Supporting business process management: clone detection and recommendation techniques

Every organization, whether an enterprise or a government institution, performs a series of coordinated activities—collectively known as business processes—to deliver products or services to its customers. These processes transform initial triggers into desired outcomes, and their structure and execution are commonly captured through business process models. These models serve as structured representations that document how operations are carried out, enabling organizations to analyze, optimize, and communicate their workflows effectively.

This thesis begins by investigating service composition languages, which define how distributed services are orchestrated to support complex system functionality. Among the notations used in service composition, business process modeling languages such as BPMN stand out for their widespread adoption in both industry and academia. As organizations increasingly rely on process modeling, the need for scalable, maintainable, and reusable models becomes critical—especially in environments managing hundreds or even thousands of models.

To address this need, this thesis introduces a novel approach for clone detection in business process models. The proposed method is capable of identifying both full-model and fragment-level clones, enabling better reuse and reducing redundancy across large process model repositories. This contributes to more consistent and efficient model management by helping practitioners understand where similar behavior already exists and avoid unnecessary duplication.

Building on this, an extensive empirical study investigates the landscape of business process models in open-source repositories, focusing particularly on those hosted on GitHub. The study reveals how these models are used, evolved, and reused across different domains. It highlights the prevalence of model and fragment-level cloning and offers insights into modeling practices, tooling, and contributor behavior. These findings emphasize the growing complexity of managing large-scale model repositories and reinforce the importance of automated support mechanisms.

Motivated by the widespread reuse and cloning observed in open-source models, the thesis explores the role of recommender systems in business process modeling. In particular, it evaluates how traditional similarity-based tools and emerging large language models (LLMs) can assist modelers in completing partially defined process fragments. A comparative evaluation shows that while LLMs offer strong performance on larger fragments, traditional similarity tools generate more meaningful recommendations for smaller subprocesses—suggesting that hybrid or context-aware approaches may offer the best support.

In conclusion, this thesis presents a coherent set of contributions aimed at improving the quality, reusability, and efficiency of business process modeling. Through clone detection, empirical characterization, and intelligent recommendation, it provides practical solutions and tools that advance the state of the art in business process management. These contributions have the potential to benefit both researchers and practitioners working with process models at scale.