In October 2013, Lars Backstrom (Facebook Inc.) and Jon Kleinberg (Cornell University) released a study suggesting a new method by which to measure and classify relationship strength (or “tie strength”) between Facebook users. The novelty of this method is its focus on the principle coined as “dispersion.”
Dispersion looks at a pair of users’ mutual friends in terms of network structure and measures the level of connectivity between each of these mutual friends (i.e. how many mutual friends do the mutual friends share). The tie between two users is deemed to have “high dispersion” if the mutual friends are not well-connected with one another. Within the absolute model of “high dispersion,” the user pair serves as the only node through which the mutual friends are connected to one another. As such, the user-pair’s relationship with one another is seen as the basis for whatever mutual friendships they share.
This measure of tie-strength classification is posited as the most accurate means of identifying romantic partnerships among users through network structure analysis. Previously, the principle of “embeddedness” has been a standard measure for assessing tie strength.
Embeddedness looks at the overlap of “social circles” between two users as a means of assessing tie strength. An analysis using embeddedness equates a high number of mutual friends with a strong user-user tie. One major critique of this logic is that large networks of mutual friendships tend to form within foci such as workplace or school. Membership in such environments may allow two people to form a high number of mutual friendships without necessarily having strong ties to one another.
We can picture how such a model might not align with the network formation surrounding a romantic relationship. In contrast, a highly dispersed group of mutual friends can reflect how a “couple” comes to share a diverse, unconnected group of peers, by means of meeting people through one another.
This being said, the accuracy of identifying a romantic partnership by sole means of searching a user’s network for patterns of dispersion is very low. An analysis combining dispersion, embeddedness, and other data, including profile pictures, tagged photos, profile views, and event attendance is obviously superior. It is interesting, though, to see how the nature of a relationship is reflected in its surrounding network.
I believe it is important to stay “informed” (as best we can) about how our online relationships are being processed as data. These processes, and the conclusions they draw about which users are “most important” to us, directly impact the content that appears in our newsfeeds, and the people with whom we are encouraged to connect. Even a cursory understanding of the measures by which websites such as Facebook go about curating our social lives can help us to re-evaluate our conduct, interactions, and sense of community within and without social media.
For those interested in reading the full study, see here.