Matching Processes in Context: Can Large Language Models improve matching Complex Process Model Correspondences?

Organizations often maintain multiple process models that describe similar or related business processes. Identifying correspondences between these models — known as process model matching — is an important prerequisite for comparing and integrating processes. This is particularly relevant in process harmonization and standardization initiatives, for example during large-scale cloud transformations, where organizations need to understand how existing processes relate to each other and to a common target process.

A wide range of process model matching techniques has been proposed, drawing on activity labels, linguistic similarity, model structure, and process context. While identifying direct correspondences between individual activities is often comparatively straightforward, matching becomes considerably harder when the same process semantics are represented by multiple activities, different structures, or different levels of abstraction. Such complex correspondences cannot be identified by considering activities in isolation: they require understanding what groups of activities represent in the broader context of a process.

Large Language Models (LLMs) make it worthwhile to revisit this challenge. Their ability to interpret natural language and reason over contextual information may provide precisely the capabilities needed to recognize correspondences across larger process fragments. This raises the central question of this thesis: Can LLMs leverage process context to identify complex process model correspondences that are difficult to capture with established matching techniques?

In this thesis, you will develop an LLM-based approach to process model matching and systematically evaluate it against existing approaches. A particular focus will be on how process context can be represented and provided to an LLM, and whether incorporating this context improves the identification of complex correspondences. Depending on the scope of the thesis, you may investigate different representations of process context, prompting and matching strategies, or different LLMs. Beyond overall matching performance, you will analyze which types of correspondences benefit from contextual information and where the limitations of LLM-based matching remain.

What You Will Do

• Review the literature on process model matching, with a particular focus on complex correspondences, contextual matching techniques, and recent applications of LLMs to business process management.
• Design an LLM-based matcher. Develop an approach for representing process models and their context to an LLM and identifying correspondences between process elements and fragments.
• Implement the approach as a prototype that can automatically identify correspondences between process models using one or more LLMs.
• Evaluate the matcher against existing matching approaches, with particular attention to its ability to identify complex correspondences.
• Investigate the role of context. Analyze how different forms and amounts of process context affect matching and which types of correspondences benefit from contextual information.
• Analyze successes and failures to understand where LLM-based matching provides advantages and where limitations remain.
• Write up the findings, discussing the implications for process model matching and the potential of context-aware matching for process harmonization and standardization.

Your Profile

• Genuine interest in business process management, process modelling, and applications of Large Language Models
• Ability and willingness to implement and experimentally evaluate an LLM-based prototype
• Experience with Python or willingness to become proficient with it
• Interest in working with LLM APIs, prompting, and structured data representations
• Prior exposure to BPMN, process modelling, or natural language processing.

The topic can be scoped for a Bachelor's or a Master's thesis. A Bachelor's thesis may focus on implementing and evaluating a selected LLM-based matching approach, while a Master's thesis can investigate alternative representations of process context and matching strategies in greater depth and conduct a more extensive analysis of complex correspondences. The exact research questions and scope will be aligned during the proposal process before registration.

Please get in touch with a short email including you’re a Motivation for writing a thesis at our chair (in the body of your email), your CV, a current transcript of records, and the planned start/finish date: rehse@ifi.uzh.ch.