, 2024). In our exploration of social media feeds as assemblages, some of these existing approaches have proven helpful. Our framework is particularly indebted to authors who have used assemblage thinking in approaching social media. Bucher (2013) proposed thinking of Facebook as an assemblage that facilitates friendship and has since extended this approach in her work on algorithmic power and politics (Bucher, 2018). Drawing on Bucher’s work and assemblage thinking, Gerrard and Thornham (2020) have described social media as “sexist assemblages” and examined discriminatory aspects of search functions, content moderation, and algorithmic recommendations. Are (2024: 4) has characterized practices of “flagging and de-platforming” on social media as “moderation assemblages”. Other authors have used the notion of assemblage in relation to social media without explicitly drawing on the strand of assemblage thinking connected to Deleuze and Guattari (Gerlitz and Lury, 2014; Lupton and Southerton, 2021). Nonetheless, these texts employ and foster a relational style of thinking in social media and platform studies. Our own theoretical work builds on these and other approaches to outline a conceptual framework for exploring the contributions of assemblage thinking to research on social media platforms in general and the role of feeds in particular.

In its most abstract sense, the term assemblage refers to “wholes whose properties emerge from the interactions between parts” (DeLanda, 2006: 5). Assemblages are socio-technical ensembles of people, things, spaces, practices, etc., whose specific properties are not only the sum of their parts, but emerge from the relations between their heterogeneous elements. In assemblage thinking, these networks of relations are conceived neither as random nor as static, but as in continuous movement.

Each assemblage brings together a broad range of both human and non-human elements. “Elements” (Nail, 2017) or “components” (DeLanda, 2006) are alternating terms used in assemblage thinking to denote any entity within an assemblage – here, we only use the former term. These elements are characterized by “distributive agency” (Bennett, 2005: 462). This means that both human and non-human elements actively participate in building relations and connections within the assemblage. It is not only the properties of elements that determine their function within an assemblage, but the capacities that emerge in their relation to other elements (DeLanda, 2006: 11). These relations between elements of an assemblage are “relations of exteriority” and not “relations of interiority” (DeLanda, 2006: 18; Nail, 2017: 23). This means that assemblages are never closed systems. Rather, their elements can also exist independently of the assemblage and/or be part of several assemblages at the same time. Consequently, “relations of exteriority” also imply that assemblages can simultaneously exist on different micro and macro-levels of scale. In assemblage thinking, individual micro-assemblages can come together and constitute more comprehensive macro-assemblages – and conversely, macro-assemblages can comprise numerous smaller micro-assemblages (Bareither, 2024: 8; DeLanda, 2006: 16–17).

The feed-as-assemblage

We propose that social media feeds be thought of as assemblages. Feeds are complex sociotechnical ensembles whose properties emerge from the interactions and relations between heterogeneous elements: human actors and their practices, technological infrastructures and algorithms, data and content, as well as interfaces and their affordances. On the micro-level, a single feed of a single user (e.g., a user’s For You Feed on TikTok or Home/Explore Feed on Instagram) is a micro-assemblage that brings together a multiplicity of elements and emerges through the relations between them. At the same time, the individual feed continuously interacts with the other feeds (i.e., the other micro-assemblages) of the user and those of millions of other users. Social media platforms are algorithmic systems designed to create connections and generate a continuous flow of exchange between feeds, making it impossible to determine where one feed ends and another begins. When feeds come together, they form much larger macro-assemblages, for example, all TikTok feeds are elements of an assemblage that contains all TikTok’s content, all its users, all the platform’s data and meta-data, and the complete global infrastructure of the platform. In short, individual social media feeds are micro-assemblages that are always connected to other feeds while simultaneously being shaped by and part of global macro-assemblages.

One of the advantages of assemblage thinking is that it encourages us to think about how different assemblages (at similar or different levels of scale) continuously interact with each other. This is essential for understanding what social media feeds are: complex wholes that emerge in constant processes of relational becoming. In the following, we use the concept “feed-as-assemblage” to refer to the ever-changing entanglements connecting sociotechnical elements of social media feeds at the micro- and macro-levels. That is to say, the feed-as-assemblage simultaneously comprises individual feeds, the sum of all feeds on a given platform, and the relations between them.

The feed-as-assemblage is not just an ensemble of different elements; it is also a productive force. In her study of social media, Bucher argues that the assemblage of Facebook, “by organizing heterogeneous relations in a specific way, constitutes a productive force: it makes new relations possible. The concern is not so much with what the assemblage is, but rather with what it can do, by bringing heterogeneous elements together. . .” (Bucher, 2013: 481). For our work, we take this concern with the assemblage of social media as a starting point to ask: What does the feed-as-assemblage do? The answer is curation.

