The more items a user is associated to, the more chance that the user overlaps in several items with other users. This is necessary to compute any similarity with other users. And then, since any items *those* users are associated too are eligible for recommendation, it increases the number of probably quality of recommendations.
So, if I have no associations to any items, nothing can be recommended. Even if I have a couple items, if nobody else has any association to them, there can be no recommendations either. This is a general boot-strapping problem. You can't really make recommendations until you know a few things about the user. I would consider encouraging users to express more associations, and, have some default set of recommendations ready for the case where nothing can be recommended. On Mon, Jul 20, 2009 at 2:50 PM, Laya Patwa<[email protected]> wrote: > Hello! > I am using TanimotoCoefficientSimilarity to generate user based > recommendations. It gives good results(recommendations) for users having > considerable sized profiles but no recommendations or very less > recommendations for users with very short profiles. > > I want to ask why the size of profile matter? Because I guess that user > based recommendations are generated from the profiles of similar users and > if there is a list of similar users then the asked number of recommendations > should be generated irrespective of the size of the profile. > > Cheers, > Laya >
