JP7349125B2 - Beauty evaluation method, sensory evaluation method, and system - Google Patents

Description

The present invention relates to a beauty evaluation method, a sensory evaluation method, and a beauty evaluation system for evaluating the beauty needs and skin type of customers and subjects based on interview data.

2. Description of the Related Art Technologies disclosed in Patent Documents 1 to 4, for example, are known as techniques for performing a beauty judgment that evaluates a customer's or examinee's skin type based on interview data.

Patent Document 1 (Patent No. 6404677) discloses that skin-related interview data and image data of keratin and/or sebum are subjected to data processing to obtain diagnostic data, and these diagnostic data are calculated to output skin diagnosis results. A skin condition determination system is disclosed. However, with this diagnostic system, it takes time for the examinee to prepare the image data, and the server device that processes the image data to create diagnostic data, performs calculations on the diagnostic data, and outputs the skin diagnosis results. The disadvantage is that program development is costly.

It is desirable that cosmetic skin condition evaluations in beauty salons, cosmetics stores, etc. be carried out simply by interviewing subjects or customers and based on their self-judgment, using a method that does not require much effort or cost. In this interview method, in addition to mere listening, various ideas are disclosed in Patent Document 2 and the like.

Patent Document 2 (Japanese Unexamined Patent Publication No. 2002-157469) requires respondents to answer a checklist regarding age, gender, living environment, skin moisture content, sebum content, skin concerns, skin sensitivity, and skin texture preferences. discloses an invention of a cosmetic purchasing system having means for automatically determining a user's skin quality. Here, the user's skin type is classified into sensitive, all skin types, mixed skin type, normal skin type, oily skin type, and dry skin type. Further, the system stores cosmetic data including cosmetic product names associated with information regarding the user's skin, and can transmit this cosmetic data to the user terminal.

Incidentally, when selecting cosmetics, care products, and beauty services suitable for each individual, skin type judgment is an important determining factor for consumers and customers, regardless of whether they are conscious or objective. The above-mentioned interview method is suitable for beauty evaluation of this skin type because it is simple and can be implemented at low cost. However, in beauty evaluations that directly ask about the dryness and oiliness of the skin, as seen in the checklist items of Patent Document 2, the inventors believe that I found it difficult to understand. In other words, there is not much correlation between the subject's self-evaluation of the skin type based on direct questions and the corresponding physical quantity measured by the measuring device. Regarding dryness, oiliness, etc., which determine skin type, it is considered desirable to make a comprehensive judgment based on answers to multiple questions, rather than solely based on answers to direct questions.

Patent Document 3 (Japanese Unexamined Patent Publication No. 2017-120595) discloses an invention of a method for improving response accuracy when evaluating the application state of cosmetics using a questionnaire. In this invention, evaluation scores are calculated based on evaluation values that are answers to evaluation terms (corresponding to the above question items), and evaluation scores are aggregated for each category of generic evaluation terms (corresponding to the above skin types). and output the evaluation results including the aggregated results. However, the disclosure of the invention is limited to the state of application of cosmetics, and what evaluation terms and generic evaluation terms should be used when evaluating the beauty needs and skin type of a subject. Furthermore, it is completely unclear how evaluation scores should be set based on evaluation terms in the above case.

In Patent Document 4 (Japanese Patent No. 5419004), in which the patented invention by the present inventor is described, basic data is collected by interviewing a large number of subjects according to their age, and from the basic data, a factor analysis method is used to determine the sebum excess factor. Disclosed are inventions such as a method for extracting skin dryness factors, and when interviewing a subject of a specific age, two-dimensionally classifying the subject's skin type based on the interview data. There is.
The inventor realized two things in the process of implementing this patented invention. The first thing we noticed is that there is room to improve the accuracy of skin type determination through factor analysis by reviewing the question set, such as by reducing the number of direct questions. The second thing I noticed is that the above two-dimensional classification of skin types based on excess sebum factor and dry skin factor has a strong correlation with basic cosmetics groups such as face wash, cleansing lotion, emulsion, and milk in cosmetics classification. . In addition to basic cosmetics, there are other types of cosmetics, such as aesthetic cosmetics such as serums and creams, base makeup cosmetics such as foundation, and point makeup cosmetics such as lipstick. If factors that are highly correlated with each cosmetic product can be found through factor analysis, a comprehensive model of customer psychology in purchasing cosmetics etc. can be established. That way, we can accurately understand the customer's skin type and beauty needs, and provide beauty advice such as recommending cosmetics and beauty services that are suitable for the customer, and suggesting how to use them, based on a comprehensive purchasing psychology model. , it may be possible to do this more accurately. Although the number of subjects in the basic data in Patent Document 4 was 4227, it is expected that the accuracy of the above model will be improved by using larger-scale big data as the basic data.

Patent No. 6404677 Japanese Patent Application Publication No. 2002-157469 JP 2017-120595 Publication Patent No. 5419004

As mentioned above, conventional techniques for performing beauty judgments that evaluate the skin type and beauty needs of customers and test subjects based on interview data require high-cost or labor-intensive means other than interview data; There is a possibility that the accuracy of determining the skin condition of the examinee can be further improved by reducing the number of questions asked, and that the skin type and beauty needs to be determined are only those that are suitable for a specific type of cosmetic group in the cosmetics classification. There were issues such as the lack of correlation.

In view of the above-mentioned problems, it is an object of the present invention to make a beauty judgment based on a more comprehensive interview based on classification of beauty needs objectively obtained by statistical methods based on big data, and preferably classification of skin type. An object of the present invention is to provide a simple, low-cost, and highly accurate beauty evaluation method, sensory evaluation method, and beauty evaluation system that can perform the following.

The present invention has been made to solve the above problems, and a first aspect of the present invention is based on basic data, which is the answers of many subjects to three or more skin-related questions, and the customer's A beauty evaluation method for determining the beauty of the customer by processing interview data, which is answers to questions, by executing a computer control program, and the computer control program performs at least the following from the basic data by a factor analysis method. a factor extraction means for extracting an EST factor related to aesthetics and an MB factor related to base makeup that is orthogonal to the EST factor; and an EST factor score S _EST of the customer based on the interview data;
a beauty needs determining means for determining the beauty needs of the customer from the position of a point (S _EST , S _MB ) on a beauty needs identification surface having an XY coordinate axis by calculating a score S _MB of the B factor; It is a beauty judgment method characterized by having the following.

