Implementation of Smell Detection Queries on Enterprise Architecture Knowledge Graphs

Abstract

Enterprise Architecture (EA) encompasses various domains, such as business architecture, information system architecture, and infrastructure architecture, providing a holistic view of an enterprise. Implementing EA can be complex, especially as enterprises grow, leading to challenges in design and architecture. Recently, the metaphor of Enterprise Architecture Debt (EAD) has been introduced to address the negative consequences of EA. It is derived from Technical Debt (TD), which covers the enterprise’s technical and business aspects. Since this definition does not provide a way to identify and measure possible debts, the concept of EA Smell has been introduced. EA Smell is a means to identify the debt in EA. A catalog consisting of sixty-three EA Smells has been published. However, since the catalog is in its early stages of development, most Smells still do nothave precise definitions and concrete approaches for detection.EA models are crucial for enterprise success. They are widely used in research and industry. Yet, more attention needs to be paid to analyzing EA models and addressing the model flaws in a more formal way. ArchiMate is a de facto standard modeling language for EA, but does not offer quality analysis tools. Examining EA Smells in ArchiMate models will assist enterprise architects in identifying design flaws and deficiencies in the model. However, manually analyzing large models can be challenging due to their complexity, which may lead to miscalculations and overlooked details. In contrast, automated analysis can assist architects in avoiding such errors. Therefore, an automated analysis tool for detecting EA Smells can ease the identification of model flaws and raise awareness about those shortcomings and negative consequences.A novel approach to interpreting EA models as Knowledge Graphs has been introduced, which facilitates model analysis via KG queries and graph algorithms.Based on the Knowledge Graph (KG) approach, our thesis analyzed the EA Smellsand defined KG queries to identify those Smells in EA models. We were able to define KG queries for thirty-six EA Smells. For each of those Smells, we briefly described the underlying Smell and how we transferred and detected it in the EA domain, particularly the ArchiMate modeling language. Additionally, we enhanced the EA Smell catalog by including the defined queries and supplementary extensions to clarify the semantics of EA Smells where necessary. Finally, we implemented a detection platform to automatically identify EA Smells in KG representing ArchiMate models. We tested our detection platform and evaluated our approach on a large set of EA models. The results were promising. Our platform was able to detect many EA Smells from real-world models. Finally, we evaluated an EA model given by an EA expert containing different EA Smells.We used precision and recall metrics to evaluate the correctness of our approach and achieved a high ratio for both metrics, which indicates that our approach is very accurate.