SketchOscent: towards a knowledge-based model and interactive visualisation of the odour space

Angélique Villière (1, 2) , Catherine Fillonneau , Carole Prost , Fabrice Guillet

1 GEPEA-MAPS2 - Matrices Aliments Procédés Propriétés Structure - Sensoriel
2 ONIRIS - École nationale vétérinaire, agroalimentaire et de l'alimentation Nantes-Atlantique

The representation of an odour space is a challenging issue for scientists and professionals working in odour-related activities. Many food and non-food sectors have created their own classification system and representations of the odorant space to manage objective communication about odours. More generally, many representations of the odorant space have been proposed. These initiatives turn the odorant space into easy-to-digest representations classifying hierarchically a restricted set of odorants in specific categories that provide a common frame of reference to professionals or sensory analysis judges. Nevertheless, the odour set is often restricted and these representations do not reflect the continuum of current odours which are not frozen in single section but on the contrary closely overlap. Thus, the related visualisations face with difficulties to represent such odour relationships. This work based on knowledge graphs from semantic web aims to propose a readable, interactive and publicly available representation of the aroma space, dedicated to olfactory and sensory practices. This representation relies on a large set of odours and takes into account overlapping categories. The knowledge-based model thus produced includes about 380 odours which are distributed in 20 main categories, which are decomposed into 117 subcategories. Seventy-seven odours and 26 categories obviously belong to multiple categories depending either on the multidimensional perception they are suggesting or on a conceptual positioning. The odour space is fully encoded into an rdf/owl Knowledge Graph, a semantic web standard language, which can be easily edited, stored, shared and requested. A hierarchical visual representation is derived to get a friendly and readable representation. The flexibility introduced by this representation enables to easily find out the relevant term(s) needed to describe an odorant perception. This approach, applied to food application, can be expanded to non-food sectors such as odour environment for example. SketchOscent is freely and publicly available at https://oniris-polytech.univ-nantes.io/sketchoscent.

Mots clés: odour space representation; human knowledge-based model; semantic representation; interactive visualisation; olfaction

Domaines: Sciences cognitives; Alimentation et Nutrition; Intelligence artificielle [cs.AI]; Algorithme et structure de données [cs.DS]; Biochimie [q-bio.BM]; Ingénierie des aliments; Chimie analytique; Machine Learning [stat.ML]

Angélique Villière, Catherine Fillonneau, Carole Prost, Fabrice Guillet. SketchOscent: towards a knowledge-based model and interactive visualisation of the odour space. Proceedings of the 16th Weurman Flavour Research Symposium, 2022, ⟨10.5281/zenodo.5948778⟩. ⟨hal-03556965⟩