You are on page 1of 10

Food Quality and Preference 15 (2004) 383–392

www.elsevier.com/locate/foodqual

Sensory mapping using Flash profile. Comparison with a


conventional descriptive method for the evaluation of the
flavour of fruit dairy products
Julien Delarue*, Jean-Marc Sieffermann
Laboratoire de Perception Sensorielle et Sensométrie, Ecole Nationale Supérieure des Industries Agricoles et Alimentaires,
1 Avenue des Olympiades, 91744 Massy Cedex, France

Received 15 June 2002; received in revised form 2 January 2003; accepted 7 June 2003

Abstract
Flash profile is a quick sensory profiling technique designed to meet industrial needs. It is based on the combination of free choice
profiling and a comparative evaluation of the whole product set. This study aimed at comparing Flash profile with conventional
profiling. This has been done with two product spaces: strawberry blended yoghurts and apricot ‘‘fromages frais’’, both from the
French market. Multidimensional data treatment such as Generalised Procrustes Analysis and Canonical Variate Analysis (CVA)
allowed us to assess the correspondence between the sensory maps obtained with both methods. In the case of strawberry yoghurts,
very similar results were obtained with the two methods. On the other hand the sensory positioning of the apricot fresh cheeses
somewhat differed between the methods. For both product sets, Flash profile was slightly more discriminating than the conven-
tional profile. Advantages and limits of Flash profile are discussed.
# 2003 Elsevier Ltd. All rights reserved.
Keywords: Descriptive analysis; Procrustes; Canonical variate analysis; Method comparison

1. Introduction process ideally aims at providing a subject-independent


description, free of hedonic judgments. Quantification is
Sensory profiling methods are used in the industry to performed on descriptive attributes that are clearly
draw up the sensory ID of a set of products. This can be identified. Since human subjects are not equally sensi-
done, for example, to investigate the sensory properties tive to sensory stimuli nor equally discriminant regard-
of a range of existing products from a given market or ing the attributes, a panel of judges is used in order to
to assess the sensory impact of a new formulation, a stabilise the generated description.
change of fabrication process or packaging. More gen- Since the first developments of sensory profiling
erally speaking, sensory profiling is increasingly viewed methods in the 1950s, sensory scientists from academia
as a way to explain and possibly anticipate consumer and the food and the flavour industry people have
preferences. In this respect, the primary purpose of sen- developed variations of the original techniques.
sory profiling is implicitly to describe and quantify Recently, Sieffermann (2000, 2002) suggested to com-
inter-product sensory differences. Carrying out a sen- bine free choice profiling with a comparative evaluation
sory profile therefore aims at positioning the products in of the product set in a technique named Flash profile.
a multivariate sensory space defined by a relevant com- The goal of our experiment was to compare Flash pro-
bination of sensory attributes. file with a well-proven conventional profiling method.
Contrary to an evaluation performed by a single Flash profile has been previously run to describe a set of
expert (e.g. an oenologist or a perfumer), the profiling red fruit jams and proved to be satisfactory as com-
pared to conventional profile (Dairou & Sieffermann,
* Corresponding author. Tel.: +33-1-6993-5010; fax: +33-1-6993-
2002) which makes it a potentially useful tool for rapid
5174. market analyses. However, in that study the descriptive
E-mail address: delarue@ensia.inra.fr (J. Delarue). task was quite easy to achieve due to large differences
0950-3293/03/$ - see front matter # 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0950-3293(03)00085-5
384 J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392

