On the Importance of a Research Data Archive

Benedict Wright, Oliver Brunner, Bernhard NebelAAAI 2018 · DOI

Introduction

In modern academia, research data often suffers from "digital rot." When staff members depart or projects conclude, critical datasets frequently vanish into a fragmented landscape of personal laptops, flash drives, and unorganized network drives. This lack of structured preservation leads to a local reproducibility crisis: labs may possess data but lose the ability to locate it or understand the specific code versions that produced certain results. Furthermore, major funding agencies, such as the DFG and the European Commission, now mandate that research data be preserved for at least a decade. To address these challenges, we introduce OntoRAIS, an archival system designed specifically for researchers to link code, data, and publications into a durable, retrievable knowledge base.

System Architecture and Methodology

OntoRAIS prioritizes the researcher's workflow by utilizing Ontology-Based Data Access (OBDA). Unlike traditional archival tools designed for librarians, OntoRAIS uses semantic ontologies to define relationships between objects.

OntoRAIS System Architecture

As shown in the architecture above, the system integrates several components:

  • Semantic Linking: Using the QUEST library, files are linked via properties such as isImplementedBy or isAuthorOf. This allows the system to perform automated reasoning, such as flagging entries missing essential documentation.
  • Dual-Interface Approach: A web interface handles metadata annotation, while a desktop client (utilizing SFTP and Git) facilitates high-speed transfers for large datasets.
  • Project-Centric Organization: The system mirrors real-world lab structures, organizing data into projects and sub-projects, complete with milestone tracking to simplify reporting to grant agencies.

The Researcher Workflow

The core design philosophy of OntoRAIS is to lower the barrier to entry for non-experts. The archival process follows a streamlined path designed to integrate with typical research activities.

The OntoRAIS Archival Workflow

The workflow consists of five key stages: uploading data, annotating with metadata, assigning to specific projects, linking to related files (e.g., connecting a plot to the source code that generated it), and finally sharing the data according to defined access tiers. These tiers range from Restricted (author-only) to Public (open access).

Evaluation and Key Results

The system was evaluated through a pilot study involving seven research groups across diverse fields, including epilepsy research, robotics, and computer science. Feedback from 21 participants (professors, researchers, and students) highlighted several key findings:

  • Usability: The interface was praised for being intuitive and streamlined, successfully reducing the technical overhead for non-archivists.
  • The Incentive Gap: Despite the improved UI, users noted that manual metadata entry remains a significant workload. The study concluded that successful adoption requires either "top-down" mandates from supervisors or stronger "bottom-up" incentives.
  • Validation through Reasoning: The ontology-based "sanity checks" proved effective. By prompting researchers to link results to source code or experiments, the system ensures the long-term quality and utility of the archive.

Conclusion and Future Work

OntoRAIS demonstrates that a researcher-centric, semantically-linked archive can effectively combat data loss in academic labs. Future development will focus on reducing the manual workload through automated metadata ingestion and gamification. Additionally, the team plans to implement full-text search across the archive to further enhance data discoverability.

Summary updated

BibTeX
@inproceedings{wright2018importance,
  author = {Benedict Wright and Oliver Brunner and Bernhard Nebel},
  title = {{On the Importance of a Research Data Archive}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.1609/aaai.v32i1.11395},
}
Figure