06 October 2008

Machine-optimised 'explore'

each time we rate
any item of content
our software ought to adjust
its predictions about
other content's ratings

unseen content
by the same author
should be rated similarly

other raters
who rated that content similarly
should be rated similar
(ditto for un-similar)

unseen content
rated highly by
these similar raters
should be 'rated up'

every moment is a choicepoint
(which next?)
(which to queue for later?)

nextbestunseen by sameauthor

bestunseen by similarraters

bestunseen overall

newifany by knownfaveauthors

newifany by knownfaveraters

faves of newauthor/rater

subscribe to authorstream/raterfaves