between the products. The authors therefore pointed (Stone, Sidel, Oliver, Woosley & Singleton, 1974), the
out that it would be necessary to compare the two Quantitative Flavor Profiling (Stampanoni, 1993) and
approaches with more similar products. In the present the SpectrumTM method (Meilgaard, Civille, & Carr,
study, we chose to work with two product spaces that 1999).
would be similar to those encountered in industry, that The time necessary to carry out such an analysis may
is to say branded products from the French market that vary widely and depends in the first place on the objec-
belong strictly to the same family of products (i.e. direct tives of the study in terms of precision and sensitivity. It
competitors’), and which were thus expected to be very also depends on the task difficulty, which is conditioned
similar. The dairy product sector is amongst the most by the product space (products may be well distin-
competitive in the French food industry, providing us guishable or very confusable) and by the sensory fields
with many eligible types of products. We came to the under investigation. It is indeed usually easier to
choice of two product spaces: strawberry blended fruit describe texture differences than aroma differences for
yoghurts on one hand and apricot ‘‘fromages frais’’ (fresh which our concepts are fuzzier. In any cases, the first
cheeses) on the other hand. In addition, we decided to two steps (language development and training) are the
restrict the evaluation to the description of flavour. We most time consuming. They may last from few weeks to
used two distinct panels that were respectively dedicated several months.
to the conventional profile and to the Flash profile.
The comparison between the methods will be based 1.1.2. Limitations of the conventional approach
on the following criteria: A weakness inherent in regular descriptive analysis
is related to the use of attributes that are common to
1. Degree of similarity between the sensory spaces. all the judges. Indeed, the ability of subjects to commu-
In other words: is the relative product position- nicate their sensory perceptions on the same vocabulary
ing identical from one method to the other? bases is sometimes doubtful, even after thorough train-
Consequently, the discriminating power of each ing. It is well-known for example that a single stimulus
panel will also be investigated. can be perceived quantitatively and qualitatively
2. Richness of the semantic information yielded in different from one subject to the other, especially when
both cases and correspondence of the descriptive chemical senses are involved (Lawless, 1999). Con-
components. sequently, the forced use of a common language may in
3. Practical aspects of the techniques. These aspects the worst case result in the situation where a given
directly result from the methodological choices attribute actually stands for differing individual con-
and are therefore discussed in the introduction cepts. Alternatively one attribute may represent the lowest
below. common denominator for all the judges, impoverishing
greatly the sensory information.
1.1. Presentation of the profile methods Other limitations come from the use of these tech-
niques in the industrial environment. First, it is not
1.1.1. Conventional profile always possible for companies’ sensory analysts to set
Traditionally, performing a sensory profile includes up as many panels as there are kinds of products to
the following key steps: analyse. Second, the competitive environment often
means that new competitors come out on the market
1. Development of a common language for the and that recipes are subject to modifications. It is
description of the studied products. The con- then necessary to adapt the vocabulary to the pro-
stitution of a unique list of attributes is achieved duct space as it changes. Eventually and maybe most
by consensus as a group process. important, running a conventional profile is time-
2. Training of the judges for the evaluation of the consuming and for this only reason, sensory eval-
attributes. This includes flavour identification and uation is often dropped out when results are needed
quantification (intensity measurement). In order urgently.
to facilitate this step and to increase the group
performances, each attribute comes with a defini- 1.2. Flash profile
tion and physical references can be provided.
3. Evaluation itself is performed individually, Flash profile was initially developed as a flexible
usually in a monadic sequential way and it the- method that aims at providing a quick access to the
oretically comprises several repetitions as for any relative sensory positioning of a set of products. In
instrumental or sensory measurement. order to save time, the phases of familiarisation with the
product space, attribute generation and rating have
Various procedures based on this scheme have been been integrated into a single step. Several methodologi-
proposed. Amongst the most widely used are the QDA1 cal choices made it possible:
J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392 385

