There are two key theories that seem to combine for massive social network analysis: Gestalt theory and graph analysis. Gestalt theory has been “in development” since the 1930’s and is one psychological theory addressing how social networks behave. Graph theory is one non-neural net way of rendering flow patterns. Empath implemented a graph where the nodes were web site interaction events and the arcs were the actions used to reach the next event. Overlapping the two, Gestalt and graph theory, I had shown at a major OEM that it was possible to model user behavior so that it was relatively predictable within the scope of a user’s intended interaction across the various interaction mediums available. The business impact when these two could be coupled, was 2-5 fold beneficial to the user and to the pilot business case and a direct result of users having inquiries answered without the need for follow-on contacts. Gestalt theory comes in to reassure us that behavioral patterns are predictable.
So what does all this have to do with anything? Turns out that the massive volume of contacts through internal social media sites like communities or forums, as well as Facebook, Twitter, etc. efficiently feed a graph-forming algorithm. In my experiments in 2002-3, the whole of 50M contacts within a consumer web site, support site, email and phone could be modeled in a graph data structure that could be then browsed with a Pentium 4 class laptop. The whole graph with metadata, one year’s worth of contacts over email, phone and web, was just a few megabytes. I tried to make the graph data structure visual for back-office analysis by contact management professionals, but the visualization was difficult and the users were not gamers. I eventually ran out of funding.
What I realized in the process, is that visualization of the graph itself was out of reach. What I should have recognized is that graphs have been the mainstream of phone company real-time routing protocols: which switch network is best to complete the call on-hand given the cost, load and profit weights assigned to each arc of the network graph.
If I could rewind, I would have assigned weights to the contact-medium graph, namely the contact graph I had built, so that contacts would get routed to the best opportunity for a successful answer. This is skill-based routing. This is what major voice telephony companies and others have been trying to do for a long time.
I firmly believe there is a way for SaaS providers to achieve skill-based routing, with interfaces to a number of contact mediums, both incoming and outgoing, so that a query, topic, inquiry, etc. reaches the best-available individual or destination site who can address the topic of interest.
This project is currently pending with the USPO.