Dynamic Network Representation Based on Latent Factorization of Tensors

A dynamic network is a complex system composed of nodes and links, where interactions can be quantified as weights. Due to the high dimensionality and incompleteness of these networks, studying how to build efficient and effective representation learning models is essential for acquiring useful knowledge. This book proposes four representative LFT-based network representation methods to address this challenge, including methods that integrate biases, consider non-negativity, and utilize an alternating direction method of multipliers framework.

Format: Paperback / softback
Length: 80 pages
Publication date: 08 March 2023
Publisher: Springer Verlag, Singapore

A dynamic network is a prevalent occurrence in numerous real-world industrial applications, including the Internet of Things. It comprises a multitude of nodes and extensive, real-time interactions among them, where each node represents a specific entity, each directed link signifies a contemporaneous interaction, and the intensity of an interaction can be quantified by the weight of a link. As the number of involved nodes significantly grows, it becomes impractical to observe their complete interactions at every time slot, resulting in a high-dimensional and incomplete (HDI) dynamic network. Despite its HDI nature, an HDI dynamic network with directed and weighted links holds abundant knowledge about the behavior patterns of the involved nodes. Consequently, it is crucial to explore the development of efficient and effective representation learning models to acquire valuable insights.

In this book, we initially transform a dynamic network into an HDI tensor and introduce the fundamental latent factorization of tensors (LFT) model. Subsequently, we propose four representative LFT-based network representation methods. The first method amalgamates the short-time bias, long-time bias, and preprocessing bias to accurately represent the volatility of network data. The second method employs a proportion-al-integral-derivative controller to construct an adjusted instance error to achieve a higher convergence rate. The third method accounts for the non-negativity of fluctuating network data by enforcing latent features to be non-negative and incorporating the extended linear bias. The fourth method adopts an alternating direction method of multipliers framework to construct a learning model for implementing representation to dynamic networks with high precision and efficiency.

Weight: 152g
Dimension: 235 x 155 (mm)
ISBN-13: 9789811989339
Edition number: 1st ed. 2023