In a second aspect of the present invention, the question includes at least one item related to firmness, elasticity, brightness, transparency, skin texture, and youthfulness, which are closely related to aesthetics, and base makeup The beauty evaluation method includes at least one of the following items, which are closely related to makeup smearing, poor makeup application, and coarse texture.

In a third aspect of the present invention, the factor extraction means extracts at least an O factor related to excessive sebum and a D factor related to skin dryness in addition to the EST factor and the MB factor, and the computer control program , based on the interview data, calculate the O factor score S _O and the D factor score S _D of the customer, and calculate the point (S _O , S _D ) The beauty determining method includes skin type determining means for determining the customer's skin type from the position of the customer.

In a fourth aspect of the present invention, the computer control program displays the degree of contribution of the customer's answers to each question to the customer's EST factor score S _EST and/or MB factor score S _MB . The beauty judgment method has a factor score answer contribution display means.

A fifth aspect of the present invention is a beauty determination method that determines whether or not at least one beauty product g should be recommended to the customer in addition to determining the customer's beauty needs and/or skin type; The questions include questions asking whether the subject uses beauty product g, and the computer control program calculates the customer's n factor scores S=(S ₁ , S ₂ , . . . , S _n ); The usage rate is calculated based on the number of beauty test subjects who have factor scores included in a certain region of n-dimensional vector space R ⁿ , including the customer's n factor scores S. The beauty judgment method is calculated based on the proportion of subjects using product g.

A sixth aspect of the present invention is a beauty determination method, in which, in addition to determining the customer's beauty needs and/or skin type, it is determined by machine learning whether at least one beauty product g should be recommended to the customer. Yes, the question includes a use/non-use question that asks whether the subject uses beauty product g, and the computer control program calculates the factor scores of the subject or the customer, Alternatively, it has a prediction function output value calculation means for calculating the output value of the prediction function, which takes a numerical value representing the answer to the question as an input value, and uses a predicted value of the numerical value representing the answer to the use/absence question as an output value. In the beauty judgment method, the prediction function is optimized by machine learning using basic data.

A seventh aspect of the present invention is the computer control program.

An eighth aspect of the present invention is a computer-readable storage medium, characterized in that a parameter of the computer control program or a parameter defining the prediction function is recorded.

A ninth aspect of the present invention is a sensory evaluation method for a beauty product using the beauty evaluation method, in which a cooperator who cooperates with the evaluation uses the beauty product both before and after using the beauty product for a certain period of time. Sometimes, the cooperator is interviewed, and based on the cooperator's answers to the questions, one or more factors among the EST factor, MB factor, O factor, D factor, and other factors. This sensory evaluation method is characterized in that the score is calculated both before and after the period, and the improvement effect is measured from the amount of change in the score during the period.

A tenth aspect of the present invention is composed of an independent computer or a server and a terminal that can communicate with each other in two directions, and the basic data that is the answers of a large number of subjects to three or more skin-related questions and the customer This is a beauty evaluation system for determining the beauty of the customer by processing the interview data that is the answer to the question of a factor extraction means for extracting at least an EST factor related to aesthetics and an MB factor related to base makeup that is orthogonal to the EST factor _; , MB factor score S _MB , and determines the beauty needs of the customer from the position of a point (S _EST , S _MB ) on a beauty needs identification surface having an XY coordinate axis. It is a beauty judgment system characterized by having the following.

An eleventh aspect of the present invention is an e-learning system in which students can learn about determining beauty needs of customers, and the system is composed of an independent computer or a server and a terminal that can communicate bidirectionally with each other; Judgment result display means configured to be able to execute a control program, and for displaying the judgment result of the beauty judgment by the beauty judgment system regarding real or virtual customer interview data, or the judgment result and the student. This is an e-learning system characterized by having a judgment result comparison means for comparing the judgment result with the e-learning system itself.

According to the first aspect of the present invention, basic data, which is the answers of a large number of subjects to three or more skin-related questions, and interview data, which is the customer's answers to the questions, are executed by a computer-controlled program. The computer control program calculates at least an EST factor related to aesthetics and a base makeup factor orthogonal to the EST factor from the basic data using a factor analysis method. a factor extracting means for extracting an MB factor related to the customer; and a factor extracting _means for extracting _{an MB} factor related to A beauty needs determining method can be provided, comprising: beauty needs determining means for determining the customer's beauty needs from the position of the point (S _EST , S _MB ) on the beauty needs identifying surface provided.

Here, "a large number of subjects" means, but is not limited to, approximately 1,000 or more subjects, preferably 8,000 or more subjects, and more preferably 16,000 or more subjects. means. The larger the basic data consisting of responses from subjects who belong to a sample group randomly sampled from the population of the target customer group, the greater the relative position of each customer in the sample group as understood by the scores of each factor. This reduces the deviation from the true relative position of the customer in the population, improving the accuracy of determining beauty needs and skin type.

In this embodiment of the present invention, skin type and beauty needs are determined not only from the customer's answers to direct questions regarding excessive sebum and dry skin, but also by integrating answers to all questions using factor analysis. Since the customer's scores for each factor are calculated and the customer's skin type and beauty needs are determined based on the scores of these factors, the accuracy of the determination can be improved. Furthermore, in the present invention, the customer's beauty needs are visually expressed as the position of a point (S _EST , S _MB ) on the beauty needs identification surface with XY coordinate axes, so both the customer and the seller can quickly understand the customer's beauty needs. You can understand and share information on beauty needs of people.

The EST factor is an extremely important factor that explains the largest part of the variance in the answers to each question in the basic data. Its explanatory power in the question set of Examples described below is greater than the O factor regarding excess sebum and the D factor regarding skin dryness. The cosmetics group that has a strong correlation with the EST factor is the aesthetic cosmetics group, and the fact that the number of products of this cosmetics group on the market is extremely large is also considered to indicate the importance of the EST factor. Furthermore, the MB factor is an important factor that explains a large part of the variance in the answers to each question in the basic data. Its explanatory power in the question set of Examples described below is second only to the EST factor, the O factor related to excess sebum, and the D factor related to skin dryness. The cosmetics group that has a strong correlation with the MB factor is the base makeup cosmetics group, and the large number of products of this cosmetics group on the market is considered to indicate the importance of the MB factor. According to this embodiment of the present invention, it is possible to more accurately determine and understand the beauty needs of the customer based on the scores of the EST factor and the MB factor.

