Trusted Facts: Triplifying Primary Research Data Enriched with Provenance Information

Kai Schlegel, Sebastian Bayerl, Stefan Zwicklbauer, Florian Stegmaier, Christin Seifert, Michael Granitzer, Harald KoschEswc (satellite Events) 2013 · DOI

Abstract

In the current research landscape, primary data—raw numbers and benchmark results—is frequently trapped in unstructured silos such as PDFs and spreadsheets. This lack of interoperability forces researchers to manually normalize data, leading to inefficiencies and potential contradictions. This paper introduces a "triplification" pipeline designed to convert unstructured primary research data into structured, Linked Data. By combining automated extraction with semi-automatic semantic enrichment and comprehensive provenance tracking, the proposed system facilitates the creation of a "Linked Science Data Cloud" where research results are comparable, reliable, and verifiable.

Introduction

The accessibility of primary research data is a significant hurdle in scientific progress. Data published in formats like PDF or basic HTML is difficult to compare across studies, creating a "time sink" for researchers who must manually extract and standardize information. Furthermore, the absence of standardized data formats can lead to "contradicting facts" where discrepancies arise from non-standardized reporting rather than actual scientific disagreement. To address this, we propose a method to transform isolated tables into machine-readable RDF formats while maintaining a clear record of data provenance.

Methodology

The proposed system utilizes a multi-stage processing chain to "lift" data into the semantic web, as illustrated in Figure 1.

The proposed triplification pipeline for primary research data.

The pipeline consists of the following stages:

  1. Extraction and Transformation: Data is harvested from unstructured sources (XLS, CSV, PDF, HTML) and converted into an intermediate HTML table "pivot format."
  2. Semantic Enrichment: This stage involves two parallel tasks: concept disambiguation (identifying specific entities and metrics) and structural description (defining dimensions and scales).
  3. Crowdsourcing: To handle the nuances that automated tools might miss, the system incorporates human-in-the-loop refinement to manage the imprecision of automatic routines.
  4. Triplification: Enriched tables are transformed into RDF Data Cubes, a W3C standard compatible with Online Analytical Processing (OLAP), allowing for multi-dimensional querying.
  5. Provenance Tracking: Using the W3C PROV ontology, the system records the entities, actions, and agents (human or algorithmic) involved in every transformation, ensuring a transparent digital paper trail.

Results and Implementation

The authors have developed the CODE Data Extraction prototype, hosted at the University of Passau. This prototype successfully implements the "five-star deployment scheme" for Linked Data, ensuring high interoperability. A key contribution is the introduction of the CODE Provenance Vocabulary, which extends the standard W3C PROV-O model to include workflow-specific semantics. The system is intended for integration with the TIRA (Testbed for Information Retrieval Algorithms) platform to manage evaluation results from CLEF challenges.

Discussion and Future Work

While the integration of crowdsourcing enhances accuracy, its scalability for massive datasets remains a challenge. Future work will focus on the full integration with the TIRA testbed and exploring ways to further automate the pipeline without sacrificing the "trust" established through human refinement and provenance tracking. By treating research results as OLAP-compliant data cubes, the scientific community can perform meta-analyses with the same efficiency as business intelligence operations.

Summary updated

BibTeX
@inproceedings{schlegel2013trusted,
  author = {Kai Schlegel and Sebastian Bayerl and Stefan Zwicklbauer and Florian Stegmaier and Christin Seifert and Michael Granitzer and Harald Kosch},
  title = {{Trusted Facts: Triplifying Primary Research Data Enriched with Provenance Information}},
  booktitle = {Eswc (satellite Events)},
  year = {2013},
  doi = {10.1007/978-3-642-41242-4\_37},
}
Figure