1. The main specificity of Flash profile is the com- All the samples respectively belonged to the same
parative procedure of evaluation of the whole fabrication lot for the totality of the experiments. Both
product set. Judges simply rank the products panels (conventional and Flash profile) therefore eval-
(with ties allowed) attribute per attribute. This uated strictly the same products. Before each evaluation
forces them to focus on the perceived differences session, samples were transferred into blank plastic
and to solely use discriminant attributes. There is containers (125 ml), coded and stored at 5  C prior to
consequently no longer need for a step of famil- the beginning of the session.
iarisation with the product set because the
panellists have a simultaneous access to all the 2.2. Panels
samples and their differences.
2. Second, Flash profile is based on the use of Free 2.2.1. Conventional profile
Choice Profiling (Williams & Langron, 1984) The panel dedicated to the conventional profiling
that allows judges to use their own list of attri- method pre-existed the present study and was composed
butes. This frees us from the time-consuming of 10 judges in the case of the strawberry yoghurt
constraints related to the seeking of the con- experiment and nine of them participated in the apricot
sensus and concept alignment (O’Mahony, 1991). fresh cheese experiment. These judges had been exten-
This therefore reduces the need for a trained sively trained for 6 months to 1.5 years. They received
panel. between 25 and 80 h of training (depending on indivi-
3. Finally, sensory evaluation experts—not neces- duals) to the evaluation of fruit flavour in blended fruit
sarily product experts—are selected to participate yoghurts. They were especially well trained in the eval-
in the panel. By sensory evaluation experts, we uation of strawberry aroma and slightly less for the
mean subjects who have previously participated apricot aroma. Since they were not specifically trained
in several descriptive evaluation tasks and who to evaluate the flavour of fresh cheeses, three additional
are able to understand panel leader’s instructions sessions have been devoted to an updating of the attri-
and generate discriminant and non-hedonic bute list and to further training with this product space.
attributes. This way, we can benefit from the
judges’ experience in communicating their per- 2.2.2. Flash profile
ceptions quantitatively. These judges do not need The panel dedicated to the Flash profile comprised
to be trained on a specific product set. In this nine judges for the strawberry yoghurt experiment and
respect, the selection of judges can be compared 10 judges for the apricot fresh cheese experiment. Eight
to that in the QFP procedure, in which the judges judges were common to both experiments. These judges
who are trained before participation in the test were all experienced in sensory evaluation and were
protocol actually become sensory experts selected for their motivation and for their general apti-
(Stampanoni, 1993). tude to understand and adapt to descriptive analysis
tasks. They were not specifically trained to the eval-
uation of strawberry or apricot aroma, yet they had
2. Materials and methods participated in the descriptive evaluation of fruit
yoghurts and fresh cheeses in the past. Care was taken,
2.1. Samples however, that they had not been involved in the eval-
uation of such products during the few months preced-
This study was split into two experiments respectively ing the present study.
conducted with two product spaces: strawberry blended
fruit yoghurts on one hand and apricot fresh cheeses on 2.3. Sensory procedures
the other hand. This was done to test the two profiling
methods with more than one kind of product. Fresh The sensory procedures were identical for both
dairy products were chosen because their short shelf-life experiments A and B. The evaluation was restricted to
made them good clients for a rapid descriptive method. the evaluation of flavour. For both experiments, tasting
Besides, fruit yoghurts and fresh cheeses are amongst sessions were conducted on the same days with the two
the largest sales of fresh food on the French market. panels. In both cases, the evaluations took place in
In experiment A, six strawberry yoghurts from the standardised individual booths.
French market were evaluated, including two yoghurts The panel of the conventional profile used the con-
(YOG1a and YOG1b) sold under the same brand but sensual attributes routinely employed for 1 year. In the
made in different plants. In experiment B, five apricot case of the apricot fresh cheese experiment, some attri-
fromages frais from the French market were analysed. butes were discussed within the group and redefined
Fromage frais is a type of quarg and will be denoted during the additional training sessions. Two replica-
here as fresh cheese. tions of the evaluation sessions were performed for the
386 J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392

