Open Collections

Shahbandeh Vayghan, Mobina

University of British Columbia

Automated generation of end-to-end web tests is challenging due to the complexity and dynamic nature of web application features. This involves verifying that a web application correctly implements its intended features across diverse user interactions and scenarios. Current state-of-the-art web application exploration methods, such as WebCanvas, are not designed for comprehensive feature-based exploration; they rely on specific, detailed task descriptions, which limit their adaptability in dynamic web environments. We introduce NaviQAte, which generates action sequences to validate features without requiring detailed parameters. Our three-phase approach leverages advanced large language models like GPT-4o for complex decision-making and cost-effective models, such as GPT-4o mini, for simpler operations. NaviQAte focuses on feature-guided web application navigation, integrating multi-modal inputs such as text and images to enhance contextual understanding. Evaluations on the Mind2Web-Live real-world web applications dataset demonstrate that NaviQAte achieves a 36.5% success rate in user task execution and a 38.5% success rate in feature validation, representing a 15% and 34% improvement over WebCanvas. These results underscore the effectiveness of our approach in advancing automated web application navigation.

Text

2025-01-02

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/90011

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/