Entering the Demos-sphere
Investigating Facebook Publics
The limitation to the hyperlink approach to mapping the publics is quite apparent. Only certain central and powerful actors are represented and the hyperlink structure guiding us towards important and authoritative actors is filled with assumptions about their visibility in a broader public. We propose to use Facebook activity data to map and explore other publics.
Facebook and social media has become a dominant source of information to many people. Until now we have explored the publics constructed by the google algorithm and the logic of the issue crawler. Now we want to map how users of social media co-construct different publics interacting with the Facebook algorithms of Sharing, Commenting and Liking. The data is big and many analysis are possible. We propose an analysis focusing on mapping and quantifying uncertainty of an issue, the variety of experts consulted and the construction of experts on Facebook and patterns in affiliations between different experts constituting different publics. We have a unique view into the vision of each Facebook-user, how the public plays out before him or her and how he or she co-produces it. With the Facebook-activity data, we have access to the actions that Facebook uses to estimate the 'collective intelligence' that affect the information flow to the user, creating the feed and the recommendations that constitute the individual users access to public information and controversy. |
The dataset evolves around the more general uncertainty connected to the concept of health. As shown with the tag clouds health is a dominant word involved in all the websites that we have collected.
We choose to sample danish users. The choice reflecting a tradeoff between the need to reduce the data size from hundreds of millions of users to the current 135,000 active danish users, and the fact that Facebook forums and constructions of publics are compared to other web based information networks highly local. We choose to search for all danish groups and Facebook-pages related to health, diets, and in part training. The data was collected using Netvizz, which extracts not the amount of likes each page has, but the concrete activity of either liking, sharing or commenting on a post on a Facebook-page. In this way the dataset is only the tip of the iceberg with regards to the people affected by the issue, but have strong data indicating how actors are involved in the creation and discussion of the issue. Pages and groups were found using varies strategies, from lists of diets, to general queries on health, slimming and training in the Facebook search and to the recommendations provided by Facebook. When liking the page you can gain access to the activity data through Netvizz (mediating the Facebook API). We choose to sample all activity since 2010-01-01. |
Untangling the Data-Monster
Above is a representation of the full dataset consisting of 135,000 users liking, commenting or sharing posts on 'Health'-sites, 550,000 times, as an undirected graph. The posts have been aggregated into pages. The size of the nodes and the size of font represents the degree (amount of links). But what to do with this monster? By the time the data size reaches this level, the spatialization algorithms start having big problems, because the position of a node is not an exact solution, but found iteratively where each random choice of 'attraction-repulsion'-relation can take the process in many different directions. The same goes with the community detection algorithm, so the point is that colors and positions are not easy to interpret, and only to make the visualization edible.
We need to ask relevant questions to the data, that will filter noise.
Above is a representation of the full dataset consisting of 135,000 users liking, commenting or sharing posts on 'Health'-sites, 550,000 times, as an undirected graph. The posts have been aggregated into pages. The size of the nodes and the size of font represents the degree (amount of links). But what to do with this monster? By the time the data size reaches this level, the spatialization algorithms start having big problems, because the position of a node is not an exact solution, but found iteratively where each random choice of 'attraction-repulsion'-relation can take the process in many different directions. The same goes with the community detection algorithm, so the point is that colors and positions are not easy to interpret, and only to make the visualization edible.
We need to ask relevant questions to the data, that will filter noise.
|
Controversy and Uncertainty
In our mappings we take theoretical interest in the concept of controversy. This makes us look for concerns, discussion, and debate, different positions and as defined by Nortje Marres, the intersection of different positions. In our case, controversy is not just the existence of different takes on the idea of personal health, or how to lose weight, or what is the natural lifestyle, but that these different positions actively engage in publics together. That they take stands towards each other. Using the facebook data, we will be interested in how users engage in the issue across different fora; groups and pages. To the left you can see and interact with the first filter, which has been named uncertainty. Users consulting and engaging in different fora. At the same time it shows how the activity of most users (78%) is tied to a single forum, which has been named: Loyal Users. It has to be stressed that we are only looking at the tip of the iceberg, because it is only the ones interacting with concrete posts on the facebook page, and not all of the exposed. Most interactions are with users that hold the same views or consult the same experts. But the 22,615 'uncertain' users are represented in all the fora and thereby also involved in the construction of the feed. We have chosen to take the uncertain users for further analysis. |
The Creation of Complex Publics
From the below illustration we can try to navigate in the complex ways that actors engage and recommend each other different information. We propose to use a distinction between Simple Publics and Complex Publics.
Simple publics represents to which extent the pages form unconnected and consistently delimited pockets of communication. Information recommendations are from peers, and the information flow is suspected to be very low.
Complex Publics on the other hand, is highly interconnected with none or very little consistency in the affiliations. The diversity of information is high, and the ways in which the information is mediated is broadly involving many different actors. Users situated near the center would be expected to be involved in more complex publics.
In the coming segment we try to quantify this notion on the facebook data, using various visualizations.