conventional profile. Judges evaluated the samples in a measure of the similarity between the sensory spaces of
monadic sequential way using the 11-point intensity the two panels was given by the RV coefficient (Schlich,
scale ranging from 0 (attribute not perceived) to 10 1996). It was computed on the table of multivariate
(very intense). components from the GPA run on each panel’s data set.
Flash profile consisted of four sessions. During the Regarding the description of the products, PCA plots
first session, judges were asked to individually generate of the attributes on the circle of correlations was used to
attributes, which should be sufficiently discriminant to describe inter-product differences when they exist.
allow a ranking of the samples. All the generated attri- All multidimensional data treatments were performed
butes were then pooled by the experimenter and, at the using the XLSTAT1 add-in for Microsoft Excel1
beginning of session 2, the judges were asked to take a (Fahmy, 2000), with the exception of GPA that was
moment to read the panel’s list and to update their own performed on Statistical Analysis System Institute
list if desired. Judges then proceeded to the evaluation (SAS1, 1999) with the macro written by Schlich (1989).
itself on a ranking mode. Ties were allowed and judges
could re-taste the samples as much as they liked. Ses-
sions 3 and 4 were replicates of the evaluation step. 3. Results and discussion
Note that the judges did not necessarily participate in
the tasting sessions at the same time. The only restric- 3.1. Experiment A: strawberry yoghurts
tion for the panel was to wait until everybody had
completed the first session before proceeding to the sec- 3.1.1. Comparison of sensory maps
ond. This was to ensure that everybody’s attributes were The PCA plots of the data from the two panels (Fig. 1)
communicated to the rest of the panel. show similar structures. In both cases, YOG1a and
Evaluation sessions lasted approximately 30–40 min YOG1b are opposed to other products on the first axis
for the conventional profile whereas evaluation took and axis 2 separates YOG3 and YOG4. When limiting
longer for Flash profile (40–75 min). In the latter case, the comparison to the first two principal components,
pauses were allowed between the evaluations of two products YOG2 and YOG3 that are more centrally
attributes when desired by the judges. located, appear to be slightly differently positioned
depending on the profile type. However, not all the
2.4. Data analysis sensory information is taken into account on this plot.
Another way to assess the correspondence between the
The data from the two types of profiles (respectively two data sets is to apply a procrustean analysis to the
rating scores and ranks) were collected on Microsoft two consensus solutions obtained by applying GPA
Excel1 spreadsheets. Judge repeatability was first tested respectively to the conventional profile data and to the
by Principal Component Analysis (PCA) and by check- Flash profile data. In this case, data from the conven-
ing for each judge, the inter-session correlation coeffi- tional profile are also treated like free choice profiling
cients attribute per attribute. Poorly performing judges data (simulated FCP). That is, as if judges used their
were excluded from the data set. This was necessary own attributes. The result of this two-level Procrustes
because judges’ averaged data were used for further analysis is plotted as a PCA on Fig. 2. It can be seen in
analyses. After this step and for experiment A, nine this figure, that the Flash profile data perfectly matches
judges were retained for the conventional profile and six the conventional profile data. This is confirmed by the
judges for the Flash profile. For experiment B, eight normalised RV (NRV) coefficient between the two panel
judges were retained for the conventional profile and averaged data sets. The higher this coefficient, the more
seven judges for the Flash profile. Besides, the attribute similar the two panel configurations. The NRV coeffi-
‘‘rancid’’ was withdrawn from the data of the conven- cient was 3.07 (P < 0.01).
tional profile for its lack of repeatability. In order to characterise this correspondence, a GPA
As it is usually done, we first analysed the average was also performed on the overall data set—that is the
data from the conventional profile by PCA. Generalised productattribute (judge) matrix—and a CVA was run
procrustes analysis (GPA) (Gower, 1975) was applied to with the resulting matrix of sample coordinates on all
the data from Flash profile to assess the consensus principal components. Twelve predicted groups were
between assessors’ sensory maps. It was also used to defined as the 12 combinations: productprofile type.
compare the results of the two profiling methods. In As expected from Fig. 2, the distance matrix showed
addition to this, canonical variate analysis (CVA) was that all the products were different pairwise (P < 0.01)
used in order to determine the quality of product dis- with the exception of the pairs of identical products
crimination. This was based on the calculation of the F evaluated by differing profile methods (Table 1).
statistic (Rao’s approximation), on inter-group Mala- Besides, the software correctly reattributed 100% of the
hanobis distances and on the percentage of correct re- observations to the six groups of products indepen-
attribution to the various groups of products. A dently from the type of profile. This confirmed that the
J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392 387

Fig. 1. Experiment A—evaluation of strawberry yoghurts: comparison between the 1 and 2 factorial maps from the PCA of the conventional profile
averaged data (left hand side) and from the GPA of the Flash profile data (right hand side).

profiling method made no difference regarding the


structure of the product space as was observed by
Dairou and Sieffermann (2002).

3.1.2. Discriminating power


The F-approximation computed with the CVA gives
an indication of the discriminating power of the corre-
sponding panel. The higher the F, the better the dis-
crimination. The obtained F was very significant in both
cases, indicating that both panels were very dis-
criminant. Remarkably, products YOG1a and YOG1b
that were expected to be very similar, were dis-
criminated by both panels. Nevertheless, a markedly
lower F was obtained for the conventional profile
(F40, 182=4.03, P < 0.0001) than for the Flash profile
(F25, 98=40.32, P < 0.0001). However, we arrive at
Fig. 2. Experiment A—evaluation of strawberry yoghurts: plot of the F25, 165=52.73 (P < 0.0001) when a GPA is applied to
Procrustes analysis performed on the consensus data from the two
the conventional profile data. Therefore it seems that a
panels respectively dedicated to conventional profile (filled symbols)
and to Flash profile (open symbols). panel’s discriminating power is decreased when judges
are forced to use the same attributes. This result agrees
with the criticisms levelled at this type of procedure.

Table 1 3.1.3. Semantic comparison


Malahanobis’s square distances between the products (strawberry The attributes used by the two panels are summed up
yoghurts) evaluated either by conventional profile or by Flash profile in Table 2. Eight attributes were used for the conven-
Products (DMalahanobis)2 Associated F p tional profile. Two of them were related to taste and six
described the aroma. In the case of Flash profile, judges
YOG1aconv.–YOG1aflash 3.28 1.64 0.2453
YOG1bconv.–YOG1bflash 1.92 0.96 0.4910
used between five and eight attributes. That is, for the
YOG2conv.–YOG2flash 2.29 1.14 0.4059 panel, a total of 19 attributes kept for data analysis. It
YOG3conv.–YOG3flash 1.20 0.60 0.7039 can be seen that judges belonging to the Flash profile
YOG4conv.–YOG4flash 1.62 0.81 0.5730 panel used some very different words such as cereal,
YOG5conv.–YOG5flash 1.44 0.72 0.6276 alcohol, floral and cheese to describe the aroma.
Distances were calculated after GPA was applied on the whole data Besides, all the attributes from the conventional profile
set. have also been used by the judges who participated in
388 J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392