Curation

The term curation comes from the Latin curare, which roughly translates to taking care. During the Roman Empire and the Middle Ages, it denoted the process of attending to people or material infrastructures (Balzer, 2014: 30; Kathke et al., 2022: 72–73). Only later was it adopted by museums and artists before eventually becoming a catch-all term for any activity loosely associated with skillful selection and presentation. Social media platforms now use the term in their own marketing language. TikTok, for example, describes its For You Feed as “a stream of videos curated to your interests” (TikTok, 2020). Though the notion has become something of a buzzword, it still holds analytical value as a concept. In museum and artistic practice, curation is described as an “activity of putting together” (von Bismarck et al., 2012: 8), a practice whose task “is to make junctions, to allow different elements to touch” (Obrist, 2014: 1). While these definitions emphasize the function of curatorial practices in creating relations between things, others focus on the aspect of selection. For example, in the context of social media, sociologist Jenny L. Davis suggests thinking of curation as a “discriminate selection of materials for display” (Davis, 2017: 771). Combining these two perspectives provides a better understanding of what the feed-as-assemblage does: it purposefully selects elements and arranges specific relations between them.

However, this is just the first step of conceptualizing curation. In museum literature, the objects of curation – what is selected and arranged – are often artworks or historical artifacts. Outside of these contexts, however, the object of curation can be anything: “Curation as a concept has expanded to encompass the sphere of everyday life” (Kathke et al., 2022: 72; see also Reckwitz, 2020: 214). The concept has become especially popular in digital cultures (Davis, 2017; Snyder, 2015). But what distinguishes curation from other processes of selecting and arranging relations between elements? If these processes of curation are purposeful, what exactly is their purpose? In both museum literature and public discourse, curation is often associated with some form of value creation. Curation intends to make a certain arrangement of elements valuable for someone (or something). To call the feed-as-assemblage curatorial is to say that it is a sociotechnical ensemble that selects elements and arranges the relations between them in order to generate value.

This idea resonates with the approaches of authors who stress how different kinds of value production are essential for how social media platforms operate (Bucher, 2018; Helmond, 2015; Gerlitz, 2016). Notably, a 2021 leak of an internal document from the company that owns TikTok showed that the platform itself operates with and through different types of “value” (Meßmer et al., 2024). Our idea of the feed as a curatorial assemblage situates the role of multi-layered value production within the dynamics of the feed. We distinguish roughly four types of value for the analytical purposes of this paper. The first type of value is experiential value, a term we use to describe and summarize the type of personal value that users derive from emotional and aesthetic experiences with social media content, as well as from the informational and epistemic insights it provides. The second kind of value we call social value, which functions as a kind of “social capital” in the sense of Bourdieu (1986: 21–24): users and creators gain recognition and attention from friends or the public for their content or their interaction with content (e.g., through “likes”). The third kind of value is the economic value produced by users’ practices of consumption for stakeholders and the companies behind social media platforms. Creating experiential value for “you” is only the target of the feed because “you” are the generator of “behavioral surplus” (Zuboff, 2019), which is used to create economic revenue for platform companies via targeted advertisements or for individual content creators via, say, product placement or “promotional politics and authenticity labor” (Arnesson, 2023). A fourth and final type of value is what we call analytical value. This is the value that is produced by capturing and analyzing specific user practices and by transforming this information into data profiles that are used by the platform to interpret human behavior. Whereas analytical value mainly supports the platform in generating economic value via personalized advertisements, it can also have value in itself, as when the analysis of user data helps stakeholders (e.g., marketing specialists, politicians, entertainment companies, researchers, and scientists) in achieving their ends. Producing these four types of value is the goal of the feed’s curatorial process, but it is also the driving force behind the assemblage and the central dynamic through which it continuously comes into being. Without curation, the assemblage would disappear, because users would lose interest and platforms could not generate economic revenue. The feed-as-assemblage has no choice but to engage in indefinite curation. Accordingly, when we speak of the feed-as-assemblage, we always mean the feed as curatorial assemblage.

Curatorial relations and dynamics

Now that we have outlined the basic concept of the feed-as-assemblage, we can ask: How does this assemblage do curation? How does it select specific elements and arrange relations between them in order to generate value? For assemblage thinking, there is rarely an all-encompassing answer to this question. Rather, assemblage thinking encourages researchers to “zoom in” on specific relations and dynamics. We have identified three heuristic access points for studying such relations and dynamics in the context of social media feeds. Each focuses on a central element in the curatorial process: (1) recommender systems, (2) interfaces, and (3) user practices. Below we describe how each element carries out curatorial functions in the way it relates to other elements (e.g., data, content, human bodies), and how they produce different kinds of value in the process.

Recommender systems

An essential dynamic of the feed-as-assemblage is determined by underlying algorithmic and AI-based systems that function as “curatorial code” (Davis 2017: 776) and are often referred to as “sorting algorithms” or “recommender systems” (Burke et al., 2011; Gillespie, 2016). Platforms such as TikTok and Instagram advertise these systems as powerful curatorial tools. Meta states that “the AI system has models that help it make predictions about content you’ll find most relevant and valuable” (Meta, 2023c). TikTok, in a similar way, emphasizes that due to its recommender system “all feeds are unique” and “tailored to you” (TikTok, 2023a). Although their general purpose is widely known, the exact way in which algorithmic sorting systems function remains largely opaque. Both Meta and TikTok have conducted transparency initiatives for several years (Meta, 2023a, 2023b, 2023c; TikTok, 2023a, 2023b), but these have yet to provide much detail about the company’s proprietary algorithms and many questions remain open (Grandinetti, 2023). From what we can distill from available information, the recommender systems of the two platforms apply two types of processes: collaborative recommendation and content-based recommendation (Burke et al., 2011).

For collaborative recommendation, TikTok and Instagram first capture what they call “signals” (Meta, 2023a; TikTok, 2020) sent by users via clicks, likes, follows, shares, or simply by watching a post for a certain length of time. These signals are used to create data profiles, which represent on an abstract level a user’s interests and tastes. The data profiles are then checked for similarities with other profiles. In assemblage thinking terms, they are embedded in a web of relations between different data points where similarities between data points or user profiles play a key role. This web of relations is the blueprint for the selection of more content that matches a specific user’s interests and tastes and is likely to create engagement. A short statement on the TikTok website describes this process in accessible language: “For example, if User A likes videos 1, 2, and 3 and User B likes videos 1, 2, 3, 4, and 5, the recommendation system may predict User A will also like videos 4 and 5” (TikTok, 2023a).

The second approach is content-based recommendation. Whereas collaborative recommendation operationalizes relations between user’s data profiles, content-based recommendation concentrates on relations between users and specific types of content. To identify specific types of content that an individual user might like, platforms need to categorize content in such a way that it can be matched to the preferences of the user. Platforms have shared few details about how this process works. The evidence available suggests that TikTok identifies a number of high-level and sub-level “interests” associated with the content (Semenova et al., 2024), whereas Instagram operates with a set of more than 500 “concepts” or “labels” that are assigned scores for each image (West et al., 2024).

This process of classification underlying content-based recommendation is possible only by using AI-based classifiers that are trained on datasets generated either externally or by the companies behind the platforms. For example, West et al. (2024: 11) show that of the more than 500 concepts applied by Instagram, 87 match labels used by ImageNet, one of the largest image datasets used for the training of AI-based classifiers. This means that Instagram’s recommender system integrates categories and structures developed as part of other “data assemblages” (Kitchin and Lauriault, 2018). This is important because structuring the data that underlies recommender systems ultimately shapes the curatorial process, and it is the reason why an assemblage thinking approach requires us to ask how the data that fuels recommender systems is collected and categorized. It also points our attention to the server parks and other electronic infrastructure needed to run social media platforms (Vonderau, 2018). This infrastructure is the material foundation of any feed-as-assemblage, and it also shapes the curatorial process. For example, the laws and policies that apply at the location of the servers storing social media data can influence whether and how a platform uses the data in the curation process (Kumar and Thussu, 2023).

From an assemblage perspective, then, the principles of collaborative and content-based recommendation are not simply linear processes of recommending individual pieces of content. Rather, the recommender system builds a complex web of valuable relations between users (collaborative recommendation) and between users and content (content-based recommendations) that is closely entangled with the data assemblages, the material infrastructures, and the policies they are embedded in. The relations crafted by the recommender system are the foundation of the curatorial process for generating multiple forms of values. The recommender system is responsible for capturing user practices that have analytical value for the platform, allowing it to match content to a user’s interests and tastes. At the same time, this process creates experiential value for each individual user, because the feed contains content that offers desirable experiences or information. The experiential value can be closely linked to the social value produced by the recommender system, because the system makes the content or interactions of users visible to others and hence invites the creation of new social connections. Finally, these forms of value are why users pay attention to the feed, which generates economic value for the platform through personalized advertisements or for individual creators through product placement, sponsorships, etc. By facilitating these forms of value generation, and by allowing different kinds of value to be exchanged or converted, collaborative and content-based recommendation acts as the engine at the heart of the feed-as-assemblage.

Interfaces

The second key access point in investigating the feed-as-assemblage is the user interface. As curatorial agents, interfaces regulate the relations and data flows between layers of software and hardware on the one hand and human actions on the other.

The first curatorial function of the user interface in the feed-as-assemblage is to generate experiential value by making the feed visible to its human users. As designed surfaces (Drucker, 2014), user interfaces not only materialize binary data in the form of texts, images, videos, icons, or sounds but also give the feed a specific “look and feel.” For instance, the layout of Instagram’s graphical user interface plays an active role in the user’s experience: by placing users in the middle of the “home feed” when they open the app and letting videos play directly by default, the app creates a sense of immediate involvement. In this way, the user interface provides a stage for platform spectatorship, where the content selected for each user by the platform’s recommender system is displayed in a carefully crafted way. The spectatorship keeps users engaged in the attention economy of the platform and leads to an “affective scroll” (Anikina, 2021), where looking at videos and images and scrolling become one and the same experience. This “tactile vision” (Cooley, 2004) is a central feature of the smartphone, where practices of seeing are always closely entwined with movements of the hand.

Apart from showcasing content and constructing a tactile economy of attention, user interfaces determine what is visible and what is not (Schubbach, 2007). It makes a difference whether quantitative information such as the number of friends, likes, and comments is visible for users, as the Facebook Demetricator, a free web browser extension that hides all visible metrics, demonstrates (Gerlitz and Helmond, 2013; Helmond, 2015: 109–111). Visible numbers can work as a strong incentive for other users to like or share something (Bucher, 2012), which shapes the creation of experiential and social value. For example, an interface that makes the popularity of particular pieces of content visible by displaying the number of likes influences how users emotionally engage with the content and how much social value they assign to it or its creator.

