Intelligent Methods for Software Architecture and Business Logic Recovery: Current State and Prospects for Application

Abstract

This article critically examines intelligent methods for recovering software architecture and business logic in polyglot enterprise systems. The central issue is evidence control. Existing approaches can identify relationships between software artifacts, but these relationships often differ in origin, reliability, and analytical meaning. A direct structural dependency is not the same as a semantic association. A configuration reference is not the same as a language-model-generated hypothesis. The review covers classical architecture recovery methods, semantic code representations, large language models, graph-based program representations, and graph neural network approaches. Each group contributes a useful but limited form of knowledge. Classical recovery methods provide traceable structural dependencies; semantic models reveal potentially related artifacts; large language models formulate interpretive explanations; and graph-based models place artifacts within a broader system context. Yet none of these approaches is sufficient on its own when business logic is scattered across implementation, configuration, data-access, and integration layers. The article argues that business logic recovery should be treated not only as relationship detection, but as the organization of heterogeneous evidence. This distinction is crucial. If confirmed dependencies, semantic similarities, configuration links, and generated interpretations are merged without separation, the recovered model may appear complete while losing analytical transparency. The review therefore identifies the need for a traceable semantic-graph representation, contributing to the field of systems analysis and mathematical modeling of complex software systems.