Evaluating Large Language Models in Process Mining: Capabilities, Benchmarks, and Evaluation Strategies
Alessandro Berti, Humam Kourani, Hannes Hafke, Chiao-Yun Li, Daniel Schuster
TL;DR
The paper addresses the challenge of evaluating the utility of large language models (LLMs) for process mining (PM) tasks by outlining the minimal capabilities required, proposing benchmarking strategies, and detailing evaluation methods for LLM outputs. It synthesizes PM tasks and LLM implementation paradigms, then identifies core capabilities (e.g., long context, visual prompts, coding, factuality) and a spectrum of benchmarks (traditional, domain knowledge, visual, and text-to-SQL) relevant to PM. It further proposes evaluation approaches—automatic, human, and self-evaluation—to address issues like hallucination and to support PM-specific benchmarking. The work aims to enable reliable PM-on-LLMs research and foster comprehensive benchmarks that account for PM-specific tasks and visualization challenges, ultimately supporting trustworthy and effective deployment of LLM-enhanced PM solutions.
Abstract
Using Large Language Models (LLMs) for Process Mining (PM) tasks is becoming increasingly essential, and initial approaches yield promising results. However, little attention has been given to developing strategies for evaluating and benchmarking the utility of incorporating LLMs into PM tasks. This paper reviews the current implementations of LLMs in PM and reflects on three different questions. 1) What is the minimal set of capabilities required for PM on LLMs? 2) Which benchmark strategies help choose optimal LLMs for PM? 3) How do we evaluate the output of LLMs on specific PM tasks? The answer to these questions is fundamental to the development of comprehensive process mining benchmarks on LLMs covering different tasks and implementation paradigms.