I was checking out the IMDB the other day and noticed their Bayesian weighted ranking system
weighted rank=(votes cast/(votes cast+minimum votes to be in top 50)) x mean rating + (minimum votes to be in top 50 /(votes cast+minimum votes to be in top 50)) x mean vote across the vote population
I think I've got this right! Thinking about recommendations engines, I then thought I'd go and sort out my personal Amazon recommendations (as most of my Amazon purchases have been corporate) by ranking items that I own and/or have read/viewed/used. I was pretty surprised at how (simply)comparative the engine appears to be: most of my recommendations so far are for authors I have read before and/or other editions of books I have read/CDs/DVDs and just a little bit of "others who purchased this..." Maybe I'm mis-calling Amazon on this one, as I know other people who say that Amazon second-guesses their purchases, but I would have thought ranking a couple of hundred items would give slightly more interesting recommendations. Oh well, I shall continue ranking and see if I'm more satisfied. Anyway, I seem to remember that there is a recommendations engine expert at Glasgow University (I think he's involved with the Information Retrieval Group) - if I can remember his name, I might send him an email to find out a bit more about the nuts and bolts of recommendations (hopefully it will be comprehensible to someone without undergrad maths).