Learning to Disambiguate Queries with Respect to Business Processes

November 7, 2025

Process mining plays a critical role in extracting insights from organizational processes by analyzing event log data. However, it requires the engagement of domain experts and process analysts. While Large Language Models have enabled conversational agents for process mining tasks —reducing dependence on analysts — their adoption introduces new challenges. Users without an understanding of process mining often formulate ambiguous or ill-defined queries due to limited specific knowledge, hindering accurate analysis. To address this, we present an approach that integrates Retrieval-Augmented Generation Large Language Models and Chain-of-Thought reasoning. By retrieving a given event log and by reducing ambiguity with reasoning steps, human interaction with conversational agents becomes more intuitive, further bridging the existing gap. This work details the design, methodology, and implementation of the system, offering a novel framework for tackling ambiguity in process mining, thus improving user query resolution while fostering more intuitive user interaction. We also introduce datasets Query-PM-LLM and AmbQuery-PM-LLM, which can be used as benchmarks for future conversational agents capable of solving process mining tasks.

Authors in the community:

• 

  L

  Lucas Fortunato Das Neves

  ist195620

Supervisors of this institution:

• 

  A

  Alessandro Gianola

  ist429767

• 

  C

  Chrysoula Zerva

  ist429001

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico