Abstract—Identifying the existence of relationship between the entities across multiple domains is fundamental to scientific discovery. The entities in different domains are usually described through different structured or semi-structured data. It is interesting to integrate structural affinity of entities in individual domains in relational learning because the structured data collected from various domains are often complementary to each other. However, this is a difficult task and has not yet been solved well. In this paper, we propose a novel approach to deal with this challenge that hampers relational learning across multiple domains. The key idea of our approach is to integrate multiple structural affinities through graph product operations that can map the structural affinities in individual domains onto a single integrated structural affinity. Furthermore, the proposed approach allows for emphasizing the function of the known relationships during integration. The problem of relationship discovery between entities is subsequently cast as a problem of node classification on the product graph. Since the integrated structural affinity captures higher order relationships of the entities from different domains, it is no surprise that we obtain more reliable results of relational learning. We validate the new approach on a real-world dataset. Experimental studies show that the proposed approach outperforms its competitors on the benchmark dataset.
Index Terms—Graph product, information fusion, relational learning, structured data.
All authors are with the School of Software Engineering, Huazhong University of Science and Technology, Wuhan, Hubei, China (e-mail: m201776139@hust.edu.com, m201776184@hust.edu.com, qiudehong@163.com).
Cite: Xuwen Lang, Pin Zhang, and Dehong Qiu, "Integrating Structural Affinity via Graph Product for Relational Learning," International Journal of Machine Learning and Computing vol. 12, no. 3, pp. 107-112, 2022.
Copyright © 2022 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).