The patterning of the discursive space in search for the #goodlife: A network analysis of the co-occurrence of Instagram hashtags

Abstract Stories about what living well means are critical both to the maintenance of existing ways of living and to the possibility of envisioning and transitioning toward fairer and more sustainable futures. The implications of the stories told on social media for the possibility of such futures have yet to be explored. In this article we explore how the use of hashtags on Instagram shapes the visibility and recognizability of understandings of the good life in the discursive field created by #goodlife on the platform. Using network analysis, we map the co-occurrence of hashtags in 793 posts tagged #goodlife to explore the formation of hashtag-based narrative and hyperlink patterns. The visibility and recognizability of narrative patterns within this discursive space are shaped by interactional conventions and by algorithmic infrastructure, favoring corporate interests over sustainable and fair livelihoods. However, we also identify themes that could support fairer and more sustainable understandings of living well and reassert their ongoing importance.


Introduction
In this article we engage the concept of good life, or the life worth living (Mackay 2013). Human beings have always lived on the basis of some understanding of "what is a better, more desirable or worthier way of being in the world" (Christopher 1999, 141). These understandings inform the development of constantly evolving social norms or shared ideals about what living well could mean (McMahon 2006).
The stories humans tell, the concepts they use, and the metaphors they live by influence their behavior (Helne and Hirvilammi 2017;Lakoff and Johnson 2003). What is understood to be a desirable life plays a crucial (although not strictly determining) role in how people lead their lives, as discourses do not merely describe existing realities but help bring them into being (Gottweis 2003), and constrain what can be thought about and how it can be thought about (Douglas 1976;Elliott and Wattanasuwan 1998).
In the contemporary societies of Europe, North America, and Australasia, the consumerist understanding of the good life is the dominant and/or institutionalized one (Bauman 2002;Jackson 2006;McCracken 1990). Consumerism differs from consumption. While consumption is a necessary aspect of human existence (e.g., food, water, shelter), consumerism is a cultural condition of consumer societies. In these societies, an increasing number of social and cultural functions are carried out and satisfied through the consumption of commodities (Evans and Jackson 2008), and people live, to varying extents, under the understanding that purchasing consumer goods will make their lives better in manifold ways. As this understanding of living well has been found to be linked with high environmental impacts (Ivanova et al. 2016), unstable economies and social injustice (Jackson 2017), the need for new stories of what living well means that can support just, economically stable, and ecologically sustainable societies is heightened (Evans 2017;Jackson 2017). There is a growing acknowledgement of the importance of powerful stories and engaging utopias (Levitas 2010;Mair, Druckman, and Jackson 2018) to initiate large-scale system change.
Alongside the development of new stories, the mechanisms that currently constrain their creation and contribute to the maintenance of less sustainable understandings of living well need to be understood.
Specifically, in this article we address the roles played by Instagram user practices and by the platform's own algorithms in shaping the salience and recognizability of good life stories available on the platform. Here, the communicative affordances of the platform infrastructure and the tacit user-generated rules of participation are understood as factors that mediate our interactions with each other, giving rise to a specific interactional environment. Mediating factors are known to both enable and constrain the constitution of understandings (Eberle 2014), and interactional environments encourage certain language and interaction choices while hindering others (Georgakopoulou 2015). As new media technologies and the digital access they afford have become pervasive in everyday life, especially those of many young people (Statista 2014, although 1 see Eynon and Geniets 2019), they are likely to play an important mediating role in the construction of understandings of the good life.
Instagram currently hosts over a billion users and is known to have a young user base (Duggan 2015). Additionally, its reputation for being a space of the hyperreal (Borges-Rey 2015; Gallagher 2015) and of the aspirational (Abidin 2016;Anggawi 2016), make the understandings of the good life circulating on the platform especially relevant. While Instagram's good life content is likely to mostly be created and consumed by younger people, the contributions of this content to our understandings of the good life are likely to have impact beyond Instagram users, as social understandings are created in the interactions of different demographic groups and in different spaces. In this context, Instagram content is under-researched in comparison to other social media platforms, such as Facebook or Twitter (Lee and Chau 2018).
Previously, we have approached Instagram posts holistically and, following Douglas and Isherwood (1996), presented participation on the platform as part of a social conversation which provides moral judgements on how one ought to live one's life and what ought to be good about it (Loukianov, Burningham, and Jackson 2020). In this article we focus our attention on a distinct aspect of Instagram posts-the hashtags-and the hyperlink infrastructure that they create. We argue that this hyperlink infrastructure plays a key role in shaping the visibility and recognizability of understandings of the good life in the discursive space created by #goodlife. Using network analysis, in this article we explore the patterning of the discursive space surrounding the concept of the good life on the platform by analyzing the relationships between themes present in the hashtags used in Instagram posts tagged #goodlife. We seek to make clear the processes at play in promoting some narrative patterns over others, paying attention to how hyperlink patterns simultaneously constrain and enable the discursive flow. In particular, we discuss how third-party applications by the software developer Prilaga, disproportionately contribute to shaping the discursive field surrounding #goodlife and unwittingly contribute to the promotion of some understandings of living well over others.
The article proceeds as follows: it begins with a review of the existing literature on hyperlink infrastructures. Then, it provides an overview of the use of network analysis for exploring the narrative patterns formed by hashtags on Instagram. This is followed by a descriptive account of the analysis of the hashtags present on 793 Instagram posts tagged #goodlife and the identification of narrative patterns. The article ends on a discussion of the implications of the analysis for sustainable and fair futures.

