Abstract
In this paper, we propose a generalized collaborative filtering algorithm. Recent web-based technology has been
quickly developed toward the user-friendly environment and one
of the important technical challenges of personalization is a
recommendation to end users. Collaborative Filtering is one of
the techniques of information-filtering using user's profiles and
is applicable to recommendation engines. Users' profiles are, in
many conventional studies, expressed as vectors. With vector-
based profiling, recommendation algorithms can be mathematically clear and easy to use. However, there are some cases
which vector-based profiling is inadequate to express users.
Thus, we generalize user's profiles to function-based ones and
extend conventional collaborative filtering algorithms for the
usage of our new formats of the profiles. The proposed algorithm
employs a general regression neural network (GRNN) to compose
function-based profiles. To verify the algorithm, we apply the
proposed algorithm to a web color preference estimation.