As these examples already imply, user interfaces also carry affordances of interaction (Bucher and Helmond, 2018). This is their second curatorial function. Through “operative iconicity” (Krämer, 2022: 268), they offer predefined actions (like, save, share) or forms of navigation (scrolling, swiping, pausing) that refer back to a longer history of interactive screen media, human-computer interaction, user interface design, and platform policies. In this sense, user interfaces are dispositives for “handling” computer technology (Wirth, 2024: 71–114). As “zone[s] of affordances organized to support and provoke activities and behaviors” (Drucker, 2014: 147–148), user interfaces offer “pre-structured grammars” (Gerlitz, 2016: 26) for users to engage with content and other users. On Instagram, posts from accounts marked as “favorites” are ranked higher in the home feed and displayed more frequently. Individual posts can be marked with a heart (“liked”) or “bookmarked,” that is, saved in a personal archive. “Not interested” flags allow users to hide certain content or participate in Instagram’s “sensitive content control” by hiding or reporting posts that either violate the community guidelines or are perceived by users as disruptive. Easily accessible, these icons or interface features “design” habits: they enable, prompt, or restrict specific kinds of interactions in the form of movements (such as scrolling) or actions (such as liking) that can create analytical value for the platform.

While prompting specific user actions, interfaces also allow the platform to capture these practices for algorithmic curation. Accordingly, the third curatorial function of user interfaces is to make user interactions calculable and algorithmically processable to assist the platform and its recommender system in creating analytical value and, by extension, economic value. Every interaction of the user with the interface triggers a series of additional interface interactions. A “like” or “save” is fed into the platform’s infrastructure as a machine-readable value, is linked to the platform’s comprehensive data collections and computational “intelligence” via Cloud services, is analyzed in relation to larger aggregations of data, and is finally fed back to the user interface as a newly compiled feed (Bratton, 2015).

But the interface for end users is just one of several interfaces to consider within the curatorial assemblage. For business users and professional accounts, Instagram offers a user interface equipped with analytics tools and a customizable dashboard, allowing users to analyze content performance in various ways. Additionally, social media platforms offer application programming interfaces (APIs) for developers, marketers, data analysts, and other user groups. These software infrastructures are a key strategic element in today’s platform economy (Bucher and Helmond, 2018: 247; van der Vlist et al., 2022). APIs allow platforms to share data with third parties or to integrate functionalities with other services. For instance, although Instagram, Facebook, and WhatsApp appear separate to end users, they are all part of the parent company Meta Platforms Inc. and are connected by a shared back-end that generates its own ecosystem of data-based apps and services through constant adaptations of APIs and standardizations of the technical infrastructure (van der Vlist et al., 2022: 6). By applying pre-defined “back-end grammars” (Gerlitz, 2016: 28), APIs enable advertisers to access large volumes of data about users and their behavior, on whose basis they can place micro-targeted ads in users’ feeds. APIs provide different points for user groups to access a platform’s data collections and thus help stakeholders convert this information into analytical and monetary value.

Overall, interfaces are “threshold condition[s]” (Hookway, 2014: 17) that draw together heterogeneous and seemingly incompatible elements such as technical processes and user practices. Interfaces relate users and algorithmic processes to each other so that they constitute each other mutually (Schulz and Matzner, 2020: 154). As we have argued, the curatorial agency of interfaces unfolds at different levels: user interfaces materialize the feed, making it visible for end users and affording specific user interactions. At the same time, they enable recommender systems to monitor user actions and transform them into calculable units, which ultimately allows platforms to make additional connections and expand data practices. Together, user interfaces assemble a diverse range of human and technological elements that generate multiple forms of value.

User practices

The third element of the feed-as-assemblage we want to discuss is user practice. How do user practices relate to other elements of curation? A central practice in this context is the creation of content through “curatorial labor” (Scolere and Humphreys, 2016). Of course, there is no clear boundary between users who consume social media content and those who produce it. Many people who interact with social media are “prosumers” (Ritzer and Jurgenson, 2010) and participate in both practices simultaneously. Nonetheless it makes sense to think about content creators as a particularly influential group in any curatorial assemblage (Carah, 2014; Davis, 2017: 772–773) because they engage in practices of creating and selecting, of editing and contextualizing, of sharing and making visible, and of embedding and circulating social media content. All are essential to the process of curation that constitutes the feed-as-assemblage.

By producing content that relates to everyday practices and performing these practices in an often stylized and aestheticized way, content creators integrate our material and social worlds into the macro-assemblages of social media platforms, where they reach into the micro-assemblages of individual feeds. Through content creation, curatorial assemblages come to draw relations between the feed-as-assemblage and the “curated lives” (Reckwitz, 2020: 214) of users outside the feed.

In this process, creators provide content that has experiential value for other users, which leads to social, analytical, and economic value down the line. For example, a social media post needs to create an emotional or aesthetic experience or have some informational value so that it can catch a user’s attention. Only then can it have social value (such as when a user recognizes that they share mutual feelings with the creator), analytical value (such as when the platform registers that a user is interested in a specific topic), or economic value (such as when knowledge about the user is used to display ads).

