denote the degrees of satisfaction of the rule predicates di1 is A1, etc.,
and aggregation functions are used to replace the AND and OR operations. For
instance, a user whose profile indicates a preference for comedies and action films
might have a recommendation rule “IF the film is a comedy OR an action THEN rec-
ommend it.” 3 Each genre can be represented as a fuzzy set with fuzzy connectives
used to aggregate the degrees of satisfaction. The OR- and AND-type behavior are
usually modeled by disjunctive and conjunctive aggregation functions respectively.
In recommender systems, it has been shown that the property of noble reinforce-
ment is desirable [44, 9]. This property allows many strong justifications to result
in a very strong recommendation, or a number of weak justifications to reduce the
recommendation if desired.
Functions that model (22.5) can be used to match items to profiles or queries in
CB, UB and KB. In some demographic RS, items will be generically recommended
to everyone in a given class, making the classification process the primary task of the
RS. It may be desirable to classify users by the degree to which they satisfy a number
of stereotypes, and in turn describe items in terms of their interest to each of these.
For instance, a personal loan with an interest-free period could be very attractive to
graduating students and somewhat attractive to new mothers, but of no interest to
someone recently married. A user could partially satisfy each of these archetypes,
requiring the system to aggregate the interest values in each demographic. This leads
to rules similar to (22.5). “IF the item is interesting to students OR interesting to
mothers THEN it will be interesting to user u” or “IF user u is unmarried AND
either a student OR mother, THEN recommend the item”.