FONTO : Une nouvelle méthode de la fuzzification d'ontologies

Houda Akremi, Sami Zghal, Vianney Jouhet, Gayo DialloSPIRE 2016

Abstract: International audience

Introduction

In the Semantic Web, ontologies serve as the structural backbone for knowledge representation. However, classical ontologies are typically built on "crisp" logic, where an entity either belongs to a class or it does not. This binary approach is often insufficient for domains like medicine, where boundaries are frequently imprecise. For example, the transition from a "normal" to a "high" blood sugar level is not a discrete jump but a gradual shift. Experts often use linguistic terms like "slightly elevated" or "severe pain" to describe these nuances. To address this, the FONTO (Fuzzy for ONTOlogy) framework was developed to transform standard, rigid ontologies into fuzzy ontologies capable of handling real-world vagueness.

Methodology

The FONTO method automates the transition from crisp to fuzzy logic through a three-phase architecture designed to bridge the gap between rigid semantic structures and imprecise data.

The FONTO framework architecture, illustrating the three-phase process of ontology fuzzification.
  1. Extraction of Fuzzy Concepts: FONTO utilizes the WordNet lexical database to scan the ontology for adjectives and adverbs (e.g., "Normal," "High," "Severe"). If a concept is associated with a synset in WordNet that implies a degree or quality, it is flagged for fuzzification. This reduces the need for exhaustive manual intervention by domain experts.
  2. Determination of Fuzzy Classes: Once a concept is identified, the system defines its mathematical membership function. FONTO supports various shapes, including triangular, trapezoidal, Gaussian, and monotonic functions. These functions map raw data (e.g., a temperature of 38.5°C) to a "degree of truth" between 0 and 1.
  3. Calculation of Membership Degrees: In the final phase, the system processes real-world instances—such as patient data—and applies the selected membership functions to determine the specific degree of membership for each data point within a fuzzy category.

Evaluation and Results

The framework was evaluated using the Rhumato ontology, a domain-specific resource for Rheumatology (specifically Rheumatoid Arthritis) containing 468 concepts and over 2,400 axioms. The study focused on fuzzifying lab results, such as C-Reactive Protein (CRP) levels.

In a comparative analysis of patient reports:

  • Bilan 1: A CRP value of 4.1 resulted in a membership degree of 0.0 for the "Normal" category, indicating it fell outside the defined normal range for that specific model.
  • Bilan 2: A CRP value of 26.0 resulted in a membership degree of 1.0, indicating total membership in the "Normal" category based on the model's parameters.

By using WordNet, FONTO achieves a higher level of automation in identifying fuzzy candidates compared to previous methodologies that relied more heavily on manual expert analysis.

Discussion and Future Work

FONTO represents a significant step toward bridging the gap between semantic structures and imprecise data. By treating linguistic variables as bridges, the system allows for "semantic enrichment" using the fuzzyOwl tag, ensuring compatibility with standard OWL structures while allowing machines to reason with "shades of gray."

However, some limitations remain. Currently, human experts are still required to define the numerical bounds (parameters) for membership functions. Future research aims to integrate machine learning to automatically derive these values from historical patient records. Additionally, the authors plan to explore fuzzy ontology alignment to resolve heterogeneity between different medical knowledge bases and incorporate broader patient contexts, such as age and medical history, into the fuzzification process.

Summary updated

Abstract

International audience
BibTeX
@inproceedings{akremi2016fonto,
  author = {Houda Akremi and Sami Zghal and Vianney Jouhet and Gayo Diallo},
  title = {{FONTO : Une nouvelle méthode de la fuzzification d'ontologies}},
  booktitle = {SPIRE},
  year = {2016},
}
Figure