Intellectual model for generating adaptive WEB selectors based on GNN

The article discusses the problem of DOM selector instability, which is one of the key reasons for flaky tests in modern automated web resource testing systems. It is shown that traditional approaches to selector formation, both manual and automated, do not take into account the global context of the DOM structure and are ineffective in the case of dynamic interface changes. The feasibility of using graph neural networks (GNN) as a tool for modelling the DOM tree in the form of a directed graph, taking into account the semantic, attributive, and structural features of nodes, is justified. A theoretical model for building adaptive DOM selectors, combining GNN with a decision-making agent, is proposed. A mathematical representation of the DOM graph, a system of node features (one-hot tag encoding, binary attributes id/class/data-testid, depth, number of descendants, position among neighbours) has been developed, and selector evaluation metrics have been formalised: uniqueness, accuracy, completeness, F1-measure and length. An agent reward function has been formed that optimises the balance between accuracy, stability, and compactness of the selector. It has been shown that the combination of DOM (Document object model) graph representations and a decision optimisation mechanism allows the creation of selectors that are resistant to structural changes in web pages and reduces the need for manual test maintenance. The proposed model forms the theoretical basis for the creation of intelligent automated testing systems capable of adapting locators during DOM changes without the need for a complete analysis of the interface by the tester.