Using Word Embeddings to Learn a Better Food Ontology

Area of Science:

• Food Science
• Computational Linguistics
• Bioinformatics

Background:

• Creating and maintaining food ontologies is labor-intensive and often lacks alignment with food science principles.
• Existing methods struggle with the complexity and scale of food knowledge representation.

Purpose of the Study:

• To develop a semi-supervised framework for automated food ontology population.
• To leverage word embeddings for enhanced food knowledge representation and ontology construction.

Main Methods:

• Utilized a semi-supervised framework for ontology population from an existing scaffold.
• Employed word embeddings trained on the Wikipedia corpus to capture food characteristics.
• Evaluated the generated ontology against the expert-curated FoodOn ontology.

Main Results:

• The word embeddings effectively captured latent relationships and characteristics of foods.
• Achieved an 89.7% improvement in precision compared to FoodOn (0.34 vs. 0.18).
• Reduced path distance (hops) by 43.6% between predicted and actual food instances (2.91 vs. 5.16).

Conclusions:

• High-dimensional food representations can effectively populate ontologies.
• The proposed method offers a more precise and efficient approach to food ontology construction.
• This work enables learning ontologies integrating contextual information from diverse sources.