Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Finetuning LLMs for Automatic Form Interaction on Web-Browser in Selenium Testing Framework

Nguyen-Khang Le, Hiep Nguyen, Ngoc-Minh Nguyen, Son T. Luu, Trung Vo, Quan Minh Bui, Shoshin Nomura, Le-Minh Nguyen

TL;DR

The paper tackles the challenge of generating robust Selenium test scripts for web form interaction using large language models. It introduces a dataset-driven fine-tuning pipeline that combines synthetic and human-annotated HTML forms, paired with a workflow to produce executable test scenarios and Selenium code, optimized for syntax correctness, runtime executability, and input-field coverage. Through extensive evaluation across multiple model families and with filtered vs full fine-tuning settings, the approach yields significant gains in executability and coverage while maintaining high syntax correctness, outperforming strong baselines such as GPT-4o. By releasing datasets and establishing clear evaluation metrics, the work provides a practical foundation for advancing LLM-based web testing and form-interaction automation in real-world pipelines.

Abstract

Automated web application testing is a critical component of modern software development, with frameworks like Selenium widely adopted for validating functionality through browser automation. Among the essential aspects of such testing is the ability to interact with and validate web forms, a task that requires syntactically correct, executable scripts with high coverage of input fields. Despite its importance, this task remains underexplored in the context of large language models (LLMs), and no public benchmark or dataset exists to evaluate LLMs on form interaction generation systematically. This paper introduces a novel method for training LLMs to generate high-quality test cases in Selenium, specifically targeting form interaction testing. We curate both synthetic and human-annotated datasets for training and evaluation, covering diverse real-world forms and testing scenarios. We define clear metrics for syntax correctness, script executability, and input field coverage. Our empirical study demonstrates that our approach significantly outperforms strong baselines, including GPT-4o and other popular LLMs, across all evaluation metrics. Our work lays the groundwork for future research on LLM-based web testing and provides resources to support ongoing progress in this area.

Finetuning LLMs for Automatic Form Interaction on Web-Browser in Selenium Testing Framework

TL;DR

The paper tackles the challenge of generating robust Selenium test scripts for web form interaction using large language models. It introduces a dataset-driven fine-tuning pipeline that combines synthetic and human-annotated HTML forms, paired with a workflow to produce executable test scenarios and Selenium code, optimized for syntax correctness, runtime executability, and input-field coverage. Through extensive evaluation across multiple model families and with filtered vs full fine-tuning settings, the approach yields significant gains in executability and coverage while maintaining high syntax correctness, outperforming strong baselines such as GPT-4o. By releasing datasets and establishing clear evaluation metrics, the work provides a practical foundation for advancing LLM-based web testing and form-interaction automation in real-world pipelines.

Abstract

Automated web application testing is a critical component of modern software development, with frameworks like Selenium widely adopted for validating functionality through browser automation. Among the essential aspects of such testing is the ability to interact with and validate web forms, a task that requires syntactically correct, executable scripts with high coverage of input fields. Despite its importance, this task remains underexplored in the context of large language models (LLMs), and no public benchmark or dataset exists to evaluate LLMs on form interaction generation systematically. This paper introduces a novel method for training LLMs to generate high-quality test cases in Selenium, specifically targeting form interaction testing. We curate both synthetic and human-annotated datasets for training and evaluation, covering diverse real-world forms and testing scenarios. We define clear metrics for syntax correctness, script executability, and input field coverage. Our empirical study demonstrates that our approach significantly outperforms strong baselines, including GPT-4o and other popular LLMs, across all evaluation metrics. Our work lays the groundwork for future research on LLM-based web testing and provides resources to support ongoing progress in this area.

Paper Structure

This paper contains 15 sections, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Training Data Creation
  5. Test Scenario and Selenium Generation
  6. Experiments
  7. Models
  8. Datasets
  9. Evaluation Metrics
  10. Results
  11. Main Results
  12. Analysis of Error rate in different field types
  13. Analysis of Causes of Errors
  14. Examples of Faulty Selenium Snippets and Implications
  15. Conclusions

Figures (3)

  • Figure 1: Overview of the data creation process.
  • Figure 2: Statistics of the synthetic HTML in training data.
  • Figure 3: Percentage of errors across field types for various LLMs. "Filtered SFT" refers to models fine-tuned with our filtering step, where unexecutable Selenium code is excluded, while "Full SFT" denotes fine-tuning without filtering, using all Selenium code for training.