According to the second aspect of the present invention, the question includes at least one item related to firmness, elasticity, brightness and transparency, skin texture, and youthfulness, which are closely related to aesthetics, and It is possible to provide the beauty evaluation method that includes at least one of the following items, which are closely related to base makeup, such as crumbling makeup, poor makeup application, and coarse texture. In this embodiment of the present invention, since the question includes at least one of the above items, the customer's EST factor score and/or MB factor score can be appropriately calculated to accurately determine the customer's beauty needs. Can be judged and understood. In order to more accurately judge and understand customers' beauty needs, it is preferable that the above-mentioned questions include at least two of the above-mentioned items that are closely related to aesthetics, and that the above-mentioned questions include at least two of the above-mentioned items that are closely related to base makeup. It is desirable that the question includes at least two items, and more preferably, the question includes all of the items that are closely related to aesthetics, and it is desirable that the questions include all of the items that are closely related to base makeup.

According to the third aspect of the present invention, the factor extraction means extracts at least an O factor related to excessive sebum and a D factor related to skin dryness in addition to the EST factor and the MB factor, and The program calculates the customer's O factor score S _O and D factor score S _D based on the interview data, and calculates a point (S _O , It is possible to provide the beauty determining method, which includes a skin type determining means for determining the customer's skin type from the position of S _D ).

Surprisingly, customers' subjective perceptions of their skin's dryness and oiliness have a low correlation with instrument measurements. In the present invention, instead of determining skin type and beauty needs only from the customer's answers to direct questions regarding dry skin and excessive sebum, the present invention reduces the number of questions that directly ask about the degree of dryness and oiliness of the skin. Using the factor analysis method, the answers to all questions are combined to calculate the customer's O factor score related to excess sebum, D factor score related to dry skin, and other factor scores. Since the customer's skin type and beauty needs are determined directly and indirectly, the accuracy of the determination can be improved. In addition, in this embodiment of the present invention, in addition to the O factor score related to excess sebum and the D factor score related to skin dryness, the EST factor score related to aesthetics and the MB factor score related to base makeup are used for determination. , it is possible to comprehensively understand the customer's skin type and beauty needs and make more accurate and detailed decisions. Furthermore, in this embodiment, the customer's skin type is visually expressed as the position of the point (S _O , S _D ) on the skin type identification surface with XY coordinate axes, so both the customer and the seller can quickly identify the customer. You can understand and share information about your skin type.

According to the fourth aspect of the present invention, the computer control program displays the degree of contribution of the customer's answers to each question to the customer's EST factor score S _EST and/or the MB factor score S _MB . It is possible to provide the above-mentioned beauty judgment method, which includes a factor score answer contribution degree display means for displaying factors. In this embodiment of the present invention, since the computer control program has the factor score answer contribution display means, the customer and/or the seller can determine whether the answer to each question by the customer is the customer's EST factor score S _EST and/or the seller. Or, you can know the contribution of the MB factor score S to how much it contributes to _MB . Therefore, according to this embodiment, the customer and/or seller, etc.
It is possible to understand the breakdown of customers' beauty needs in more detail, which can be used as a reference for purchasing and recommending beauty products.

According to a fifth aspect of the present invention, in addition to determining the beauty needs and/or skin type of the customer, it is also a beauty determination method that determines whether or not at least one beauty product g should be recommended to the customer. , the questions include questions asking whether the subject uses beauty product g, and the computer control program calculates the customer's n factor scores S = where n is an integer of 1 or more. (S ₁ , S ₂ , . . . , S _n ); and a usage rate standard recommendation determining means for determining whether beauty product g should be recommended to the customer based on at least one usage rate calculated from (S 1 , S 2 , . . . , S n ). , the utilization rate is calculated based on the number of subjects whose factor scores are included in a certain region of n-dimensional vector space R ⁿ , which includes the customer's n factor scores S. , it is possible to provide the beauty determination method that is calculated based on the proportion of subjects using beauty product g.
The usage rate standard recommendation determination means of this embodiment is configured to use the beauty product g by a test subject who has a factor score close to the customer's factor score calculated based on the interview data in the n-dimensional number vector space R ⁿ . This form of the present invention is based on the factor analysis of the basic data, which is big data, because the usage rate is calculated from the basic data and it is determined whether or not the beauty product g should be recommended to the customer based on the calculated usage rate. It is possible to provide a beauty evaluation method that can objectively recommend beauty products based on legal statistical analysis and customer interview data.

According to the sixth aspect of the present invention, in addition to determining the customer's beauty needs and/or skin type, the beauty judgment determines by machine learning whether at least one beauty product g should be recommended to the customer. The method includes a usage/non-use question that asks whether or not the subject uses beauty product g, and the computer control program is configured to determine whether or not the subject or the customer uses the beauty product g. Prediction function output value calculation means for calculating an output value of a prediction function, the input value being a score or a numerical value representing the answer to the question, and the output value being a predicted value of the numerical value representing the answer to the use/non-use question. The prediction function can provide the beauty judgment method that is optimized by machine learning using basic data.
The prediction function output value calculation means of this embodiment takes as an input value a numerical value representing a customer's factor score or an answer to a question in the medical interview data, and takes as an output value a predicted value of a numerical value representing an answer to the use/non-use question. The output value of a prediction function is calculated, and the prediction function is optimized by machine learning using basic data. Therefore, this embodiment of the present invention combines machine learning using basic data, which is big data, and customer interview data. Based on this, it is possible to provide a beauty evaluation method that can statistically and objectively determine the recommendation of beauty products.

According to the seventh aspect of the present invention, the computer control program can be provided.