Table 2 ciated to the wild strawberry note. Product YOG2 that


Flavour attributes used for the evaluation of the strawberry yoghurts was centrally positioned on map 12 could be said to be
with both methods of profile
more balanced for these attributes. It is better repre-
Flash profile Number of occurrences Conventional sented on the third axis, on which it is related to the jam
in Flash profile profile note, but is less vanilla-like than YOG4 and YOG5.
Sour 4 Sour
The PCA of the Flash profile data naturally involves a
Bitter to sweet 1 larger number of attributes, which renders the descrip-
Astringent 1 tion less straightforward. Nevertheless, it can be seen
Sweet (doux) 2 that products YOG1a and YOG1b are characterised by
Sweet (sucré) 4 Sweet (sucré) a set of attributes related to sourness and green aroma
Alcohol 3
Cereal 2
notes. They are opposed on the first principal compo-
Jam 3 Jam nent to the other products (specially YOG3, YOG4 and
Creamy 4 Creamy YOG5) by attributes sweet, cream, vanilla and jam.
Creamy/vanilla 1 Besides, YOG3—and YOG5 to a lesser extent—have
Cooked 1 been rated more ‘‘cereal’’ by two judges. On the other
Wild strawberry 2 Wild strawberry
Floral 2
hand, there is no consensus between the two judges who
Strawberry 1 used the attribute ‘‘wild strawberry’’, neither is there a
Typical to atypical 1 consensus between the three judges who used the attri-
strawberry bute ‘‘alcohol’’.
Cheese 1 As can be seen, many of the same descriptive char-
Vanilla 1 Vanilla
Green 3 Green
acteristics were obtained with both methods. However,
Green to vanilla 1 the wild strawberry note of products YOG1 (a and b)
0 Overall aroma that was spotted with the conventional profile no longer
intensity appears with the Flash profile. In return, new informa-
tion is brought to light by the Flash profile with the use
of the attribute ‘‘cereal’’. The description provided by
the Flash profile with the notable exception of the the Flash profile is thus satisfactory as compared to
attribute ‘‘overall aroma intensity’’. This is typically the conventional profile. The same conclusions can indeed
case of an attribute of interest for the experimenter, be drawn independently from both profile outputs.
which is however not spontaneously used by the judges.
This may be explained either by the fact that it is not 3.2. Experiment B: apricot fresh cheeses
intuitive enough or because there are no differences
among the products for this particular attribute. That is 3.2.1. Comparison of sensory maps
to say, the attribute is not discriminating enough Fig. 3 represents the sensory maps obtained with both
towards the evaluated product space. The latter panels. Contrary to what was observed in the case of
hypothesis is plausible since a CVA applied to the con- strawberry yoghurts, the two PCA plots are markedly
ventional profile data indicates a poor correlation of the different. The first two dimensions explained 78% of the
variable ‘‘overall aroma intensity’’ with the first dis- total variance in the case of conventional profile
criminant axes (the correlation coefficient is 0.31 with whereas they explained up to 89% of the total variance
axis 1 and 0.12 with axis 2). This agrees with the in the case of Flash profile. Particularly, it can be seen
desire, within the frame of Flash profile, to use only that FCH2 and FCH3 are not identically positioned.
discriminating attributes. Although axis 2 separates FCH3 from other products
on the PCA of the conventional profile, axis 2 isolates
3.1.4. Description of the products the product FCH2 on the PCA of the Flash profile. This
As often with Free Choice Profiling techniques, the is somewhat balanced by the third axis of both PCA
loading plots yield rather complex graphs gathering all (Fig. 3B). As it was done for experiment A, a two-level
individual attributes. These figures as well as the loading Procrustes analysis was used to facilitate the comparison.
plots for the conventional profile are thus not presented The consensus configuration of each panel was first com-
here. However, main descriptive facts are reported in puted by GPA and subsequently compared by the way of
the following. The PCA of the conventional profile data a second Procrustes analysis. The results are plotted as a
indicates that products YOG1a and YOG1b are PCA on Fig. 4. This figure shows that a good correspon-
strongly characterised by green and sour notes that dence between the two consensus configurations can be
oppose them to products YOG3, YOG4 and YOG5. obtained. It is however not as good as it was for experi-
These latter products were rated sweeter, with jam, ment A as indicated by the lower percentage of explained
vanilla and creamy notes. Products YOG1a and variance with the first two dimensions (77.5% to be
YOG1b had a stronger overall aroma that was asso- compared to the 78 and 89% for the PCA of the two
J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392 389

