3 ms·

Thanks for the great feedback! An actual recommendation engine would be a lot of fun to build. It would be awesome to recommend products based on subreddit clu

Thanks for the great feedback! An actual recommendation engine would be a lot of fun to build. It would be awesome to recommend products based on subreddit clusters, and similarities between users. But I don't really want to build a giant AI that sifts through the entire Amazon catalog. I feel that it's a nicer user experience when you just offer a small selection of manually curated products, and some studies show that limiting the number of choices can make people more likely to buy something [1]. I used to list a few CDs and MP3 downloads, but deleted those after I decided that there are already much better music recommendation engines, and they're subscription based. My line of thinking is that I should just start doing a few categories really well, and then expand later. [1]: http://blog.ted.com/2012/07/18/does-having-choice-make-us-happy-6-studies-that-suggest-it-doesnt-always/ http://blog.ted.com/2012/07/18/does-having-choice-make-us-ha...

OafTobark 13y ago
You're arguing apples and oranges here. Yes, offering a smaller selection can increase conversion but that has nothing to do with how the curation process works on the backend to present the selection. Whether you do it by your method or by going full on AI, you can still present a limited selection. The only difference is how algorithmic the recommendations can be on an individual level
- nathan_f77 13y ago
  That's a good point. I guess my adverse reaction to building a proper recommendation engine is because I've never experienced one that works really well. I've never seen a decent product suggestion from Amazon and Google. But maybe the problem is that their suggestions are based on my search history, instead of my specific interests.