Hashtagging, visibility, algorithms, and social conventions
First introduced by Chris Messina on Twitter (Dorsch 2020), hashtags are now widely used across a diversity of social media platforms. Simply put, a hashtag is a string of characters-typically forming a word or string of words-that is preceded by a # sign. On Instagram, users may place them in the caption of their posts, their profiles, and their stories, or in comments on their own and other users' posts.
Hashtags have a particular status as they are at once linguistic segments and hyperlinks (Giaxoglou 2018;Paveau 2013;Zappavigna 2015). As linguistic segments, hashtags contribute to the establishment of a user-generated system of classification of online content into different categories. As hyperlinks, they enter content into the attention economy by making posts searchable (Zappavigna 2015) and facilitate users' engagement with each other's content (Marwick 2015). The hyperlink function of hashtags enables the creation of a network that weaves together "connected and unconnected narratives" (Giaxoglou 2018, 13), bringing together a range of posts into a same discursive space. By searching for any hashtag on Instagram, users can access all of the public posts that feature it. Motivations for using hashtags include self-promotion, social interaction, documentation, and creativity (Sheldon et al. 2017). Hashtags have been conceptualized as vehicles for networked topic framing (Meraz and Papacharissi 2013). In practice, hashtags can lead to the formation of communities or ad hoc publics around particular topics or events (Bruns and Burgess 2015;Highfield and Leaver 2015). On Instagram, users have the possibility to follow hashtags, further facilitating engagement with topics of interest. Additionally, by making content searchable, hashtags can contribute to increasing the visibility of one's posts and attract likes, comments, and followers. This feature is particularly important to influencers (popular users) and businesses as careful selection of hashtags enables them to gain and sustain the attention of other users (Abidin 2016). One of the aims of this accumulation of social capital is to convert it-through advertising revenue and data mining-into financial gains. Although the last few years have also seen some influencers abandon the use of hashtags entirely (Manovich 2017), hashtags are still widely used, with 18.8 M posts tagged #goodlife at the time of writing.
The norms of hashtags use on Instagram are shaped both by the social conventions prevailing on the platform and by the platform's interface and design (Laestadius 2017). While, technically, users can input up to 30 hashtags in a single post, social etiquette regarding the use of hashtags is constantly evolving. Use of hashtags has been shown to follow cultural (Sheldon et al. 2017) as well as gendered (Ye et al. 2018) scripts. Opinions differ on the number of hashtags one should use and where to place them (Cotter 2019), and user guides provide advice for businesses and lay users alike on hashtagging best practice (Dorsch 2020).
Thus far, studies on Instagram have primarily focused on user activity on the platform itself. Yet, as Skeggs andYuill (2016, 1358) point out in relation to Facebook's sustained expansion of user behavior tracking, social media platforms are not "self-contained [entities] set apart from the rest of the Internet" or indeed from broader "social conversations." Users increasingly make use of a priori external resources to improve their performance on particular platforms (Cotter 2019). Since the creation of Instagram in 2010, users have adopted a range of third-party applications from photo-editors such as VSCO and Lightroom, to scheduling tools such as Later which enable auto-posting at pre-determined times. Of particular interest to this article are third-party applications such as Top Tags, InstaTag, and inTags which offer tools for generating hashtags that promise heightened engagement.
These tools make use of the platform's Application Programming Interface (API) and are bound by Instagram's terms of use. Additionally, as Bucher (2013) and Skeggs and Yuill (2016) observe, APIs provide extremely controlled access to information. In other words, tools that are a priori external to the platform and supposedly operate according to external logics, are in fact regulated by the priorities and vested interests of the platform's creators (Skeggs and Yuill 2016). In general Hartley (2012) argues that online platforms and databases must be understood as systems that organize themselves, notably through their software. As infrastructures can play significant roles in shaping the availability of online content and the landscape of possible actions, Hine (2015) and others (Beer 2013;Bishop 2018) warn that these infrastructures may be carrying out invisible structuring work.
While user intentionality and motivations in hashtag use cannot be easily presumed (Highfield and Leaver 2015), the dual function of hashtags results in the creation of a system in which linguistic concepts are used and made to relate to each other in particular ways, notwithstanding the intentions of individual users. Such organization defines systems of discourse (Lakoff 1990). For Paveau (2013), hashtagging constitutes a "technodiscursive act" which shapes the interactional environment beyond the confines of an individual post. Due to their fleeting and disarticulated character (Carah and Shaul 2016), individual Instagram posts struggle to establish a lasting narrative. However, hashtags regroup multiple posts that sustain similar themes and convey similar stories, forming discernible narrative patterns.
Instagram researchers have thus far conducted multiple content analyses of posts tagged with particular hashtags of interest, such as pregnancy-related hashtags (Tiidenberg and Baym 2017), #fitspiration (Tiggemann and Zaccardo 2018), #happy (De Paola, Hakoköngäs, and Hakanen 2020), #sustainabletourism (Palazzo et al. 2021), and our own exploration of #goodlife (Loukianov, Burningham, and Jackson 2020). These analyses provide important insights into the social construction and esthetic norms of particular concepts on the platform, but they cannot shed light on the processes at play in shaping these constructions.
The analysis of the relationships between hashtags that co-occur on posts tagged #goodlife provides a novel approach to study understandings of the good life on Instagram. There are obvious limitations to this approach as some themes may only appear in images and captions rather than hashtags. However, researchers agree that hashtags tend to relate to the content of the posts (Dorsch 2020; Giannoulakis and Tsapatsoulis 2016;Highfield and Leaver 2015). Additionally, the extensive use of hashtags in Instagram posts, their acknowledged status as narrative resources (Giaxoglou 2018), and their contribution to the creation of a hyperlink infrastructure suggests that relationships between hashtags are worth exploring in their own right.
Although hyperlink networks have generated scholarly interest for over two decades (Park 2003), the focus on hashtag networks is relatively new. Research on hashtag networks has been primarily developed with Twitter data (e.g., Eddington 2018;Giaxoglou 2018;Türker and Sulak 2018;Shi et al. 2020), but recent years have also seen the application of this approach to Instagram. For instance, Ichau, Frissen, and d'Haenens (2019) have combined a content analysis of #jews (and related hashtags) with hashtag network analysis of the tag, Buente et al. (2020) have looked at the hashtag networks in relation to #betelnut, and Pilař et al. (2020) have explored hashtag networks derived from posts simultaneously tagged #food and #sustainability. These early works indicate that hashtags actively contribute to shaping the social construction of concepts on the platform.
In this study, we contribute both to this growing body of work on Instagram hashtag networks, and to the existing literature on understandings of living well by exploring the hashtag network formed in relation to #goodlife. In the process, we discover the unwitting role played by Prilaga's third-party hashtag optimization applications, in shaping the discursive space of the good life on the platform. As developers create third-party applications and Instagram users make use of them, both developers and users contribute to the creation of added value for the platform. Participatory opportunities of Web 2.0 may fuel a creationist capitalism that "feeds off of people's self-expression, social aspirations, and problem-solving skills in economic profit-making" (Ruckenstein 2011(Ruckenstein , 1061, also see Abidin 2016; Arvidsson 2005; Bishop 2018).
However, the relationships between corporations and social media users are less straightforward than they may first appear. The communicative and creative opportunities of social media platforms empower users, while their algorithms enable platform owners to steer and profit from users' activities (Duffy and Hund 2015;Leaver, Highfield, and Abidin 2020;Uldam 2018). Uldam (2018) shows that social media platforms have afforded unprecedented levels of visibility for activists, while simultaneously enabling corporations to monitor and silence campaigns unfavorable to their interests. The entanglements between corporate steering of social media platforms, and user engagements with said platforms gives rise to new social practices and reshapes existing ones, as Lewis and Phillipov (2018) show in their study of the relationships between food and digital media.
For Ruckenstein (2011), both parties utilize the platforms to develop their intertwined projects of value creation which may be in conflict or in harmony with each other. It is with this understanding of Instagram-namely as a social media platform which is ambiguously and unequally shaped both by individual users and by corporate interests-that we begin our exploration of the patterning of the discursive space created by #goodlife through co-occurrence of hashtags on Instagram posts and our discussion of the implications of this patterning for fair and sustainable futures. As Van Dijck and Poell (2013) argue, disentangling the entanglements, identifying the connections between the different actors involved, and exploring the mechanisms and principles which shape social media activity are crucial to our understanding of the role played by social media activity beyond the platforms themselves.

