Blog Closed

This blog has moved to Github. This page will not be updated and is not open for comments. Please go to the new site for updated content.

Monday, January 21, 2008

Amazon and Recommendations's recommendation feature is certainly an interesting one, and it can give pretty remarkably good results if you give it enough data. Of course providing enough data, and enough of the right kinds of data, can be difficult and time-consuming. Although, it would seem that giving it too much data can be problematic as well. Let me try to explain.

I don't know the exact algorithm that Amazon uses to make recommendations, although I assume that they are using data based on past user histories. For instance, if a user buys both book X and book Y, when I say that I like book X, the system then also recommends book Y to me. I may be nowhere near the correct explanation with this, and to be fair it doesn't matter at all.

For school, I tend to buy computer-related engineering books from Amazon. For my pleasure, I also buy books on computer programming languages, philosophy, and literature. For a long time, this is all I bought on Amazon, and these were the only kinds of books that Amazon recommended to me. I found a few gems through the recommendation process, but in general it was slim pickings because i was so narrowly defined inside the database. So, looking to get better recommendations, I began to expand my horizons. I listed some CDs and DVDs on my wishlist. I added some books about chess, and some other casual reading books. I added videogames too, since I'm still a player (although hardly so avid a player as I once was).

The recommendations did diversify, although the CD and DVD recommendations were not numerous nor particularly diverse. Eventually, I stopped getting recommendations for CDs and DVDs altogether. Maybe people just aren't buying those kinds of things on Amazon, so there isn't a lot of recommendations to be made. When I saw videogames that I owned, I would tell amazon that I already owned them, and so I would be recommended more videogames.

Eventually, Amazon got just the right information it needed, because it hit the motherload: it began recommending to me games that I already owned by the dozens. nearly every recommendation was spot-on, and in short order I had input more data about videogames then about any other type of media. The double-edged to this sword comes because now Amazon only recommends videogames, related memorabilia, accessories, and strategy guides. It can take several pages of searching before I find things like books in my list of recommendations, and I dont want to take the time to mark "not interested" on all the millions of games and accessories that the system wants me to look at.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.