The complementary practices to content creation are the practices of content consumption. Like the former, the latter also actively participate in the curatorial process. In scrolling through the feed and looking at content – and by liking, following, sharing, etc. – users are sending signals to the recommender system and participating in a process of “consumptive curation” (Davis, 2017: 773–775).

The curatorial function of consumption practices can be explicit or implicit. An example for explicit curation is the practice of liking content via the platforms’ “social buttons” (Gerlitz and Helmond, 2013). Users who like content are usually well-aware that their likes will be used by the social media platform to “understand” their taste. Accordingly, they might like a type of content that they want to see more frequently represented in their feed, and they might not like content that they want to avoid in the future. In this way, the practice of liking, associated with the consumption of content, becomes an explicit curatorial practice. The practice is shaped by users “algorithmic imaginaries” (Bucher, 2017; Ytre-Arne and Moe, 2021), that is, “the way in which people imagine, perceive and experience algorithms and what these imaginations make possible” (Bucher, 2017: 31). If a user imagines that a like will have certain effects on the composition of future feeds, it will influence how the user applies the practice of liking. These imaginaries are also highly influential in controversies around shadowbanning and algorithmic (in)visibility, shaping how creators situate themselves and their agency within the feed-as-assemblage (Are, 2023; Duffy and Meisner, 2023; Savolainen, 2022). Acknowledging the role of such imaginaries for the feed-as-assemblage is important because it demonstrates that the relations between user practices and recommender systems is not linear. It is not only recommender systems that capture and interpret users’ behavior. Users also try to understand and interpret what algorithms do. The curatorial relations between both sides thus depend on processes of mutual interpretation (Schulz, 2023).

An example of more implicit curation is the practice of scrolling. Both TikTok and Instagram measure the time that a user stays with a specific piece of content and processes this data through their recommender systems (Blank and Xu, 2016; TikTok, 2020; see also TikTok, 2023a). The rhythmic scrolling and halting while looking at the feed is a curatorial practice that influences which content will appear in the feed in the future. For this kind of curation to take place, it is of secondary importance whether or not users are aware of how recommender systems track their behavior. Assemblage thinking allows and encourages us to consider how users, their bodies, and their experiences fulfil curatorial functions in the assemblage even if they are not intentionally participating in curation. While halting the scroll and viewing a piece of content for a longer time might not be an intentional curatorial decision, it is still an implicit decision made by a user’s incorporated “know-how and a certain way of wanting and feeling” (Reckwitz, 2002: 254). Another concept to describe this is “taste” (Paßmann and Schubert, 2021). Within the curatorial assemblage, taste can lead a user to implicitly make certain curatorial decisions that shape the feed so that it “matches” the user’s taste as much as possible and thus creates a strong experiential value.

In this section, we have discussed the role of content creation and content consumption and how these create essential curatorial relations in the feed-as-assemblage. Through the creation of content, users curate relations between selected elements of their everyday lives and social media feeds. At the same time, content consumption constitutes both explicit and implicit curatorial practices. These are shaped by algorithmic imaginaries as well as by users’ bodies and tastes. We have shown, in sum, how users, their bodies, and their practices relate to the elements of the feed-as-assemblage and are central parts of the curatorial process.

Conclusion

In this paper, we have used assemblage thinking to develop the concept of the feed-as-assemblage, a complex sociotechnical ensemble whose properties emerge from the interactions between its heterogeneous elements. The feed-as-assemblage consists of both individual feeds, the sum of all feeds of a given platform, and the relations between them at the same time. Further, the feed-as-assemblage is characterized by the constant curation of specific elements and relations between them to create different forms of value. For our purposes, we have focused on experiential, social, analytical, and economic value. The goal of the first part of the paper was to offer a theoretical conceptualization of social media feeds. Our framework is meant to capture these highly complex objects and approach them at multiple micro- and macro-levels at the same time. In this way, we hope to move forward an interdisciplinary discussion of what feeds are and of how to conceptualize them in a way that is accessible and productive for interdisciplinary thinking.

A challenge we face is that assemblages can never be studied or described holistically. We will never be able to describe the feed-as-assemblage in its entirety, with all its elements and relations. In the second part of this paper, we thus showed how with the feed-as-assemblage framework we can explore different heuristic access points and different elements of the assemblage in their contribution to the curatorial process. We have shown how recommender systems, interfaces, and user practices act as curatorial agents that relate to other elements of the assemblage and enable various kinds of value production, transformation, and exchange. We believe that these three access points, and the relations they are embedded in, are central for all social media feeds. However, other elements such as material infrastructures, data, content, aesthetics, taste, bodies, imaginaries, etc. (all elements we have touched on) promise to be equally productive as access points to the feed-as-assemblage, depending on the questions addressed by specific research projects.

