Content-based recommendation
Dietmar Jannach, Markus Zanker, Alexander Felfernig, GERHARD E. FRIEDRICH · Cambridge University Press eBooks · 2010
From our discussion so far we see that for applying collaborative filtering techniques, except for t he user ratings, nothing has to be known about the items to be recommended. The main advantage of this is, of course, that the costly task of providing detailed and up-to-date item descriptions to the system is avoided. The other side of the coin, however, is that with a pure collaborative filtering approach, a very intuitive way of selecting recommendable products based on their characteristics and the specific preferences of a user is not possible: in the real world, it would be straightforward to recommend the new Harry Potter book to Alice, if we know that (a) this book is a fantasy novel and (b) Alice has always liked fantasy novels. An electronic recommender system can accomplish this task only if two pieces of information are available: a description of the item characteristics and a user profile that somehow describes the (past) interests of a user, maybe in terms of preferred item characteristics. The recommendation task then consists of determining the items that match the user's preferences best. This process is commonly called content-based recommendation . Although such an approach must rely on additional information about items and user preferences, it does not require the existence of a large user community or a rating history – that is, recommendation lists can be generated even if there is only one single user.