Dataset Reuse: Toward Translating Principles to Practice.

Area of Science:

• Data Science
• Information Science
• Computer Science

Background:

• The web hosts millions of datasets, but their reuse potential is often limited by a lack of empirical understanding.
• Existing guidelines for dataset reusability lack practical, actionable insights for data publishers and tool designers.

Purpose of the Study:

• To empirically investigate features that enhance dataset reusability.
• To develop a predictive model for dataset reusability using machine learning techniques.
• To bridge the gap between theoretical principles and practical implementation of data reuse functionalities.

Main Methods:

• Conducted a literature review to identify potential dataset reuse features.
• Created a corpus of over 1.4 million data files from more than 65,000 repositories on GitHub.
• Utilized GitHub's engagement metrics as proxies for dataset reuse and employed deep neural networks to build a predictive model.

Main Results:

• Identified key features correlating with dataset reuse on GitHub.
• Developed an initial deep neural network model capable of predicting dataset reusability.
• Demonstrated a practical approach to quantifying and predicting data reuse.

Conclusions:

• GitHub engagement metrics can serve as effective proxies for dataset reuse.
• A predictive model for dataset reusability is feasible and valuable for data publishers.
• Actionable insights are needed to translate data reuse principles into practical functionalities.