If you have a linear distinction like this in mind, then the best option is probably to just build a classifier from examples you have in mind.
On Thu, Feb 17, 2011 at 9:36 PM, Lance Norskog <[email protected]> wrote: > I would like to uncover the vector that has "chick flix" at one end > and Star Trek at the other. How would one attempt to cluster line > segments instead of points? > > Suppose we isolate a set of "I like this / I hate that" pairs. > Representing each item as a point, the line segment between the points > is what I want to cluster. A cluster would be a longer line segment, > all co-directional but not necessarily near each other. > > > > -- > Lance Norskog > [email protected] >
