Method for modelling sensory appraisal of cosmetic products, tool for assisting with design of cosmetic products, and method for designing cosmetic products

Abstract

The invention relates to a method for modelling sensory appraisal of a group of cosmetic products (R1, R2), the cosmetic products being characterized by sensory properties defined by sensory descriptors, and by a set of meta-variables (MV1-MV7) defined by combining the sensory descriptors, that comprises a multivariate statistical analysis, namely decorrelation of the meta-variables into principal components (MVMAP), and generation of the sensory appraisal model based on the principal components.

Description

[0001] DESCRIPTION

[0002] TITLE: Method for modelling sensory appraisal of cosmetic products, tool for assisting with design of cosmetic products, and method for designing cosmetic products.

[0003] Embodiments of the invention relate to a computer-implemented method for modelling sensory appraisal of cosmetic products; to a tool for assisting with design of cosmetic products employing the modelling method; and to a method for designing the chemical composition of cosmetic products using the assisting tool.

[0004] Preferring to use one cosmetic product, such as a moisturizer, rather than another may seem quite simple, but it is actually a much more complex behaviour.

[0005] Due to their functional and sensory benefits, cosmetic products often generate a strong emotional reaction.

[0006] In this context, the consumer’s evaluation is therefore not limited to her or his appraisal but may also include other dimensions relating to concepts with emotional connotations ("Linking sensory characteristics to emotions: An example using dark chocolate", Thomson et al., Food quality and Preference, Volume 21, Issue 8, 2010). Indeed, some authors have observed that these dimensions may be based on multiple sensory determinants, and on physiological, psychological and social factors ("Understanding creaminess", Bom Frost & Janhoj, International Dairy Journal, Volume 17, Issue 11, 2007; "Impact of refreshing perception on mood, cognitive performance and brain oscillations: An exploratory study", Labbe, Martin, Le Coutre, & Hudry, Food Quality and Preference, Volume 22, Issue 1, 2011). In addition to this, these dimensions would also appear to have a hedonic tone ("Beyond sensory characteristics, how can we identify subjective dimensions? A comparison of six qualitative methods relative to a case study on coffee cups", Masson, Delarue, et al., Food Quality and Preference Volume 47, Part B, 2016). Consequently, some authors have called for account to be taken of these dimensions influencing consumer choices ("Theories of food choice development", Koster & Mojet, 2007).

[0007] However, neither a purely sensory description nor a hedonic test are able to account for these percepts.

[0008] Conventionally, descriptive methods are used to describe and quantify the sensory characteristics of a product. They make it possible to establish the sensory profile of the product, i.e. a fixed image giving the average values of the perceived levels of specific sensory attributes. The most common methods are quantitative descriptive analysis (QDA®) ("Sensory evaluation by quantitative descriptive analysis" Stone, Sidel, Oliver, Woolsey, & Singleton, Food Technology, Volume 28, 1974) and the Spectrum method ("Sensory Evaluation Techniques Third Edition" Meilgaard, Civille & Carr, CRC Press, 1999). They consist in detecting, describing and evaluating sensory characteristics of a specific portion of the product, often a mouthful of food or drink in the agri-food field, or an application in the field of cosmetics. The sensory characteristics studied are called descriptors. These two methodologies must be carried out by a panel of trained judges.

[0009] Complementary as they are, sensory characterization and hedonic measurement may be linked in order to better understand consumer judgement. A number of modelling methodologies have been developed.

[0010] External preference mapping ("MDPREF", acronym of "multidimensional preference analysis") is the best known methodology ("Preference mapping in practice" Greenhoff & MacFie, 1994). It aims to analyse preference data for a set of products depending on their sensory characteristics. Products are differentiated via their sensory characteristics, which will be linked with consumer preferences. A plurality of so-called "nested" models may be used to regress the consumer preference data: the vector model, the circular model, the elliptical model and the quadratic model (Schlich, 1995). These models make it possible to predict the potentially most popular products within a sensory space for a given group of consumers.

[0011] Another approach that is also possible to understanding preferences is partial- least-squares (PLS) regression ("Partial least-squares regression on design variables as an alternative to analysis of variance", Martens, Izquierdo, Thomassen, & Martens, Analytica Chimica Acta, Volume 191, 1986). It allows a block of variables to be explained (the preferences) to be related to a block of quantitative explanatory variables (the sensory characteristics).

[0012] However, it would be desirable to improve these conventional methods. Specifically, a number of criticisms may be made of these methods.

