A carregar...
Publicação
Leveraging Large Language Models for Process Analytics Assistants: Assessing Accuracy in Process Mining Tasks
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.94 MB | Adobe PDF |
Orientador(es)
Resumo(s)
The integration of Large Language Models with Process Mining has the potential to enhance
business outcomes by improving data-driven decision-making. The employment of visual
analytics allows the translation of Process Mining insights for non-technical users, enabling
them to make deeper and well-informed decisions, while also supporting integration with
decision support systems. This study introduces a framework designed to evaluate the current
state of research in this field by employing the latest Large Language Models and prompt
engineering techniques, assessing their capability to perform visualisation tasks aligned with
Process Mining, from process discovery to predictive analytics. Among the most recent models
provided by OpenAI and Anthropic, GPT4.1 demonstrated the strongest performance,
achieving an LLMs-as-Judges score of 8,18 ± 2,3 and a Visualisation Error Rate of 25%.
Although these results are promising, the findings also indicate that the existing Large
Language Models still face challenges in handling domain-specific libraries such as PM4Py,
having an increased performance when using widely adopted libraries. This highlights the
need for further improvements of these models to fully support automated visual insight
generation in Process Mining contexts.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business Analytics
Palavras-chave
Large Language Models LLMs-as-Judges Process Mining Prompt Engineering Visual Analytics Visualisation Error Rate SDG 8 - Decent work and economic growth SDG 9 - Industry, innovation and infrastructure