What new things could we discover if social network analysis took time and space into account, in addition to the raw connections between people? In most cases, social network analysis today is limited to discovering friend connections, community leaders and outlines, influential people and personal friend recommendations - in a static or snap-shot kind of way. If new factors could be taken into consideration, specifically changes over time and space, then social network analysis could discover things like emergence or decay of leadership, changes in trust over time, migration and mobility within particular communities online. That's very valuable information that the social web has barely begun to tackle capturing.
That's the topic of discussion in a new paper by Shashi Shekhar and research assistant Dev Oliver, spatial data scientists at the University of Minnesota, titled Computational Modeling of Spatio-temporal Social Networks: A Time-Aggregated Graph Approach (PDF). The paper was highlighted on the blog GIS and Science today. We've excerpted and put in context key points below.
The Impact of Space and Time on Networking
Space and time are big factors in determining the diverse friend connections that different people form. Or, as Shekhar and Oliver put it, "Spatio-temporal constraints (e.g., geographic space, travel, schedules and diurnal [daily] cycles) play a major role in determining baseline homophily due to reasons like opportunity and minimization of cost and effort."
In other words, if we don't pay attention to the way space and time factor into our lives, if we just see who knows whom and who chatters about what, then we'll have a very blunt understanding of the world. Diurnal cycle in, and diurnal cycle out (!) businesses and software users seem likely to call for more clarity than that in the future.
Consider the diagram below, for example, from Shekhar and Oliver's paper. It might look a little intimidating, but if you follow it step by step from left to right, it's not. Today we might look at a group of four participants in some network and just see the final timeframe of connections (t10). You can see who knows who and who doesn't know who. But imagine if time were taken into consideration as it is here. You can see how this particular mini-network unfolded from the first connections, through the end-point. That's a much richer understanding of this group.
Look at poor little Node 3, for example, in the bottom of the square. It took them longer to go from Visitor to Friend than it took anyone else, in each of the relationships Node 3 formed. Node 2 at the top, on the other hand, looks friendly and effective at building strong relationships quickly.
Social network analysis services see these differences in the way people interact already and they see the way changes in spacial relations impact them, but it's a very nascent field.
"We see the spatio-temporal effect manifest in Twitalyzer data during conferences, tradeshows, and live events (e.g., SXSW)," says Eric Peterson, creator of professional Twitter analysis service Twitalyzer.
"The results are pretty obvious in our data: individuals who exhibit an otherwise 'normal' level of Impact, Influence, and Engagement in our data go 'off the charts.'
"The simplest explanation is that there is a compressing effect on an individual's network when they are more spatially proximate (e.g., can grab a drink and interact face to face.) When this happens, especially when it happens over a short period of time, people's scores change, their networks expand (typically, although we have seen contraction), and their 'chatter' (in your words) becomes more focused."
As Yet Unanswered Time-based Questions of Value for Any Community
- How is trust or leadership changing over time?
- Who are the emerging leaders in a group?
- What are the recurring changes in a group?
- How long is the tenure of a leader in a group?
- How long does it take to elevate the level of trust such as a relationship changing from visitor to friend?
Now imagine that kind of temporal and spacial analysis being performed on much larger networks, over greater periods of time.
"This added dimension or set of data points is out there and generally widely available as 'exhaust data', so to harness it and factor it in with the rest of the social graph would be truly valuable," says Eileen Burbidge, a London angel investor at White Bear Yard and a former product manager at Skype, among other positions.
"From an investment (i.e. value creation) point of view, spatio-temporal data has the potential to add an element of value, context and relevance to otherwise 'flat' data points.
"Take LinkedIn for example. I use it to see who in my network knows (and might endorse) whom, but I'm often cross-referencing/checking a person's contacts by their work history to discern if a specific contact was established at one spatio-temporal point vs another (ie relevancy)... The ability to build this into social network analysis would be extremely valuable as space and time offer tremendous context and relevance to social connections and relationships. "
The Hard Parts
That could very well be the kind of sophisticated social network analysis that service providers aspire to in the future and that their customers seek. Identifying some basic opportunities doesn't mean it will be easy to get there, however.
Shekhar and Oliver say this points to the need for "a central role for computation and computational models, not only to scale up to the large and growing data volumes, but also to address new spatiotemporal social questions related to change, trends, duration, mobility, and travel.
"The need for computational efficiency conflicts with the requirement for expressive power of the model and balancing these two conflicting goals is challenging."
If each historical moment of our relative connective history becomes no longer exhaust data, but points on a chart, that sure is going to be a challenge in the computational efficiency department! Presumably, it will just be timestamped changes of state that will be preserved.
Scaling that analysis is one challenge, finding new ways to recognize, tell and leverage the stories unearthed by analysis of that data is a whole other challenge.
Shekhar and Oliver cast themselves into the abyss of that data-centric future with their conclusion: "We welcome collaboration towards identifying datasets and use-cases to evaluate the potential of TAG [the time aggregated graph model] to address spatio-temporal questions about social networks."
Good luck guys, may you help open up this whole new frontier to us all. You've certainly articulated something that a whole lot of people are going to be very interested in in the future.
Right: A slice of Veronica Belmont's closest Twitter buds, per Mailana, a system without reference to time. From The Inner Circles of 10 Geek Heroes.
Title photo: Blinded by the Light, by Flickr user Jule_Berlin