How appropriate is modeling individual’s online behavior in social networks? Imagine, you get access to Facebook data. Can you model the online behavior of the users to come up with suggestions to marketers?
I am skeptical. And here is why – there are too many unknown parameters about which we are talking. For example, I have more than 300 friends on Facebook and I group them in categories such as Family, School, Binghamton, Work etc. Consider the ‘Work’ category. I am close to some of them, but not so close that I categorize them as ‘Good Friends.’ Others in ‘Work’ category are not as close as others. So, within the ‘Work’ category I have plenty of variation when it comes to ‘closeness.’ I have several such categories. And this is just my way of doing it. You may be doing it some other way. I can easily imagine people using different schemes to categorize their friends. So, each person, in theory, can have a separate categorization scheme. Now, this is just for categorizing friends. There are several other rules I may follow such as commenting, visiting profiles, checking pictures, sending messages, playing games, sharing the information about the games, and so on.
I think there is too much information at the personal level but worse still, there are too many unknown parameters. Then there is enormous endogeneity, but we will talk about it later! Such modeling will then require putting too much structure by making restrictive assumptions, which would lose the usefulness of the models to marketers.
I had a great discussion with my friend Eduardo yesterday. He works with FICO and knows more modeling than I will ever know. Among several topics that we discussed, we also talked about the above problem. Guess what, he already knew that modeling social networks is futile!
If you have opinions countering my arguments above, please feel free to comment.