Containing Homophily

Network analysis and social media platforms assume and construct homophily: they use similarity to breed connection. These projects question homophily’s axiomatic status by exploring its history, questions, methods and platforms—and by exploring and imagining other modes of connection. Heterophily was coined in the same article as homophily; mutual indifference is key to cities and infrastructures. In particular, these projects ask:

  • How has the concept traveled across scientific disciplines? How did homophily transform from questionable finding from unpublished data to a core principle of network epistemology?
  • How can we re-conceptualize social media platforms by exploring and operationalizing mutual indifference, invisibility, and heterophily?
  • How can social media alternatives contest network logics of distinction and segregation?
  • To what extent do our current social media networks prescribe the homophily they claim to only describe?

Facilitators: Wendy Chun, Richard Rogers, Michael Stevenson

Group 1. How has homophily traveled?

Project members

Intro and Research Question(s)


Findings and visualizations

Shared similar Keywords in Google News for "Homophily"

Fig.XXX Similar shared keywords for Homophily (first) and Heterophily (second) on Google News and for Homophily on PubMed (third).

Group 2.[Sub-project title here]

Project members

Petra Audyova Ben Blackwell Andra Irina Cristina Ene Jiyoung Ydun Kim Cengiz Salman

Intro and Research Question(s)

We aim to explore heterophily through the indifference present across social media platforms and in social networks. We ask how exclusion and erasure may be ingrained in the infrastructure of social media platforms, and whether they are necessary for the formation of networks. This begs the question: how might one detect invisible users in a network? How might we visualise the ‘invisibles’?

Most research focuses on the strongly expressed positive or negative connections between individuals, with network science ultimately oriented toward finding and analysing clusters. These clusters are defined by their rejection of that which lies outside of their boundaries. In our project we intended to map mutual indifference, which we defined as a lack of (expressed) affective regard for the other or object encountered in a shared space. We gesture toward one potential method for detecting and measuring indifference across social media platforms: measuring and mapping “fuzzy clusters” or sets of edges that are relatively homophily-resistant.


To begin, we conducted short interviews with experts to define the metrics for indifference per platform. These interviews led us to conceptualise a split between indifference from the perspective of users, and from the perspective of the content. Indifferent users were seen to be the ones who showed a low level of engagement with infrequent original posts whilst indifferent content was found to be the posts with low levels of views, comments and other signifiers of engagement. Indifference, as seen in these perspectives, would most obviously be presented as the absence of data, which poses difficulties to the researcher aiming to detect and visualise it. We were then faced with the task of tackling the subject of indifference in a way which could be traced through interactions by and between users and content. In other words, we thought it was important to identify “mutual indifference” within datasets that capture a matrix of interactions between users and use these relations to detect homophilous communities.

Findings and visualizations

According to Newman, a network of neatly clustered communities can be evaluated according to a measure of modularity, which is based on the normalized difference between a the supposed random number of edges one would expect to appear between clusters and the number of edges that a network scientist actually observes.

“A good division of a network into communities is not merely one in which there are few edges between communities; it is one in which there are fewer than expected edges between communities” (Newman 2006, 8578).

A positive value indicates that there are fewer links between clusters than one would expect, suggesting a network consisting of relatively discrete communities. Positive values, that is, indicate a greater degree of homophily. In contrast, a negative modularity score suggests a that there are more links between clusters than you would expect, indicating the presence of “fuzzy clusters” or sets of relations between nodes that resist easy clustering.

We collected data with the Twitter Capturing and Analysis Toolset (TCAT) from DMI’s “american politics” dataset for June 21, 2018, a day with a smaller volume of twitter activity to inverse the event and controversy-based approach that network scientists often take to identify homophilous activity online. Because we used the “social graph by mentions” module in TCAT, identified clusters represent communities of twitter users engaging in conversation and citing one another. Using Matthieu Latapy’s “Partition Analysis Supplement” script, available at Graph Recipes, we identified modularity scores for each partition/cluster/community in our dataset to identify a potential “fuzzy cluster.”

Group 3. Engineering Homophily: what is alternative in ''alternative'' social networks?

Project members

Loes Bogers, Serena Coppolino Perfumi, Anu Masso, Silvia Semenzin, Dan Xu

Intro and Research Question(s)

Which features in “alternative” social media platforms can be seen to enable forms of heterophily and homophily? How and to what extent can they be seen as engines of homophily? A walkthrough of Path, Mastodon and Peanut.

