Finetuned language models can turn messy organizational text into clean event logs for process mining.

Organizations want to study and improve how their work gets done using process miningprocess miningA technique that analyzes business processes by looking at data from event logs., which analyzes records of business activities. However, this tool requires neatly organized event logsevent logsStructured records of activities, like who did what and when, used to study processes.. Today, human specialists must build these logs by hand. Because this takes so much work, valuable records like written incident tickets, manuals, and reports get ignored. To fix this, researchers tested whether Large Language Models could act as automatic translators. They built a new text-to-log training dataset and used it to finetune, or specifically train, the models. The finetuned models successfully pull accurate, high-fidelity event logs straight from plain text. The team discovered that finetuningfinetuningThe process of further training a pre-trained AI model on specific data to improve its performance on a particular task. beats simple zero-shot or few-shot prompting by a large amount, making specialized training essential to produce reliable event data.

Why it matters

This method reduces the manual work needed to create event logs, freeing up experts for higher-level tasks. It allows organizations to use previously ignored text data to understand and optimize their processes better.

Questions answered