Ontology-Based Data Access: A Survey

Guohui Xiao, Diego Calvanese, Roman Kontchakov, Domenico Lembo, Antonella Poggi, Riccardo Rosati, Michael ZakharyaschevIJCAI 2018 · DOI

TL;DR

This survey provides a comprehensive overview of Ontology-Based Data Access (OBDA), a paradigm that uses ontologies to provide a user-friendly semantic layer over complex data sources like relational databases. It details the core OBDA framework, query answering techniques, mapping management, and various extensions, highlighting its importance for making data accessible and actionable for non-expert users.

What Problem Does This Paper Address?

In today's data-rich environment, accessing and integrating information from diverse and often complex data sources (like large relational databases) is a significant challenge. Users, especially those without deep technical expertise in database structures and query languages (like SQL), struggle to formulate queries that retrieve the specific information they need. This is exacerbated by the fact that databases are typically designed for application support, not for intuitive domain-level querying, leading to data redundancy, obscurity, and incompleteness. OBDA aims to bridge this gap by introducing a conceptual, semantic layer (an ontology) that users can query using familiar terms, abstracting away the underlying data complexity.

Methods

The paper outlines the fundamental OBDA framework, which consists of:

  • Ontology (O): A conceptual model of the domain, providing a high-level vocabulary.
  • Data Source Schema (S): The schema of the underlying data sources (e.g., relational database schemas).
  • Mapping (M): Declarative rules that connect the ontology's concepts and properties to the data source's structures.

The core process involves:

  1. Query Formulation: Users pose queries against the ontology.
  2. Ontology-Mediated Query (OMQ) Rewriting: The OBDA system translates the user's ontology query into an equivalent query over the ontology's vocabulary.
  3. FO-Rewriting: This OMQ is then rewritten into a First-Order (FO) logic query that can be executed directly over the data source. This is a crucial step, aiming for "FO-rewritability" to ensure query answering can be done efficiently.
  4. Unfolding: The FO query is "unfolded" using the mapping rules to produce a query in the native language of the data source (e.g., SQL).
  5. Execution: The native query is executed against the data source.
  6. Result Assembly: The results are returned to the user in a semantically meaningful way.

The paper also discusses various extensions and techniques, including:

  • OWL 2 QL: A profile of OWL 2 designed for OBDA, balancing expressiveness with the guarantee of FO-rewritability.
  • Mapping Languages: Such as R2RML, for defining connections between relational data and ontologies.
  • Query Answering Algorithms: Including techniques like PerfectRef and optimizations to handle the complexity of rewritings.
  • Mapping Management: Tools and techniques for analyzing, debugging, and maintaining mappings, including consistency and redundancy checks.
  • Extensions: For handling SPARQL, temporal data, spatial data, and more expressive ontologies.

Datasets

Not discussed in the paper. The paper focuses on the theoretical framework and general techniques, using examples like the Statoil (Equinor) subsurface exploration scenario to illustrate concepts rather than specific datasets for evaluation.

Key Insights and Ideas

  • Abstraction is Key: OBDA's primary value lies in abstracting away the complexity of underlying data sources, allowing domain experts to query data using familiar, high-level concepts.
  • The Challenge of Rewritability: Ensuring that ontology-mediated queries can be translated into executable queries over the data (FO-rewritability) is a central theoretical and practical challenge. This often requires careful selection of ontology and mapping languages.
  • Trade-offs in Expressiveness: More expressive ontologies or mapping languages can capture richer semantics but may sacrifice the guarantee of efficient, FO-rewritable query answering.
  • Mapping is Crucial (and Difficult): The mapping layer is the bridge between the ontology and the data. Its creation, maintenance, and analysis are complex but critical for the system's success.
  • Beyond Virtual Access: While OBDA is often "virtual" (no data is stored beyond the original sources), materializing parts of the mapping can improve performance.
  • Evolving Landscape: OBDA is a dynamic field with ongoing research into handling streaming data, spatial/temporal aspects, and integrating with advanced analytics.

Key Results

The paper is a survey and does not present novel quantitative results. Instead, it synthesizes key theoretical results and practical considerations:

  • FO-Rewritability Guarantees: For certain ontology languages (like OWL 2 QL) and mapping types, Conjunctive Queries (CQs) are guaranteed to be FO-rewritable, enabling efficient query answering with data complexity akin to standard database queries (AC⁰).
  • Complexity of Rewritings: Rewritings can be exponentially large in the worst case, even for decidable fragments, necessitating optimization techniques.
  • Mapping Analysis Complexity: Deciding consistency and redundancy of mappings can be done with complexity comparable to standard ontology reasoning tasks.
  • Practical Adoption: OBDA systems have been adopted in industrial projects (e.g., Statoil, Siemens) and use cases, demonstrating their real-world applicability.

Limitations and Future Work

Limitations: * Expressiveness vs. Complexity: A persistent challenge is balancing the expressive power of ontologies and mapping languages with the tractability of query answering. Highly expressive languages can lead to non-FO-rewritable queries or computationally expensive reasoning. * Practical Implementation: While theoretical guarantees exist, the practical performance of complex rewritings and the efficiency of current RDMSs in handling them remain areas for improvement. * Read-Only Assumption: Most OBDA frameworks have historically been considered read-only, with limited support for handling data updates or changes to the ontology/mappings.

Future Work: * Data Quality Assessment: Extending OBDA to formally assess data quality dimensions beyond consistency (e.g., completeness, currency). * Handling Updates: Developing robust mechanisms for OBDA systems to react to insertions, deletions, and modifications of data and axioms. * Benchmarking: Creating more challenging and tuneable benchmarks to rigorously evaluate OBDA systems and optimization techniques. * Non-Uniform OBDA: Developing systems that can handle expressive ontologies and mappings while still providing efficient query answering, potentially through a combination of rewriting strategies. * Streaming Data Integration: Enhancing OBDA to effectively process and integrate real-time streaming data. * Data Analytics Integration: Extending OBDA to support not just data access but also advanced data analytics tasks on numerical and temporal data. * Automated Mapping Generation: Further research into bootstrapping ontologies and mappings from databases.

Summary updated

BibTeX
@inproceedings{xiao2018ontology,
  author = {Guohui Xiao and Diego Calvanese and Roman Kontchakov and Domenico Lembo and Antonella Poggi and Riccardo Rosati and Michael Zakharyaschev},
  title = {{Ontology-Based Data Access: A Survey}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.24963/ijcai.2018/777},
}
Figure