Lazersfeld and Merton (1954) coined the terms homophily and heterophily as key concepts to understand friendships as social process in what were at the time racially segregated neighbourhoods in the US. Since the 50s, homophily and the idea that “birds of a feather flock together” stubbornly remained at the core of ideas of friendships and social networks. Heterophily however has not taken off to the same extent as far as understanding social attraction goes, although it would help understand how “opposites attract”, something commonly believed to be as prevalent as its birds-of-a-feather counterpart.

Online news consumption through untraditional sources such as Facebook and Twitter, have produced a coupling between news consumption and personalization: algorithms learn about users’ behaviours and curate content accordingly. Such “content” can by anything suggested by the platform to the user, from news to products to people they might like and want to know more about. These algorithmic suggestive practices are both are said to lead to socalled algorithmic filter bubble (Pariser 2011) or echo chambers (Flaxman, Goel & Rao 2016) and can be seen as a threat to the social fabric and even democracy as they decrease people’s exposure to diverse perspectives and difference. In the context of social networks, algorithmic suggesting might occur on the level of news content, but also in terms of people and groups to befriend or link with. Homophily is the operating axiom here: where such suggestions are made based on likeness, meaning they are similar to what/who a user has interacted with before.

The infrastructure of the technical system and the way it curates content and constantly suggests lines of action are both prescriptive and performative (Chun 2018); it constantly nudges users into performing certain actions (and not others). Each user action conversely, also confirms the system’s predictions about that user, training it its algorithm to recognize likeness even more accurately. If including and promoting difference within social networks can be seen as a social and political imperative, we need to engineer its affordances in such a way, but our understanding of networks is too strongly anchored in axioms of homophily (Chun 2018a). In this project we have set out to recognize and identify moments in which heterophily is actively engineered inside of “alternative” social media platforms such as Path, Mastodon and Peanut. In doing so, we have also started to identify moments in which systems’ patterns for user interaction and other affordances promote homophily instead. The outcomes of this project could be a starting point for developing concrete UI/UX design principles for platform features that can work as engines for heterophily (as well as homophily), with the ultimate goal to envision social networks that can expose us to difference and dissent without disconnecting from otherness, or in Chun's words (2018a) can we move from correlation to co-relation and explore how we are all entangled, to explore difference rather than similarity?


We used the walkthrough method (Light, Burgess & Duguay 2016) to examine the different platforms. The method is developed for critical analysis of apps, building on STS and cultural studies). By engaging with the interface, looking at screens, features and flows of activity, technological mechanisms as well as embedded cultural references can be examined to stake out the app’s “environment of expected use” (Light, Burgess, Duguay 2016: 1). A walkthrough consists of three stages: 1) login and registering; 2) mimicking everyday use, and 3) suspending or logging out or deleting the account. The researcher records field notes and observations throughout, logging screenshots or screencasts. Light, Burgess and Duguay also suggestion looking at additional materials such as business models, revenue sources, economic and political interests, employee recruitment materials, press releases, policies of use, terms and conditions etcetera, while paying attention to mediator characteristics such as UI arrangements, textual content and tone, functions and features, and symbolic representations (2016: 10-15).

Key concepts
Chun pointed out that Lazarsfeld and Merton fail to give instructions as to how one might operationalize the examination of heterophily (2018b). In the context of the Summer School where this project was developed, Chun has suggested four modes of heterophily that might be identified inside networks:

  1. Weak ties (as developed by Granovetter 1977, 1983) can be understood as features that leverage outer circles, rather than already strong ties;
  2. Antagonism or anti-homophily, being moments in which a systems features afford forms of disagreement or dissent;
  3. Mutual indifference: moments in which actors are connected by being equally indifferent to eg. content or an algorithmic suggestion.
  4. Affinity politics: organising around a general agenda or cause. This idea was developed by Donna Haraway (1985) as alternative to identity politics (modes of organising around issues and causes that circumvent mobilization along lines of identity labels which still hold the premise of “sameness” in terms of eg. being queer, POC, differently abled).

