A matrix alignment approach for link prediction

This paper introduces a new discriminative learning technique for link prediction based on the matrix alignment approach. Our algorithm automatically determines the most predictive features of the link structure by aligning the adjacency matrix of a network with weighted similarity matrices computed from node attributes and neighborhood topological features. Experimental results on a variety of network data have demonstrated the effectiveness of this approach.

• Type: Conference
• Year: 2008
• Where: ICPR
• Authors: Jerry Scripps, Pang-Ning Tan, Feilong Chen, Abdol-Hossein Esfahanian
• Keywords: Computer Vision, Discriminative Learning Technique, ICPR 2008, Matrix Alignment Approach, Weighted Similarity Matrices