Intent Based E2E Automated Test Case Generation for Web Applications Using LLM

Ensuring reliability in dynamic web applications is like hitting a moving target, where content constantly changes in response to user interactions and real-time updates. Therefore, testing these applications automatically remains challenging. Additionally, elements, which render graphical content outside the DOM and lack semantic metadata, makes it difficult to automatically extract and validate their content, leading to incomplete testing in existing methods. This paper proposes a novel approach named E2E-TestGen, which based on intent-driven testing, leverages Large Language Models (LLMs). E2ETestGen interacts dynamically with the DOM by triggering JavaScript-driven updates and tracking real-time state transitions, ensuring comprehensive validation across all dynamic elements and interactions to improve coverage. For canvas-rendered graphical components, we integrate an image-based object detection and text extraction module to extract meaningful data from visual content, enabling the generation of realistic test case scenarios guided by user intent. The experiments on multiple open-source web applications are conducted, which show E2E-TestGen achieving 88.5% and 80.1% code and branch coverage, respectively, along with a 27.91% improvement in fault detection over baseline approaches.