[0013] Firstly, in the case of preference mapping, the models take into account only two external dimensions (the first two dimensions of the considered space). Thus important information present in higher dimensions is omitted ("Simple improvement of consumer fit in external preference mapping", Faber, Mojet, & Poelman, Food Quality and Preference Volume 14, Issues 5-6, 2003). This omission may lead to models that are less relevant to a non-negligible number of consumers. Moreover, increasing the number of sensory dimensions in regression models does not seem to solve the problem, as this may lead to over-fitting of the preference scores when circular, elliptical or quadratic models are used, because only a small number of degrees of freedom are available to estimate the parameters.

[0014] Secondly, the conventional methods make large assumptions, and in particular assume that consumer preferences will be explained by the sensory characteristics of the products. However, in descriptive analysis, panellists are asked to break down the sensory characteristics of a product into one-dimensional descriptors, but the result is not necessarily representative of what a consumer will perceive (Jaeger, Wakeling, and MacFie, 2000). Specifically, a consumer's sensory perception is not necessarily an aggregation of one-dimensional attributes.

[0015] Thus, it is proposed, according to the modes of implementation defined below, to improve conventional techniques for modelling appraisal of cosmetic products, via an approach based on meta-variables in particular allowing a priori knowledge to be introduced with a view to achieving better relevance.

[0016] This a priori knowledge, applied specifically to sensory appraisal of cosmetic products, olfactory appraisal of cosmetic products for example, may result from the union of several fields, for example: professional knowledge (i.e. know-how) of perfumery; sensory knowledge of olfactory sensory teams; and knowledge contained in the literature (often applied to the agri-food industry, and even in particular the citations mentioned above).

[0017] A computer-implemented method for modelling sensory appraisal of a group of cosmetic products is thus provided, the cosmetic products being characterized by sensory properties defined by sensory descriptors, and by a set of meta-variables defined by combining the sensory descriptors, the method comprising a multivariate statistical analysis, namely decorrelation of the meta-variables into principal components, and generation of the sensory appraisal model based on the principal components.

[0018] The meta-variables are for example elaborated so as to simplify statistical analysis of the sensory data. Specifically, instead of working with a high number of variables, it is possible to focus on a limited set of meta-variables, this facilitating application of multivariate analysis techniques. Use of meta-variables allows a sensory appraisal model to be generated that may be used to evaluate in an apposite way whether or not a cosmetic product corresponds to consumer preferences.

[0019] According to one embodiment, the cosmetics are further characterized by respective preference scores, and generation of the appraisal model comprises regression of the preference scores to the principal components. The regression may advantageously follow a quadratic model.

[0020] According to one embodiment, the principal components are configured to orient a plane containing respective positions for each cosmetic product, the metavariables being elaborated so that the principal components orient the plane in a manner that groups the positions of the cosmetics depending on their respective sensory appraisals.

[0021] According to one embodiment, the set of cosmetics belongs to a database containing the definition of each cosmetic product by the sensory descriptors, and a preference score assigned to the products.

[0022] According to one embodiment, appraisal is modelled for at least one type of consumer, and at least one of said meta-variables is dedicated to each respective type of consumer.

[0023] According to one embodiment, the sensory appraisal is olfactory, the sensory descriptors defining properties relating to olfaction.

[0024] According to one embodiment, statistically, the meta-variables are discriminant, continuous and follow a normal distribution.

[0025] Moreover, the sensory descriptors are advantageously determined by a team of judges trained to identify and evaluate them, and the meta-variables are computed from the sensory descriptors via combination and / or aggregation.

[0026] Also provided is a computer-implemented computational tool, in particular a software tool, for assisting with design of a cosmetic product, comprising employing the sensory appraisal model obtained via the method such as defined above, to generate a decision tree comprising a sequence of conditions on values of the meta-variables culminating in a theoretical prediction of the appraisal of the cosmetic product.

[0027] Such a tool may be used to simplify and make less time-consuming the design of a cosmetic product corresponding to consumer preferences.

[0028] According to one embodiment of this tool, the theoretical prediction of the appraisal of the cosmetic product is made for at least one type of consumer, and the sequence of the decision tree and said conditions are at least partly dedicated to each respective type of consumer.

[0029] Also provided is a method for designing a cosmetic-product chemical composition and adjusting the composition with chemical compounds having sensory properties, comprising using the sensory appraisal model obtained via the method defined above, or using the decision tree generated by the tool defined above.

[0030] Such a method makes it possible to simplify and make less time-consuming the design of a cosmetic product corresponding to consumer preferences.

