Since its first premiere in 2011, the HBO produced tv show Game of Thrones has become one of the most watched, illegally downloaded, critically acclaimed, put simply, one of the most anticipated tv series presently available. We may henceforth ask: in what ways and by what means does this show shape and influence contemporary culture? On what grounds and in what forms is it becoming a meaningful part of people’s lives?
To do so, we can start by revisiting Raymond Williams’ conceptualization of a selective tradition . For Williams, culture is shaped and created through selective operations. Through the active engagement of people and institutions with cultural objects and experiences. An active engagement in which certain aspects of a culture once lived are selected to be of value and emphasis while others are rejected. Those of value are recorded and preserved, be it in the form of objects or practices. This recorded culture then shapes and conditions the grounds upon which a contemporary culture emerges, determining what meaningful experiences can look like in the future.
These selective operations hence don’t start from scratch. They are part of a selective tradition constantly evolving. Within this tradition, selective operations are pursued across a variety of linked ecologies, all of which are invested into culture following different objectives, visions, and desires, different histories of their own (often accompanied by conflict and controversy).
We can see this for instance in the Game of Thrones tv show being an adaption of the same-titled George R.R. Martin’s fantasy novel series. Game of Throne’s cultural significance always already has been part of a larger historical trajectory of selective operations. And in doing so, the ecologies of publishers, marketers, literary critics, ‘ordinary’ readers, and fans are expanded. Production companies, tv networks, a whole other set of marketers and critics, ‘ordinary’ viewers, and dedicated fans of the show are all now also part of the selective operations surrounding this cultural conjuncture.
Trying to gain hold onto this continuous development, we can look at two areas within this tradition that we might describe as particularly new within our present digital society. Namely Game of Thrones video montages and their algorithmic ordering.
These montages highlight those parts of a participatory culture in which people negotiate and discuss their experience of the show. But further, and more crucially, they do create spaces in which the show can be experienced and re-experienced in different forms and rhythms. On the one side a discourse and collective memory, on the other a sui generis experience of the show. Their combination creating in the digital an experientially different ground to record and preserve parts of a culture once lived.
However, in being distributed on social media platforms like YouTube, the flow of these contents is regulated by the platform’s inherent logics. While this creates opportunities to connect and make accessible different contents surrounding Game of Thrones, it also raises questions of control and power. For have herein non-human entities such as algorithms, by ordering and recommending contents, become central actors carrying out selective operations, also making determinations of what parts of Game of Thrones and its experience are of value and to be emphasised.
Given the vast storylines featured within Game of Thrones, we will take House Stark as case to inquire these developments. Accordingly looking through the field of video montages related to Stark characters, we can identify four main areas House Stark as such is remembered by and presented along.
Firstly, there are montages of Ned, Benjen, and Robb Stark. They are often used to highlight themes of honour and loyalty, themes seen as integral to House Stark. While less frequent in number of videos, these montages are produced in high quality and have a broad reach and acclaim among viewers. They are edited to appear very dramatic, feature slow music yet forward paced and fast-moving images. Their focus is on action and they display lots of fighting scenes. Honour and loyalty are herein not presented as abstract concepts but as put into practice. The dramatic edit seeking to engage people into how the meaningfulness of House Starks comes to matter. Yet further these montages also seek to highlight narrative ties within the larger storyline. For instance, through cross-cuts and voiceovers, they make present connections between scenes that before where not related in the original edit of the show.
Secondly, we find montages of Brandon, Rickon, and Catelyn Stark. They are often used to showcase themes of courage and strength associated with House Stark. While these videos are seldom of high quality, they are also not that frequent. However, their existence is acknowledged when you look at what people say about them, for do they show aspects often marginalised and overlooked according to many. Courage and strength are herein presented in slow and dragging edits that retell the story of the characters. They have a particular focus on close-ups, seeking to highlight a deep reading of the characters’ emotions and their overlooked meaningfulness for the story.
