Abstract
Industrial applications of semantic technologies - in particular ontologies - include the integration of heterogeneous information sources and the management of information resources and services. Currently, the adoption of these technologies in large-scaled applications within enterprises is slowed down by their failure to meet some basic requirements of commercial applications. One of those is the support of sophisticated security policies for protecting the knowledge contained in ontologies and their instances from unauthorized use. In this paper, we propose a model-driven approach to enable access control for lightweight ontologies based on a role-based security model and well-known standard technologies.
Introduction
As industries like automotive manufacturing increasingly adopt semantic technologies to integrate massive, heterogeneous data sources, they encounter a significant roadblock: security. While relational databases benefit from mature, fine-grained access control mechanisms, ontologies often lack a standardized way to restrict access to specific "knowledge" for unauthorized users.
The challenge is twofold. First, industrial ontologies are typically "lightweight" but large-scale, necessitating high-performance reasoning. Second, the "inference problem" complicates security; even if a specific fact is hidden, a user might infer it from other visible data. Most existing solutions either ignore inference or require deep, custom modifications to complex reasoning software, which hinders adoption in real-world commercial settings. This paper proposes a model-driven framework to bring fine-grained, role-based access control to industrial ontologies by intercepting and rewriting SPARQL queries at a security proxy.
Methodology
The proposed approach employs a Model-Driven Architecture (MDA) to bridge the gap between data modeling and security policy.
Lightweight Metamodeling
The authors define a simplified metamodel for a subset of OWL (specifically OWL Lite⁻,ᵖ). This metamodel maps effectively to Datalog and relational structures, ensuring high performance in industrial settings.
Security Extension
The metamodel is extended with concepts from the eXtensible Access Control Markup Language (XACML). This allows administrators to define Role-Based Access Control (RBAC) policies (Permit/Deny) that refer directly to ontology elements like Classes, Properties, and Individuals.
SPARQL Query Rewriting
Instead of modifying the reasoner, a Security Proxy is introduced. When a user submits a SPARQL query, the proxy identifies the user's role, fetches relevant security policies, and rewrites the query by injecting FILTER and OPTIONAL clauses. For instance, if a user is forbidden from seeing a "RestrictedClass," the proxy adds a filter that checks the rdf:type of return variables and removes results that the reasoner identifies as belonging to that class.
Key Insights
- Delegating Inference to the Reasoner: By rewriting queries to include
OPTIONALandFILTERclauses, the system forces the reasoner to determine if an individual belongs to a hidden class. This avoids the need to pre-calculate every possible inference in the security proxy. - Model-Driven Flexibility: Utilizing metamodels (via the Eclipse Modeling Framework) allows security policies to be managed as Java objects and transformed into various formalisms like XACML or F-Logic.
- Conflict Resolution: The system implements a "Deny-Overrides-Permit" strategy to prevent accidental data leaks during policy maintenance.
- Formal Verification: The framework includes formal verification rules to assist administrators in identifying logical contradictions within security policies.
Results
The paper provides a conceptual and algorithmic proof of concept for inference-aware filtering. It demonstrates that the approach successfully hides individuals even when their membership in a restricted class is only discoverable through logical inference (e.g., via sub-classing or property domains). Crucially, the architecture is non-invasive, sitting in front of the database as a proxy, which allows any standard SPARQL-compliant reasoner to be used without modification.
Conclusion and Future Work
While query rewriting offers a clean, non-invasive solution, the addition of complex FILTER and OPTIONAL clauses may impact performance. Future work will focus on quantitative benchmarks for large-scale datasets and addressing potential information leaks from Cartesian product queries. Additionally, the authors plan to decompose the security proxy into standard XACML components (Policy Enforcement and Decision Points) to better align with enterprise IT architectures.
Summary updated