A visit to the Tivologist
Another IT conversation thought provoker: this time it was listening to Noshir Contractor’s great talk at Mesh Forum. Mesh is a gathering that discusses the impact of social networks on the world and current developments within them, all very interesting and extremely relevant to the current ‘Web 2.0’ explosion.
Noshir made a comment that “Amazon thinks I’m pregnant, and my Tivo thinks I’m gay” based on the recommendation technology they employ on his book buying and TV watching. Obviously he meant this comment because he isn’t gay and definately isn’t pregnant, but then a thought struck me; what if his Tivo knows him better than he knows himself? I mean, Tivo has watching habits collected across thousands of people and surely some of them at least are gay. This could mean this if, on average, Dr Contractor’s watching habits match those of the gay Tivo watchers then, just perhaps, he may be gay at some level? Either that or he should cut down on watching those house renovation programs.
Obviously Dr Contractor’s comments were based on Tivo recommending shows that he perceived as ‘gay’, hence his comment. Tivo has no form for entering your sexual preference as far as I’m aware. However just imagine if it did. All this interesting social behavior, at least on TV watching habits, would be collected and you could pay Tivo for the value added service of analysing your choice of recordings and presenting a report on your deepest, darkest psyche.
The strangeness of some Amazon recommendations was something that I’d laughed at years ago whilst running the customer relationship practice at Quidnunc. One of our sales guys, who shall remain anonymous, was asking exactly what recommendation engines were – so we pulled up Amazon on his computer and pointed to the section ‘Amazon Recommends for you’. Apparently he should have been buying ‘How to please your man: sex tips for gay men from a straight women’… after a bit of red faced blustering this transpired to be due to a book he’d bought “as a joke” for his girlfriend. We all laughed at that as he flipped his hair back, left the room and went to the gym – again. In this case, Amazon had thought he was gay rather than Tivo (which didn’t exist back then) – all of which was an issue with Amazon making assumptions based on implicit actions rather than explicit actions – ie, Amazon assumed he bought that book for himself, rather than someone else.
The fad of recommendation engines seems to be making a bit of a resurgence at the moment, for example last.fm recommends ‘bands you will like’ purely on entering a single band. Pointless. Average recommendations are just wrong (on average). Even with more detailed source information recommendation engines don’t take into account why you like something – the emotional context in which your preference exists, which runs from where you first heard it, to who told you to try it through to having a great guitar sound. Where web 2.0 kicks in is with social groups that share recommendations – people you like and bond with suggest things you might like based on their own preferences. Sounds like the real world to me.