Fig. 3. Experiment B—evaluation of apricot fresh cheeses: comparison between the 12 (A) and 23 (B) factorial maps from the PCA of the
conventional profile averaged data (left hand side) and from the GPA of the Flash profile data (right hand side).

panels, respectively). Furthermore, it can be seen on the the conventional profile and F16, 83=26.96 (P < 0.0001)
13 factorial map (Fig. 4B) that products FCH4 and for the Flash profile. Products were thus much better
FCH5 are not well paired. The NRV coefficient between discriminated using the Flash profile. However, apply-
the two panel averaged data was 1.49 (NS), which ing a GPA to the conventional profile data prior to the
reflected the lower degree of similarity. CVA gives a better discriminating power (F16, 98
As was done for experiment A, a GPA followed by =62.38, P < 0.0001) which is higher than the Flash pro-
CVA was applied to the overall productattribute (judge) file one. It is clearer than for the strawberry yoghurt
matrix that gathered the data from the two panels. This experiment, that the discriminating power of the con-
was done to characterise the correspondence between the ventional profile panel was potentially higher than it first
two panels. Table 3 shows that three products were well appeared. Once again, this may be attributed to the
paired whereas two of them (FCH4 and FCH5) were dis- forced use of consensual attributes and particularly to
tinctly positioned, depending on the type of profile. This the following data treatment, which implicitly considers
confirms the graphical interpretation above. Besides, the that these common attributes refer to the same under-
CVA indicates that differing products were discriminated lying sensory concepts for all panel members. The larger
and were correctly reattributed to their corresponding discrepancy observed here is probably due to the fact
clusters, regardless of the profile type. that judges from the conventional profile panel were less
trained to the evaluation of apricot fresh cheeses than
3.2.2. Discriminating power they were for strawberry yoghurts. As a conclusion, the
As before, the F-approximation was computed from use of a free choice profiling method seems to be more
the CVA in order to compare the discriminating power indicated when the panel has not been, or cannot be,
of each panel. We obtained F40, 100=1.74 (P=0.014) for trained enough to align concepts among the judges.
390 J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392

Table 4
Flavour attributes used for the evaluation of the apricot fresh cheeses
with both methods of profile

Flash profile Number of occurrences Conventional


in Flash profile profile

Sour 1 Sour
Astringent 1
Sweet (sucré) 5 Sweet (sucré)
Fruity apricot 1
Artificial 2
Banana 1
Artificial candy 1
Caramel 2
0 Candy—
confectionery
0 Jam
Creamy 2 Creamy
Cooked 1
Spicy—cinnamon 1
Floral 4
Cheese 1
Fresh cheese to 2
crème fraıˆche
Fruity 1
Liquor 1
Apricot skin 2
0 Overall aroma
intensity

(jam and overall aroma intensity) were not used for the
Flash profile. A CVA applied to conventional profile
data indicates that the attribute ‘jam’ was poorly dis-
criminating (the correlation coefficient with the first
discriminant axis is 0.15 and 0.12 with the second axis).
Fig. 4. Experiment B—evaluation of apricot fresh cheeses: plot of the This might be an explanation for the fact that it was not
12 (A) and 13 (B) factorial maps derived from the Procrustes ana- used by the judges in the case of Flash profile. On the
lysis performed on the consensus data from the two panels respectively
dedicated to conventional profile (filled symbols) and to Flash profile
other hand, and contrary to what was found with the
(open symbols). strawberry yoghurts, the attribute ‘overall aroma inten-
sity’ was highly discriminating (the correlation coeffi-
cient with axis 1 of CVA is 0.65 and 0.39 with axis 2).
Table 3
Malahanobis’s square distances between the products (apricot fresh We could therefore hypothesise that this attribute was
cheeses) evaluated either by conventional profile or by Flash profile not intuitive enough to be used by judges who partici-
pated in the Flash profile. It is also possible that ‘overall
Products (DMalahanobis)2 Associated F p
aroma intensity’, which is a fairly synthetic concept, was
FCH1conv.–FCH1flash 1.14 0.82 0.5435 systematically broken down into more accurate attri-
FCH2conv.–FCH2flash 2.89 2.08 0.1590 butes. Generally speaking, the vocabulary employed by
FCH3conv.–FCH3flash 3.15 2.26 0.1344
the Flash profile panel was richer than the standardised
FCH4conv.–FCH4flash 6.84 4.91 0.0188
FCH5conv.–FCH5flash 16.62 11.94 0.0008 list of attributes from the conventional profile as can be
seen from Table 4 and the main sensory concepts from
Distances were calculated after GPA was applied on the whole data the conventional profile were also yielded by the Flash
set.
profile.

