Banca de QUALIFICAÇÃO: GABRIEL MARTINS SPÍNOLA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : GABRIEL MARTINS SPÍNOLA
DATE: 06/03/2026
TIME: 10:30
LOCAL: Google Meet
TITLE:

From Reactive Execution to Structured Planning: E2E Test Automation Guided by Navigational Graphs and Multimodal LLMs

KEY WORDS:

End-to-end test automation; End-to-end tests; Large language models (LLMs); Flaky Tests; Multimodal agents.

PAGES: 54
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUBÁREA: Metodologia e Técnicas da Computação
SPECIALTY: Engenharia de Software
SUMMARY:

End-to-end (E2E) test automation is fundamental to ensure the quality of modern Web applications by validating complete user flows from the end-user’s perspective. However, traditional script-based strategies incur high maintenance costs and high fragility, often resulting in flaky tests due to their strong coupling to mutable details of the Document Object Model (DOM). Although large language models (LLMs) have demonstrated the ability to generate tests from natural language descriptions, existing approaches mostly rely on direct translation of these descriptions into scripts, without maintaining a persistent structural representation of the interface, which limits knowledge reuse.

In this work, we analyze the impact of using navigational models that represent the application interface as a semantically enriched graph of routes for E2E test generation. We propose a strategy that separates the knowledge discovery stage from test synthesis, exploring the interface to build a transition graph whose states are annotated with semantic information obtained by a multimodal agent that assesses both the structural and visual aspects of the interface. Subsequently, we map natural language test cases to paths in this graph, aiming at generating more robust executable scripts. The approach is evaluated through a comparative empirical-experimental study on open-source Web applications against direct-translation baselines. With this, we expect to reduce the fragility typical of DOM-coupled tests and promote greater knowledge reuse throughout the application lifecycle.

COMMITTEE MEMBERS:
Presidente - 2274774 - EIJI ADACHI MEDEIROS BARBOSA
Interno - 1644456 - UIRA KULESZA
Externa ao Programa - 1709820 - ROBERTA DE SOUZA COELHO - UFRN