The study: A network analysis of the co-appearance of hashtags in Instagram posts tagged #goodlife
We introduce our data analysis processes below, starting with the harvesting of the dataset, through to importing the dataset and filtering our data, and visualizing thematic clusters. Throughout, we introduce important vocabulary (see Table 1) that will be used in the remainder of the article.

Harvesting the dataset
On the 10th of October 2018, we queried Netlytic, a free social media content collection software, to harvest 2,000 Instagram posts tagged #goodlife from public Instagram accounts. The oldest post was created on the 30th of January 2018 and the most recent one on the 10th of October 2018, with most of the harvested posts having been created between September and October 2018. Some users contributed multiple posts to the sample.
For every one of the 2,000 posts collected, Netlytic harvested the hyperlink, the publication date, the author's username, the caption, the "like" count, and the potential use of filters. For each post, we sought to collect the original hashtags used by the post creator. Original hashtags are typically located in captions, although users may also place original hashtags in a comment (Abidin 2016). Accordingly, our first step upon obtaining the Netlytic dataset was to manually process the 2,000 posts to ensure that the original hashtags were recorded for each post. However, some Instagram users deleted their posts, the comments containing the hashtags, and their accounts, removed the hashtags from the caption, or switched to private accounts, making the posts and/or hashtags inaccessible. The deletion/privatization of content between data collection and analysis phases can be interpreted as a withdrawal of consent, and good practice demands the removal of these posts from the dataset (Franzke et al. 2020). After removing these unusable posts from the dataset, only 793 posts were left. The ways in which hashtag deletion is treated by Instagram is unclear, making it difficult to assess the impact of these unusable posts. We expect that they did not have undue repercussions on our analysis, given that the hashtags that they contained no longer played a role in shaping the hashtag infrastructure. Each remaining post contained up to 30 hashtags, adding up to a total of 6130 hashtags in the dataset. Most of these remaining posts were still images, either taken by the user or reproduced from a different source.

