5 ms·

“We think you’d also like…” and the Math of Suggestion - Part 2

28 points by wheels 18y ago

wheels 18y ago
This is the continuation from here: http://news.ycombinator.com/item?id=548584 http://news.ycombinator.com/item?id=548584 http://www.gruenderszene.de/it/we-think-youd-also-like-and-the-math-of-suggestion-–-teil-1/ http://www.gruenderszene.de/it/we-think-youd-also-like-and-t... As noted in the last round, this is a pretty light introduction to recommendations targeted mostly at only moderately technical folks (the audience of the blog that published it).
sgrove 18y ago
It's a well written, if short, piece. I've never thought of recommendation systems as fun, but I feel like trying out some basic stuff with it now. Anyone know a good way to get started - especially simple test data?
- wheels 18y ago
  If you just want data sets drop me a line, though most of the ones we test with are probably prohibitively large for just playing around with (millions of links / ratings). O'Reilly's Programming Collective Intelligence gives some decent background information, but in my opinion doesn't really cover enough to build a real system. A couple papers I often suggest with a lot of practical content are: Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions http://portal.acm.org/citation.cfm?id=1070751 http://portal.acm.org/citation.cfm?id=1070751 That one's fairly readable for people outside of the field. The Google News paper, which has some insights on doing large scale recommendations on a fairly dense user to item matrix, is a little more jumping into the deep end, but is worth glancing at even just to follow the references it sites: Google News Personalization: Scalable Online Collaborative Filtering http://www2007.org/papers/paper570.pdf http://www2007.org/papers/paper570.pdf (The paper itself doesn't mention being problematic on sparse rating sets, but I've implemented something very similar and found that to be the case.)

28 points by wheels 18y ago

This is the continuation from here: http://news.ycombinator.com/item?id=548584 http://news.ycombinator.com/item?id=548584 http://www.gruenderszene.de/it/we-think-youd-also-like-and-the-math-of-suggestion-–-teil-1/ http://www.gruenderszene.de/it/we-think-youd-also-like-and-t... As noted in the last round, this is a pretty light introduction to recommendations targeted mostly at only moderately technical folks (the audience of the blog that published it).

sgrove 18y ago

It's a well written, if short, piece. I've never thought of recommendation systems as fun, but I feel like trying out some basic stuff with it now. Anyone know a good way to get started - especially simple test data?

wheels 18y ago
If you just want data sets drop me a line, though most of the ones we test with are probably prohibitively large for just playing around with (millions of links / ratings). O'Reilly's Programming Collective Intelligence gives some decent background information, but in my opinion doesn't really cover enough to build a real system. A couple papers I often suggest with a lot of practical content are: Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions http://portal.acm.org/citation.cfm?id=1070751 http://portal.acm.org/citation.cfm?id=1070751 That one's fairly readable for people outside of the field. The Google News paper, which has some insights on doing large scale recommendations on a fairly dense user to item matrix, is a little more jumping into the deep end, but is worth glancing at even just to follow the references it sites: Google News Personalization: Scalable Online Collaborative Filtering http://www2007.org/papers/paper570.pdf http://www2007.org/papers/paper570.pdf (The paper itself doesn't mention being problematic on sparse rating sets, but I've implemented something very similar and found that to be the case.)