A syntactic-semantic analysis method for automatic API connection points discovery in systems-of-information systems

Abstract

Establishing interoperability links is a significant challenge in systems-of-information systems (SoIS) engineering. Even with constituent systems (CS) interfaces documentation, achieving such links is a difficult, time-consuming, and error-prone task that requires attention from CS developers, especially if it is performed manually. To contribute to that task, this thesis presents a semi-automatic method using both syntactic and semantic similarity analysis of application programming interface (API) descriptions to identify potential connection points among CS. That method consists of three interconnected steps: (i) requirements definition, where the motivations and justifications for creating the method are outlined; (ii) core method design, where all the phases and components necessary to obtain the API connection points are detailed; and (iii) core method implementation, where the technological details of the tool developed to support the method are presented. Through that tool, two evaluations were performed. In the first, a controlled experiment was executed to evaluate the tool’s performance and empirically define the best similarity algorithms and thresholds for the task of identifying connection points between two well-known API. In the second, the tool was used in a case study applied to a real-world SoIS of a global computer manufacturer. The study covered three scenarios, involving seven CS API, and considered the perspectives of syntactic analysis and syntactic-semantic analysis of API. Then, semi-structured interviews were conducted with five developers who worked on the SoIS to evaluate the tool’s efficiency. The results demonstrated that the tool is successful, with the syntactic and semantic similarity analysis of API proving to be efficient in identifying interoperability links between CS. From a practical point of view, the syntactic algorithms demonstrated notable performance in local environments. At the same time, semantic analysis, which is more robust, requires advanced infrastructure due to its high computational complexity. From a research perspective, the use of short text semantic similarity (STSS) is still a promising field, and the use of artificial intelligence with Large Language Models (LLM) can generate new results and discoveries of interoperability links in SoIS.