Thirdly, a key strand focusses its attention on the characters Arya Stark, Sansa Stark, and Jon Snow. Doing so, they less explicitly direct attention towards House Stark, but instead focus on the themes of becoming and identity within the characters’ storylines. However, it is this development of finding and reclaiming a place within the family and world that is often seen as characteristic element of House Stark’s significance within the Game of Thrones experience. These montages seek to do so in dramatic edits. Both music and images are uprising, they are very rhythmic. Not retelling the original storyline, but rather telling a story on its own within. They vest the identity struggle and development of the characters with a rhythmic movement of images. Doing so, they primarily seek to engage people into the becoming of the characters.
Lastly, a major strand of videos focus on House Stark as such, looking at the themes of becoming and home in particular. These montages are very frequent and most of them are of high quality. They are forward paced, feature lots of cross-cuts and voiceovers, are often action laden, and incorporate a large variety of musical genres and styles that are in many cases unrelated to Game of Thrones (most of the character montages use pieces from the original score or a similar style of music). Taking this form, they less seek to retell the story of House Stark, but rather make present the feeling and mood of it; the emotional journey of finding back home. While they show key moments of and parallels between the different Stark characters, they mainly showcase the fall and rise of House Stark. Do so less in explaining the contexts surrounding this trajectory, but rather in rhythmically engaging people into this dynamic and existentially challenging movement.
Following these four areas of montages, we are confronted with a space of experience on the platform. It is within these community-created video montages that people not only highlight and negotiate different aspects of the show and why they are of value (for them an others). But rather, in that these montages are audio-visually engaging contents, they create a themed space in which the mood associated with House Stark in particular and the show in general can be made present again. Even further, they create a space in which Game of Thrones can be experienced in a form that is unique from the original tv series. A unique experience that allows people to be engaged and attuned.
While video montages are one form of engaging with the show in the digital, they are part of a larger content stream available on the platform. Contents that are ordered and made accessible via YouTube’s related video network . Looking at the network for our case of House Stark, we are confronted with around 37.000 unique videos that were classified as related. Using computationally supported content analysis techniques and network modelling , we can identify twelve clusters or themes of videos within the overall network.
The most interesting things to observe here are the four big clusters in the centre of the network and the four clusters at the edges in the top left. These more marginal, yet still big clusters are interesting in that they feature tutorials and displays of practices that seek to recreate or materialize aspects of the show. We find a lot of videos that are piano tutorials that teach people how to play Game of Thrones themes but also many other musical scores. We can find many videos that are tutorials showing how to recreate certain symbols and artefacts from the show using woodworking or origami techniques. But there also is a group of videos in which people recreate and stage battles from the show within video games.
Moving to the centre, we have three big clusters surrounding one medium sized one. This medium sized cluster entirely consist of reaction videos, i.e. people capturing their reaction while watching certain, often very dramatic scenes of the show. The three surrounding clusters can be divided in community, commercial, and show related videos. Community related contents are especially videos that discuss the show in more depth and make connections between it and the novel series. Commercial videos feature the official behind the scenes materials, interviews with cast and production staff and connect towards the broader entertainment industry in form of late night tv contents or music videos. The show related cluster features video montages like those discussed above but also unedited clips and scenes from the show.
Put simply, looking at the related video network shows us a larger variety of different forms and formats along which House Stark and Game of Thrones (can) become a meaningful part of people’s lives. Within these clusters different ecologies are at work, following different logics, highlighting and emphasising different aspects of the show to be of value. Yet in addition to these ecologies that are directly invested into Game of Thrones, YouTube now is too. We may hence try to understand in what ways their algorithmic recommendations play a role in shaping contemporary culture.
Focussing our attention to the three central clusters featuring the community, commercial, and show related videos, we can see that YouTube is not ‘neutral’. Videos that overall discuss Game of Thrones and interviews with the cast are placed without emphasis. They are placed neutrally. However, videos that focus on specific characters, instead of the show at large, are pushed (placed more centrally in the network, are more likely to be recommended to people). Especially those video montages concerned with Jon Snow and Sansa Stark discussed above are emphasised here. However, it also pushes community videos that discuss and make predictions about the upcoming final season of the tv show.
