SysEvoRecomd: Network Reconstruction by Graph Evolution and Change Learning

Abstract

We introduce a System Evolution Recommender (SysEvoRecomd) algorithm that uses a novel algorithm Graph Evolution and Change Learning (GECL) to do system network reconstruction. Internally, GECL uses Deep Evolution Learner (DEL) to learn about evolution and changes happened over a system state series. The DEL is an extension of the deep learning algorithm, which uses an Evolving Connection Matrix (ECM) representing temporal patterns of the evolving entity-connections for training incremental states. The DEL generates a Deep System Neural Network (Deep SysNN) to do network (graph) reconstruction. The SysEvoRecomd extracts the evolving characteristic of graph with deep neural network techniques. It aims to learn the evolution and changes of the system state series to reconstruct the system network. Our key idea is to design three variants of GECL based on three remodeled deep learning techniques: Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), and denoising Autoencoder (dA). Based on proposed SysEvoRecomd algorithm, we developed a SysEvoRecomd-Tool, which is applied on different evolving systems: software, natural language, multisport event, retail market, and IMDb movie genre. We demonstrated the usefulness of intelligent recommendations using three variants of GECL based on RBM, DBN, and dA. © 2007-2012 IEEE.