3.2.3. Semantic comparison 3.2.4. Description of the products


Nine attributes were used for the conventional profile. The PCA of the conventional profile indicates that
Two of them were related to taste and eight described the product FCH3 is separated from other products along a
aroma. In the case of Flash profile, judges used between fruity-apricot and jam dimension. Product FCH4
four and nine attributes. That is, for the panel, a total of appears to be sweeter and more cheesy than other pro-
18 attributes kept for data analysis. It can be seen from ducts. It is still close to product FCH2 that has a
Table 4 that two attributes from the conventional profile stronger creamy note. These two latter products have a
J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392 391

lower overall intensity than the rest of the products. On whose flavour contrasted to the rest of the product set
the other hand, product FCH1 is characterised by floral because of their green/vegetal note. This strong sensory
and candy notes and by its sourness. Product FCH5 is characteristic largely contributed to the configuration of
centrally positioned and therefore cannot be easily the sensory maps with both types of profiles. Besides,
characterised. when the product space was more homogeneous—as in
As was mentioned above, the Flash profile involves a the case of the apricot fresh cheese—the two profile
larger number of differing attributes, which renders the methods resulted in slightly differing positioning.
description more delicate. Still, some general descriptive The second observation made in the present study is
trends can be given. The first axis of the PCA is posi- that products were better discriminated by the use of
tively correlated to attributes ‘‘sweet’’ and ‘‘cream’’ Flash profile than by the use of conventional profile.
(product FCH4) and negatively correlated to attributes This was observed with the two tested product sets. This
such as ‘‘floral’’, ‘‘apricot peel’’, ‘‘artificial’’ and could be linked to the evaluation mode as well. Yet, the
‘‘astringent’’ (product FCH1, and FCH3 to a lesser application of GPA to the conventional profile data
extent). FCH2 seems to be discriminated from other showed that the treatment usually applied to the panel’s
products for its fruity note. As observed with the con- averaged data masks inter-subject differences in their
ventional profile, product FCH5 is equally balanced for understanding of the attributes and therefore reduces
most of the attributes and was thus difficult to describe the panel’s discriminating power. It is our opinion that,
using this technique. when product positioning and discrimination are of
Obviously, some descriptive elements are common to prime concern, profiling techniques based on the free
both profiles, but the comparison on these bases choice of attributes are preferable to those involving the
remains, once again, awkward and no further conclu- development of consensual attributes. Flash profile
sion can be drawn. would be specially indicated when the available panel
cannot be trained for a sufficient time.
On the other hand, Flash profile is less adapted to the
4. General discussion and conclusion description of the products than conventional profiling.
This was expected since Flash profile was not designed
The use of Flash profile for the evaluation of a set of to give an accurate description of the products. How-
strawberry yoghurts allowed a rapid access to their ever it is noticeable that we clearly lack tools to evaluate
sensory positioning. It was found to be identical to that and compare the quality of semantic descriptions. The
obtained by a conventional method. However, we latter point should be addressed in future research.
observed that the match was not perfect for the second The main benefit of Flash profile is its rapidity and
product space studied (the apricot fresh cheeses). The ease of use. Each evaluation session may take much
reasons for the discrepancy between the results of the time than it would in conventional profiling. Never-
two panels are not clear. It is worth noting that the theless, we found, as shown in a previous study (Dairou
interpretation is inherently limited by such a compar- & Sieffermann, 2002), that the total amount of time
ison methodology. It is indeed impossible to separate necessary to achieve a Flash profile is much lower than
the sources of variations (i.e. the panel and the profile with conventional profiling. It is also less time-intensive
type). This holds for all the results presented here. One for the experimenter. In addition to this, all sessions can
way to partly solve this problem would have been to be carried out individually, on separate days and sepa-
replicate experiments with several different panels to rate locations, providing that all samples are available.
increase the number of reproduced profiles and hence to Indeed, the simultaneous evaluation by ranking
increase the power of the test. It was unfortunately not obviously implies that all the samples must be available
possible to proceed this way within the frame of the at the same time. Consequently, no supplementary
present study. It is hypothesised however that observed sample can be evaluated afterwards without re-profiling
differences are likely to be linked to the differing eval- the whole sample set. This may be a strong restriction
uation modes (comparative vs. monadic sequential) that on the use of Flash profile depending on material cir-
automatically result in differing contexts of evaluation. cumstances. As pointed out by Dairou and Sieffermann
For Flash profile, each product is evaluated by direct (2002), this methodology cannot be used for product
comparison with the rest of the product set whereas for stability tests or quality control.
the conventional profile, the monadic sequential mode In conclusion, we think it would be misleading to
implies that each product is evaluated in comparison consider Flash profile as a substitute for conventional
with judges’ representation of the product space, which profiling, which is certainly the most adapted and accu-
is either taught and/or based on judges’ previous rate profiling technique to date. Furthermore, these two
experience. Thus, memory is not involved in the same methods do not fulfil exactly the same objectives. We
way in both cases. Besides, it should be noted that the rather propose to consider Flash profile as a convenient
set of strawberry yoghurts comprised two products sensory mapping tool for conducting preliminary phases
392 J. Delarue, J.-M. Sieffermann / Food Quality and Preference 15 (2004) 383–392