Contents that are suspended by the algorithm (placed less centrally, are less likely to be recommended to people) mainly follow two strands. On the one side general major media content such as music videos or bits and clips from tv shows. On the other side, primarily community videos that establish and discuss connections between tv show and the novel series.
This is not to say that YouTube and its algorithms are engaging in the discussion and selective tradition with a clear and purposeful intention. But rather do they follow a logic inherent to the platform. Currently, this logic, at its core, seeks to maximize time-spent on the platform per user. In observing and quantifying their activities, they make predictions what contents will keep people interested in watching videos. Yet further, creating user-profiles from this data, they auction targeted ad space and monetize activities unfolding on the platform.
There sure is a need for more research into the impact and implication of these recommender systems. Yet while I think that them having an impact is without question, the degree to which most present research raises alarming voices about the amplification, captivation, and control that comes with such systems seems questionable (even if there are cases where I would agree). For does it feel like that this often entails doing the same mistake mainstream media studies did decades ago. Looking at what media do to the people instead of what the people do to the media.
It is thus with little surprise that scholars like Sonia Livingstone critically ask the field to take seriously the experiences people make within such digital worlds. Not to undermine the power imbalances and domains of control presently at work. However, in doing so be able to better understand, rethink, and reclaim these systems. As Nick Seaver, studying recommender systems through an anthropological lens of traps, argues, we should not aspire to free us from such systems that (seek to) captivate us. Rather, thinking about ways to reclaim control over them and the ways they (could) enrich our lives is what we need to do. Something, as Fred Turner argues, might also entail establishing democratic governance structures, laws and institutions dealing with these sites of our contemporary culture; for have, according to Turner, the ideals of connecting the world and simply amplifying and broadcasting what people do failed as did the communes of the 1960s.
This is not to argue against those sites running on a commercial basis. However, to argue in favour for them to more acknowledge the impact they have on shaping contemporary culture. The way YouTube currently is taking part in the selective tradition is one option. However, returning to our case, just as people feel that the experience of courage and strength in the characters of Brandon, Rickon, and Catelyn is often overlooked and undervalued, so do the algorithms also overlook and undervalue them.
As platform operator, it is easy to position oneself as plain mediator here. Yet, we may ask, is this the right thing to do? Is this really the best option out there, given all the data and computational resources at hand? Not that I am particularly anxious about the socio-economic futures these systems help create. But rather that I am slightly sad and disappointed about the chances for those futures they seem to miss and be unaware of day after day.
Parts of this piece were presented at the 1st International Popular Culture Conference, University of Seville, Spain, December 2018.
 Williams has most notably discussed this idea in The Long Revolution (1961). A more concrete application can for instance be found in Williams’ The Country and the City (1973). In tracing the persistence and historicity of the rural-urban divide in literary works since the 16th century, Williams highlights not only those selective operations that emphasised and rejected the lived experience of the country-side, but also how these shaped future experiences and socio-economic contexts of industrialisation and urbanisation accordingly.
 In order to personalize the experience on the platform, YouTube recommends each user videos based on a variety of factors, often with the intend to increase the likelihood to watch another video, i.e. increase time spent on the platform. Accordingly, most of the videos people watch were recommended to them by YouTube. The backbone for doing these personal recommendations is the related video network. It consists of videos that are classified as related or similar to one another based on different factors (content data, metadata, activity data) following a dynamic modelling process. This backbone dataset is then used to, based on a person’s activity data, guess what videos might be of interest to be watched next. Implicit feedback, the activity data gathered by YouTube for each user using the site and each video being watched/used on the site is hence of crucial importance here (Baluja et al. 2008, Davidson et al. 2010, Bendersky et al. 2014, Covington et al. 2016).
Reconstructing related video networks (the data is partially and with limited access available publicly) hence allows us to understand three sides of user-generated content platforms like YouTube: Firstly, what variety of contents there are available, i.e. contents created by individuals and organizations. Secondly, how the experience of watching videos related to a certain topic can look like for a broader variety of people. Lastly, how YouTube positions certain videos to be more accessible than others within a given topic, i.e. how YouTube takes part in selective operations of emphasising and rejecting contents.
