Much of my time on the web is spent looking for new books, music, or movies. This is a slow and often fruitless process - I haven't found any recommendation services that make "good" recommendations so my usual method is a long, rambling search that results in one or two new things to try. This surprises me because recommendation algorithms are a somewhat popular (okay, well, not obscure) topic in Computer Science. There are a few systems out there but none of them seem to work really well.

For music, Audioscrobbler is the best I've found so far. I've not yet found a new artist that I like by digging in the Audioscrobbler data but at least its findings are useful. There are other things like U of Illinois' iTunes system, which includes songs by Flock of Seagulls, Hanson, and Dolly Parton in the list of 10 Most Recommended Tracks. I won't be trying that one.

Although IMDB is the definitive resource on movies, the content that is created by users (votes, reviews, message boards) is not very useful for a "try it if you like ______" type inquiries.

Amazon has a ton of useful recommendation information on books (and music and movies). I almost always make a new discovery when I devote some time to digging around Amazon. I find the "your recommendations" section pretty useless but the Listmania! lists, "customers also bought", "customers also viewed", and "similar items" screen are all very helpful in finding new books that I might like.

Before I start.. I don't know what Amazon's software looks like and I'm sure (I hope) that they've spent time working on the suggestion engine. My ideas below may already be part of the system or I may be missing something important. The problem is this - my Amazon recommendations are never good and I have to spend a lot of time searching for books to read next.

My main complaint with Amazon is this - too much "garbage" information is tossed into the mix that is used for creating suggestions and the good information is not given enough weight. My purchase history contains plenty of Christmas gifts, birthday gifts, and books that I later regretted buying. (There is a way to rate and enable/disable the items that drive your recommendations but it still doesn't seem to work properly). Wish lists, on the other hand, should be very useful and I'd like to see Amazon suggest items based on the wish lists or their users. I wonder how the current system would act if everything that drew on past purchases looked at past and current wishes instead. Bestselling items and Amazon's own financial interests also dirty things up a lot. *Nobody* needs to be recommended the next Harry Potter book or Bill Clinton's biography. Also, people with tastes that closely match mine sometime go out and buy the Da Vinci Code for themselves or for someone else. This kind of behavior should be noticed as "out of character" and treated accordingly. Listmania is great for finding things and lists that contain highly rated books, a variety of authors and a small proportion of big sellers should play a key role in the suggestion system. (Something to think about - I've seen quite a bit of Listmania spam from small authors and/or rabid fans. Does Amazon know that this is happening?) Reviews and ratings can also be very useful. The body of books that a reviewer has commented on can be just as valuable as a Listmania list and could be treated similarly.

Amazon has an excellent API and Amazon Web Services has been winning awards since it's inception. I have yet to see a useful application built on top of Amazon Web Services. I looked through the newest API docs and it looks like the services that they provide are complete enough (in a roundabout way) for someone to build a useful recommendation engine on top of Amazon. You can't see inside people's orders with the API, but I think that the purchase history is more of a problem than a useful resource. I wish that I had time to play with this...