4 ms·
An algorithm would have to be really complicated to discover the fact that just because so many people are interested in New Moon, or Tiger Woods, that doesn't
by jgrahamc 17y ago
An algorithm would have to be really complicated to discover the fact that just because so many people are interested in New Moon, or Tiger Woods, that doesn't mean I am.
A Naive Bayesian text classifier could easily learn that. If you fed it news stories that you are interested in it would quickly discover that you never click on stories about Tiger Woods.
It would then classify incoming news for you.
In fact, my old email project POPFile has been adapted to support things like RSS and NNTP filtering with ease.
- NathanKP 17y agoThat would work to eliminate "noise" info but such algorithms still do a poor job of helping one find new content and subjects that you have never expressed interest in before. I still trust groups of real people to help me in this regard more than I would trust an aggregator.
- derefr 17y agoTheoretically, you could have a (lazily-evaluated) subscription to the entire universe of discourse. It would start out completely hidden, but would gradually become visible in your reader as your filters trim down the stream to a tractable level. Thus, you wouldn't so much be saying "tell me something I don't know," but rather "tell me everything, and leave out the 99.9999% I don't care about."