[Paper Review] Learning Vague Concepts for the Semantic Web

Paolo Pareti, Ewan Klein|arXiv (Cornell University)|Oct 19, 2014

Semantic Web and Ontologies32 references3 citations

TL;DR

This paper proposes VAGO, a system that models vague concepts in ontologies using adjustable numerical adaptors in Description Logic, enabling automatic learning and local repair of inconsistencies during ontology evolution. By treating threshold values as learnable parameters, VAGO dynamically updates vague concepts like 'close' or 'expensive' through training axioms, achieving 96% accuracy in adapting to target thresholds with minimal deviation from optimal values.

ABSTRACT

Ontologies can be a powerful tool for structuring knowledge, and they are currently the subject of extensive research. Updating the contents of an ontology or improving its interoperability with other ontologies is an important but difficult process. In this paper, we focus on the presence of vague concepts, which are pervasive in natural language, within the framework of formal ontologies. We will adopt a framework in which vagueness is captured via numerical restrictions that can be automatically adjusted. Since updating vague concepts, either through ontology alignment or ontology evolution, can lead to inconsistent sets of axioms, we define and implement a method to detecting and repairing such inconsistencies in a local fashion.

Motivation & Objective

• To address the challenge of representing and evolving vague concepts—such as 'near' or 'expensive'—in formal ontologies used in the Semantic Web.
• To overcome the limitations of static fuzzy or probabilistic models by modeling vagueness as a dynamic, learnable property.
• To detect and resolve inconsistencies that arise when updating vague concepts during ontology alignment or evolution.
• To develop a method that modifies adaptor values rather than deleting axioms, preserving semantic integrity during repair.

Proposed method

• Vague concepts are modeled using adaptors—metavariables representing threshold values in DL axioms, such as 'cost > X' for 'expensive'.
• The system uses a learning mechanism that iteratively adjusts adaptor values by validating the ontology against training axioms.
• Inconsistencies from combining original and training axioms are interpreted as misalignments, prompting targeted adaptation of threshold values.
• A sigmoid function is applied during learning to reduce the influence of outliers on adaptor evolution.
• The method operates within OWL 2 and Description Logic, using datatype and cardinality restrictions to express vague concepts.
• Repair is performed locally by modifying adaptor values without deleting axioms, ensuring preservation of new knowledge.

Experimental results

Research questions

• RQ1 How can vague concepts in formal ontologies be represented in a way that allows for dynamic, context-sensitive adaptation?
• RQ2 What mechanism enables automatic learning of threshold values for vague concepts like 'close' or 'expensive'?
• RQ3 How can inconsistencies arising from ontology updates be detected and resolved without discarding axioms?
• RQ4 To what extent can the system align two ontologies by evolving vague concept definitions through training data?

Key findings

• The system achieved a final adaptor value for 'CloseDistance' that differed from the correct value by only 4% of the standard deviation, indicating high accuracy in learning.
• The adaptor for 'profitable' (number of bookshops) deviated by 162% of the standard deviation from the correct value, suggesting room for improvement in convergence.
• The learning process successfully reduced the impact of outlier data points on adaptor evolution through the use of a sigmoid function.
• The method effectively repaired inconsistencies by modifying adaptor values rather than removing axioms, preserving semantic content.
• Simulations demonstrated that VAGO can evolve vague concept definitions to align with training data, improving extension-based ontology alignment.
• The approach enables consistent, incremental evolution of ontologies with vague concepts, avoiding the pitfalls of rigid or destructive repair strategies.