New paper: Friendship Networks and Conversation Structures in Facebook Groups

Summary of our new paper presented at the 9th International Conference on Web and Social Media


THE GIST: To what extent do friendship ties influence the conversation structure in open groups? And how do individuals leverage their existing friendship relations when speaking to a large and often heterogeneous audience. In this paper we examine the effects of several structural metrics at both the actor and group level on the structure of conversations in 30 Facebook groups. The paper is presented at ICWSM 2015 in Oxford.

The Internet is often touted for its ability to enable interaction among otherwise disconnected actors. On Twitter, disparate users can coalesce over a topic using hashtags. On Reddit, users employ subreddits to discuss everything from crochet to conspiracy theories. Facebook has similarly embedded a feature to allow collective discussions beyond one’s friendship circle: the Facebook group. Community associations, student unions and all manner of subcultures host Facebook groups for discussions. Yet, while these groups can be accessible to all (or say all members of a student union), it is unclear how much the underlying friendship structure plays a role in social feedback.

I have first started to analyse this problem during my Master of Science programme at the Oxford Internet Institute in 2012/2013. Even though it originates from my MSc dissertation, this paper is a collaboration between myself and my supervisor and colleague Dr Bernie Hogan.

In this paper, we assess the extent to which the underlying friendship structure of university discussion groups influences the turn-taking and social feedback behaviours of group members. We are interested in assessing whether groups function as inclusive spaces, or whether such groups reproduce existing social hierarchies. We frame the process of engaging with others as “conversational agency”. Insofar as latent friendship structure is reproduced in the social feedback and conversational initiation practices within an open group, we can assert that some group members have more agency than others. We present findings from an analysis of 30 student groups drawn from a stratified sample of UK universities.

Some numbers from this paper: Overall we analysed 30 Facebook groups with 10,820 group members and 67,555 friendship ties between them. Of these, 2,148 were active members who produced 3,895 wall posts, 5,336 comments and 5,680 likes within the studied groups.

Our conclusions presented in the paper suggest that friendship structure pervades group conversation structure. This suggests a need to consider new features to make ostensibly “open” groups more inclusive to new members and more deliberate strategies to manage feedback within groups.

While social science researchers have long studied conversations in social groups, this paper provides granular insight into one particular aspect of this question: how do group members’ network positions and the overall patterning of personal ties in online social groups relate to their prospects of conversational agency?

Friendship network of a typical Facebook group (connections indicate friendship ties, size of nodes indicates number of shared friends in the same Facebook group, colours indicate communities)
Conversation network of a typical Facebook group (connections indicate in-group communication e.g. comments or likes, size of nodes indicates number of initiated posts on the group wall, colours indicate communities)

Degree and eigenvector centrality in the friendship network were both found to be positively associated with conversational in-degree and out-degree, as well as the total number of initiated wall-posts. Following this rationale, actors with more friends in the group may feel empowered and motivated by their presence, kindling a greater sense of agency. On the other hand, group members who react and participate may make themselves attractive as friends to fellow group members.

Being in a position mostly connected to low-degree alters may not contribute to eigenvector-centrality scores, but may render actors more powerful, because their surrounding neighbours depend on them. For instance, peripheral actors may rely on more central alters to access new information and to arrange introductions to distant members.

Finally, the group-level variables from our analysis posit that actors integrated in strongly cohesive Facebook groups face a different set of constraints and resources than those who are not embedded in such networks. Across all social groups, the conversational measures are positively associated with group density, while being negatively related to modularity, and the number of components. Particularly density is found to be substantial for conversational out-degree and in-degree.

In view of the results, actors appear to be particularly receptive to group cohesiveness when initiating new wall-posts. This provides evidence for the hypothesis that more cohesive groups constitute group solidarity, creating more encouraging environments for conversational agency and uptake. From past research we know that a more cohesive network could increase the likelihood of activating social resources. Accordingly, highly cohesive groups are better arranged to generate feedback.

From a practical perspective, our findings indicate that group managers can facilitate activity in online communities by recognising the links between relational structures of the friendship network and conversational agency. For example, by encouraging the formation of social ties and incentivising the invitation of friends from outside, they might cultivate a better climate for conversation. Similarly, by improving the visibility of low-degree members, managers might enhance the agentic prospects of hitherto silent group members. In addition to that, group managers need to recognise the role of influential group members in facilitating dialogue and integrating newcomers into the group. These and other measures can develop greater group cohesiveness and nurture more engaged online communities.


A pre-print of the conference paper and abstract can be accessed below:

Polonski, V. W. & Hogan, B. (2015). Assessing the Structural Correlates between Friendship Networks and Conversational Agency in Facebook Groups. In Proceedings of the 9th International AAAI Conference on Web and Social Media, 674–677.


About the authors: Vyacheslav ‘Slava’ Polonski (@slavacm) is a doctoral candidate at the Oxford Internet Institute. He researches the psychology of technology adoption and is a Global Shaper at the World Economic Forum.

Bernie Hogan (@blurky) is a research fellow at the Oxford Internet Institute with research interests at the intersection of social networks and media convergence. He is also working on a number of methodological issues, including reliable capture of online networks and social science software development.

Like what you read? Give Dr Vyacheslav Polonski a round of applause.

From a quick cheer to a standing ovation, clap to show how much you enjoyed this story.