According to an eighth aspect of the present invention, it is possible to provide a computer-readable storage medium in which analyzed parameters that are parameters related to the computer control program or parameters that define the prediction function are recorded. . For example, in the case of machine learning, parameters defining a prediction function are optimized by machine learning based on basic data, but machine learning often requires computation time. Therefore, by recording analyzed parameters, which are optimized learned parameters, in a computer-readable storage medium, machine learning that requires calculation time does not need to be performed on other computers, servers, or terminals. The beauty judgment according to the present invention can be performed by reading previously analyzed parameters from the storage medium and executing the computer control program. The same applies to the case where a factor analysis method is used; in this case, the parameters related to the computer control program are factor loading coefficients, factor score coefficients, variables and independent variables obtained by factor analysis using basic data. Refers to analyzed parameters such as the covariance matrix of factors.

According to a ninth aspect of the present invention, there is provided a sensory evaluation method for a beauty product using the beauty evaluation method, in which a cooperator who cooperates with the evaluation uses the beauty product before and after using the beauty product for a certain period of time. At both times, the cooperator was interviewed, and based on the cooperator's answers to the questions, any one or more of the EST factor, MB factor, O factor, D factor, and other factors. It is possible to provide a sensory evaluation method characterized in that the scores of the factors are calculated both before and after the period, and the improvement effect is measured from the amount of change in the scores during the period.

Normally, in sensory evaluation during the development phase of cosmetics, etc., taking the level of "moistness" as an example, subjects are asked to test use the cosmetics, etc. for a short time in a laboratory, and A method is adopted in which the patient is interviewed by direct questions, the selected evaluation options are used as a moisturizing scale, and the improvement effect is measured by the amount of change in the moisturizing scale before and after the test. On the other hand, in this embodiment of the present invention, the improvement effect is measured as a change in score before and after a certain period (for example, several days or more) using the dryness factor score S _D calculated from answers to questions that are not necessarily direct. This embodiment is a sensory evaluation method that does not use a measuring device, and can provide a sensory evaluation method that can be used in the actual use of cosmetics or beauty services in accordance with actual usage conditions outside the laboratory.

According to the tenth aspect of the present invention, the computer is composed of an independent computer or a server and a terminal that can communicate with each other in two directions, and the basic data that is the answers of a large number of subjects to three or more skin-related questions is collected. , interview data that is answers to the customer's questions, and is processed by executing a computer control program to determine the customer's beauty, and the computer control program performs a factor analysis method from the basic data. factor extraction means for extracting at least an EST factor related to aesthetics and an MB factor related to base makeup that is orthogonal to the EST factor, and a score S of the EST factor of the customer based on the interview data. _EST and the score _SMB of the MB factor are calculated, and the beauty needs of the customer are determined from the position of the point (S _EST , S _MB ) on the beauty needs identification plane having the XY coordinate axes. It is possible to provide a beauty judgment system characterized by having a judgment means.

According to an eleventh aspect of the present invention, there is an e-learning system in which students can learn about determining beauty needs of customers, and the system is composed of an independent computer or a server and a terminal that can communicate with each other in two directions, Judgment result display means configured to be able to execute the computer control program and display the judgment result of the beauty judgment by the beauty judgment system regarding real or virtual customer interview data, or the judgment result. It is possible to provide an e-learning system characterized by having a judgment result comparison means for comparing the judgment result and the judgment result by the student himself.

FIG. 1 is a table showing the analysis results using the conventional factor analysis model [13], where FIG. 1A shows the factor loadings and FIG. 1B shows the factor score coefficients. FIG. 2 is a graph plotting the above factor score coefficients. FIG. 3 is a table showing the analysis results using the factor analysis model [21] according to an embodiment of the present invention, where FIG. 3A shows factor loadings and FIG. 3B shows factor score coefficients. FIG. 4 is a table showing correlation coefficients between factor scores in the analysis results using the above two models. FIG. 5 is a graph plotting the factor score coefficients in the analysis results using the above factor analysis model [21]. FIG. 6 is a table showing specific examples of questions and customer answers according to an embodiment of the present invention. FIG. 7 is a graph diagram illustrating the beauty needs and skin type of the customer on the beauty needs identification surface and the skin type identification surface in one embodiment of the present invention. FIG. 8 is a graph diagram illustrating the degree of contribution of the customer's answers to each question to the score of each factor in an embodiment of the present invention. FIG. 9 is a table showing factor loadings in the analysis results by the factor analysis model [34] according to an embodiment of the present invention. FIG. 10 is a table showing factor score coefficients in the analysis results using the above factor analysis model [34]. FIG. 11 is a table showing an example of a display of a recommendation determination result for determining whether each beauty product should be recommended to a customer in an embodiment of the present invention. FIG. 12 is a configuration diagram showing two configuration examples of a neural network according to an embodiment of the present invention. FIG. 13 is a flow diagram of a beauty determination method according to an embodiment of the present invention. FIG. 14 is a detailed flow diagram of each of the beauty determination step, the determination result display step, and the recommendation determination result display step in the flow diagram of FIG. 13. FIG. 15 is a block diagram of an embodiment of the present invention comprised of an independent computer (eg, a personal computer). FIG. 16 is a configuration diagram of a computer control program according to an embodiment of the present invention. FIG. 17 is a configuration diagram showing two embodiments of the present invention in which a server and a terminal are connected via a network. FIG. 18 is a table showing an example of classification of cosmetics. Figure (19A) is an explanatory diagram showing the classification of skin types by the conventional factor analysis model [13], and Figure (19B) is an explanatory diagram showing the classification of beauty needs according to the present invention.

Embodiments of the beauty judgment method and system according to the present invention will be described in detail below with reference to the drawings.

<1. Conventional beauty assessment to identify skin type>
(Summary of Patented Invention) In the patented invention disclosed by the present inventor in Japanese Patent No. 5419004, a factor analysis method is used to determine the factor of excess sebum from interview data of women by age using questions regarding beauty and skin condition. Skin drying factors are extracted. The excess sebum factor and dry skin factor can be explained by a ``one-dimensional model'' in which they are inversely correlated in younger age groups, and a ``two-dimensional model'' in which they are almost uncorrelated in middle-aged and older age groups. As shown in Figure (19A), the customer's scores for these factors are distributed on a two-dimensional discrimination surface defined by the X-axis related to excess sebum and the Y-axis related to dry skin, and the customer's scores for these factors indicate Skin types are classified into four quadrants: dry skin, normal skin, oily skin (or skin with excessive sebum), and dry skin with excessive sebum (or secretly dry skin). The axis obtained by rotating the X-axis by 45 degrees around the origin is defined as the A-axis, and the axis obtained by rotating the Y-axis by 45 degrees is defined as the B-axis. The A-axis passes through the third quadrant (normal skin) and the first quadrant (excessive sebum-type dry skin), and the B-axis passes through the fourth quadrant (oily skin) and the second quadrant (dry skin). The B-axis corresponds to a one-dimensional model for the younger age group, and the A-axis and B-axis correspond to the two-dimensional model for the middle-aged and older age groups. Furthermore, the coordinates of the A-axis indicate the level of skin dissatisfaction regarding sebum and skin dryness. The patented invention is a beauty judgment method for classifying skin types using the above-mentioned identification surface.

