The Smell Network

Abstract

Abstract. The smell of a molecule is subjective, because there is a variablity in its representative language. The reporting is done according to the vocabulary repertoire of human subjects and researchers concerned. The olfactory databases thus consist of molecules and their smell characteristics defined by words. In this paper, we have demonstrated a network based approach based on the words to understand the perceptual universe. We defined perceptual communities based on the normalized co-occurrence network and hence propose the perceptual classes. We find the characteristics of this perceptual social network. We have also proposed a generative LDA-based topic modeling approach for topic detection in olfactory databases. This is for the first time that an objective approach to defining perceptual classes has been carried out which confirms with many subjective analyses that has been done till now. This work may open new avenues towards understanding the relationship between language and olfaction besides objectively defining perceptual classes.