[0031] Also provided is a computer program product comprising instructions that, when the program is executed by a computer, cause the latter to implement the modelling method or assisting tool such as defined above.

[0032] Also provided is a computer-readable recording medium comprising instructions that, when they are executed by a computer, cause the latter to implement the modelling method or assisting tool such as defined above.

[0033] Other advantages and features of the invention will become apparent on examining the detailed description of completely non-limiting embodiments, and the accompanying drawings, in which:

[0034] [Fig. 1] illustrates a conventional preference map of a set of cosmetic products, based on sensory descriptors;

[0035] [Fig. 2] illustrates a preference map of a set of cosmetic products, based on meta-variables according to the invention;

[0036] [Fig. 3] illustrates a relationship between the preference map according to the invention and the preference scores given by users;

[0037] [Fig. 4] illustrates one example of a decision tree obtained using the preference model obtained from the map according to the invention;

[0038] [Fig 5] illustrates one embodiment of a computer system allowing a tool for assisting with design of cosmetic products to be executed;

[0039] [Fig. 6] illustrates a computer-implemented method for modelling sensory appraisal of a group of cosmetic products;

[0040] [Fig. 7] illustrates a method for designing a cosmetic-product chemical composition.

[0041] Figure 1 illustrates one example of a representation PRMAP of appraisal ratings given by users to a set of cosmetic products. The appraisal ratings may be assigned to a plurality of distinct parameters in order to quantitatively evaluate the appraisal, for example based on sensory evocations detected by the questioned users. The evaluated evocations may for example include: freshness; sensuality; well-being; heaviness; old-fashionedness; comforting; addictive; chemical; etc.

[0042] For example, positive appraisal ratings correspond to a preference, whereas negative ratings may correspond to a rejection.

[0043] In order to plot the graph PRMAP of Figure 1, a multivariate statistical analysis, namely decorrelation into principal components, was performed on the appraisal ratings.

[0044] This statistical analysis technique, usually referred to as "principal component analysis" and by the acronym "PCA", is a method for analysing data, which allows correlated variables to be converted into new decorrelated variables called "principal components". The principal components are for example defined so as to best explain the variance of the data. This makes it possible to summarize information by reducing the number of variables and thus to represent the statistical population in a readable space, typically of two dimensions defined by the first two principal components.

[0045] The graph PRMAP obtained via the principal component analysis makes it possible to distinguish between products rejected by consumers (located in a first region Rl) and products preferred by consumers (located in a second region R2), in a plan oriented by the first two principal components.

[0046] Consistently in the graphs PRMAP, SENSMAP, MVMAP of Figures 1, 2, and 3, products preferred by users have been represented by circles, whereas products rejected by users have been represented by triangles.

[0047] Figure 2 illustrates one example of a conventional map SENSMAP of a set of cosmetic products, once again obtained via principal component analysis.

[0048] The principal component analysis was carried out on sensory descriptors DI, D2, D3 and D4, also called "sensory characteristics", defining sensory properties of the cosmetic product. The sensory descriptors may in particular be olfactory descriptors.

[0049] For example, sensory descriptors may include the following descriptors: lemon; lime; honeysuckle; aldehyde; lily of the valley; orange blossom; cucumber; coumarin; vanilla; vinegar; oily; honey; cherry; tiare flower; linden tree; candy; melon; apricot; rose; etc. The descriptors may also be grouped by families, such as for example the following families: citrus; aromatic; freshness; fruity; floral; green; aldehyde; aqueous; moreish; powdery; spicy; woody; amber-scented or musky; animalic or leathery; etc.

[0050] Trained judges are considered to be capable of characterizing each cosmetic product by assigning to each sensory (in particular olfactory) descriptor a strength score varying for example from 0 (unperceivable) to 100 (very strong), the averages of the scores assigned by all of the judges characterizing the sensory properties of the cosmetic product.

[0051] However, for the "PREFMAP" model (PREFMAP standing for preference mapping) to be relevant and predictive, a hypothesis must be confirmed: consumer perceptions must be aligned with the sensory characteristics defined by a panel of trained judges. However, this hypothesis is rarely confirmed.

[0052] Indeed, by comparing the graphs PRMAP and SENSMAP it may be seen that they do not coincide. In the graph PRMAP, it may be seen that the points indicating a negative appraisal (triangles) are all located on the right of the map, whereas points indicating a positive appraisal (circles) are all located on the left of the graph. In the graph SENSMAP, points indicating a negative appraisal (triangles) are jumbled up with points indicating a positive appraisal (circles): for example, two points that are very close sensorially (the solid black circle and triangle) may have two very different appraisals (it is a question of a circle corresponding to products preferred by users and of a triangle corresponding to products rejected by users). The graph SENSMAP does not therefore reflect consumer perceptions well.

