Optimization of Regression Testing Using Graph-Based Dependency Models: A Theoretical Review

Abstract

The paper provides a theoretical justification for optimizing regression testing using graph-based dependency models under conditions of frequent releases and high test execution costs. The relevance of the study is that the growth of software systems and the load on continuous integration infrastructure make full regression testing a chronic bottleneck, requiring formalized approaches to selecting, ordering, and reducing test suites at a controlled level of risk. The aim of the review is to synthesize the results of eight peer-reviewed studies and to construct a unified conceptual framework in which regression testing is interpreted as a risk-management problem over a dependency structure rather than as a mechanical re-execution of everything. The scientific novelty lies in interpreting the dependency graph as a carrier of a risk field, where edge and node weights reflect the strength of influence, interaction frequency, and node criticality, and an affected test is defined through reachability within an influence subgraph, subject to depth constraints and significance thresholds. The review systematizes three main classes of graph-based models (source-code level, component/service level, and bipartite tests–code graphs) and relates them to the operations of selection, prioritization, minimization, and time-budgeting, supplementing them with requirements for interpretability, reproducibility, and accounting for the total cost of ownership of the model. As a promising trajectory, the paper outlines a transition towards adaptive multimodal code–test–requirement–incident graphs enriched with observability data and learnable estimates of regression probability. The paper is intended for researchers and practicing software quality engineers who design or deploy methods for optimizing regression testing in scalable CI/CD processes.