Collaborative filtering is like amazon's "people who bought xxx also bought yyy" but it gets better: "people who bought w x and y (as you did) also bought z" And that's great, unless you were just shopping for your grandmother...
Actually collaborative filtering has a lot of thought and history behind it. I can see a benefit to looking at reviews of an item as meta-data. Then it could be related to how well you trust the person who published the review.
no more blockbusters
more links
An interesting idea would be to use the context in which others are looking at the page you are currently looking at to inform your context... Kind of like "people who are looking at the page you are looking at often came here from x and went to y"... maybe not too useful. But maybe "people who looked at the last 10 pages you looked at also looked at: z"
some inspiration for the above
That seems like it would have a bunch of noise input. i.e. pages you visited recently that were irrelevant to your current purpose. The answer to cutting out this noise often takes the for of more user involvement. Yet less user involvement seems to be the key for good collaborative filtering.
Case in point: you can educate amazon's recommendations, but who would bother? Either it gives you good recommendations based on your past purchases or it isn't much good.
As long as I'm rambling, aren't paths (in real life) collaborative filtering? You're following cues left by other human beings about which way to go. I've even heard of planners who build the buildings and plant lawns around them but create no sidewalks. Where the lawn gets worn away, pathways are later paved.
bring that idea online
Dated: 01/09/2003