TL;DR
This paper introduces a novel architecture for persistently storing and querying fuzzy knowledge in the Semantic Web. By serializing fuzzy information into standard RDF triples and integrating a fuzzy reasoning engine (FiRE) with an RDF store (Sesame), they enable expressive fuzzy queries on large-scale knowledge bases, demonstrated with a real-world casting application.
What Problem Does This Paper Address?
The Semantic Web aims to make web data machine-understandable, often relying on ontologies (formal representations of knowledge) expressed in languages like OWL. However, real-world information is frequently vague or uncertain. While fuzzy Description Logics (DLs) offer a way to represent this imprecision, existing RDF storage and querying systems struggle to handle it effectively. This paper addresses the gap of persistently storing and querying expressive fuzzy knowledge bases within the Semantic Web framework, moving beyond in-memory reasoning to practical, scalable solutions.
Methods
The authors propose a three-pronged approach:
- Fuzzy Information Serialization: They introduce a novel way to represent fuzzy information as standard RDF triples without relying on extensions to RDF stores or specific datatypes. This involves using blank nodes to represent fuzzy assertions, including membership degrees and inequality types.
- Fuzzy Reasoning Engine Integration: They integrate FiRE, a fuzzy DL reasoner for the f-SHIN language, with the Sesame RDF store. This allows FiRE to leverage the storage and querying capabilities of Sesame while providing fuzzy reasoning services.
- Expressive Fuzzy Querying: They extend existing query languages (like SPARQL) to support fuzzy queries. This includes translating fuzzy assertions into RDF triples and enabling queries that consider membership degrees and thresholds, such as Conjunctive Threshold Queries (CTQs) and General Fuzzy Conjunctive Queries (GFCQs).
Datasets
The paper evaluates their proposed architecture using a real-world industrial application scenario: a database of 2140 human models from a production company. This database contained information on various characteristics like height, age, body type, etc. The authors fuzzified certain numerical attributes (age and height) and used this enriched dataset to demonstrate their system's capabilities.
Key Insights and Ideas
- Bridging the Gap: The core insight is that fuzzy knowledge, often handled in-memory by specialized reasoners, can be effectively managed using standard RDF triple stores by clever serialization. This opens the door to leveraging existing Semantic Web infrastructure for fuzzy data.
- Expressiveness without Extensions: The serialization method is crucial because it allows existing RDF tools to store fuzzy information without requiring them to be modified or extended, promoting interoperability.
- Practical Fuzzy Querying: The integration with Sesame and the extension of query languages demonstrate that complex fuzzy reasoning and querying can be performed in a practical setting, not just in theoretical research.
- Real-World Applicability: The casting scenario highlights how fuzzy information is inherent in many real-world applications and how a system like this can unlock valuable implicit knowledge that would be hard to extract otherwise.
Key Results
The proposed architecture was successfully implemented and evaluated. The system demonstrated its ability to:
- Store a fuzzy knowledge base of 29460 explicit assertions and 2430 implicit assertions (extracted via GLB tests) in a Sesame repository, resulting in approximately 530,000 triples.
- Perform fuzzy queries, such as finding "young female models with beautiful legs and eyes as well as long hair," which would be difficult with traditional databases.
- Achieve query response times that, while potentially longer than crisp querying, are significantly faster than manual browsing by producers, making it a practical tool for industrial applications. For example, a query involving multiple fuzzy conditions took between 894 and 4543 milliseconds depending on the repository size and type.
Limitations and Future Work
The authors acknowledge several limitations:
- Incompleteness: The system is not a sound and complete query answering system for f-SHIN. Queries involving transitive or inverse roles are not fully evaluated, as these are complex reasoning tasks even for crisp DLs.
- Performance: While promising, query answering times can still be several seconds longer than crisp querying due to post-processing steps for GFCQs or lengthy SPARQL queries for CTQs.
- Future Work: The authors plan to further investigate different ways of performing queries using expressive fuzzy description logics and compare their architecture with approaches using fuzzy DL-lite ontologies and approximation methods.
Summary updated