Our exploration of recommender systems, interfaces, and user practices has been limited to the scope of a single paper. A deep dive into the feed-as-assemblage would require a far more detailed analysis. We believe that such an analysis could greatly profit from interdisciplinary approaches. Ethnographic and qualitative/quantitative approaches in anthropology, communication studies, and social science are well suited for understanding user practices and experiences vis-a-vis technological affordances; media studies approaches can help understand the dispositives and media environments that shape digital infrastructures and interfaces; and computer science can build tools that can partly reconstruct how recommender systems participate in the curatorial process, even if they remain unable to crack the “black box” of proprietary platforms. Ultimately, our framework is intended to show that understanding the feed-as-assemblage can benefit from interdisciplinary approaches. Conversely, it can foster the interdisciplinary study of social media feeds by allowing different disciplines to bring their particular strengths to bear in studying the same complex sociotechnical assemblage.

Footnotes

Acknowledgements

We would like to thank our colleagues Tim Gollub, Lisa Rein, Benno Stein, and Ann-Marie Wohlfarth for their work as part of the DFG-funded research project "Curating the Feed: Interdisciplinary Perspectives on Social Media Feeds and their Curatorial Assemblages," which provided the basis for this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The research project “Curating the Feed: Interdisciplinary Perspectives on Social Media Feeds and their Curatorial Assemblages,” on which this article is based, was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 421299207.

ORCID iDs

Christoph Bareither

Sabine Wirth

References

Anderson

Kearnes

McFarlane

, et al. (2012) On assemblages and geography. Dialogues in Human Geography 2(2): 171–189.

Anikina

(2021) Automating platform spectators: Algorithmic montage and affective scroll in TikTok. Digital Culture & Society 7(2): 119–138.

Are

(2024) The assemblages of flagging and de-platforming against marginalised content creators. Convergence: The International Journal of Research into New Media Technologies 30(2): 922–937.

Arnesson

(2023) Influencers as ideological intermediaries: Promotional politics and authenticity labour in influencer collaborations. Media, Culture & Society 45(3): 528–544.

Balzer

(2014) Curationism: How Curating Took Over the Art World and Everything Else. Toronto: Coach House Books.

Bandy

Diakopoulos

(2023) Facebook’s news feed algorithm and the 2020 US election. Social Media + Society 9(3): 1–15.

Bareither

(2023) Museum-AI assemblages: A conceptual framework for ethnographic and qualitative research. In: Thiel

Bernhardt

(eds) AI in Museums: Reflections, Perspectives and Applications. Bielefeld: transcript, pp.99–113.

Bareither

(2024) Cultures of artificial intelligence: AI assemblages and the transformations of everyday life. Zeitschrift Für Empirische Kulturwissenschaft. Journal for Cultural Analysis and European Ethnology 2024(1): I–XIX.

Bennett

(2005) The agency of assemblages and the North American blackout. Public Culture 17(3): 445–465.

10.

Blank

(2016) More articles you want to spend time viewing. Available at: https://about.fb.com/news/2016/04/news-feed-fyi-more-articles-you-want-to-spend-time-viewing/#:~:text=For%20example%2C%20we%E2%80%99ve%20found%20that,the%20post%20or%20the%20headline (accessed 22 April 2025).

11.

Bourdieu

(1986) The forms of capital. In: Richardson

(ed.) Handbook of Theory and Research for the Sociology of Education. New York: Greenwood, pp.241–258.

12.

Bratton

(2015) The Stack: On Software and Sovereignty. Cambridge, MA: MIT Press.

13.

Buchanan

(2021) Assemblage Theory and Method: An Introduction and Guide. London, New York: Bloomsbury Academic.

14.

Bucher

(2012) Want to be on the top? Algorithmic power and the threat of invisibility on Facebook. New Media & Society 14(7): 1164–1180.

15.

Bucher

(2013) The friendship assemblage: Investigating programmed sociality on Facebook. Television & New Media 14(6): 479–493.

16.

Bucher

(2017) The algorithmic imaginary: Exploring the ordinary affects of Facebook algorithms. Information, Communication & Society 20(1): 30–44.

17.

Bucher

(2018) If. . .Then: Algorithmic Power and Politics. New York: Oxford University Press.

18.

Bucher

Helmond

(2018) The affordances of social media platforms. In: Burgess

Marwick

Poell

(eds) The SAGE Handbook of Social Media. London, New York: SAGE Publications Ltd, pp.233–253.

19.

Burke

Felfernig

Göker

(2011) Recommender systems: An overview. AI Magazine 32(3): 13–18.

20.

Carah

(2014) Curators of databases: Circulating images, managing attention and making value on social media. Media International Australia 150(1): 137–142.

21.

Carlson

Eggel

Franken

, et al. (2021) Approaching code as process: Prototyping ethnographic methodologies. Kuckuck. Notizen Zur Alltagskultur 21(1: Code): 13–17.

22.

Cooley

(2004) It’s all about the fit: The hand, the mobile screenic device and tactile vision. Journal of Visual Culture 3(2): 133–155.

23.

Davis

(2017) Curation: A theoretical treatment. Information, Communication & Society 20(5): 770–783.

24.

DeLanda

(2006) A New Philosophy of Society: Assemblage Theory and Social Complexity. London; New York: Continuum.

25.

DeLanda

(2016) Assemblage Theory. Edinburgh: Edinburgh University Press.

