I have a friend, (hard to believe I know, but it’s true) who is the only person I have ever met that can accurately predict whether or not I will like a particular band, album or piece of music. He sends me links and information on bands I’ve never heard of and would likely never discover and makes accurate predictions on the extent to which I will like them even though he knows I’ve never heard of them, never mind listened to them.
It’s a great service for me and saves me no end of time (but equally costs me a lot of money, because when he’s right, he’s absolutely right and I end up buying the entire back catalogue) but what’s really interesting about this, is that it represents a principle of a very different type of social search to the one we’ve been discussing recently around the aggregation of “sentiment” (Facebook “likes” in our case) around a given topic.
The importance of this new approach is that it deals with the assumption that “your friends are like you” that is implicit in the current method of introducing the social signal to search. Although at a high level, this assumption may be broadly right, at a more granular level it’s often completely wrong. My social network is made up of friends, family, colleagues (old and new) and a few other random acquaintances – to make the assumption that all these people are “like” me is, generically, probably true, but at a more specific level it is hideously wrong – for example, one of my brothers supports West Ham United and listens to the Smiths. Not following WHU is probably self-explanatory, but like Mitch, I must confess I never went through a Smiths phase. But I digress, my point is that just because someone in my network likes these things, does that mean I do? Of course not, generically you might infer that our connection may imply I like football and 80’s indie music but to be explicit about it would just be silly.
A new piece of research from our friends in MSR Cambridge is focusing on this principle, using a technique called “prediction extraction” to solicit opinions from friends as to whether they think the individual in question would like the item in question before they have even experienced it.
This approach is based on the observation that “although your friends are not you and may not have the same tastes as you, they are likely the people who actually know you best”.
You can read the detail of the approach here but essentially it offers a number of advantages, primarily around the accuracy, quality and coverage that the harnessing of this tacit knowledge brings, the real trick however is how to extract this information in a way that is easy and rewarding for the contributor and seamless for the consumer.
Predicting your friends opinions is nothing new (Mum _always_ knew best, right?, and it augments rather than replaces the “wisdom of the crowds”, but it does offer a new way of providing accurate, insightful predictions around the relevancy of a given topic or item to the individual. Going forward we’re going to need a range of these techniques if we are to truly humanise the way search provides us the answers we’re looking for.
My friend and music sensei doesn’t like the same music as me (can you believe he only has _one_ Men They Couldn’t Hang album?) but he does know me well, and he loves music – this combination alone could save HMV’s fortunes (and likely bankrupt me!).