People Who Read This Article Also Read... | Hacker News Reader

People Who Read This Article Also Read...

spectrum.ieee.org

8 pointsslackerIII9 comments

I wonder how much shared accounts complicate recommendation engines like these. My wife and I both use the same account for netflix and for amazon purchases. I can't imagine that an outsider looking at our movies and purchases could make much sense of it.

What other people read/buy/like alone is near useless data. Garbage in, garbage out.

I can't believe people are still trying to wave their hands and pretend they have some amazing algorithm to turn this data into half useful recommendations.

Well it works pretty well for Amazon.

And since I discovered lats.fm I have never listened music elsewhere, since their recommendations hit my (somewhat obscure) taste in music perfectly.

There's definitely data to be mined, and meaningful results to be extracted. The problem is that the people doing the algorithms are not good enough at what they do. Admittedly it is not an easy problem.

Indeed, it's the discovery problem that all of social media is trying to solve. The web it good at finding something you're looking for, not so great at finding stuff you would like if you knew it existed. Lots of bucks to the people that figure it out (last.fm'a acquisition is a good example of that)

Admittedly I haven't had much experience in this field, but it doesn't seem all that hard. What am I missing? Is it one of those "the devil is in the details" things?

Take books for example. I read lots of books, from all different genres (computer, fiction, history, economics, business, math, etc). How do I find books to read? If I look on Amazon, it suggests books similar to books I've already read, but that doesn't help me find books in different categories. Recommendations on Hacker News will give me a well rounded set of computer and entrepreneur-related books, but what else? I hear about books in other fields when they become popular enough to get press on NPR or IT Conversations or something (like The World Without Us) but mostly it's just random things mentioned here or there. I actually use an Amazon Wishlist to keep track of books I hear mentioned, because it might be a year or two before I read it.

As an example of how you can do an excellent job but miss most of the picture, try the Pandora music service. You choose a song you like and they make you a radio station full of songs with similar characteristics to the one you picked. I tried "Fade to Black" by Metallica and it played other metal and rock songs, but it had no idea I liked classical, Hawaiian, reggae, old school hip hop, or anything else.

I haven't encountered any recommendation service that was both accurate and broad. The best way to get recommendations I've found is to know and interact with different kinds of people (the kind who wouldn't know each other). That helps fill in the holes.

Those are not "half useful." Perhaps a 1/4 of what they could be with better input data and a trivial recommendation algorithm.

Last.fm just seems like music ordered by plays and filtered by tags (often poorly-done tags). Am I missing something? If you like the really really popular stuff then it's alright, otherwise, no. Further, the plays are skewed by what was hot at the times when last.fm and audioscrobber got big publicity hits.

I'm working in this area so I'd be really interested on anyone's opinions on current music sites.

My taste in music is cafe del mar tracks, ibizahouse, goa, classical, jazz and deephouse. I don't think that's a particularly normal taste in music - but last.fm seems to locate great music for me anyway, and rarely misses. Might all be done with tags though.

Reply on news.ycombinator.com