Investigating temporal sensory data via a graph theoretic approach

Abstract

A graph theoretic approach is applied to investigate perception dynamics in the Sensometrics 2018 Data Analysis Workshop data sets. First, temporal check-all-that-apply (TCATA) data are investigated: ten sensory attributes are modelled as vertices and concurrent perception as edges, the size and width of each corresponding to the momentary elicitations and the concurrent elicitations, respectively. A temporal dominance of sensations (TDS) data set does not lend itself as readily to this type of analysis because the task for each assessor involves selection of only one dominant attributes at a time. For this reason, TDS data sets considered in this workshop are (re-)analyzed within localized temporal ranges, considering dominance elicitations within localized temporal ranges to be adjacent-dominant. In TDS by Modality (M-TDS), assessors taste the sample to evaluate the five taste and flavour attributes via TDS, then retaste the sample to evaluate the five texture attributes via TDS. M-TDS results from these two separate tasting timelines are adjoined as if they arose from a single evaluation timeline, which ignores systematic biases that may be introduced by the evaluation protocol (e.g. texture attributes are always evaluated with carryover effects from the first tasting). Data from each of these temporal sensory method are then analyzed using a graph theoretic approach. Cliques and attribute structures are identified. TCATA and TDS results are more similar to one another than either is to M-TDS results. Implications of method choice and various data pre-processing decisions are discussed.