Alternatives To Collaborative Filtering

Why can’t we build some interesting techniques other than collaborative filtering for recommendation systems that recommends users based on their unique style rather based on what other users like?

CF does the following:

A purchased chocolates.
B purchased chocolates and a whip.
A gets recommended a whip.

This doesn’t seem effective!

One idea would be to get another input from users when they make a purchase, perhaps the reason why they are making that particular purchase. Based on this reason other users with the similar reasons might get recommended a better product.

Jeremy gave an example of when buys all the Harry Potter books why would the system recommend buying a book he already has. It is just lazy code. Compare recommendation with bought items remove remove matches then make recommenations. There becomes a cost benefit analysis which says a recommendation cost 1 cent where as doing the work costs 2 c and the users incurrs the wasted cost of ignoring the recommendation.

Regards CF

1 Like