Social search and collaborative filtering

By August 9, 2006 January 6th, 2007 6 Comments


Social search and collaborative filtering are hot areas/buzzwords at the moment. 

So what are they?

Defining social search is tricky – this ZD net blog comes at it from a very techie angle and argues social search is difficult because putting social input into search algorithms is very hard.  I see things a little differently and would argue that social search is best defined as a complement to traditional search.  E.g. when I search for books on “Web2.0” traditional search algorithms do the first cut and are augmented by social aspects – e.b. people like you have recommended this.  So the search is made “social” by incorporating consciously given social input – e.g. reviews.

Collaborative filtering then becomes the anonymous counterpart to social search.  By mining the stats of what I have e.g. listened to and other people have listened to collaborative filters can augment traditional search with a ‘people who listen to what you listen to also listen to this’ filter.  Note in this case all that is required is listening history, or book purchase history – no names or other aspects of profile.

Social search and collaborative filtering will bump up against each other and the distinction here is given solely to aid understanding.

And where is the value? 

To my mind there are two things that people want: 

  1. To quickly find something when they are looking for it
  2. At other times to have things suggested to them that they might like

I would suggest that social search is more useful in 1. and collaborative filtering is more useful in 2.

Items people consciously search (and research) for are typically bigger ticket items – you don’t buy them very often and they last a long time, so you want to get it right.  New blogging software is the most recent example for me (thanks WordPress) and before that a lawnmower.  Reviews and recommcendations are very helpful, particularly from trusted sources.  To me there is huge value for the company that finds a way to productise this concept.  I’ve recently come across two cool companies Crowdstorm and ThisNext who are in the early stages of trying to do just that, although the space is getting competitive already – see this post on Mashable.  Local search companies are getting into this game to via recommendations in what is becoming known as networked search.

Collaborative filtering is more of a long tail play.  As per Chris Anderson supply and demand in the long tail need help to find each other and collaborative filtering is a leading technology for solving this problem and is at the heart of the success of a long tail companies like Amazon, Pandora, and lastfm.