[0053] It is therefore difficult, if not impossible, to predict the appraisal of a product based on this graph SENSMAP.

[0054] It is thus proposed to construct meta-variables, for example from sensory descriptors, that offer a description of cosmetic products that is closer to the perceptions and appraisal of consumers.

[0055] In this regard, reference is made to Figure 3.

[0056] Figure 3 illustrates one example of an advantageous map MVMAP of a set of cosmetic products, once again obtained via principal component analysis.

[0057] The principal component analysis was carried out on meta-variables MV1- MV7 that may be determined by combining or aggregating sensory descriptors.

[0058] The meta-variables are elaborated in order to provide synthetic and abstract information on the sensory evaluations made by trained judges, and in particular allowing a priori knowledge to be introduced with a view to achieving better relevance. This a priori knowledge for example results from the union of several fields, for example: professional knowledge (i.e. know-how) of perfumery; sensory knowledge of olfactory sensory teams; and knowledge contained in the literature.

[0059] The meta-variables are advantageously elaborated in such a way as to offer a closer description of preferences than conventional descriptors.

[0060] The meta-variables are further for example elaborated so as to reduce dimensionality, by summarizing a plurality of measurements into a smaller number of synthetic meta-variables. Specifically, the raw sensory data may be bulky and complex, this making it difficult to interpret and analyse them.

[0061] The meta-variables are further for example elaborated so as to provide higher- level information that facilitates interpretation of the results of the sensory analysis, making it possible to understand general trends, and similarities and differences between the evaluated samples.

[0062] The meta-variables are further for example elaborated so as to communicate the results of the sensory analysis in a concise and understandable manner, simplifying presentation of the data and facilitating comparison between the various samples.

[0063] The meta-variables may further highlight relationships or correlations between the various measured sensory characteristics. They may thus make it possible to detect hidden relationships that would not be obvious if the raw data were examined.

[0064] The meta-variables are further for example developed so as to define synthetic quality indices, thus making it possible to estimate the overall quality of a product from a plurality of sensory measurements.

[0065] The meta-variables are further for example elaborated so as to simplify statistical analysis of the sensory data. Specifically, instead of working with a high number of variables, it is possible to focus on a limited set of meta-variables, this facilitating application of multivariate analysis techniques.

[0066] In summary, the meta-variables are advantageously elaborated in order to simplify interpretation, reduce the dimensionality of the data, detect hidden relationships, and / or improve the effectiveness of communication of results. This allows deeper understanding of sensory evaluations and facilitates use of this information in decision making.

[0067] The meta-variables are for example elaborated in the context of knowledgegeneration activities, via statistical analyses carried out on a database of cosmetic products, containing the definition of each cosmetic product by the sensory descriptors, and a preference score assigned to the products by users.

[0068] When a plurality of meta-variables have been identified, defined and computed, a statistical analysis is carried out to validate the relevance of each metavariable.

[0069] Specifically, as explained above with respect to the prior art, it is advantageous for the meta-variables to meet the following statistical criteria:

[0070] - they are discriminant (criterion evaluated for example using a one-way "ANOVA" technique - ANOVA standing for ANalysis Of VAriance);

[0071] - they are continuous and follow a normal distribution (criterion evaluated for example by means of the Shapiro-Wilk test);

[0072] - they express a consumer perception.

[0073] For example, among meta-variables obtained using the method described above, the following may be defined:

[0074] MV5 - olfactory complexity: corresponds to the number of olfactory notes perceived by the panellists, i.e. for example to the number of descriptors having an average above a perception threshold, for example set at 6%.

[0075] MV3 - head profile: corresponds to the maximum strength perceived among descriptors belonging to the families citrus, aromatic and freshness.

[0076] MV4 - heart profile: corresponds to the maximum strength perceived among descriptors belonging to the families fruity, floral, green, aldehyde, aqueous and moreish.

[0077] MV6 - background profile: corresponds to the maximum intensity perceived among descriptors belonging to the families powdery, spicy, woody, amber- scented / musky, animalic / leathery.

[0078] MV7 - longevity: corresponds to 0.75 x background profile + 0.25 x heart profile.

[0079] Meta-variables MV1, MV2, MV3, MV4, MV5, MV6, MV7 were obtained in this way and allowed a sensory map MVMAP (Figure 3) close to consumer preferences to be generated.

