High Fidelity Synthesis of Dynamic Social Networks using Measurement-based Calibration

We propose to develop a systematic approach of synthetically generating diverse types of large-scale, dynamic and high fidelity social network data via measurement calibration. Social network research needs access to realistic social media datasets for testing social network theory and developing network algorithms. However, the available social network data are either small, static, or cannot be readily collected or shared due to anonymization, privacy, and complexity issues. We will design and implement scalable algorithms, methods and software tools to generate realistic social network data with respect to multi-dimensional network analytical functions and validate it with a comprehensive set of statistical, temporal, and topological metrics, and application-level benchmarks. Our key innovative solution is based on capturing dynamic (visible or latent/hidden) interactions, patterns/anomalies, information flows/cascades along with the coupled dynamics of network structure and data content (topics, sentiments, and memes) to synthesize social media. Our approach generates fully anonymous social network data that can be distributed for public use by researchers. By leveraging advanced tools from graph-theoretic, statistical, topological and time-series analysis, we will provide an integrated architecture and technology base to support diversity of emerging social media types and deploy strategic social network applications for high fidelity social network synthesis.