Simple publics represents to which extent the pages form unconnected and consistently delimited pockets of communication. Information recommendations are from peers, and the information flow is suspected to be very low.
Complex Publics on the other hand, is highly interconnected with none or very little consistency in the affiliations. The diversity of information is high, and the ways in which the information is mediated is broadly involving many different actors. Users situated near the center would be expected to be involved in more complex publics.
In the coming segment we try to quantify this notion on the facebook data, using various visualizations.
The graph represents the subgraph of all users engaging in at least two different fora. The visualization has sized each node and font by their degree, and although the contrasts are not very sharp the font is color coded to show the share of 'uncertain users' each group/page has, the whiter the larger share. See Christian Bitz in the Corner for a greyish font, representing low degree of overlapping users. Another Data-Monster. Now only the users engaging in multiple fora. The spatialization algorithm and community detection find more stable results. And the affiliations and overlapping users between groups can be explored visually. Users are mostly placed in the center of the graph, indicating many different links. We can now interpret different semantic groupings: The Before and After Situated in the middle to the right we see that many consult or take interest in the two groups that represent individual sharing their struggle to lose weight; 'Fies vaegttab' and 'Sofias Kamp'. The page serves entertainment purposes and does motivational speeches and rolemodels. Health Tips 'Lukmundenandletroeven' and 'Sunde-tips-hverdag' is placed in the center, where small tips and recipes are shared. Paleo Community We see that the users involved in our subcontroversy are highly affiliated with training philosophies of Crossfit which also propagates ideas about the natural movements of man(Butcher's Garage, Butcher's Lab and Copenhagen Crossfit). Together they form the component weakest connected to the rest. |
|
Gender performing Publics Looking at the top of the graph we see how gender affects the construction different publics, from the middle to to the top. The training magazine (Iform.dk) training center and a celebrity dietist Christian Bitz, cluster in a pink (unintended heuristic) cluster. Guided by the pink heuristics we choose to make a visualization of the gender data . To the left you can explore the share of females in the different groups and pages. |
Quantifying Complex Publics
Given that the flow of information and the activity on a facebook page is co-created by page and users the amount of users engaging in several fora could express the diversity of a the public. This has been investigated by calculating the percentage of users active in more than one forum per facebook page. In the below visualization this can be explored for each page.
The interactive visualization above let's you investigate how different sites have different distribution of 'uncertain users'. The complexity of how the groups will mobilize to make recommendations is increasing on the y-axis and on the size of the dot. Along the x axis the size of the public increases. Fresh Fitness and Fitness World users, are both fairly large, and with high complex overlaps. In the network spatialization from before this will be shown by the inability to distinquish different clusters when looking at our Graph, both Fresh Fitness and Fitness World are situated in centered positions.
To open up even further in how different forums have relative complex constituations, one should look at the distribution of different combinations. Using the network analytics from the fan (Entering the Paleo Diet) this would be shown by the count of 'Out of community Links'. However we have chosen to look at another population. Given our interest in the Paleo controversy we have filtered out once again from the full Facebook network. Now we keep only the users active in Paleo and investigate the complex affiliations of the Paleo 'Ego-Network'.
To open up even further in how different forums have relative complex constituations, one should look at the distribution of different combinations. Using the network analytics from the fan (Entering the Paleo Diet) this would be shown by the count of 'Out of community Links'. However we have chosen to look at another population. Given our interest in the Paleo controversy we have filtered out once again from the full Facebook network. Now we keep only the users active in Paleo and investigate the complex affiliations of the Paleo 'Ego-Network'.
Paleo Affiliations
This visualization tells us which publics affect the Paleo page information flow to the users, and at the same time where active Paleo users are engaging and performing other publics. The different size of the node shows the size of equal engagement on the Paleo page simultaneously with the other pages.
Let's take a closer look.
First of all it is apparent that users liking the Paleo page also engage in other pages representing lifestyle issues.
The affiliation to pages like Butcher's Garage, Buther's Lab and Copenhagen Crossfit is not so surprising, given that they represent fitness options, which are already a big part of the Paleo lifestyle. What becomes quite interesting is the affiliation to the page Lukmundenogletroeven. The page belongs to the Danish nutrition expert Christian Bitz who officially is quite antagonistic to the Paleo Diet. There is clearly an uncertainty among the users, who choose to engage in both pages, because of the conflicting interests of the two camps.
Let's take a closer look.
First of all it is apparent that users liking the Paleo page also engage in other pages representing lifestyle issues.
The affiliation to pages like Butcher's Garage, Buther's Lab and Copenhagen Crossfit is not so surprising, given that they represent fitness options, which are already a big part of the Paleo lifestyle. What becomes quite interesting is the affiliation to the page Lukmundenogletroeven. The page belongs to the Danish nutrition expert Christian Bitz who officially is quite antagonistic to the Paleo Diet. There is clearly an uncertainty among the users, who choose to engage in both pages, because of the conflicting interests of the two camps.