<2. Beauty judgment of the present invention that identifies beauty needs>
(Discovery of the third factor) As a result of re-analyzing the data accumulated in the process of implementing the above patented invention, in addition to the excess sebum factor (oily factor) and the dry skin factor (drying factor), an independent third factor was discovered. (hereinafter referred to as "Factor No. 3"). The inventors newly interviewed 20,000 women between the ages of 20 and 69 using the questions described below, and obtained valid responses from 19,468 women. In this specification, the above discovery and the present invention will be explained using as an example the results of an analysis performed using a factor analysis method using the above valid answers as basic data.
Figure (1A) shows the selection of two-choice answers (0 or 1) to a set of 13 questions useful for skin type analysis from the above basic data, and the selected answers are analyzed and extracted using factor analysis. FIG. 3 is a table showing the factor loadings of the three factors. The reason why the number of factors to be extracted is set to 3 is because among the eigenvalues of the covariance matrix of the above answers standardized to have an average of 0 and a standard deviation of 1, the number of eigenvalues that are 1 or more is 3. The maximum likelihood method was used to estimate the factor loadings. The factor loadings shown in Table (1A) are the factor loadings after six rotations were performed until convergence using the varimax method without Kaiser's normalization. Furthermore, FIG. 1B is a table showing factor score coefficients obtained by analysis using the above factor analysis method. Regression method was used to calculate factor scores. Furthermore, Figure (2A) is a graph in which the factor score coefficient of the greasy factor is plotted on the horizontal axis and the factor score coefficient of the drying factor is plotted on the vertical axis, and Figure (2B) is a graph plotting the factor score coefficient of the No. 3 factor and the drying factor. 2C) is a graph in which the factor score coefficients are similarly plotted for the No. 3 factor and the greasy factor. As can be seen from the table in Figure (1B) and the graph in Figure 2, the third No. 3 factor is related to makeup dissatisfaction (skin quality dissatisfaction) such as "makeup comes off,""makeupdoesn't go well," and "the texture is rough." It was found that there is a high correlation with related question items. Hereinafter, the above model will be referred to as a factor analysis model [13].

Table 1 shows the correlation coefficients between the scores of each factor in the factor analysis model [13]. It can be seen that the correlation between the scores of different factors is small; in other words, the factors are almost orthogonal. This point can also be read from Figure 2. For example, in Figure 2A, questions related to makeup dissatisfaction (skin quality dissatisfaction), such as "makeup crumbles," "makeup doesn't apply well," and "texture is rough," are plotted near the origin surrounded by a dotted line. . In other words, when considering the three diagrams in Figure 2 as projection diagrams of one spatial diagram in three orthogonal directions, the above question items that are highly related to No. 3 factor are is plotted near the axis of factor score coefficients (referred to as the Z axis). This indicates that the axes (X-axis and Y-axis) of the factor score coefficients of the oiliness factor and the dryness factor are substantially orthogonal to the Z-axis.

(Search for the fourth factor) Furthermore, by discovering a fourth factor that is orthogonal to factor No. 3 and also orthogonal to the oily factor and the dryness factor, and obtaining a new discrimination surface in a higher dimension, we will be able to utilize the knowledge of the above patented invention. We considered and repeatedly considered the possibility of a new beauty evaluation method that includes the following. Specifically, using the above basic data, we combined the above 13-item question set constructed in Patent No. 5419004 with a new set of interview items, and in addition to the dry factor and the oily factor, the third and , we repeated the process of extracting the fourth factor that forms the pair through analysis using the factor analysis method. As a result, the above objective was achieved by adding eight new interview items. In other words, five factors were extracted by factor analysis using a question set with (13+8) items. Hereinafter, this model will be referred to as a factor analysis model [21].

Figure (3A) shows two-choice answers (0 or 1) for the 13-item question set and 5-choice answers (-2 (don't care), -1) for the additional eight question items, based on the basic data above. (Not really concerned about it), 0 (Usually), +1 (A little concerned), +2 (Concerned)) responses were converted into binary values (positive values were converted to 1, and other values were converted to 0). It is a table showing the factor loadings of five factors extracted by analyzing the selected answers using a factor analysis method. The eight additional question items are indicated by adding "R" to the end of the question item name, such as "Kime R." The reason why the number of factors to be extracted is set to 5 is that among the eigenvalues of the covariance matrix of the above answers standardized to have an average of 0 and a standard deviation of 1, the number of eigenvalues that are 1 or more is 5. The maximum likelihood method was used to estimate the factor loadings. Moreover, the factor loadings shown in FIG. 3A are the factor loadings obtained after six rotations are performed until convergence using the varimax method without Kaiser's normalization. FIG. 3B is a table showing factor score coefficients obtained by analysis using the above factor analysis method. Regression method was used to calculate factor scores. Further, FIG. 4 is a table showing correlation coefficients between factor scores in the conventional factor analysis model [13] and the new factor analysis model [21].

Table 2 shows the sum of squares of factor loadings for each of the five factors extracted in the factor analysis model [21] and the proportion of variance in responses explained by each factor.
Factor (21-1) is the factor with the highest explanatory power and is the fourth objective factor. This factor will hereinafter be referred to as the EST factor.
Factor (21-2) has a large correlation coefficient of 0.97 between the greasy factor in the factor analysis model [13] and the factor score, and is a factor that roughly corresponds to the above X axis. In the following, this factor will be referred to as the O1 factor.
Factor (21-3) is similar to, but different from, the oily factor, and is presumed to be involved in skin satisfaction due to oily skin. In the following, this factor will be referred to as the O2 factor.
Factor (21-4) has a large correlation coefficient of 0.92 between the No. 3 factor in the factor analysis model [13] and the factor score, and is a factor that can be almost identical to No. 3 factor. Hereinafter, this factor will be referred to as the MB factor.
Factor (21-5) has a large correlation coefficient of 0.90 between the dryness factor in the factor analysis model [13] and the factor score, and is a factor that approximately corresponds to the above Y axis. This factor will be referred to as the D factor below.
The EST factor discovered as the "fourth factor" has a correlation coefficient of 0.07, -0.01, and 0.10 with the MB factor, O1 factor, and D factor, in order of 0.07, -0.01, and 0.10. It has a nearly orthogonal relationship with these four factors and is suitable for the purpose of the search.