[0080] This may for example be seen by comparing the map MVMAP obtained with the meta-variables and the map PRMAP (Figure 1) obtained based on consumer perceptions. Specifically, the map MVMAP thus obtained by principal component analysis based on meta-variables makes it possible to position the products rejected by consumers in a first region Rl, quite distinctly from the products preferred by consumers, which are positioned in a second region R2, in the plane oriented by the first two main components.

[0081] In other words, the principal components are configured to orient a plane containing respective positions for each cosmetic product, and the meta-variables are elaborated so that the principal components orient the plane in a manner that groups Rl, R2 the positions of the cosmetics depending on their respective sensory appraisals.

[0082] Furthermore, in the context of a method usually called "PREFMAP" (standing for "preference mapping"), it will further be possible to perform a regression, to said principal components obtained by PCA, of consumer preference information such as mentioned with reference to Figure 1.

[0083] The regression will allow a mathematical model to be constructed of consumer preferences with respect to the principal components, and they may therefore be linked to the initial sensory descriptors.

[0084] The "PREFMAP" model allows the preferences of an individual or group to be plotted and analysed and allows the preferences of individuals in respect of various choices, options or characteristics to be visualised and understood.

[0085] In summary, it is proposed to model the sensory appraisal of a group of cosmetic products, via a computer-implemented method. The cosmetic products are characterized on the one hand by sensory descriptors, and by a set of meta-variables defined by combinations of the sensory descriptors, and on the other hand by respective preference scores. The method comprises a multivariate statistical analysis, namely decorrelation of the meta-variables into principal components, and generation of the sensory appraisal model, this comprising regression of the preference scores to the principal components.

[0086] Thus, the map MVMAP (Figure 3) allows preference scores to be analysed for a set of cosmetic products differentiated by meta-variables based on their sensory descriptors, which meta-variables are related, by regression, to the preferences of consumers.

[0087] A plurality of so-called "nested" models may be used to regress the consumer preference data: the vector model, the circular model, the elliptical model and the quadratic model (Schlich, 1995). For example, with X and Y the first two principal components of the map MVMAP, and with P a modelled preference value, the quadratic model may comprise a regression of the form "P = a + bX + cY + dX2+ eY2+ fXY".

[0088] The goal is to predict the most appreciated products within a sensory space, for a given group of consumers.

[0089] Thus, on the basis of the reliable results of the meta-variable-based model described above, it is possible to provide a computer-implemented computational tool, in particular a software tool, for assisting with design of cosmetic products, comprising employing the sensory appraisal model to generate a decision tree comprising a sequence of conditions on values of meta-variables culminating in a theoretical prediction of the appraisal of the cosmetic product.

[0090] In this regard, reference is made to Figure 4.

[0091] Figure 4 illustrates a decision tree comprising a sequence of conditions <Sk; >Sk on values of meta-variables MVk, with (k=[l; 2; 3; 4; 5; 6; 7]), culminating in a theoretical prediction of the appraisal of the cosmetic product lying between +1 and - 1 (where +1 indicates prediction of a preference in the appraisal, 0 indicates prediction of a neutral appraisal, and -1 indicates prediction of a rejection in the appraisal).

[0092] Specifically, when the quadratic regression model of the PREFMAP is applied between consumer perceptions (appraisal ratings of all the products) and metavariables (computed from sensory descriptors of all the products), it is possible to estimate, for each meta-variable MV1-MV7, respective thresholds S1-S7 for their values allowing appreciation to be maximized.

[0093] In order to then create this ideal product based on these results, the metavariables must be converted into specific characteristics: the meta-variables are translated into clear specific characteristics.

[0094] For example, if the model predicts an ideal complexity of 4, the ideal product will have the characteristic of having 4 olfactory notes, etc. An ideal -product synopsis may thus be created on the basis of these predictions, in practice taking the form of a sequence of conditions forming a decision tree such as illustrated in Figure 4.

[0095] Thus, it is proposed, in a method for designing a cosmetic-product chemical composition, to adjust the composition with chemical compounds having sensory properties, so as to satisfy a sequence of conditions on meta-variables, said sequence of conditions being contained in a decision tree generated by the tool for assisting with design of cosmetic products, so as to culminate in a preference in the appraisal. Of course, a practical evaluation with adjustments will then possibly be carried out to verify whether the prediction is reliable and optionally to adjust the conditions of the decision tree.

