Heuristics for Social Network Analysis

点击翻译按钮获取中文摘要

Social networks have been used to model a comprehensive range of real-life phenomena. Social networks analysis is the study of such networks to discover common structural patterns and explains their emergence through computational models of network formation. The complexity and dynamics are essential properties of such networks. Since these networks evolve quickly over time through the appearance or disappearance of new links and nodes, the connections become stronger and weaker with underlying network structure changing over time. Researchers are embarked on examining various issues in social network analysis such as classification of nodes, detection of communities, formation of teams, and prediction of links between nodes. Ubiquity and growing popularity of social networks present the challenging task to analyze the massively increasing datasets of social interactions in an efficient manner. Vast amounts of knowledge can be extracted and presented into the hands of domain experts to help make important decisions. However, without timely and accurate predictions, opportunities are missed. This research explores Deep Learning methods including hybrid approaches as well as population-based evolution algorithms for pioneering efficient community detection and link prediction frameworks with increased accuracy in large, complex, and dynamic social networks. For instance, by studying common neighbors based subgraphs of a target link and using feature matrices for learning the transitional pattern for a given dynamic network, we effectively transform the dynamic link prediction to a video classification problem and enable the use of Convolutional Neural Networks . On another front, classification of nodes induces the formation of clusters. The prediction of links brings about correlations and/or ties formation. By studying edge sampling techniques, and examining an architecture with distinct representation learning layers, we exploit feature-extraction and dimensionality-reduction to enhance automatic extraction of facts from the network structure. By partnering with domain experts from other disciplines we can effectively develop these frameworks into decision support models of real-world value and validate them in their respective domain. This research pioneers link prediction and community detection methods for social network analysis by stepping up innovation, seeking efficiency and accuracy, while handling dynamic and fast increasing network data. The program further aims to tackle team formation problems and health networks by capitalizing on new innovations in population based evolutionary heuristics, as well as deep learning approaches. Practical outcomes include the adaptation of novel algorithms from benchmark tests and embedding them into agent-based models to serve as useful tools for decision support that can be used by domain experts from other disciplines such as healthcare leaders.