FIG. 5A is a graph plotting the factor score coefficient of the EST factor on the horizontal axis and the factor score coefficient of the MB factor on the vertical axis for each question item on a plane. When looking at EST factors from the perspective of cosmetics categories, there is a strong relationship with questions related to aesthetics (firmness, brightness, transparency, skin texture, youthfulness). Furthermore, when looking at the MB factor from the perspective of the cosmetics category, there is a strong correlation with questions related to base makeup (makeup coming off, makeup not applying well, rough texture).

FIG. 5B is a graph plotting the factor score coefficient of the O1 factor on the horizontal axis and the factor score coefficient of the D factor on the vertical axis for each question item on a plane. When looking at the O1 factor from the perspective of the cosmetics category, there is a strong relationship with question items related to oiliness (acne, acne scars). In addition, when looking at factor D from the perspective of the cosmetics category, the question items related to skin dryness (
dryness, prevention of dryness).

As a result of the fact that the scores of the EST factor, MB factor, O1 factor, and D factor are almost uncorrelated, in other words, these factors are almost orthogonal, the question items related to oiliness in Figure (5A) are The factor score coefficients for questions related to skin dryness (acne, acne scars) and skin dryness (dryness, preventing dryness) are small and plotted near the origin. In addition, in Figure (5B), questions related to aesthetics (firmness, brightness and transparency, skin texture, youthfulness) and questions related to base makeup (makeup crumbling, poor makeup application, poor texture) are included. The factor score coefficients for (coarse) are small and plotted near the origin.

As mentioned above, the factor scores for the EST factor and the MB factor are almost uncorrelated.In other words, these two factors are almost orthogonal, so we can interview customers using the questions above and find out the answers to the questions. The customer's score for each factor is determined, and as shown in Figure (19B), for example, a two-dimensional plane (beauty needs identification plane) has two orthogonal coordinate axes: horizontal axis = EST factor score, vertical axis = MB factor score. It is possible to plot the factor scores of the customer on the top of the screen, and to express and understand the beauty needs of the customer based on the position of the plotted points. Customers with high scores on the EST factor are considered to have strong interest and expectations regarding aesthetics, and customers with high scores on the MB factor are considered to be highly dissatisfied with base makeup.
Similarly, since the O1 factor and the D factor are almost orthogonal, for example, a two-dimensional plane (called a skin type identification plane) with two orthogonal coordinate axes, the horizontal axis = score of the O1 factor and the vertical axis = score of the D factor. The customer's factor scores are plotted on the top of the screen, and the customer's skin type can be expressed and understood by the position of the plotted points.

In the table showing the correlation coefficients in Figure 4, in addition to the scores for each factor, "MB-E composite" and "DO composite" are listed as items. These are quantities defined by the following formulas.
MB-E composite = (MB factor score + EST factor score)/2 ^0.5
D-O composite = (D factor score + O1 factor score)/2 ^0.5
The "MB-E composite" is a quantity that represents the size of a customer's beauty needs and the height of their interest in beauty. Customers with a high MB-E composite have high interest in and expectations for the use of beauty products, that is, the purchase and use of cosmetics, and the use of beauty services. Furthermore, the "DO composite" is a quantity corresponding to the coordinates of the A axis in the above patented invention, and is a quantity representing the level of customer's skin dissatisfaction regarding sebum and skin dryness.

The factor analysis model [21] using (13+8) question items includes the knowledge of the patented invention of Patent No. 5419004 related to beauty determination of skin type. According to the factor analysis model [21], in addition to the customer's visit/drop-in needs, specifically the customer's skin type and skin dissatisfaction, the customer's (1) strength of beauty needs, and (2) center of gravity of beauty needs (3) To quickly classify the psychological tolerance for the use of beauty products, realize effective customer response, minimize the burden on customers, and provide prompt, accurate, and comprehensive beauty counseling. This makes it possible to increase sales opportunities for beauty products and encourage customers to become loyal customers. According to the factor analysis model [21], a beauty evaluation method, a beauty evaluation program, a sensory evaluation method, and a beauty evaluation system that include the above-mentioned patented invention and have improved practicality in the sense that they are directly connected to cosmetics sales. can be provided.
As mentioned above, it is desirable to reduce the number of direct questions and estimate the scores for the O1 factor and D factor from the answers to indirect questions. Further, in order to be able to appropriately extract EST factors and MB factors and to speed up judgment, it is desirable that the number of question items is not too large. The optimal set of questions that can obtain factor scores with high accuracy using the minimum number of questions, which were selected taking these conditions into account, are the 21 questions of the factor analysis model [21]. It is.

Note that although the factor scores were calculated by the regression method in the above embodiments, the method for calculating the factor scores in the present invention is not limited to the regression method. Any method for calculating the score of a factor, such as the Bartlett method or the Anderson-Rubin method, in which the score of the factor can be expressed using a linear combination of observables, can be used in the present invention.

