Cascade Recommendation Methods

Abstract

In this chapter, we address the problem of recommendation by developing a two-level cascade classification architecture. The first-level classification step involves the incorporation of a one-class classifier which is trained exclusively on positive patterns. The one-class learning component of the first-level serves the purpose of recognizing instances from the class of desirable patterns as opposed to non-desirable patterns. On the other hand, the second-level classification step is based on a multi-class classifier, which is also trained exclusively on positive data. However, the second-level classifier is trained to discriminate among the various (sub-)classes from which positive patterns originate.