Replication Package for Synthesizing Static Rules for Code Smell Detection

14,739 code smell annotations from 522 repositories in the MLCQ benchmark, alongside 1,840 developer evaluations and 40 interview transcripts. Zijie Huang published this replication package on figshare in April 2026 under a CC-BY-4.0 license. The package includes datasets, source code, and scripts for reproducing research on automated code smell detection.

Use Cases

• Training and benchmarking machine learning models for code smell detection based on the MLCQ benchmark dataset.
• Conducting developer studies on the effectiveness of detection rules based on the 1,840 evaluation instances.
• Performing qualitative analysis of developer perceptions based on the 40 interview transcripts.
• Reproducing statistical analyses and effect size calculations for research validation.

Strengths

• Includes a substantial benchmark with 14,739 annotations across 522 software repositories.
• Contains human evaluation data from 1,840 instances and 40 detailed developer interviews.
• Provides a complete replication package with source code (~5.3 KLOC), models, and analysis scripts.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small in file size (9.1 KB), suggesting it contains metadata and pointers rather than the full raw data.
• Row counts for individual CSV files are unknown, which may limit suitability assessment.

Provenance

Source

figshare, authored by Zijie Huang.

Collection Method

Likely collected from software repositories (MLCQ benchmark) and developer studies, as described in the replication package structure.

Time Range

null

Freshness

Last updated 2026-04-20 07:41:59

Geography

null