Acceptance Test Generation with Large Language Models: An Industrial Case Study
2025 IEEE/ACM International Conference on Automation of Software Test (AST)
This research explored the power of Large Language Models (LLMs) in automating acceptance testing for web applications. We developed a two-step approach: AutoUAT generates Gherkin-based test scenarios from user stories, while Test Flow converts these scenarios into executable Cypress scripts. Integrated into an industrial setting, these tools improved test coverage, reduced manual effort, and enhanced software quality. With a 95% user approval for AutoUAT and a 92% accuracy rate for Test Flow, our study highlights the potential of AI-driven testing in agile development.
DownloadDecomposing Complex Text Classification Tasks through Error Analysis: A Study on Genocide-related Court Hearings
Discovery Science 2023
This study explored the challenges of classifying legal transcripts from genocide-related court hearings using Natural Language Processing (NLP). By analyzing errors in existing models, we uncover the limitations of traditional Transformer-based approaches and demonstrate how targeted pretraining can enhance classification accuracy. Our research identifies key linguistic and contextual challenges in the Genocide Transcript Corpus (GTC) and proposes novel solutions, ultimately improving state-of-the-art results. This work bridges the gap between legal text analysis and modern NLP methodologies, offering insights into better model selection and fine-tuning strategies.
Download