| As
the market of electronic commerce grows explosively, it becomes
more and more important to provide the recommendation system which
suggests the pre- ferred items for consumers using the large-scale
customers database. In this paper, we discuss the algorithms and
their performances of the recommendation systems using the collaborative
filtering in the case of the Netflix database: they are, 1) memory-based
system (k-nearest neighbor using the correlation coefficients),
2) model-based system (matrix decomposition), and 3) the combination
method. When the customer-item matrix is a sparse matrix like the
Netflix database, the matrix decomposition method shows better
performance than the k-nearrest neighbor; in addition, it is found
that the combination method of the two methods provide a much better
performance.
|
|
|
|
|
|
Netflix, collaborative filtering, k-nearest
neighbor, matrix decomposition, singular-value decomposition,
combination method.
|
|
|