<3. Determining and displaying customer beauty needs>
<3-1. An embodiment for determining and displaying customer beauty needs>
An embodiment of the present invention that determines and displays a customer's beauty needs based on a factor analysis model [21] will be described. Basic data, which is big data, is analyzed in advance using a factor analysis method to extract at least the EST factor related to aesthetics and the MB factor related to base makeup, and preferably also the O1 factor related to oiliness and the D factor related to skin dryness. . In addition, calculate the scores for each factor of the subject included in the basic data, divide the distribution of factor scores into multiple intervals for each factor, and record the factor scores at the boundary points of adjacent intervals. . There is no particular limitation on how to determine the score of a boundary point factor, but to give one example, if the score of a certain factor is divided into five sections with approximately equal number of data, then The factor scores for the point, 0.4 quantile, 0.6 quantile, and 0.8 quantile are calculated in advance. The factor score for each quantile may be the minimum value of the factor scores belonging to the interval with the larger factor score among the adjacent intervals, or the minimum value of the factor scores belonging to the interval with the smaller factor score. The maximum score may be used, or the average or weighted average of both may be used. In addition, even if the distribution of factor scores for each quantile is not normal, the average value m of the test subject's score for each factor in the basic data can be calculated based on the position of the quantile in the case of a normal distribution. Based on the standard deviation σ, for example, m-0.84σ (0.2 quantile), m-0.25σ (0.4 quantile), m+0.25σ (0.6 quantile), m+0 It may be determined as .84σ (0.8 quantile).

Next, the customer is asked several questions out of the 21 questions in the factor analysis model [21], and the customer is given two choices (I care (1) or I don't care). (0)) Obtain the answer value. An example is shown in FIG. Figure (6A) shows the appearance of the completed medical questionnaire, showing the selected 20 question items and a certain customer's answers to those question items. In this example, since the question is multiple-choice, the answer is made by checking or not checking the checkbox. Figure (6B) is a table showing that the value of the answer derived from the customer's answer is subsequently processed by the computer control program as 1 (with a check mark) or 0 (without a check mark). It is not visible to the customer.

Then, for the customer, at least the EST factor score and the MB factor score are calculated, and preferably the O1 factor score and the D factor score are also calculated based on the values of the above answers. Then, for each factor, it is determined to which interval the score of each factor of the customer belongs based on the factor score of each quantile. For example, if the EST factor score is divided into five sections, the customer's EST factor score is placed in the first, second, third, fourth, or fifth section in descending order of score. It is determined whether it belongs to an interval, and the number of that interval is called the EST degree. The EST degree of the customer, calculated from the customer's answers illustrated in FIG. 6A, is 2. Similarly, when the MB factor score is divided into five sections, the MB degree, which is the number of the section to which the customer's MB factor score belongs, is 5. Furthermore, when the O1 factor score is divided into five sections, the oiliness level, which is the section number to which the customer's O1 factor score belongs, is 3, and when the D factor score is divided into five sections, the corresponding customer's O1 factor score is 3. The dryness level, which is the number of the section to which the customer's D factor score belongs, is 1.

FIG. 7A is a graph diagram in which the EST degree and MB degree of the customer are plotted on a beauty needs identification surface, which is a two-dimensional coordinate plane with the EST degree on the horizontal axis and the MB degree on the vertical axis. From this diagram, you can read at a glance the strength and center of gravity of the customer's beauty needs. FIG. 7B is a graph in which the customer's oiliness and dryness are plotted on a skin type identification surface, which is a two-dimensional coordinate plane with oiliness on the horizontal axis and dryness on the vertical axis. From this diagram, it is possible to read at a glance the customer's skin type and the level of skin dissatisfaction.

Here, we will supplement the calculation method of the "factor score" of each factor for the customer. Generally, the score f _i of the i-th factor can be calculated using the following equation (1).
(1) f _i = Σa _ij × (x _j - m _j )/σ _j
Here, a _ij is the factor score coefficient estimated by the factor analysis model [21] based on the basic data, x _j is the value of the customer's answer to the jth question, m _j and σ _j are the jth These are the average value and standard deviation of the values of the test subjects' answers to the questions. Σ means the sum of j. In equation (1), the sum for j is generally taken over all 21 questions. However, when interviewing the customer by selecting some questions from 21 questions as in the example shown in FIG. 6, the sum of j in equation (1) is Covers the questions selected in . In addition, in order to determine EST degree, MB degree, oiliness degree, and dryness degree, it is necessary to calculate the "factor score" for each quantile based on basic data in advance. When calculating a person's "factor score" using equation (1), the sum for j is taken over the questions selected in the interview.

Figure (8A) is a bar graph showing the degree of contribution of the customer's answer to each question to the score of each factor presented to the beauty advisor or customer in an embodiment of the present invention. 8B) is a graph diagram illustrating the degree of contribution using a radar chart. Here, the degree of contribution c _ij of the answer to the j-th question of the customer to the score of the i-th factor is defined by the following equation (2) for simplicity.
(2) c _ij = a _ij ×x _j /σ _j
In other words, the contribution degree c _ij is calculated from the score of the i-th factor calculated based on the value of the answer to each question of the customer, if the value of the answer to the j-th question of the customer is 0 (check mark (none), and the score of the i-th factor is subtracted when the values of the answers to other questions remain unchanged. In the case of this example where the answer to the question has two values, 0 and 1, it is simple to define the above contribution as shown in equation (2). However, in general, the following equation (3) ) may define the above contribution c _ij .
(3) c _ij = a _ij × (x _j - m _j )/σ _j
In Figures (8A) and (8B), the graphs titled Dryness, Oiliness, EST, and MB indicate the contribution of each question to the score of the D factor, O1 factor, EST factor, and MB factor, respectively. Show degree.

In the case of the customer illustrated in Figures 7 and 8, first of all, looking at Figure 7, the MB degree is 5, and the EST degree is not so large, at 2, so the customer has stronger beauty needs for base makeup than for aesthetics. Next, looking at Figure 8, it can be seen that the strength of the customer's beauty needs for base makeup is derived from the customer's answers to questions such as ``makeup comes off easily'' and ``coarse texture.'' . According to this embodiment, the customer or the beauty advisor generally knows the customer's EST degree and MB degree, in other words, the customer's position on the beauty needs identification plane, and the customer's answers to each question with respect to the EST factor and MB factor scores. It becomes possible to objectively judge and understand the customer's beauty needs and their center of gravity and details based on the degree of contribution.
Similarly, according to the present embodiment, the customer or the beauty advisor generally determines the customer's oiliness and dryness, in other words, the customer's position on the skin type identification plane, and the customer's respective questions regarding the O1 factor and D factor scores. It becomes possible to objectively determine and understand the customer's skin type, the magnitude of skin dissatisfaction, and its center of gravity and details based on the contribution of the answers.

