On Fri, Sep 20, 2002 at 08:46:27AM +0100, Simon Wistow wrote: >So, anybody got any smart ideas on how to do it. I wondered about Markov >Chains but they have a concept of order as far as I know. Am I >overengineering? Is there some really simple way of doing this?
Standard correlation coefficient will do it. Just regress the ratings for A against the ratings for (B, C, D...) and pick the ones with the highest correlation. See http://fonsg3.let.uva.nl/Service/Statistics/Correlation_coefficient.html for a formula if you need one. (You can even do an "if you liked this, you are likely to dislike..." as well. Funny how nobody ever does.) Roger