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

We propose to develop a high fidelity approach of synthesizing measurement-calibrated dynamic social networks. Social network research needs access to realistic graph datasets for testing social network theory and developing network algorithms. However, the available social network data are either typically small or cannot be readily shared due to privacy issues. To address these challenges, a unifying robust framework is proposed to incorporate novel techniques for high fidelity synthesis of social network data with dynamic graph measurement calibration. Online social network (OSN) data will be leveraged to form social interaction graphs. This approach provides the flexibility to capture link formations dynamically as functions of interaction intensities by adding temporal dynamics. We propose to apply measurement-based calibration of diverse graph models. Different network analytical functions (including user attributes, interaction, subgraph populations and temporal variations) can be represented by statistical time-varying descriptions in these graph models. Network synthesis results will be thoroughly evaluated in terms of similarity with OSNs by using diverse fidelity criteria such as statistical metrics, application-level benchmarks and geometric/topological properties. The accuracy-complexity trade-offs in multi-dimensional fidelity objectives will be balanced by introducing a closed-loop control mechanism that can automatically update model selection and optimize parameters in graph fitting.