Knowledge Transformation by Cross-Domain Belief Propagation

Belief propagation is an iterative algorithm for computing marginals of functions on a graphical model most commonly used in information retrieval. In this paper, we consider the problem of performing cross-domain belief propagation on multi-relational data for semi-supervised learning. We demonstrate that partial knowledge on one type of variables can help knowledge discovery on the other type of variables with cross-domain belief propagation by utilizing the existing relationships in multi-relation data. For example, in a word-document data set, information on the word domain can effectively enhance the labeling of document domain. In this paper, we explore this new area, knowledge transformation of multi-relation data for semi-supervised learning tasks. We show that partial knowledge on one data variable domain can help knowledge discovery on the other variable domain with cross-domain belief propagation by utilizing the existing relationships in multi-relation data. The experimental results on several real world data sets are presented to show the effectiveness of our method.