Importing the network into Gephi
Using Gephi, an open source software for network analysis, we plotted the co-occurrence of all pair-permutations of hashtags present on a same post. For instance, a post tagged #goodlife #travel #beautiful generated 3 pair-permutations: #goodlife-#travel; #goodlife-#beautiful; #travel-#beautiful. This process was repeated for every post of the dataset.
Each hashtag was represented by a "node" (n), and the co-occurrence of two hashtags was represented by a link between the two nodes, called an "edge" (e). The frequency of appearance of a hashtag in the dataset was represented by its node weight (ⱳ n ). The number of posts two hashtags co-appeared in indicated the strength of the relationship between these two hashtags and was represented by the edge weight (ⱳ e ).
The network of hashtags was imported as an undirected network, meaning that the edges were bidirectional: if hashtag A appears in a post with hashtag B, then hashtag B necessarily also appears in that post with hashtag A. Upon importing the dataset, our first step was to delete the hashtag #goodlife. Due to its appearance in every post of the dataset, the hashtag #goodlife had a relationship to every hashtag in the dataset, obscuring the relationships between other hashtags. 2 As we were only interested in keeping those hashtags that were used relatively often, we excluded all hashtags that appeared in one post only. We also deleted connector words (#and, #by, etc.), usernames used as hashtags, and punctuation signs (#.). We then proceeded to classifying the remaining hashtags into themes. To this aim, we considered each of the remaining hashtags in the context of the posts containing them. For each hashtag, this entailed referring to the posts in which it appeared and assessing the meaning of the hashtag in the context of the images and the captions (if present) of the posts in which it was used. The classification was carried out by the lead author. Hashtags that clearly referred to different things in different posts were deleted during the classification process, as they could not be classed in one theme or another. The remaining 1526 hashtags were classified into 33 thematic clusters. A table of the top hashtags in each cluster by frequency of appearance is provided in Appendix (Table A1).

Visualizing the network: Creating "supernodes" to represent thematic clusters of hashtags
The visualization was simplified by merging all nodes within a thematic cluster, so each thematic cluster relationship between two hashtags-represents the co-occurrence of two hashtags in a post e node weight frequency of appearance of a hashtag in the dataset ⱳ n edge weight strength of relationship between two hashtags-indicated by the number of posts in which two hashtags co-appear ⱳ e supernode merged thematic cluster of hashtags sn supernode weight estimate of amount of space taken up by a theme in the dataset ⱳ sn supernode edge weight strength of relationship between two themes-indicated by number of posts two themes co-occur in ⱳ esn density proportion of the potential connections in a network that are actual connections-a complete graph has all possible connections and a density of 1 D clique subgroup of nodes in which nodes are all directly connected to one another and no additional node exists which is also connected to all nodes of the subgroup degree of connectedness number of co-occurrences of a hashtag with any other hashtags in the dataset narrative pattern pattern formed by consistent co-occurrence of hashtag themes in posts was represented by a single node which we called a "supernode" (sn). We named the thematic clusters as follows: (1) business and entrepreneurship, (2) success, (3) education, (4) inspiration and dreams, (5) motivation and hard work, (6) mindfulness and mental health, (7) outdoors activities, (8) nature, (9) relaxation and leisure, (10) love, (11) quotes, (12)  The weight of a supernode (ⱳ sn ) is an estimate of the amount of space that a given theme takes up in the dataset. It results from the summation of the weights of all nodes classed in a cluster and is hence a function of the number of hashtags (x) within the cluster and their respective frequencies of appearance (ⱳ n ). ⱳ sn = ∑(ⱳ n1 ; ⱳ n2 ; ...; ⱳ nx ) While the size of a supernode conveys an idea of the space taken up by a given theme in the dataset, that space can be taken up in different ways (high number of hashtags, hashtags with high frequency of appearance, or combination of both).
Following the merge, the graph contained 33 supernodes, each representing a thematic cluster of hashtags, connected through a weighted edge to other supernodes in the dataset (Figure 1). The weight of the edge between two supernodes (ⱳ esn ) represents the number of posts two themes co-occur in and stands for the strength of the relationship between these two themes. We used the edge weight between the most important (frequency of appearance and number of connections) hashtags of each cluster prior merging as a proxy for ⱳ esn . Most important hashtags within a cluster typically appear on a majority of posts within that cluster, making the edge weight between two such hashtags a good approximation for the strength of the relationship between two themes. Thematic clusters that had strong relationships to each other and weaker relationships to other thematic clusters were colored, potentially revealing parallel understandings of the good life on Instagram.
As Figure 1 shows, nearly all thematic clusters are connected (D = 0.98). Hence, exploring narrative patterns, identified through cliques, required a temporary filtering out of the weaker relationships in the sample. We proceeded through incremental increases of the threshold of magnitude (Allesina, Bodini, and Bondavalli 2006), which enabled us to distinguish relatively stronger relationships from relatively weaker ones. In this study, we use two thresholds of magnitude (Figure 2: ⱳ esn > 20 and Figure 5: ⱳ esn ≥ 10) to explore both the strongest relationships in the sample and the relatively weaker but nonetheless important ones. While more sophisticated filters designed to identify the 'backbone' of a weighted network have been developed among physicists (Serrano, Boguña, and Vespignani 2009), incremental thresholding is regarded as an adequate approach for qualitative purposes.

Exploring narrative patterns: Analyzing the relationships between hashtags
In this section we explain the influence that third-party hashtag optimization applications have on the discursive space around #goodlife by approaching hashtags as hyperlinks and as semantic segments. Figure 2 illustrates the strongest relationships in the network. There are two parallel sets of relationships, respectively represented by yellow and blue supernodes. The supernodes colored green do not form any strong relationships in this sample. Additionally, two supernodes that have relatively strong relationships to only one other supernode are colored red. We do not discuss these here and focus instead on the yellow, blue, and green supernodes.

