Re:The Gentleman's Agreement
by sadlyNess on Thursday March 15, 2012 @08:09AM
(#3708006)
What we need is algorithms that can recommend like a human who knows you would. An rapid publish/editing filter engine based on my previous interactions...
[ Parent
| on HN ]
With the new project I am working on we're attempting to create a content filtering algorithim that can determine the content a user likes based off of previous interactions of upvoting and hiding content from their feed. I think with enough interactions and utilizing a user grouping system, the majority of irrelevant content could be filtered. Assuming the community is large enough and provides an adequate amount of feedback.
How do you include "serendipity"?
I subscribe to "One Story" on Kindle. I read a short story by Caitlin Horrocks. I would never have picked her book up, but because of that short story I bought her book of shorts. It's one of the best books I've read in the past few years. I have no idea if it would have been picked up by your algorithm.
I do know that "Amazon recommends" has got much less useful.
(http://www.caitlinhorrocks.com/prose.htm)
(The story was “Life Among the Terranauts”.)
What did you use as training data?
We're still collecting our baseline data at the moment, but we have implemented the design features into our community that should create more interactions of upvoting and removing content the user finds relevant.
Re:The Gentleman's Agreement
by cheatercheater on Friday March 16, 2012 @03:52AM
(#3712406)
...a neural-net processor. A learning computer.