I spent the weekend at BarCampNYC3, an unconference in the mode of the BrainJams I once attended. It was great to meet a bunch of new people, including some nextNY folks I had never met in person, and to get the chance to talk about interesting topics for a couple days.
One session I attended early in the weekend was led by Joe Fernandez, on how to measure influence in the social web. This started a great discussion, as we first had to agree what influence meant. The marketers in the room translated it into how much money can we make from this person’s recommendation? If Bob has 1,000 readers, but only 5 of them buy the product, and Alice has only 10 readers, but 7 of them buy the product, who is more influential? Bob’s got the bigger audience, but Alice has more influence, as measured by the dollars.
We also discussed social influence. What does it mean to be a thought leader? Somebody mentioned the Fast Company article on whether the idea of Influentials is valid. Somebody (not me!) brought up Clay Shirky’s new book. Rohit Khare mentioned his work on leveraging not just the social network, but also the documents as rated by that network (which makes sense when we realize that documents only have value when creating a connection). Lots of interesting ideas floating around, and Sanford Dickert suggested that we do another session to try to come up with a better definition.
On Sunday, Sanford led a session where we tried to derive an equation measuring influence. Sanford’s background is in robotics, so he was applying systems theory and feedback loops to the problem. We spent some time discussing what the equivalent concepts of inertia, friction, and dampening might be (we came up with the acceptance of the current worldview, the difficulty of forwarding a new idea/concept on, and the natural decay of interestingness of a new idea over time as possible analogues).
Sanford led a later session on “Web 3.0” where he tried to build on these ideas of influence, and what that would mean for designing social applications. One marketer in that session suggested that marketing was making potential customers want what you have. I thought that was too simplistic and Machiavellian, but it got me thinking.
I realized that this might be a good situation in which to apply actor-network theory as a framework for thinking about this problem. Actor-network theory is all about evanescent indirect connections between people that need to be re-established. It’s also about how every element in the network has an effect on every other element – all participants are “actors” in that they have an effect on the network. Objects that have no effect are not actors and can be removed from consideration. But there are rarely direct connections between network endpoints – all effects must be traced through mediators which can alter the message in surprising ways. Actor-network theory is about observing the network and tracing the connections between different actors and seeing the effects of mediators.
So I started mulling the idea of trying to trace the network between the product on one end and a customer on the other end (my notes from the session say, in contrast to the earlier claim, that “Marketing is building a connection between the product and the person”). The product has certain characteristics, the marketer advertises some of those characteristics, the newspaper reviewer might write about the product and its characteristics, a friend might read the reviewer and think highly of the product, and mention it to the eventual customer who happens to have a need for a product like that.
I wanted to write up these ideas this evening, so I went back to review my posts about actor-network theory from years past, and discovered that I had already written a post on applying actor-network theory to marketing. Clever of me, eh? Go read that post now.
One thing I don’t address in that post is how to create a mathematical model of influence. I was talking about it with Sanford later, and suggested that it’s a tricky problem because influence is such a personal thing. I may be influenced more by a famous person like Oprah or by my good friend. Also, a person’s influence is not invariant – I may trust my geek friend for a recommendation on which laptop to buy but not at all on where to eat. So the model would need coefficients of influence for each connection between nodes on each topic, with those coefficients varying depending on results.
I wonder if a neural network might be the way to model this sort of thing. Our brain can be modelled as a collection of neurons, each of which influence each other with certain coefficients that are strengthened or weakened based on how well their outputs contributed to desired outcomes. Perhaps our networks can be modelled in the same way. This would play into my idea of cognitive trust, where I suggest that once we trust other people enough, they’re just an extension of our own brain. I certainly have people like that, where I don’t even bother having opinions on certain topics like cuisine and fashion because I can always call my friends to get a more informed opinion. In some sense, my outsourcing of taste is the ultimate in influence.
I really need to find the time and energy to do some programming. It wouldn’t be that hard to create a toy model of an influence network built off a neural network model. And it would be interesting to see how that model corresponded with real world tastes. Maybe I should throw it at the Netflix data to see what happens. But that might have to wait for the summer when classes are over.
Technorati tag: barcampnyc3
I think it’s useful to think of influence as having an effect on a decision or a behavior. This makes it clearly not just a function, but a functional, but it also means you can focus on influence that relates to the behaviors/decisions you’re interested in, and just throw out all the rest.
I dunno that a neural network in the strict sense will get you anywhere interesting. They tend to be very black-boxy, and unless you restrict the topology in specific ways, it’s easy to get networks that just thrash about without doing anything interesting. But generalize it to behavioral agents that live in a network and talk to their neighbors and I think you’ve definitely got potential.
Yeah, it could be that a social atom approach is the way to go. Neural networks seemed reasonable, since they involve nodes connected to other nodes with weights and feedback. If I listen to your advice on a restaurant, and then don’t like that restaurant, then your weight gets reduced next time. That’s the same process by which a neural network works (project a result, compare to actual result, adjust weights accordingly).
Restricting it to a certain decision/behavior makes sense – although that kind of plays into a neural network model as well, since neurons have default levels of activation, and need more or less incoming signal to get pushed over their threshold. If I’m in the market to buy a laptop, I don’t need much of a push to do so, but if I’m not even looking, then no amount of pushing will convert me. Hrm.