Hashtags as hyperlinks: Strength of relationships and visibility of themes in discursive fields
The yellow supernodes form a clique containing the following thematic clusters: "business and entrepreneurship", "motivation and hard work", "inspiration and dreams", "success", and "education" suggesting that these themes regularly co-occur on Instagram posts. Most of the hashtags classified into the yellow thematic clusters were provided by one of two Prilaga applications as the posts that featured them also contained #prilaga. 3 Prilaga is the brand name of an application developer who created two third-party applications designed to provide users with hashtags optimized for content exposure: inTags and Best    Hashtags for Promotion. Together, these applications have had over 1.5 million downloads.
Both applications are free to download and provide two main features: categories of hashtags and a search tool. The first feature enables users to pick and choose from categories such as "animals" or "food" created by the developer in which optimal hashtags are classified semantically. The second feature enables users to search a keyword such as "entrepreneur" and return a list of up to 30 hashtags which are often used together on Instagram and the application tracked as currently maximizing chances of obtaining likes, comments, and followers. In both cases, these hashtags can then be copied and pasted into an Instagram post.
The relationships between the Prilaga thematic clusters are some of the strongest of the sample (Figure 3), despite the relative dissociation between "education" and "success" (ⱳ esn = 31) and "business and entrepreneurship" and "success" (ⱳ esn = 29).
The second set of relationships that becomes salient when filtering out the weaker relationships is that of the supernodes colored in blue. As they do not seem to primarily result from sets of hashtags generated by a third-party application, these relationships could be said to form more organic narrative patterns than the one created by the yellow nodes. Two cliques of blue supernodes appear in Figure 2: one formed of "love", "lifestyle", and "positive qualifiers"; another formed of "love", "travel", "photography", and "positive qualifiers. " All three narrative patterns (the yellow clique and the two blue cliques) are characterized by the strong relationships between the thematic clusters which form them. Strong relationships represent the consistency with which these thematic clusters co-appear on posts. Strong relationships support the visibility of themes in the discursive space. A given thematic cluster is more visible as part of a narrative pattern because it appears consistently rather than occasionally in a range of feeds created through hashtag search. For instance, the consistent use of a set of hashtags such as "#entrepreneur #motivated #dreambig #success #goodlife" on posts means that feeds constituted by searching for any of these hashtags are flooded by these posts.
While relatively strong in comparison with other relationships in the sample, the blue sets of relationships are weaker than the ones established between the yellow supernodes. It appears that the consistent use of the hashtags provided by Prilaga solidifies relationships between themes to a much greater extent than in cases where users select hashtags on their own. Feeds constituted through the search for hashtags such as #entrepreneur, #inspiration, #dream, but also #goodlife, are effectively flooded with content relating to entrepreneurship via the working of the Prilaga apps, making this type of content more visible in the discursive space created by #goodlife. Figure 4 presents a list of the degrees of connectedness of the supernodes. The thematic clusters called "love", "Instagram and blogging", as well as "happiness" are the most connected of the sample (see Figure 4). Some of the thematic clusters that form the Prilaga set of hashtags, such as "motivation and hard work", "success", and "inspiration and dreams" are also relatively high in the connectedness table. High degrees of connectedness may be due to strong connections to a few clusters, or due to minor connections to a multiplicity of clusters. We look at Figure 2 to elucidate the ways in which these connections are allocated.

Hashtags as linguistic segments: Establishing a consistent and recognizable vocabulary
As Figure 2 shows, "Instagram and blogging" and "happiness" do not establish strong connections. Their high degree of connectedness suggests that they establish connections with a wide diversity of clusters, but these connections do not follow consistent patterns. Hashtags relating to "happiness" and "Instagram and blogging" appear in relation to a multitude of topics, reinforcing their relevance in a range of contexts. But the popularity of these hashtags in relation to a wide range of topics translates into a diffuse use of hashtags and the failure to establish strong relationships. Conversely, the connections of clusters such as "motivation and hard work", "success", and "inspiration and dreams" are predominantly allocated to a few specific thematic clusters (business and entrepreneurship, inspiration and dreams, success, motivation and hard work, education). While they also sustain minor connections, the allocation of most of their connections to specific clusters results in strong and consistent relationships.
The lack of relatively stronger relationships with a consistent set of thematic clusters translates to a lack of consistent vocabulary which results in a situation where themes such as happiness lose ground to themes such as business and entrepreneurship in the discursive field of the good life. In a paper on narrative change, Waddock (2018) used the concept of "meme" introduced by Dawkins to explain a similar phenomenon. Memes are "a resonant set of core words, values, phrases, and ideas" (Waddock 2018, 17) on which narratives are based. The use of consistent memes makes it easier to broadcast particular narratives as they tend to be easily recognized, and hence more easily picked up upon. Waddock argues that although people interpret memes in different ways, using a set of core memes repeatedly and consistently has a greater influence on attitudes, beliefs, practices, and behaviors than the use of diffuse sets of memes.
The logic of Waddock's argument can be extended to make sense of the impact of the sets of hashtags generated by Prilaga. The strength of the relationships between the yellow nodes results in the creation of a narrative pattern that we call "entrepreneurial spirit. " Even though they are ultimately separate components which could be reconfigured into different sets of relationships-for instance, "inspiration and dreams" could be envisioned in association with "art", "significant others", or "nature"-the consistency with which this set of themes appears together makes it recognizable as a particular approach to living well. This is an important observation, as cumulatively the Prilaga sets of hashtags take up a large proportion of the discursive space. Conversely, themes such as "happiness" or "Instagram and blogging", while popular, are also less recognizable as parts of specific narratives.