 Network clusters are reconstructed using the Louvain community detection algorithm (Blondel et al. 2008) because, as YouTube builds relations based on similarity (see citations above), clusters within the network are assumed to be densely connected within themselves but not with the rest of the network. The Louvain algorithm maximizing the modularity measure henceforth comes in useful. Additionally, those nodes/videos connecting clusters with each other are assessed through the betweenness centrality measure (Brandes 2001). Lastly, a force-directed layout with scaled gravity is used for visualising the network (things central are placed centrally, which is further enhanced through more compact and comparable spatialization Bannister et al. 2013).
Computationally supported content analysis has become a prominent tool within sociology and the social sciences over the years (DiMaggio et al. 2013, Fuhse and Mützel 2011, Mützel 2015, Lewis et al. 2013, Lucas et al. 2015). Given the large number of videos available, these tools can be used in assessing themes of clusters through analysis of video titles and video tags. To do so, three steps are applied to each network cluster. Firstly, an inductive categorization of lower and upper bound of central (betweenness centrality) and influential nodes/videos (view count and engagement ratio) is undertaken from within the field. Secondly, central terms for each network cluster are extracted through co-occurrences analysis (Deerwester et al. 1990, Lund and Burgess 1996). These two steps are concluded by labelling each cluster. Lastly, in a final step, by fitting a structural topic modelling for each cluster with centrality and influence measures as covariates (Roberts et al. 2013), algorithmic ordering practices are sought to be reconstructed.
Bannister, M.J. et al. (2013). Force-Directed Graph Drawing Using Social Gravity and Scaling. In W. Didimo and M. Patrignani eds., Graph Drawing. GD 2012. Lecture Notes in Computer Science, vol 7704. Berlin: Springer, 414-425.
Baluja, S. et al. (2008). Video Suggestion and Discovery for YouTube. Taking Random Walks Through the View Graph. In Proceedings of WWW 2008, Beijing, China.
Bendersky, M. et al. (2014). Up Next: Retrieval Methods for Large Scale Related Video Suggestion. Proceedings KDD 2014, New York, US.
Blondel, V.D. et al. (2008). Fast unfolding of communities in large networks. In Journal of Statistical Mechanics, P10008.
Brandes, U. (2001). A Faster Algorithm for Betweenness Centrality. In Journal of Mathematical Sociology, 25(2), 163-177.
Covington, P. et al. (2016). Deep Neural Networks for YouTube Recommendations. In Proceedings of RecSys 2016, Boston, US.
Davidson, J. et al. (2010). The YouTube Video Recommendation System. In Proceedings of RecSys 2010, Barcelona, Spain.
Deerwester, S.C. et al (1990). Indexing by latent semantic analysis. In Journal of the American Society for Information Science, 41(6), 391–407.
DiMaggio, P. et al. (2013). Exploiting affinities between topic modelling and the sociological perspective on culture. In Poetics, 41, 570-606.
Fuhse, J. and Mützel, S. (2011). Tackling connections, structure, and meaning in networks: quantitative and qualitative methods in sociological network research. In Qual Quant, 45, 1067-1089.
Lewis, S.C. et al. (2013). Content Analysis in an Era of Big Data A Hybrid Approach to Computational and Manual Methods. In Journal of Broadcasting & Electronic Media, 57(1), 34-52.
Lucas, C. et al. (2015). Computer-Assisted Text Analysis for Comparative Politics. In Political Analysis, 25, 254-277.
Lund, K. and Burgess, C. (1996). Producing high-dimensional semantic spaces from lexical co-occurrence. In Behavior Research Methods, Instruments, and Computers, 28 (2), 203– 208.
Mützel, S. (2015). Facing Big Data: Making sociology relevant. In Big Data & Society, 1-4.
Roberts, M.E. et al. (2013). The structural topic model and applied social science. In Advances in Neural Information Processing Systems Workshop on Topic Models: Computation, Application, and Evaluation.
Williams, R. (1961). The Long Revolution. London: Chatto & Windus.
Williams, R. (1973). The Country and the City. London: The Hogarth Press.