A method for hierarchical generalization of a software product process model

Abstract

This paper proposes a method for hierarchical generalization of a software product process model formed from documentation artifacts. The initial model is derived from an end-to-end process specification, two operating procedures, a role-and-system matrix, and acceptance-test scenarios. The documents are segmented into logical fragments, candidates for actions, conditions, roles, systems, and transitions are extracted, and only elements validated or edited by a business analyst are included in the graph. The model is formalized as a typed directed graph whose attributes retain links to documentation sources. Candidate groups are generated within a local graph neighborhood and checked for the integrity of alternative branches. Structurally valid fragments are evaluated by an integral metric combining textual, contextual, and control-flow components. The method was checked on test graph. After execution of the method the numbers of visible nodes and transitions decreased. During generalization, the method preserves the control flow, branching integrity, and traceability of aggregated elements to the source documentation. The proposed approach combines the semantic characteristics of process elements, their execution context, and the control-flow structure, which makes it possible to form consistent levels of representation without violating the logic of alternative branches. The reproducibility of the construction is ensured by fixed rules for forming candidate groups, preserving links to the source documents, and successively expanding generalized elements to the atomic level. The method was evaluated on a test graph. After generalization, the numbers of visible nodes and transitions decreased. A candidate with a sufficient numerical score was rejected because it covered only part of an alternative branch. The method provides a compact multilevel view while preserving traceability to atomic process elements and the documentation from which they were obtained. The proposed approach can be applied to the multilevel representation of software product process models.