Automatic generation of ABM Narratives using Simulation Traces and LLM
Keywords: ABM Narratives, Communication Support, LLM
TL;DR: This paper presents a novel methodology for generating execution-based narratives for ABMs using simulation logs and large language models.
Abstract: Effective communication of agent-based models is essential for ensuring their usability and transparency. However, conventional documentation approaches often struggle to capture the dynamic execution details of simulations, making it challenging to convey complex processes clearly and accessibly. Moreover, ABMs frequently exhibit emergent and unexpected behaviors resulting from multiple agent interactions—dynamics that static model descriptions does not fully capture. This paper presents a novel methodology for generating execution-based narratives for ABMs using simulation logs and large language models. By integrating process mining, Business Process Modeling Notation, and automated narrative generation, the approach transforms raw simulation data into coherent visual and textual artifacts that faithfully reflect the model’s dynamic execution. Unlike conventional documentation—which often relies on subjective assessments and demands significant effort from modelers—this methodology minimizes subjectivity and reduces the effort required from modelers while promoting a more accessible approach to model communication. To demonstrate its expressivity, we applied the methodology to the Luneray Flu Model and successfully produced artifacts such as process maps, business process diagrams, and narrative explanations. This work offers a step toward improving transparency and accessibility in ABM verification and communication.
Submission Number: 6
Loading