Efficient usage of collective classification algorithms for collaborative decision making

Abstract

Collective classification algorithms with underlying network structure of related entities are a powerful modelling tool that can address collaborative decision making problems. The paper presents the usage of collective classification algorithms for classification problem in which unknown nodes are assigned with classes based on the classes of known nodes. In such problem the classification decision for particular node is inferred from collaborative knowledge of nodes with known classes and underlying network connections. The paper considers Iterative Classification (ICA) and Loopy Belief Propagation (LBP) algorithms applied in various network configurations for collaborative decision making. The experimental results revealed that greater number of output classes decreases classification accuracy and LBP outperforms ICA for dense network structures while it is worse for sparse networks. © 2013 Springer-Verlag Berlin Heidelberg.