黑料福利网

PhD researcher Mahdi Saeedi Nikoo addressed these challenges by developing methods to detect duplication across large repositories of process models and provide intelligent recommendations for incomplete subprocesses. His work combines empirical analysis, automated tools, and AI techniques to make process modelling more actionable and efficient. He defended his PhD thesis on Thursday, November 13.

Analyzing service composition languages

Modern organizations increasingly rely on orchestrated services to deliver complex functionality. Service composition languages are tools that help organizations describe how different services, such as software components or business processes, work together to perform complex tasks.

studied 14 of these languages, looking at their features, how widely they are used, and how their popularity has changed over time. His analysis showed that the most common approaches today either coordinate services from a central point, or combine central coordination with more flexible, distributed arrangements. Many earlier languages are no longer in use.

These insights help organizations and researchers understand which modeling approaches are most relevant today.

Detecting duplication in process models

Managing large collections of business process models can be difficult because similar or identical models often appear multiple times. To address this, Saeedi Nikoo developed a method for detecting duplication, also called 鈥渃lones,鈥� in repositories. His approach can find both entire models that are repeated and smaller fragments that appear in multiple places.

By identifying these duplicates, organizations can reduce unnecessary repetition, make better use of existing models, and keep their processes more consistent. Tests showed that his method can outperform or complement leading tools such as , depending on the situation.

Insights from open-source repositories

Saeedi Nikoo also conducted the first large-scale study of business process models in open-source repositories, including . He examined the domains, tools, ownership patterns, and duplication trends in these collections. The study revealed that clones are common at both the model and fragment level, and showed how models are reused and adapted across different contexts.

These findings give researchers and practitioners practical guidance for managing large collections of process models more effectively.

Supporting modelers with intelligent recommendations

Building on these findings, Saeedi Nikoo created intelligent recommendation systems to help modelers fill in missing steps in incomplete subprocesses. He compared traditional similarity-based tools with advanced large language models (LLMs), finding that similarity tools work best for smaller pieces, while LLMs are more effective for larger, complex fragments.

By combining both approaches in a hybrid system, modelers get the most reliable and useful suggestions, making it easier to design accurate and complete process models.