of thorough sensory studies. It could be of great help, Meilgaard, M., Civille, G. V., & Carr, B. T. (1999). Sensory evaluation
for example, in the language development step of a techniques (3rd ed.). Boca Raton, FL: CRC Press.
Moniez, N., Truchot, A., & Sieffermann, J.-M. (2001). Sensory char-
conventional profile (Moniez, Truchot, & Sieffermann,
acteristics and consumer preferences for surimi sticks (poster). In
2001). One could also use it as a screening tool for The 4th Pangborn Sensory Science Symposium, 22–26 July, Dijon,
selecting products or factors when designing a larger France.
experiment. Notice that, technically, Flash profile data O’Mahony, M. (1991). Descriptive analysis and concept alignment.
can serve as a sensory base for a preference mapping In H. T. Lawless, & B. P. Klein (Eds.), Sensory science theory
study. Besides, Flash profile is fast and flexible, which and applications in foods (pp. 223–267). New York: Marcel
Dekker.
makes professional experts such as flavourists or oenolo- SAS (1999). Version 8. Cary, NC: Statistical Analysis System Institute
gists affordable as panel members. Likewise, using Flash Inc. .
profile may be an easy way to rationalize and increase the Schlich, P. (1989). A SAS/IML program for Generalised Procrustes
value of in-house round table tasting with company peo- Analysis. In SAS European Users Group International Conference,
ple. It also gives access to the sensory evaluation of short Köln, Germany.
Schlich, P. (1996). Defining and validating assessor compromises
shelf life products in the context of experimental design about product distances and attribute correlations. In T. Noes, &
studies or when a one-shot evaluation is only possible. E. Risvik (Eds.), Multivariate analysis of data in sensory science
Generally speaking, Flash profile could be used whenever (pp. 259–306). Amsterdam: Elsevier Science.
one is interested in getting the relative sensory positioning Sieffermann, J.-M. (2000). Le profil flash—un outil rapide et innovant
of a set of products in a very short time. d’évaluation sensorielle descriptive. In AGORAL 2000, XIIèmes
rencontres ‘‘L’innovation: de l’idée au success’’ (pp. 335–340), Mont-
pellier, France.
Sieffermann, J.-M. (2002). Flash profiling. A new method of sensory
References descriptive analysis. In AIFST 35th Convention, July 21–24, Sidney,
Australia.
Dairou, V., & Sieffermann, J.-M. (2002). A comparison of 14 jams Stampanoni, C. R. (1993). The ‘‘quantitative flavor profiling’’ tech-
characterized by conventional profile and a quick original method, nique. Perfumer and Flavorist, 18, 19–24.
the Flash Profile. Journal of Food Science, 67, 826–834. Stone, H., Sidel, J., Oliver, S., Woosley, A., & Singleton, R. C. (1974).
Fahmy, T. (2000). XLSTAT (version 4.4). Paris. Sensory evaluation by quantitative descriptive analysis. Food Tech-
Gower, J. C. (1975). Generalised procrustes analysis. Psychometrika, nology, 28, 24–34.
40, 33–51. Williams, A. A., & Langron, S. P. (1984). The use of free-choice pro-
Lawless, H. T. (1999). Descriptive analysis of complex odors: reality, filing for the evaluation of commercial ports. Journal of the Science
model or illusion? Food Quality and Preference, 10, 325–332. of Food and Agriculture, 35, 558–568.

You might also like