A Tool for Agent-Based System Produced Emergent Networks (ASPEN)

Social network analysis is a powerful method for making sense of an increasingly interconnected world. Developing social network analysis tools requires datasets with which to test them, but sharing datasets is problematic because of privacy issues. Initial attempts to anonymize network datasets proved ineffective due to de-anonymization attacks that leveraged publically available information to identify individuals. Synthetic datasets provide an alternative, but it is difficult to ensure that a synthetic network is sufficiently similar to a real network for tool development. To address this issue, our system for Agent-Based System Produced Emergent Networks (ASPEN) imbues agents with social and behavioral theories, has them interact in a simulation, and generates a synthetic network from the simulation log. The simulation network is then compared to a real network using network metrics to assess their similarity or fidelity. These metrics are then fed to an evolutionary algorithm (EA) that varies the behavioral inputs to the agent-based simulation to improve the synthetic network"s fidelity. Synthetic networks thus emerge from the behavioral interactions among agents rather than being directly specified.