Implications for themes that do not establish strong relationships
Having established that strong relationships help themes gain visibility in a given discursive space through hyperlinks and create recognizable narrative patterns by connecting semantic themes with each other, we now explore which themes fail to establish any strong relationships at all in this sample. Figure 5 shows a plot displaying all but the weakest relationships in the sample. It appears that, among others, "mindfulness and mental health", "nature", "outdoors activities", "art", "relaxation and leisure", "spirituality", and "happiness" do not sustain any strong relationships in this sample. While these themes are presented as important aspects of a good life, they are overshadowed-although not entirely eclipsed-by others, such as business and entrepreneurship or travel. Of course, "happiness" and "nature" succeed in acquiring a certain level of visibility in this sample based on their sheer size (and so does "luxury and wealth"). These clusters contain hashtags with high frequency of appearance, and/or contain a large number of different hashtags which relate to these themes, hence these themes are widely represented in the sample.
But smaller clusters must be woven into consistent patterns to remain visible in the discursive field of the #goodlife. Small clusters contain hashtags with low frequency of appearance and/or contain only a few different hashtags, meaning that they occupy little space in the discursive field of the good life. But the ways in which that space is occupied impacts the visibility of the clusters in the discursive field. Small clusters which establish strong relationships consistently appear with a particular set of other clusters, meaning that their appearances accumulate in specific places in the discursive field of #goodlife, making them more visible in this field than small clusters with diffuse relationships. Indeed, diffuse relationships mean that the appearances of the hashtags that form that cluster are scattered across the discursive field. While such clusters may, theoretically, co-appear with a wider range of themes than clusters with strong relationships, their small size means that these diffuse relationships make them harder to notice in any one place.
Both "success" (ⱳ sn = 212) and "education" (ⱳ sn = 221) have cluster sizes similar to "art" (ⱳ sn = 189) and "mindfulness and mental health" (ⱳ sn = 257) but their strong relationships with other thematic clusters ensure their visibility in the discursive space of #goodlife. Indeed, many of the posts which are tagged #success appear in feeds created by searching for #entrepreneur or #motivation. On the other hand, posts which are tagged #art may appear in a wide range of feeds created by hashtag search, but those appearances will be limited to a few posts at a time.
Ultimately, thematic clusters such as "art" or "mindfulness and mental health" may have strong relationships to themes that are not captured in this sample. It is possible that, in a wider sample or in a sample created from a different hashtag search, both "art" and "mindfulness and mental health" could be large thematic clusters and/or sustain strong relationships to other themes. However, in this specific sample which only captured a subset of posts tagged #goodlife, these two clusters occupied a small proportion of the discursive space. Analyzing them showed that in the context of a feed created by hashtag search, small clusters that fail to establish strong relationships tend to be less visible than those that sustain strong relationships.

