TestHumanizer Large Artifact Package - Dependencies, Refactorings, SIDE Model, and Generated Summaries

This archive contains the large artifacts required to reproduce and inspect the experiments of the study: Humanizing Automatically Generated Unit Test Suites with LLM-Based Refactoring The associated GitHub repository contains the main replication package, including the TestHumanizer implementation, scripts, configuration files, lightweight metadata, prompts, documentation, and reproduction commands. This Zenodo archive complements the GitHub repository by hosting the large files that are required for full replication but are not suitable for direct inclusion in GitHub. The archive contains four main artifact groups: Refactored test suites Compiled benchmark dependencies and Java dependency archives SIDE model files Generated source-code summaries These artifacts support the execution and inspection of the TestHumanizer pipeline, including EvoSuite-based test generation, LLM-based refactoring, summary-based prompting, compilation/validation, and downstream metric computation. Structure Refactored_data/ SIDE_Model/ Summary_generated/ Dependancies/ Artifact contents Refactored_data This directory contains the LLM-generated refactored unit test suites produced during the study. It is divided into two subdirectories: Refactored_data/ ├── Raw_refactorings/ └── Compilable_refactorings/ Raw_refactorings/ contains the raw LLM outputs before compilation filtering. Compilable_refactorings/ contains the subset of refactored test suites that passed the compilation stage. The directory includes refactorings produced by the LLMs used in the study, including GPT-4o and Mistral. Example files: gpt4o_raw_refactored_tests.jsonl gpt4o_raw_refactored_tests_metadata.csv mistral_raw_refactored_tests.jsonl mistral_raw_refactored_tests_metadata.csv gpt4o_compilable_refactored_tests.jsonl gpt4o_compilable_refactored_tests_metadata.csv mistral_compilable_refactored_tests.jsonl mistral_compilable_refactored_tests_metadata.csv The JSONL files contain the full refactored test-suite contents.The CSV files provide flat metadata for indexing, filtering, and analysis. Dependancies This directory contains the compressed benchmark and dependency archives required to reproduce the experiments. Dependancies/ ├── DEFECTS4J.zip ├── JARS.zip └── SF-110.zip DEFECTS4J.zip contains the Defects4J source/class layout used by the TestHumanizer pipeline. SF-110.zip contains the SF110 source/class layout used by the TestHumanizer pipeline. JARS.zip contains Java dependency jars required by EvoSuite, compilation, validation, coverage analysis, and metric computation. These archives are needed because the TestHumanizer pipeline relies on compiled Java classes and project-specific classpaths when generating, compiling, and validating unit test suites. SIDE_Model This directory contains the SIDE model artifact used in the source-code summarization stage. SIDE_Model/ ├── hard-negatives-20260608T120328Z-3-001/ ├── hard-negatives-20260608T120328Z-3-001.zip └── README.md The SIDE model is used to evaluate the semantic alignment between generated natural-language summaries and the corresponding Java source code (Paper avalaible). These summaries are used by the summary-based TestHumanizer configuration, where compact class descriptions replace full source-code context to reduce prompt length and improve robustness. Summary_generated This directory contains the generated source-code summaries used by the summary-based refactoring configuration. Summary_generated/ ├── code_summary_dataset_flat.json ├── code_summary_dataset_structured.json └── code_summary_mapping_report.json The flat and structured JSON files contain the generated summaries for target classes. The mapping report documents how summaries are linked to dataset entries, projects, classes, and downstream refactoring inputs. Reproducibility The GitHub repository contains the full executable replication package: https://anonymous.4open.science/r/TestHumanizer-06FB/ The repository includes: the TestHumanizer Python package; the command-line pipeline; dataset metadata and experiment runners; configuration templates; prompts and documentation; scripts for sample and full reproduction; instructions for linking the Zenodo artifacts with the GitHub codebase. After downloading this Zenodo archive, users should extract or link the artifact folders into the expected repository layout described in the GitHub documentation. Expected high-level layout: TestHumanizer/ ├── Approache/ ├── Datasets/ ├── Prompts/ ├── Results/ ├── Refactored_data/ ├── SIDE_Model/ ├── Summary_generated/ └── Dependancies/ The main TestHumanizer tool is located in: Approache/TestHumanizer/ The dataset experiment runner is located in: Approache/TestHumanizer/scripts/run_dataset_experiment.py Files Recommended Zenodo upload file: TestHumanizer_Large_Artifacts_v1.zip Recommended archive contents: Refactored_data/ SIDE_Model/ Summary_generated/ Dependancies/ Notes This Zenodo archive does not replace the GitHub replication repository. Instead, it stores the large artifacts required by the GitHub codebase for full replication and inspection. The GitHub repository should be used as the main entry point for installation, documentation, scripts, and reproduction commands. The Zenodo DOI ensures long-term availability, versioning, and citation of the large artifacts.