PROJECT TITLE :

Heuristic 3D Interactive Walks for Multilayer Network Embedding

ABSTRACT:

Network embedding has seen widespread adoption as a solution to the challenge of network analytics. The majority of currently available methods concentrate on networks that only have a single layer of either homogeneous or heterogeneous networks. Multilayer networks, which are another term for heterogeneous networks with multiple edge/relation types, are able to naturally represent a great number of real-world complex systems. This is the case for many of the systems. A significant obstacle that arises when trying to embed a multilayer network is how to effectively capture and make use of the abundant interaction information that pertains to multiple types of relations. In order to solve this issue, we have proposed a multilayer network embedding model that is both quick and scalable. This model is given the name HMNE, and it is designed to efficiently learn and store information regarding multiple types of relations within a single embedding space. We develop a heuristic approach to a three-dimensional interactive walk specifically for multilayer networks. This approach is able to capitalize on the rich interactions that exist between the various layers of the network and successfully record important information that is contained within the layered structure. Two downstream analytic applications—node classification and link prediction—serve as the basis for our assessment of our proposed model, HMNE. The results of experiments conducted on seven different multilayer social and biological network datasets indicate that the proposed model outperforms existing competitive baselines while simultaneously reducing the amount of time and memory it occupies.