LLM-Based Multi-Agent Framework for Process Discovery (ICSOC 2025 Dataset)

This dataset accompanies the ICSOC 2025 submission on LLM-based Multi-Agent Systems (MAS) for conversational business process discovery. It includes the source code, role prompts, evaluation logs, experimental configurations, and example outputs for three process scenarios (Shop, Reimbursement, Hotel). The MAS was tested using GeminiPro, Mistral, and DeepSeek under four architectural setups (Monolithic, Duo, Manager, Team). Instructions are included for reproducibility. The data supports the evaluation of task completion, BPMN feature identification, and failure analysis.