Case studies
We selected three case studies that can be considered “alternatives” to the big social media platforms like Instagram, Facebook and Twitter. We selected Mastodon, a Twitter-like platform that emerged from the hacker community and actively positions itself against ad-based incomes and does not sell user data in anonymized nor aggregated form, where uses can program and host their own nodes (implying a rather tech-savvy user base). Secondly, we selected Path, a platform similar to Instagram aimed at a young user base, that emerged from a business start-up. Path is is promoted as putting user agency first when it comes to privacy. We had to nuance these statements very early on, as we will describe below. Thirdly we selected Peanut. A Tinder-like app and discussion platform developed for moms to meet other mom friends near them. Peanut was selected because it starts out from the idea of a small homogenous user base (young cis gender women with children), yet at the same time signifies a form of affinity politics in the way it positions itself: “Meet as mamas, connect as women”. The shared agenda her is that young mums suffer from a big change in identity after having children and can support each other in dealing with these changes that all moms seems to face to a greater or lesser extent.

Findings and visualizations

Example of an annotated screenshot from the Path Walkthrough

Example of an annotated screenshot from the Path Walkthrough

Outcomes of the project is a features matrix based on Light, Burgess & Duguay’s analytical categories. We extended their analytical categories with subcategories based on the small, but very typical or very unique features inside the social media platforms studied. We have highlight the features that afford more potential for heterophily.

Homophily works as a basic assumption to greater or lesser extent when approaching social networks, for engineers/developers/designers as well as users themselves. The idea of “sameness” as the basis for connecting is present to greater or lesser extent everywhere.

Overall, open-source platform Mastodon offers the most potential in its unique features. It explicitly distances itself from ad-based models and promotes user agency in terms of privacy, both in terms of social privacy (what you share with whom) and privacy on the back-end: user data is not sold in aggregated/anonymized form to third parties. Also, the way privacy is engineered in Mastodon, enables more heterophily as it promotes users’ agency to select content while using decentralized identities. In Mastodon you can engage in different spaces with an avatar registered under a different email address for each community you join within the platform. This practice of several identities makes it impossible to pull together user actions into one single profile, and as a result, circumvents practices of algorithmic profiling that can make predictive/performative suggestions to the user.

We can see an ongoing tension between homophilous and heterophilius “engines” insides alternative social media platforms. Some features are explicitly developed to produce serendipity for example (potential feature for heterophily), while others are latent in the system. Some features that nudge actions along lines of homophily are actively requested by users, or ironically are prevented from taking place due to a small user base (where algorithms do not have large enough “pools” of users to draw suggestions from).


Chun, W. (2018a) "Critical Data Studies, or How to Desegregate Networks. From Correlation to Co-relation." Keynote lecture, 9 July Digital Methods Initiative Summer School, Amsterdam.

Chun, W. (2018b) "Queering Homophily" Unpublished paper and personal communication during workshop, 10-13 July Digital Methods Initiative Summer School, Amsterdam.

Flaxman, S., Goel, S., & Rao, J. M. (2016). Filter bubbles, echo chambers, and online news consumption. Public opinion quarterly, 80(S1), 298-320.

Granovetter, M. S. (1977). The strength of weak ties. In Social networks (pp. 347-367).

Granovetter, M. (1983). The strength of weak ties: A network theory revisited. Sociological theory, 201-233.

Haraway, D. J. (1985). A manifesto for cyborgs: Science, technology, and socialist feminism in the 1980s (pp. 173-204). San Francisco, CA: Center for Social Research and Education.

Lazarsfeld, P. F., & Merton, R. K. (1954). Friendship as a social process: A substantive and methodological analysis. Freedom and control in modern society, 18(1), 18-66.

Light, B., Burgess, J., & Duguay, S. (2018). The walkthrough method: An approach to the study of apps. New Media & Society, 20(3), 881-900.

Pariser, E. (2011). The filter bubble: What the Internet is hiding from you. Penguin UK.

Group 4. Uncommon associations: Metrics and modes of heterophily on Twitter

Project members

Ana Pop Stefanija, Andrea Benedetti, Beatrice Gobbo, Chad Van De Wiele, Clark Powers, Mathieu Jacomy, Yarden Skop

Intro and Research Question(s)


Research protocol

Step-by-step conceptualization diagram

Step by step conceptualization diagram, from dataset to small multiples

Fig.XXX Starting from the visualization of frequency of pairs of hashtags in a co-hashtag network exported from TCAT, the result is mathematically inverted and then used to see, for each user, which pairs of hashtags are used and to then calculate their entropy value, allowing categorization between being an entropist inside the issue, or a conformist.

Findings and visualizations

Topic attachments
I Attachment Action Size Date Who Comment
researchprotocol.pngpng researchprotocol.png manage 149 K 19 Jul 2018 - 13:01 LoesBogers Research Protocol
Topic revision: r9 - 19 Jul 2018, LoesBogers
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback