Efficient label propagation for classification on information networks

Abstract

Classification on networked data plays an important role in many problems such as web page categorization, classification of bibliographic information network, etc... Most classification algorithms on information networks work by iteratively propagating information through network graphs. One important issue concerning iterative classifiers is that false inferences made at some point in iteration might propagate further causing an "avalanche". To address this problem, we propose an efficient label propagation learning algorithm based on the graph-based regularization framework with adjusting network structure iteratively to improve the accuracy of classification algorithm for noisy data. We show empirically that this adjusting network structure improves significantly the performance of the algorithm for web page classification. In particular, we demonstrate that the proposed algorithm achieves good classification accuracy even for relatively large overlap across the classes.