Random walk strategies in information network representation learning

• Main Author: Zhou, Xuwen
• Other Authors: Lihui CHEN
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/140213
• Institution: Nanyang Technological University

The data, informational objects, components interact with each other, forming Information Network (IN). Most current research papers make an assumption that information networks are homogeneous, whose nodes and links are of the same types. However, most of the real-world networks are Heterogeneous Information Network (HIN), a graph containing different types of nodes and links. Representation learning transforms input data and produces an expected result, which is also able to reduce the dimension of IN data and preserve important information of individual object in the IN. There are various representation learning methods in the graph and the relationship between nodes and links. In these methods, a large number of random walks will be extracted and then representation learning algorithms are applied. As different random walk strategies will greatly affect the learned representations, we need to find one with the best approach. In this project, DeepWalk and Metapath2vec are conducted to gain graph representations. By applying dblp dataset, it is compared and analyzed with their properties and impacts to representation learning. This report states the methodology and implementation details used in the experiments, followed by discussion and analysis.