[Paper Review] NdFluents: A Multi-dimensional Contexts Ontology

José M. Giménez‐García, Antoine Zimmermann+1 more|arXiv (Cornell University)|Sep 22, 2016

Semantic Web and Ontologies7 references3 citations

TL;DR

NdFluents proposes a generic, extensible ontology for modeling multi-dimensional contextual metadata in RDF/OWL, extending Welty and Fikes' 4dFluents temporal fluents model to support arbitrary context dimensions such as provenance, confidence, and trust. It enables complete OWL reasoning over contextualized statements by representing context as structured, reified entities, allowing assertions about entities within specific contextual slices.

ABSTRACT

Annotating semantic data with metadata is becoming more and more important to provide information about the statements being asserted. While initial solutions proposed a data model to represent a specific dimension of meta-information (such as time or provenance), the need for a general annotation framework which allows representing different context dimensions is needed. In this paper, we extend the 4dFluents ontology by Welty and Fikes---on associating temporal validity to statements---to any dimension of context, and discuss possible issues that multidimensional context representations have to face and how we address them.

Motivation & Objective

• Address the lack of a general framework for representing multiple context dimensions (e.g., time, provenance, confidence) in semantic data.
• Extend the 4dFluents ontology—originally designed for temporal fluents—to support arbitrary contextual dimensions.
• Enable complete OWL reasoning over contextualized statements by modeling context as first-class entities.
• Provide a modular, extensible ontology that supports combining multiple context dimensions in a single knowledge representation.
• Support contextual metadata not only on statements but also on entities, enabling context-aware data retrieval and reasoning.

Proposed method

• Adopt a perdurantist (4D) approach, treating entities as temporal parts, and generalize this to any context dimension by introducing context-specific 'context parts'.
• Define a generic ontology structure where each context dimension (e.g., time, source, confidence) is modeled as a distinct type of contextual part linked to the original entity.
• Use OWL 2 functional syntax to define classes (e.g., ContextPart, TemporalPart), object properties (e.g., fluentProperty, contextOf), and axioms for modularity and extensibility.
• Represent contextual metadata via reification-like structures, where a statement is asserted about a context-part rather than the original entity, preserving FOL topology.
• Introduce a mechanism to link context parts to original entities via functional properties (e.g., contextOf), enabling retrieval of all facts within a given context.
• Apply the model to a real-world use case in the WDAqua project, publishing the ontology and example data following Web of Data best practices.

Experimental results

Research questions

• RQ1 How can a single ontology model support multiple, arbitrary context dimensions (e.g., time, source, confidence) in a unified and extensible way?
• RQ2 What are the key challenges in combining multiple contextual dimensions in a semantic knowledge graph while preserving reasoning capabilities?
• RQ3 Can a generalized fluents-based model support complete OWL inference over contextualized statements, including transitive, symmetric, and inverse properties?
• RQ4 How can contextual metadata be attached not only to statements but also to entities themselves, enabling context-aware data retrieval?
• RQ5 What are the trade-offs in expressivity, reasoning support, and data size when using a multi-dimensional fluents model compared to alternatives like N-ary relations or reification?

Key findings

• NdFluents is the first known generic extension of 4dFluents to support arbitrary context dimensions beyond time, enabling modular extension for provenance, confidence, and trust.
• The model supports complete OWL reasoning over contextualized statements, including transitive, symmetric, and inverse properties, unlike reification or singleton property patterns.
• Contextual metadata can be attached not only to statements but also to entities, allowing retrieval of all facts associated with a specific entity-context combination.
• The approach enables efficient querying for all information within a given context (e.g., all facts about Paris with Wikipedia as source and high confidence), improving data provenance and trustworthiness in question-answering systems.
• While the model increases data size due to additional triples, it provides a robust, formal foundation for context-aware knowledge representation that supports complex reasoning and extensibility.