Recommender systems for music and books haven't lived up to initial expectations. Netflix and Amazon give me pretty decent recommendations, but among other things they get confused between things I buy for myself and things I buy for other people DeLong excerpts Slee to suggest a host of other problems -- mostly fraud related.
There's so much money riding on recommender systems even clumsy fraud is common, so it's rather likely that subtle fraud is also common. It's the same problem Google has had, since Google's original search approach was a form of recommender system.
Useful recommender systems may first require a good reputation and identity management infrastructure, or be based on data points that cost money to create. Not coincidentally, when I'm researching Amazon the first thing I look at is the sales ranks for the product domain I'm interested in. Then, for each product that's selling well, I look at the negative reviews first. I don't pay that much attention to the star rankings or the positive reviews.
No comments:
Post a Comment