Ontology-Mediated Query Answering: Harnessing Knowledge to Get More from Data

Meghyn BienvenuIJCAI 2016

Introduction

In the era of big data, the challenge is often not just having the data, but accessing it in a way that makes sense. Traditional database systems are literal; they only return what is explicitly stored. Meghyn Bienvenu’s 2016 survey on Ontology-Mediated Query Answering (OMQA) explores a sophisticated solution: using high-level conceptual models—ontologies—to bridge the gap between user queries and raw data structures.

The Semantic Bridge: How Ontologies Enhance Data Access

Standard databases are limited by their schema. If a medical database records "Hodgkin’s lymphoma" but a researcher queries for "cancer patients," the system will fail to return a match unless the relationship is explicitly coded. OMQA introduces an ontology—a formal set of rules and hierarchies—that sits atop the data.

Example of Description Logic axioms defining an ontology.

As shown in the figure above, an ontology defines relationships such as class hierarchies (e.g., a Professor is a member of the Faculty) and property constraints (e.g., if someone teaches, they must be Faculty). These rules allow the system to infer that a specific instance of "Hodgkin’s lymphoma" is indeed a type of "cancer," enabling more intelligent and comprehensive search results.

Overcoming Technical Hurdles

While the concept of OMQA is powerful, implementing it at scale presents two significant challenges: computational complexity and data inconsistency.

1. Computational Complexity and Query Rewriting

Combining complex logic with massive datasets can lead to an "exponential blowup," making queries prohibitively slow. To address this, researchers use Query Rewriting. Instead of reasoning over the data in real-time, the system transforms the user’s query and the ontology into a more complex query (often in SQL or Datalog). This rewritten query is then executed by standard, optimized database engines. The use of Non-recursive Datalog (NDL) has proven particularly effective in keeping these rewritings manageable in size.

2. Managing "Dirty" Data

Real-world data is rarely perfect. It often contains contradictions that violate the ontology's rules—for instance, a record might list an individual as both a "Professor" and a "Fellow" when the ontology explicitly states these roles are disjoint. Standard logic fails in the face of such contradictions.

OMQA employs Inconsistency-Tolerant Semantics to handle these scenarios:

  • IAR (Intersection of All Repairs) Semantics: A cautious approach that only returns answers that are true across every possible consistent subset of the data.
  • Brave Semantics: An optimistic approach that returns answers true in at least one consistent subset.
  • SAT Solvers: Systems like CQAPri leverage SAT solvers to efficiently navigate these logical puzzles and determine valid answers despite inconsistencies.

Real-World Applications

The theoretical advancements in OMQA are already seeing practical application in large-scale environments:

  • SNOMED CT: This massive medical ontology is used to integrate disparate patient data across healthcare systems.
  • The Optique Project: Industrial giants like Statoil and Siemens use OMQA to query complex technical data across multiple heterogeneous sources, demonstrating the scalability of these methods.

Key Insights and Future Directions

The research highlights a critical "sweet spot" in logic: finding languages that are expressive enough to be useful but simple enough to run on existing database technology. For many common ontologies, it is possible to rewrite queries into standard First-Order (FO) logic, ensuring high performance.

However, several frontiers remain:

  • Explanation Services: Users need to understand why the system returned a specific result, requiring better transparency.
  • Natural Language Integration: Bridging the gap between human language and formal queries remains a priority.
  • Automated Construction: Using machine learning to help build ontologies could reduce the manual labor currently required.

By continuing to refine these "semantic bridges," OMQA promises to make the vast oceans of digital data more accessible, reliable, and meaningful for experts and non-experts alike.

Summary updated

BibTeX
@inproceedings{bienvenu2016ontology,
  author = {Meghyn Bienvenu},
  title = {{Ontology-Mediated Query Answering: Harnessing Knowledge to Get More from Data}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2016},
}
Figure