Amazon Context Specific Recommendations
By Pete Freitag
You may have noticed in the last few weeks I have had something at the bottom of my posts called Related Books Beta. Let me explain how I did that.
I play around a lot with Amazon web services (have you seen my site to find amazon deals?). A few weeks ago I noticed a feature called TextStream where you can pass in a paragraph of text and it is supposed to return a list of related products (books, cd's, dvds, etc).
I thought that was pretty cool because I could pass it my blog entries and get a list of related books. Kind of like Google AdSense for Amazon. What I found however was that the technology does not work very well.
I quickly found that the more text I passed to it the less accurate my results became. So I just pass the title of my post to Amazon, and the results are then about 20-50% on topic.
The problem is that there is no filter for common words it seams. If I post something about Mac OS X my first results from Amazon will be about Malcom X, and X-Men. The reason for this is probably that Malcom X books sell better than Mac OS X books - but most people reading my post about Mac OS X, probably don't have an interest in Malcom X. I should point out that I'm using Amazon's default sort, I have tried sorting by relevancy but that did not improve results.
If you want to play with this technology I have setup a TextStream test web page (no longer active) that you can type paragraphs into and see the results. You will see that for most paragraphs you have no idea how they came up with the results.
Amazon can do much better here. Google AdSense (though I feel AdSense could do better too, thats for another post) does a pretty good job at allowing bloggers to make money off their content. The reason AdSense works so well is that I don't have to do any extra work to get context specific ads, I don't have to feed it keywords or tags (though it would be kind of cool if you could as well).
So what do you think is the 20-50% accuracy worth keeping on my site? or should I remove it?
I can't imagine what books it will recommend for this post!
Amazon Context Specific Recommendations was first published on March 10, 2005.