Implications for sustainable and fair futures
In this article we argued that the practice of hashtagging creates an infrastructure of hyperlinks that regulates the discursive flow, directing attention toward some themes, eclipsing others, and solidifying semantic associations through the creation of narrative patterns. Additionally, this infrastructure is unequally shaped by the various actors involved in the production of Instagram hashtags. In this discussion, we develop the implications of both of these aspects for the possibility of fair and sustainable futures.
We begin by discussing the implication of our interpretation of semantic meanings of the three narrative patterns we have identified above. The first narrative pattern, represented by the yellow supernodes and edges, relates entrepreneurship to success, hard work, education, and dreams and inspiration. It broadly relates to entrepreneurial spirit and establishes the good life as one of entrepreneurial pursuits brought to success through hard work. Patterns of association show the promotion of a meritocratic ideal, on the one hand, and a relative dissociation of success and entrepreneurship, on the other. The semantic association of "motivation and hard work" and "success" could be perpetuating a meritocratic understanding of success, which as Littler (2013) argues, maintains unequal chances in an unequal society by individualizing failure and obscuring systemic inequalities. The relative dissociation between "business and entrepreneurship" and "success" may reflect this bleaker reality.
The second narrative pattern links love, positive qualifiers, travel, and photography; and echoing Sontag's (1979) analysis, establishes the good life as one of tourism-we call this pattern "wanderlust. " The consistent hyperlink relationships and the frequency of appearance of hashtags within these thematic clusters ensure that posts relating to travel have high visibility in the discursive field of #goodlife. The consistent semantic association between travel, photography, love, and positive qualifiers also reinforce the presentation of travel/tourism as an overwhelmingly positive experience. The social and environmental sustainability of traveling and tourism is increasingly questioned (Scott, Peeters, and Gössling 2010). Yet, it remains a practice that even environmentally committed individuals are reluctant to change (Barr, Shaw, and Coles 2011). The narrative patterns evidence in Instagram may be normalizing tourism/travel as a key aspect of the good life and reinforcing its desirability.
The third narrative pattern links love and positive qualifiers to lifestyle-we call it "living the dream." These semantic associations leave the type of lifestyle unspecified, and this narrative pattern could refer to a range of lifestyles, some more sustainable than others. However, the relatively important relationships that "lifestyle" establishes with "success" (ⱳ esn = 10) and "travel" (ⱳ esn = 11) seem to narrow down the field of possible lifestyles that could be referred to ( Figure 5). The strong relationships that "success" sustains with the rest of the "entrepreneurial spirit" clique, on the one hand, and the relationships that "travel" sustains with the rest of the "wanderlust" clique, on the other, suggest that these could be the lifestyles of an entrepreneur or of a traveler. Hence, "living the dream" seems to work in support of the first two patterns.
Crucially, some of the themes relating to the good life on Instagram that fail to sustain narrative patterns may have potential to provide more sustainable and fairer approaches to living well. This is for instance the case of art, mindfulness and mental health, spirituality, significant others, and outdoors activities. Researchers show that positive relationships, religious or spiritual thinking, creative activities, sports and physical exercise, as well as focusing on others have the potential to give rise to lifestyles that have low environmental impact and are supportive of individual wellbeing (Isham, Gatersleben, and Jackson 2019). The relative lack of visibility of these themes in this Instagram sample should not detract from their importance in understandings of what makes lives worth living. Rather, the analysis points to the dynamics specific to Instagram where, as we argued, failure to form narrative patterns through hyperlinks negatively impacts their visibility and recognizability in the discursive space of #goodlife.
As on other social media platforms, the Instagram algorithms play a crucial part in the regulation of the presentation of content. On Instagram users access content either through their Feeds and Stories, the Explore tool, and more recently through Reels. In each case, a dedicated algorithm selects posts that it determines to be the most compelling to its users based on the algorithm's definition of relevancy (Kitchin and Dodge 2011; Mosseri 2021) and imposes its own valuations of content (Willson 2017). As we have focused on hashtags present in posts, only the algorithms which organize Feeds and the Explore tool are discussed in this article. Instagram is secretive about the exact weighting of the criteria each algorithm uses to determine relevancy, but the order in which they are considered has recently been clarified (Mosseri 2021). Feeds primarily present users with content from users or hashtags they follow, with timeliness being the most important criterion. Beyond timeliness (in order of importance), content popularity, poster popularity, previous user behavior, and history of interaction with poster shape the content which appears in Feeds. The Explore tool presents users with content from sources they do not yet follow, hence the criteria for relevancy differs with previous user behavior and preferences of other users interacting with the same content being the most relevant criteria. Beyond these (in order of importance), the Explore algorithm takes into account content popularity, history of interaction with poster, previous user behavior on Explore, and poster popularity.
On the one hand, both in Feeds and in Explore these algorithmic features result in the creation of opinion silos and echo chambers (Shi et al. 2020). Because the algorithms seek to connect users with content that is likely to interest them, and assess relevancy partly on the basis of previous user behavior (Rader and Gray 2015) or the behavior of other users with similar preferences (Mosseri 2021), a given user is likely to be exposed to posts which align with their existing understandings of living well. Conversely, posts which convey understandings of living well that the user disagrees with or feels neutrally about are less likely to appear on their feeds. For any given user, dataification and algorithmic processes reinforce the validity of the understandings of living well that they already value.
On the other hand, outside of the accounts that a user follows, the content that is most likely to be deemed relevant is one with commercial interests. Research suggests that ensuring content visibility on the platform is a complex task which is time-intensive and requires high level of expertise (Abidin 2016;Cotter 2019). Accordingly, users who can invest the time and build up their expertise benefit the most from the system. In practice, these users tend to either be businesses or influencers sponsored by businesses (Bishop 2018). This contributes to reinforcing the normality of carrying out social and cultural functions through the consumption of commodities and favors the promotion of a consumerist understanding of living well. Of course, commercial content may be promoting "green" or "sustainable" products and consuming differently is an important aspect of more sustainable living. But in the wealthy societies of Europe, North America, and Australasia, consuming less is central to the reduction of environmental impact, and reductions in consumption cannot be achieved while the consumerist understanding of living well is maintained.
Third-party applications such as those developed by Prilaga can support the visibility of one's posts and popularize content that does not have commercial aims. However, such third-party applications may be attractive to particular communities of users, with particular understandings of living well (e.g., "entrepreneurial spirit"), while communities of users which favor different understandings may not be drawn to the Prilaga applications. Additionally, the applications may also be nudging users toward using one set of hashtags over another, contributing to the external imposition of semantic associations between themes that can undermine organic processes of meaning-making. By offering established categories of hashtags, the developer's own semantic associations are potentially reproduced by thousands of users, solidifying these semantic associations at the expense of semantic associations that are not led by third-party apps. The feature which enables users to track hashtags that are currently optimal on Instagram tends to favor hashtags that are used by those who have more time and expertise to dedicate to Instagram, such as businesses or influencers, or more worryingly, scam artists preying on users' hopes for a good life (Brown 2018).
The algorithms behind Instagram and Prilaga's applications produce social realities (Cheney-Lippold 2011) for Instagram users that disproportionately enhance a few users' power to shape narrative patterns. In practice, these inequalities in shaping the discursive space reinforce the visibility of content with commercial aims. This does not mean that all content which achieves high visibility has commercial aims, nor does it suggest that all content which fails to achieve high visibility is sustainable. In fact, research indicates that lay users may attempt to emulate the content published by successful users (Abidin 2016). Rather, we mean to highlight that proportionally more of the highly visible content is content with commercial aims, due to the business logics of Instagram and to the time and skill intensity required for creating successful Instagram content. Furthermore, research has evidenced that influencers overwhelmingly belong to a specific social group, namely the white, educated, well-connected, and financially secure middle-class. This inequality in shaping online social conversations and wielding attention arise both from the advantage that time-rich and financially secure demographics have in creating content, and from the alignment of middle-class ways of presenting content with marketers commercial interests (Bishop 2018; Duffy 2017).
Accordingly, the consequences of these inequalities reach beyond the promotion of less sustainable narrative patterns over ones with more potential to provide sustainable approaches to living well. The inequalities themselves undermine the possibility of sustainable futures. For Hammond (2019, 69), sustainability demands a democratic political culture in which "meanings can evolve in an open and normatively driven way rather than cementing themselves structurally as a result of powerful interests.". Hammond understands sustainability as the adaptation of societies to a changing environment, while retaining socially worthwhile forms of living. As such, sustainability requires democratic participation in the process of creating new stories of living well. Indeed, to be sustainable, the meanings of living well must be continuously questioned, and different perspectives must be engaged in this public discussion (Young 1997). Instagram superficially promises the possibilities for such public discussions but in practice systematically undermines them by promoting the perspectives of a few specific users and by minimizing user exposure to content they may be unfamiliar with. Actively working to "disrupt the feed" (The Female Lead and Apter 2018) and secure access to more heterogeneous content is a first step toward public discussions of the meaning of living well. But allowing this kind of participation to flourish demands a rethinking of the business logics underpinning Instagram itself.