26.

Deleuze

Guattari

(1987) A Thousand Plateaus: Capitalism and Schizophrenia. Minneapolis, MN: University of Minnesota Press.

27.

Drucker

(2014) Graphesis: Visual Forms of Knowledge Production. Massachusetts: Harvard University Press.

28.

Duffy

Meisner

(2023) Platform governance at the margins: Social media creators’ experiences with algorithmic (in)visibility. Media, Culture & Society 45(2): 285–304.

29.

Geismar

(2017) Instant archives? In: Hjorth

Horst

Galloway

, et al. (eds) The Routledge Companion to Digital Ethnography. New Tork: Routledge, pp.331–343.

30.

Gerlitz

(2016) What counts? Reflections on the multivalence of social media data. Digital Culture & Society 2(2): 19–38.

31.

Gerlitz

Helmond

(2013) The like economy: Social buttons and the data-intensive web. New Media & Society 15(8): 1348–1365.

32.

Gerlitz

Lury

(2014) Social media and self-evaluating assemblages: On numbers, orderings and values. Distinktion: Journal of Social Theory 15(2): 174–188.

33.

Gerrard

Thornham

(2020) Content moderation: Social media’s sexist assemblages. New Media & Society 22(7): 1266–1286.

34.

Gillespie

(2016) #Trendingistrending: When algorithms become culture. In: Seyfert

Roberge

(eds) Algorithmic Cultures: Essays on Meaning, Performance and New Technologies. Abingdon: Routledge, pp. 52–75.

35.

Grandinetti

(2023) Examining embedded apparatuses of AI in Facebook and TikTok. AI & Society 38(4): 1273–1286.

36.

Hansen

Koch

(2022) Assemblage – constructing the social in empirical cultural research. Hamburger Journal Für Kulturanthropologie 14: 3–15.

37.

Helmond

(2015) The web as platform: Data flows in social media. PhD Thesis, University of Amsterdam, Netherlands. Available at: https://www.annehelmond.nl/wordpress/wp-content/uploads//2015/08/Helmond_WebAsPlatform.pdf (accessed 22 April 2025).

38.

Hookway

(2014) Interface. Cambridge, MA: MIT Press.

39.

Kathke

Tomann

Uhlig

(2022) Curation as a social practice: Counter-narratives in public space. International Public History 5(2): 71–79.

40.

Kaun

Stiernstedt

(2014) Facebook time: Technological and institutional affordances for media memories. New Media & Society 16(7): 1154–1168.

41.

Kaye

DBV

Zeng

Wikström

(2022) TikTok: Creativity and Culture in Short Video. Cambridge: Polity Press.

42.

Kim

Yun

(2023) Sociotechnical challenges to the technological accuracy of computer vision: The new materialism perspective’ technology in society 75(102388): 1–11. https://doi.org/10.1016/j.techsoc.2023.102388

43.

Kitchin

Lauriault

(2018) Towards critical data studies: Charting and unpacking data assemblages and their work. In: Thatcher

Eckert

Shears

(eds) Thinking Big Data in Geography. Lincoln: University of Nebraska Press, pp.3–20.

44.

Krämer

(2022) Reflections on ‘operative iconicity’ and ‘artificial flatness’. In: Wengrow

(ed.) Image, Thought, and the Making of Social Worlds, vol. 3. Heidelberg: Propylaeum, pp.251–272.

45.

Kumar

Thussu

(2023) Media, digital sovereignty and geopolitics: The case of the TikTok ban in India. Media, Culture & Society 45(8): 1583–1599.

46.

Latour

(2005) Reassembling the Social: An Introduction to Actor-Network-Theory. New York: Oxford University Press.

47.

Leaver

Highfield

Abidin

(2020) Instagram: Visual Social Media Cultures. Cambridge: Polity Press.

48.

Lindgren

(2023) Critical Theory of AI. Cambridge: Polity Press.

49.

Lupton

Southerton

(2021) The thing-power of the Facebook assemblage: Why do users stay on the platform? Journal of Sociology 57(4): 969–985.

50.

Macgregor Wise

(2017) Assemblage. In: Ouellette

Gray

(eds) Keywords for Media Studies. New York: New York University Press, pp.16–18.

51.

Meßmer

Degeling

Hohlfeld

(2024) Auditing TikTok: Methodological reflections on the RSBA process. Auditing TikTok. Available at: https://tiktok-audit.com/blog/2024/TikTok-RSBA-Step1/ (accessed 22 April 2025).

52.

Meta (2023a) Introducing 22 system cards that explain how AI powers experiences on Facebook and Instagram. Available at: https://ai.meta.com/blog/how-ai-powers-experiences-facebook-instagram-system-cards/ (accessed 22 April 2025).

53.

Meta (2023b) The AI behind unconnected content recommendations on Facebook and Instagram. Available at: https://ai.meta.com/blog/ai-unconnected-content-recommendations-facebook-instagram/ (accessed 22 April 2025).

54.

Meta (2023c) Instagram Feed Recommendations AI system. Available at: https://transparency.fb.com/features/explaining-ranking/ig-feed-recommendations/?referrer=1 (accessed 22 April 2025).

