CLV3W: A clustering around latent variables approach to detect panel disagreement in three-way conventional sensory profiling data

Abstract

a b s t r a c tTo detect panel disagreement, we propose the clustering around latent variables for three-way data (CLV3W) approach which extends the clustering of variables around latent components (CLV) approach to three-way data typically obtained from a conventional sensory profiling procedure (i.e., assessors rating products on various descriptors). The CLV3W method groups the descriptors into Q clusters and estimates for each cluster an associated latent sensory component such that the attributes within each cluster are as much related (i.e., highest squared covariance) as possible with the latent component. Simultaneously, for each latent sensory component separately, a system of weights is estimated that yields information regarding the extent to which an assessor (dis)agrees with the rest of the panel according to the latent sensory component under study. Our new approach is illustrated with a dataset pertaining to Quantitative Descriptive Analysis applied to cider varieties. It is shown that CLV3W, as opposed to related approaches, is able to detect differential panel disagreement on various latent sensory components.

Three‐way clustering around latent variables approach with constraints on the configurations to facilitate interpretation

“…Without loss of generality, we assume that all X j ( j = 1,…, J ) are column‐wise centered (i.e., across the elements of the first mode). Other pre‐processing steps, such as standardization or block scaling, can be applied depending on the dataset under study 11 …”

“…In the context of two‐way data, the clustering around latent variables ( CLV ) approach 8 has gained ground in sensometrics and chemometrics (for applications on spectral data, see Vigneau et al and Cuny et al 9,10 ). For clustering variables within the scope of three‐way data, Wilderjans and Cariou 11 proposed a clustering around latent variables approach ( CLV3W ) which extends CLV to a three‐way data array, with the variables referring to the second mode. The CLV3W analysis consists of grouping the variables into Q nonoverlapping clusters and determining for each cluster a latent configuration corresponding to a rank‐one Parafac model applied to the data of each cluster (i.e., the set of lateral data slices associated with the variables belonging to the cluster in question).…”

“…More precisely, CLV3W simultaneously partitions the variables and determines for each cluster of variables: (a) a latent component associated with the first mode (e.g., samples, subjects), (b) a vector of loadings reflecting the degree of closeness of each variable to this latent component, and (c) a system of weights corresponding to the third mode. CLV3W has been introduced in sensometrics in order to either optimally weight underlying sensory dimensions in the context of conventional sensory profiling data 11 or to identify consumer segments on the basis of their multivariate ratings of a set of samples 12 . With regard to longitudinal metabolomics data, CLV3W fully takes the higher‐order structure of the data into account as opposed to multiblock and multilevel approaches, which rely on unfolding the data leading to a substantial loss of information.…”

Blockwise simple component analysis via rotation, constraints or penalties, with an application to product × attribute × panelist data

“…Qannari, Wakeling, & MacFie (1995) and Bro, Qannari, Kiers, Naes and Frøst (2008) go beyond that in their proposal to use STATIS (Lavit, 1988) and Parafac (Harshman, 1970), respectively to assess individual panelists behavior. Wilderjans and Cariou (2016) follow up on this by proposing to use Clustered Parafac (Krijnen & Kiers, 1993), so as to get easier interpretable solutions. These methods can be very successful for the purpose of summarizing product features and panelist behavior.…”

Sensometric 2014: ‘Data that Works in the City that Works’

“…Special emphasis was placed on better understanding consumers and their preferences/tolerance ranges (Grygorczyk, Mhlanga, & Lesschaeve, 2015), their variability both in terms of liking behavior through an emphasis on panel disagreement (Wilderjans & Cariou, 2015) and by setting aside atypical consumers (Vigneau, Qannari, Navez, & Cottet, 2015), and their perceptions of samples in Projective Mapping (Vidal et al, 2015). With the development of social media, textual analysis is getting a lot of attention, and it is no surprise that new findings in that area were also presented (Deneulin & Bavaud, 2015).…”