[0096] Specifically, it will be noted that the method for modelling sensory appraisal of a group of cosmetic products is advantageously implemented entirely for at least one specific type of consumer, i.e. for appraisal ratings given by users having the same commercial profile.

[0097] In addition to appraisal data relating to one type of consumer, at least some of the meta-variables MV1-MV7 and their respective threshold values S1-S7 are specific to each type of consumer.

[0098] Figure 5 illustrates a computing system SYS, a computer for example. The computing system SYS comprises a processing unit UT and a memory MEM.

[0099] A computer program PRG is stored in the memory MEM. The computer program comprises instructions that, when the program is executed by the processing unit UT, cause the latter to implement a tool OTL for assisting with design of a cosmetic product, such as described above.

[0100] Figure 6 illustrates a computer-implemented method for modelling sensory appraisal of a group of cosmetic products. As indicated above, the cosmetic products are characterized by sensory properties defined by sensory descriptors, and by a set of meta-variables (MV1-MV7) defined by combinations of sensory descriptors. The modelling method comprises a multivariate statistical analysis 10, namely decorrelation of the meta-variables into principal components (MVMAP), and generation 11 of the sensory appraisal model based on the principal components.

[0101] Figure 7 illustrates a method for designing a cosmetic-product chemical composition.

[0102] The designing method comprises employing 20 a decision tree generated by a tool for assisting with design of cosmetic products, such as the tool described above, in light of meta-variables computed from sensory descriptors of an initial cosmetic product having a given chemical composition.

[0103] The designing method then comprises adjusting 21 the composition of the chemical product with chemical compounds having sensory properties, so as to satisfy a sequence of conditions on the meta-variables, which sequence is contained in said decision tree.

Claims

CLAIMS1. Computer-implemented method for modelling sensory appraisal of a group of cosmetic products (Rl, R2), the cosmetic products being characterized by sensory properties defined by sensory descriptors, and by a set of meta-variables (MV1-MV7) defined by combining the sensory descriptors, the method comprising a multivariate statistical analysis, namely decorrelation of the meta-variables into principal components (MVMAP), and generation of the sensory appraisal model based on the principal components.

2. Method according to Claim 1, wherein the cosmetic products are further characterized by respective preference scores, and generation of the sensory appraisal model comprises regression of the preference scores to the principal components.

3. Method according to Claim 1 or 2, wherein the principal components are configured to orient a plane (MVMAP) containing respective positions for each cosmetic product, the meta-variables (MV1-MV7) being elaborated so that the principal components orient the plane in a manner that groups the positions of the cosmetic products depending on their respective sensory appraisals (Rl, R2).

4. Method according to one of Claims 1 to 3, wherein the set of cosmetic products belongs to a database containing the definition of each cosmetic product by the sensory descriptors, and a preference score assigned to the products.

5. Method according to one of Claims 1 to 4, wherein appraisal is modelled for at least one type of consumer, and wherein at least one of said meta-variables (MV1-MV7) is dedicated to each respective type of consumer.

6. Method according to one of Claims 1 to 5, wherein the sensory appraisal is olfactory, the sensory descriptors defining properties relating to olfaction.

7. Method according to one of Claims 1 to 6, wherein, statistically, the metavariables (MV1-MV7) are discriminant, continuous and follow a normal distribution.

8. Computer-implemented tool for assisting with design of a cosmetic product, comprising employing the sensory appraisal model obtained via the method according to one of Claims 1 to 7 to generate a decision tree (Figure 4) comprising a sequence of conditions (<S1-S7; >S1-S7) on values of the meta-variables (MV1-MV7) culminating in a theoretical prediction of the appraisal of the cosmetic product (1; 0; -9. Tool according to Claim 8, wherein the theoretical prediction of the appraisal of the cosmetic product is made for at least one type of consumer, and wherein the sequence of the decision tree and said conditions are at least partly dedicated to each respective type of consumer.

10. Method for designing a cosmetic-product chemical composition and adjusting the composition with chemical compounds having sensory properties, so as to satisfy a sequence of conditions on meta-variables, said sequence of conditions being contained in a decision tree generated by the tool for assisting with design of cosmetic products according to either of Claims 8 and 9, culminating in a preference in said appraisal.

11. Computer program product comprising instructions that, when the program is executed by a computer, cause the latter to implement the method according to one of Claims 1 to 7, or the tool according to either of Claims 8 and 9.

12. Computer-readable recording medium comprising instructions that, when they are executed by a computer, cause the latter to implement the method according to one of Claims 1 to 7, or the tool according to either of Claims 8 and 9.

Images