On the Potential of Large Language Models to Solve Semantics-Aware Process Mining Tasks

TL;DR

This work investigates whether large language models can tackle semantics-aware process mining tasks by defining five tasks (semantic anomaly detection, semantic next activity prediction, and semantic discovery) and providing extensive benchmarking data derived from a corpus of process behaviors. It systematically compares in-context learning and supervised fine-tuning for open decoder LLMs (Llama-3 and Mistral-2) against encoder baselines (RoBERTa) and a rule-based semantic detector. The results show that out-of-the-box LLMs underperform, but task-specific fine-tuning yields strong performance across diverse processes and generations tasks, often surpassing encoder baselines. The findings highlight the potential of combining semantic reasoning from LLMs with traditional process mining techniques and point to future work on broader data, integration with existing methods, and the development of a Large Process Model. These insights have practical implications for improving semantics-aware analyses such as discovery, anomaly detection, and next-action predictions in real-world processes.

Abstract

Large language models (LLMs) have shown to be valuable tools for tackling process mining tasks. Existing studies report on their capability to support various data-driven process analyses and even, to some extent, that they are able to reason about how processes work. This reasoning ability suggests that there is potential for LLMs to tackle semantics-aware process mining tasks, which are tasks that rely on an understanding of the meaning of activities and their relationships. Examples of these include process discovery, where the meaning of activities can indicate their dependency, whereas in anomaly detection the meaning can be used to recognize process behavior that is abnormal. In this paper, we systematically explore the capabilities of LLMs for such tasks. Unlike prior work, which largely evaluates LLMs in their default state, we investigate their utility through both in-context learning and supervised fine-tuning. Concretely, we define five process mining tasks requiring semantic understanding and provide extensive benchmarking datasets for evaluation. Our experiments reveal that while LLMs struggle with challenging process mining tasks when used out of the box or with minimal in-context examples, they achieve strong performance when fine-tuned for these tasks across a broad range of process types and industries.

Figures (5)

• Figure 1: Categorization of the process models in the corpus.
• Figure 2: One-shot in-context-learning prompts.
• Figure 3: Illustration of discriminative classification with an encoder LM.
• Figure 4: Illustration of constrained generative fine-tuning of a decoder LM.
• Figure 5: Illustration of generative fine-tuning of a decoder LM.