[Paper Review] An Evidential Path Logic for Multi-Relational Networks

Marko A. Rodriguez, Joe Geldart|ArXiv.org|Oct 8, 2008

Semantic Web and Ontologies15 references3 citations

TL;DR

This paper proposes an evidential path logic framework for multi-relational networks that enables non-bivalent, non-monotonic reasoning using algebraic operations over evidence tuples. It models knowledge as a network with directed edges labeled by support and opposition evidence, allowing efficient, approximate inference through path composition and filtering, with applications to RDF and the Semantic Web.

ABSTRACT

Multi-relational networks are used extensively to structure knowledge. Perhaps the most popular instance, due to the widespread adoption of the Semantic Web, is the Resource Description Framework (RDF). One of the primary purposes of a knowledge network is to reason; that is, to alter the topology of the network according to an algorithm that uses the existing topological structure as its input. There exist many such reasoning algorithms. With respect to the Semantic Web, the bivalent, monotonic reasoners of the RDF Schema (RDFS) and the Web Ontology Language (OWL) are the most prevalent. However, nothing prevents other forms of reasoning from existing in the Semantic Web. This article presents a non-bivalent, non-monotonic, evidential logic and reasoner that is an algebraic ring over a multi-relational network equipped with two binary operations that can be composed to execute various forms of inference. Given its multi-relational grounding, it is possible to use the presented evidential framework as another method for structuring knowledge and reasoning in the Semantic Web. The benefits of this framework are that it works with arbitrary, partial, and contradictory knowledge while, at the same time, it supports a tractable approximate reasoning process.

Motivation & Objective

• To address the limitations of traditional bivalent, monotonic reasoning in Semantic Web knowledge systems like RDFS and OWL.
• To enable reasoning under uncertainty by modeling knowledge as evidence-based truth values rather than absolute truth.
• To support tractable, approximate inference in large-scale knowledge networks under resource constraints.
• To provide a formal algebraic framework for path-based inference in multi-relational networks using evidence tuples.
• To enable integration with the Semantic Web via RDF reification and named graphs for representing evidential statements.

Proposed method

• Represents knowledge as a multi-relational network $M = (V, \mathbb{E})$, where $\mathbb{E}$ is a family of edge sets with categorical labels.
• Assigns each edge a support and opposition evidence tuple $\langle w^+, w^- \rangle \in [0,1] \times [0,1]$, forming an evidential network.
• Defines two binary operations: evidence addition ($+$) and evidence composition ($\cdot$), forming an algebraic ring over the network.
• Applies path algebra to compute evidence along paths, such as $\mathcal{A}^{\texttt{wrote}}_{(t+1)} = \mathcal{A}^{\texttt{wrote}}_{(t)} \cdot \mathcal{A}^{\texttt{wrote}}_{(t)\top}$ for self-citation inference.
• Uses filters like $n(\mathbf{I})$ to exclude self-loops in coauthorship inference, ensuring no vertex is its own coauthor.
• Employs a clip function $c$ to convert non-zero evidence to $\langle 1,0 \rangle$, enabling evidence-based path propagation and revision.

Experimental results

Research questions

• RQ1 How can non-bivalent, non-monotonic reasoning be formalized in multi-relational knowledge networks?
• RQ2 What algebraic structure supports efficient, approximate inference under uncertainty and resource constraints?
• RQ3 How can path-based inference be composed and optimized in evidential networks without global computation?
• RQ4 In what ways can evidential logic be integrated into the RDF and Semantic Web stack for practical deployment?
• RQ5 How can contradictory or partial knowledge be represented and reasoned with using evidence tuples?

Key findings

• The framework supports reasoning with arbitrary, partial, and contradictory knowledge by representing truth as degrees of support and opposition.
• Evidence-based inference is computationally tractable through local path walks and filtering, avoiding expensive global matrix operations.
• The path algebra enables derivation of equivalent, more efficient inference paths, such as using $n(\mathbf{I})$ to filter self-coauthorship.
• Self-citations are used as revising evidence for the 'wrote' relation by applying the clip function to convert non-zero evidence to $\langle 1,0 \rangle$.
• Coauthorship inference is computed via $\mathcal{A}^{\texttt{coauthor}}_{(t+1)} = \left(\mathcal{A}^{\texttt{wrote}}_{(t)} \mathcal{A}^{\texttt{wrote}}_{(t)\top}\right) \circ n(\mathbf{I}) + \mathcal{A}^{\texttt{coauthor}}_{(t)}$, excluding self-loops.
• The system can be natively represented in RDF using quad-form reification, where evidence tuples are attached to statements via named graphs.