55.

Müller

Schurr

(2016) Assemblage thinking and actor-network theory: Conjunctions, disjunctions, cross-fertilisations. Transactions of the Institute of British Geographers 41(3): 217–229.

56.

Nail

(2017) What is an Assemblage? SubStance 46(1): 21–37.

57.

Obrist

(2014) Ways of Curating. New York: Farrar, Straus and Giroux.

58.

Paßmann

Schubert

(2021) Liking as taste making: Social media practices as generators of aesthetic valuation and distinction. New Media & Society 23(10): 2947–2963.

59.

Reckwitz

(2002) Toward a theory of social practices: A development in culturalist theorizing. European Journal of Social Theory 5(2): 243–263.

60.

Reckwitz

(2020) Society of Singularities. Cambridge: Polity Press.

61.

Ritzer

Jurgenson

(2010) Production, consumption, prosumption: The nature of capitalism in the age of the digital ‘prosumer’. Journal of Consumer Culture 10(1): 13–36.

62.

Rosenbaum

(2020) Algorithmic neutrality, algorithmic assemblages, and the lifeworld. AMCIS 2020 Proceedings 6: 1–10.

63.

Savolainen

(2022) The shadow banning controversy: Perceived governance and algorithmic folklore. Media, Culture & Society 44(6): 1091–1109.

64.

Schellewald

(2023) Understanding the popularity and affordances of TikTok through user experiences. Media, Culture & Society 45(8): 1568–1582.

65.

Schubbach

(2007) “. . . A display (not a representation) . . .” Zur Sichtbarmachung von Daten. Navigationen. Zeitschrift Für Medien- Und Kulturwissenschaft 7(2): 13–27.

66.

Schulz

(2023) A new algorithmic imaginary. Media, Culture & Society 45(3): 646–655.

67.

Schulz

Matzner

(2020) Feed the interface. Social-Media-Feeds als Schwellen. Navigationen - Zeitschrift Für Medien- Und Kulturwissenschaften. Filter(n) – Geschichte Ästhetik Praktiken 20(2): 147–164.

68.

Scolere

Humphreys

(2016) Pinning design: The curatorial labor of creative professionals. Social Media + Society 2(1): 1–13.

69.

Semenova

Degeling

Hess

(2024) Understanding TikTok’s for you feed. Auditing TikTok. Available at: https://tiktok-audit.com/blog/2024/For-You-Feed/ (accessed 22 April 2025).

70.

Simatzkin-Ohana

Frosh

(2022) From user-generated content to a user-generated aesthetic: Instagram, corporate vernacularization, and the intimate life of brands. Media, Culture & Society 44(7): 1235–1254.

71.

Snyder

(2015) Discourses of ‘curation’ in digital times. In: Jones

Chik

Hafner

(eds) Discourse and Digital Practices. Abingdon, England: Routledge, pp.209–225.

72.

Sülflow

Schäfer

Winter

(2019) Selective attention in the news feed: An eye-tracking study on the perception and selection of political news posts on Facebook. New Media & Society 21(1): 168–190.

73.

TikTok (2020) How TikTok recommends videos #ForYou. Available at: https://newsroom.tiktok.com/en-us/how-tiktok-recommends-videos-for-you (accessed 22 April 2025).

74.

TikTok (2023a) How TikTok recommends content. Available at: https://support.tiktok.com/en/using-tiktok/exploring-videos/how-tiktok-recommends-content (accessed 22 April 2025).

75.

TikTok (2023b) How TikTok’s for you feed works. Available at: https://www.tiktok.com/transparency/en-us/recommendation-system/ (accessed 22 April 2025).

76.

Triệu

Ellison

Schoenebeck

, et al. (2021) Implications of Facebook engagement types and feed’s social content for self-esteem via social comparison processes. Social Media + Society 7(3): 1–12.

77.

Tseng

Y-S

(2023) Assemblage thinking as a methodology for studying urban AI phenomena. AI & Society 38: 1099–1110.

78.

van der Vlist

Helmond

Burkhardt

, et al. (2022) API governance: The case of Facebook’s evolution. Social Media + Society 8(2): 1–24.

79.

Vepřek

(2024) At the Edge of AI: Human Computation Systems and Their Intraverting Relations. Bielefeld: transcript.

80.

von Bismarck

Schafaff

Weski

(2012) Introduction. In: von Bismarck

Schafaff

Weski

(eds) Cultures of the Curatorial. Berlin: Sternberg Press, pp.8–15.

81.

Vonderau

(2018) Scaling the cloud: Making state and infrastructure in Sweden. Ethnos 11(2): 1–21.

82.

West

Thiemt

Ahmed

, et al. (2024) A picture is worth 500 labels: A case study of demographic disparities in local machine learning models for Instagram and TikTok (Version 1). arXiv.

83.

Wirth

(2024) Formierungen des Interface: Zur Mediengeschichte und -theorie des Personal Computing. Bielefeld: transcript.

84.

Ytre-Arne

Moe

(2021) Folk theories of algorithms: Understanding digital irritation. Media, Culture & Society 43(5): 807–824.

85.

Zuboff

(2019) The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. New York: Public Affairs.