Introduction
As governments worldwide move toward "open data" initiatives, the public sector information (PSI) landscape remains fragmented. While vast amounts of data regarding crime, health, and politics are technically available, they are often trapped in incompatible silos. This paper explores a case study in breaking down these silos by transforming disparate UK public datasets into a unified "Linked Data" framework. By leveraging geographical administrative boundaries as a universal "join point," the researchers developed a functional dashboard that allows citizens to explore localized data simply by entering a postcode.
The Challenge of Data Silos
The primary obstacle to effective public data utilization is not a lack of information, but rather its format and structure. The authors identify three critical hurdles:
- Format Heterogeneity: Data is published in various formats such as XLS, XML, and HTML, often lacking explicit semantic descriptions.
- Integration Fragility: Traditional "mashups" typically rely on screen scraping or fixed APIs, which are prone to breaking whenever a source website's layout changes.
- The "Join" Problem: Cross-domain querying—such as correlating regional crime rates with hospital waiting times—is difficult because datasets rarely share common identifiers.
Methodology: A Linked Data Approach
To address these challenges, the researchers adopted a bottom-up Linked Data strategy, moving away from rigid top-down schemas. The core of this methodology involved:
- Normalization: Converting all datasets into the Resource Description Framework (RDF), specifically utilizing the Turtle representation for human readability and machine processing.
- Ontology Reuse: Instead of creating new vocabularies, the project utilized established standards like FOAF (Friend-of-a-Friend) for modeling political figures and Dublin Core for general metadata. Crucially, the Statistical Core Vocabulary (SCOVO) was used to model multi-dimensional statistics like crime rates and mortality.
- The "Join Point" Strategy: The Ordnance Survey (OS) administrative geography served as the authoritative "anchor." By linking all datasets to OS identifiers via
owl:sameAsrelations, the team created an interconnected web of data. - Infrastructure: The project utilized 4store, a scalable RDF triple store, to manage the resulting billions of triples.
Data Sources and Integration
The study integrated five major UK-based sources into a master geographical framework. The diversity of these sources highlights the complexity of the integration task, as they spanned different domains and technical formats.
As shown in the table above, the sources varied significantly in their native formats, from HTML and XML to Excel spreadsheets. The Ordnance Survey data acted as the master dataset, providing the necessary administrative and voting region boundaries to link the others.
Key Insights and Results
The project successfully integrated five disparate domains into a single, searchable graph. A key finding was that geography serves as a universal key; postcodes and constituencies are the most intuitive way for citizens to interact with complex data.
However, the researchers also encountered a "semantic gap" in naming conventions. For example, manual alignment was required to link "Yorkshire and the Humber Region" with the OS's "Yorkshire & the Humber," demonstrating that automated alignment still struggles with human naming variations. Furthermore, while RDF provides immense flexibility for data modeling, the researchers noted that User Interfaces (UIs) are often rigid, creating a friction point where data must be specifically re-modeled for end-user visualization.
The final result was a client application where users could enter a postcode to receive a localized view of their MP’s voting record, expenses, and local crime or health statistics. The project also "gave back" by publishing the newly linked datasets to the Linked Open Data cloud.
Limitations and Future Directions
Despite its success, the study highlighted several areas for improvement. The reliance on manual string matching for geographical alignment is unscalable for larger datasets. Additionally, temporal misalignment—where different departments publish data for different fiscal or calendar years—makes longitudinal comparisons difficult. Future work should focus on automated data mining for schema alignment and more sophisticated backlinking services to maintain the currency of the data web.
Summary updated