Conclusion
In this article, we have argued that the themes that are the most visible in the discursive space of #goodlife on Instagram are unlikely to be supportive of sustainable understandings of living well. Evidently, our identification and interpretation of these narrative patterns, as well as the implications that we draw for the possibility of more sustainable and fairer futures are subjective. The relationships between clusters are dependent on our thematic classification of the hashtags. Furthermore, our conclusions are based on a bounded snapshot of Instagram activity, taken at a particular point in time, which obviously limits our ability to discuss the rapidly changing landscape of the platform (Skeggs and Yuill 2016). Current understandings of living well appearing on Instagram posts tagged #goodlife may differ from the ones we have identified. Future research could usefully explore changes over time in the discursive space created by posts tagged #goodlife or engage with niche hashtags in which more sustainable understandings of living well are more likely to be present (e.g., #slowliving). Increasingly, research also evidences the relevance of untagged content to the social meanings of particular concepts, such as McCosker and Gerrard (2021) work on Instagram-mediated depression. Such approaches could be extended to further understand the good life resources created by and available to Instagram users. Finally, exploring which communities of users engage with the different narratives of the good life that we have identified could potentially further our understanding of their differing visibility and recognizability in the discursive space created by #goodlife.
Despite these limitations, we believe that important conclusions can be drawn from our analysis. Our research builds on existing scholarship on hashtags by demonstrating how the use of third-party hashtag optimization applications shapes discourses on hashtag feeds. We argued that the hyperlink infrastructure created by hashtags is essential in shaping the discursive flow and establishing semantic associations between themes of hashtags. The use of third-party hashtag optimization applications results in consistent use of hashtag sets, contributing to the creation of a narrative pattern and making this association of themes more recognizable and the posts which convey those narratives more visible in a given discursive space. Building on existing work on commercial platform logics and algorithmic shaping of social practices, we suggested that the visibility of posts is disproportionately determined by algorithms (Instagram and Prilaga), often to favor corporate interests (Bishop 2018). Instagram users use third-party applications for their own interests-typically increasing the popularity of their own posts-but this use impacts broader social conversations and imbues them with a commercial flavor, as we have shown with understandings of the good life. As it stands, the discursive space of the #goodlife on Instagram is dominated by narrative patterns that are less likely to support environmentally and socially sustainable understandings of the good life. Beyond our interpretation of the implications of semantic associations between particular themes, the inequalities in shaping the discursive space also undermine our ability to create more sustainable stories of living well by limiting the scope for democratic discussions.
However, more sustainable approaches to living well are present on the platform. Broadcast by users themselves, these understandings, while not necessarily presented as "sustainable", seem to already be valued in some people's lives. The potential benefits of such understandings of the good life on individual, social, and environmental levels undeniably heighten their importance in the context of ongoing inequalities within and between countries, collapsing economies, and increasing ecological destruction. Which conditions can support their shining through remains an important question of our time. We encourage future research to look beyond existing algorithmic processes and think critically and creatively of the possibilities for a democratic online space.

Eynon and Geniets explain that not all young people
have access to online digital media, even in places such as the UK where it is usually assumed they do. 2. We deleted all versions of #goodlife: while Gephi differentiates between capitals and minuscules, Instagram does not. 3. The free versions of the applications feature advertisements and automatically input #prilaga in the hashtags that they provide. Unless users type out the provided hashtags themselves, #prilaga cannot be removed. By upgrading to a paid version, users obtain the ability to remove #prilaga and block in-app advertising.

Declarations of interest
The authors declare that they have no competing interests. All materials are reproduced with their authors' consent.
To protect the privacy of Instagram users, the data supporting this study is not publicly available.
This work has been supported through a PhD studentship from the Faculty of Engineering and Physical Sciences of the University of Surrey.

Funding
This work was funded by Economic and Social Research Council (ES/M010163/1).