Reproduce “StaQC: A Systematically Mined Question-Code Dataset from Stack Overflow” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ziyu Yao, Daniel S. Weld, Wei-Peng Chen, Huan Sun
Researchers from The Ohio State University, University of Washington, and Fujitsu Labs of America developed StaQC, a systematically mined dataset of approximately 148,000 Python and 120,000 SQL question-code pairs from Stack Overflow. This dataset was created using a Bi-View Hierarchical Neural Network (BiV-HNN) that identifies standalone code solutions by processing both textual context and code content, achieving at least 15% higher F1-score and accuracy than heuristic methods.
Key insights
- The Bi-View Hierarchical Neural Network (BiV-HNN), which processes both text and code, significantly outperforms single-view models and heuristic baselines for identifying standalone code solutions.
- A hierarchical structure within the neural network proves more effective than flat models for understanding the relationship between textual context and code content.
- Incorporating question information into the code representation further enhances the model's ability to accurately identify relevant code snippets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.