1. Initial Situation

The subject-oriented process modeling language PASS is a simple, yet formal and especially executable modeling language. Due to its subject-oriented nature, the naming of different elements in PASS models is of particular importance for understanding. E.g., subject names should denote their active nature, while states in behavior diagrams should ideally be written to describe the activity as active verbs. In a previous work, a prototypical tool has been developed that checks a PASS model and informs a process modeler whether and how the naming in models could be improved. The tool works on a functional level but has not been fully optimized or tested.

2. Research Goal/Research Questions

The overall goal of this thesis is to advance the existing prototype by integrating state of the art AI techniques. The current tool relies on basic prompting strategies without leveraging more complex architectures or agent-based mechanisms. The thesis aims to identify, compare and assess different approaches like advanced prompting methods, AI agent frameworks and human- or AI-in-the-loop concepts. By systematically analyzing the theoretical foundations and implementing selected techniques in practice, the work seeks to increase reliability, accuracy and usability of the PASS model generator.

Research questions:

Theoretical:

• What state-of-the-art AI techniques and architectural paradigms are currently used
  • in general
  • for text analysis and validation tasks

and how can they be applied to naming evaluation in PASS models?

• How do different advanced AI techniques compare in terms of suitability formal modeling languages?
• How do human-in-the-loop and AI-in-the-loop concepts influence the accuracy, reliability, and transparency of automated naming assessments in PASS models?
• What are the benefits and potential drawbacks of adopting a modular system architecture for an AI-based naming evaluation tool, and to what extent does modularity facilitate the integration of future AI advancements?
• Which evaluation criteria (e.g., accuracy, consistency, explainability, transparency, computational cost, robustness, implementation effort, drawbacks of potential technology lock-ins) are most appropriate for comparing results

 

Practical:

• How do selected state-of-the-art techniques perform when implemented in an improved prototype, and what measurable improvements can be observed in comparison to the existing tool?
• To what extent do model choice and architectural configuration (e.g., single-model vs. multi-agent pipelines) impact the overall quality and reliability of naming suggestions?
• How does the inclusion of modular or loop-based components (human-in-the-loop) affect the maintainability, extensibility, and user acceptance of the revised tool?
• What design recommendations can be derived for future AI-based validation tools in subject-oriented process modeling based on the empirical findings of this thesis?

 

3. Planned Method + Planned structure  

What are some steps planned to answer the research questions? (A rough concept is sufficient)

Theoretical:

• Literature review
• Methodology for Comparative Analysis

Practical

• Implementation
• Testing