Collaborative ordinal regression

Ordinal regression has become an effective way of learning user preferences, but most of research only focuses on single regression problem. In this paper we introduce collaborative ordinal regression, where multiple ordinal regression tasks need to be handled simultaneously. Rather than modelling each task individually, we build a hierarchical Bayesian model and assign a common Gaussian Process (GP) prior to all individual latent functions. This very general model allows us to formally model the inter-dependencies between regression functions. We derive a very general learning scheme for this type of models, and in particular we evaluate two example models with collaborative effect. Empirical studies show that collaborative model outperforms the individual counterpart.

• Type: Conference
• Year: 2006
• Where: ICML
• Authors: Shipeng Yu, Kai Yu, Volker Tresp, Hans-Peter Kriegel
• Keywords: Collaborative Ordinal Regression, ICML 2006, Machine Learning, Ordinal Regression Tasks, Single Regression Problem