<3-2. Application to beauty counseling>
The embodiment of the present invention, which determines and displays a customer's beauty needs and skin type based on a factor analysis model [21], can be used to determine and display a customer's beauty needs and skin type through a network such as the Internet, face-to-face, at the time of treatment, or by the customer himself/herself. can be applied to beauty counseling. O1 factor and D factor affect the feeling of using face wash/cleansing (cleansing power, selection of surfactant, etc.), lotion (ingredients, formulation, etc.), emulsion/milk (same as above, emulsification balance, etc.) in basic cosmetics. This has a significant impact on satisfaction and, as a result, on the selection and continued use of those products. In addition, EST factors and MB factors have a great influence on the feeling and satisfaction level of using serums, creams, and foundations, and as a result, on the selection and continued use of these products. By using this technology, we can select and propose products and usage methods that suit the customer's awareness and requests, without having to choose mechanical measurements that require time and maintenance (moisture measurement, sebum measurement, etc.) as the first choice. The goal is to standardize and speed up beauty counseling. Compared to mechanical measurement, it requires less effort and cost, increases the time share of high-level consultation opportunities and proposal responses, and increases the "quality of sales" through appropriate response to customer psychology.
The set of question items in the factor analysis model [21] consists of multiple items related to current skin condition and multiple items related to skin care objectives. The set allows customers to easily ask about their skin concerns and beauty needs, so customers and sellers can share answers and results in a very short period of time, without any discomfort at the beginning of the consultation. In addition, an environment that visually displays current skin type awareness (dryness - oiliness) and beauty needs (MB-EST), and quickly and accurately judges and provides advice on the direction of counseling, product selection, and usage methods. I will provide a. For each of the beauty needs identification surface and the skin type identification surface, for example, in the case of a "5x5" map, it is possible to operate by preparing 25 basic patterns (at least 9 patterns) of beauty advice content. If detailed determination is required for each age group, prepare content for each age group multiplied by the number of patterns described above.

<3-3. Application to sensory evaluation>
The embodiment of the present invention, which determines and displays a customer's skin type and beauty needs based on a factor analysis model [21], can be applied to sensory evaluation. This technology relates to objective information conversion and sharing of subjective awareness regarding skin quality or beauty needs, and has the same purpose and direction as sensory evaluation tests. However, it must be noted that the evaluation methods (questions), evaluation scales (scoring methods), evaluation time lags (period before and after) are different, and data interpretation is also different. Normally, during the development phase, a method is adopted in which, for example, the degree of ``moistness'' is measured on a moisturizing scale using questions and evaluation options that are directly asked before and after the test, and the improvement effect is measured before and after the experiment and in the laboratory. On the other hand, in this technology, the improvement effect is measured by calculating the score of the "drying factor" calculated from the answer results as a change in score before and after a certain period (several days or more). Therefore, it is desirable that the sensory evaluation using this technology be used outside the laboratory and in the actual use of the cosmetic product. Here are some specific examples. For example, when developing a ``highly moisturizing'' product, assuming that the perceived level of skin dryness that satisfies the product's appeal is ``5'', first select a group of ``5'' test subjects through a simple interview. It is possible to provide a development evaluation protocol under actual use conditions that indicates that the product is "suitable" if the average score of the drying factor becomes "4" or less after continuous use for a certain period of time. The dryness factor score can be measured with the minimum set of items required, so by combining other items that the developer deems necessary, it is possible to measure the scores related to high moisturization, such as ``fluffy'' and ``restored texture.'' It is possible to perform sensory measurements of the effects and verify the relevance. In addition, since it is close to the actual usage conditions, it can be used to predict the sales of the formulation product after its release. Note that although we have explained here an example of displaying an ordinal scale from "1" to "5", the scores of the original factors can take continuous values or discrete values (intermittent values), and as they are, Alternatively, it is also possible to perform ordinal scaling with more detailed scales and use it as a sensory evaluation scale.

<4. Recommended beauty product discrimination>
Thus far, several embodiments of the present invention have been described for determining a customer's beauty needs and skin type. Next, an embodiment of the present invention will be described in which, in addition to determining a customer's beauty needs and skin type, it can also be determined whether a beauty product should be recommended to the customer. Note that in this specification, "beauty products" means cosmetics or beauty services, and "use of beauty products" means the purchase or use of cosmetics or the use of beauty services.

<4-1. Recommendation of beauty products by experts or customers themselves＞
In this embodiment, the method is based on the customer's beauty needs, their center of gravity, and details, which are ascertained from the customer's position in the beauty needs identification plane and the contribution of the customer's answers to each question to the scores of the EST factor and the MB factor. Then, an expert such as a beauty advisor or the customer himself/herself determines whether a particular beauty product should be recommended to the customer. Consider cosmetics as an example of beauty products. FIG. 18 is a table showing an example of classification of cosmetics. Cosmetics include basic cosmetics such as face washes, cleansers, lotions, emulsions, and milk, aesthetic cosmetics such as serums and creams, base makeup cosmetics such as foundation, and point makeup cosmetics such as lipstick. be.

The EST factor and the MB factor are closely related to the aesthetic cosmetics group and the base makeup cosmetics group. As shown in Table 3, the beauty needs identification surface is divided in advance into, for example, 5 x 5 areas corresponding to the customer's EST degree and MB degree, and each area has "B", "E", "E & B". A beauty needs code such as Each beauty needs code is associated with a list of beauty products such as cosmetics that are suitable to be recommended to customers having that beauty needs code. The expert or the customer can see the list of beauty products linked to the beauty needs code corresponding to the EST degree and MB degree of the customer and understand the beauty products that should be recommended to the customer. Furthermore, the expert or the customer may select beauty products included in the above list and/or beauty products not included in the above list based on the contribution of the customer's answers to each question to the score of the EST factor and the MB factor. , determines that it should be recommended to the customer.

Moreover, the oily factor and the drying factor are closely related to basic cosmetics. Preferably, as shown in Table 3, the skin type identification surface is divided in advance into, for example, 5×5 regions corresponding to the customer's oiliness and dryness, and each region is marked with “D” and “O”. A skin type code such as "D&O" is given. Each skin type code is associated with a list of beauty products such as cosmetics that are appropriate to be recommended to customers with that skin type code. The expert or the customer can see the list of beauty products linked to the skin type code corresponding to the customer's oiliness and dryness and understand the beauty products that should be recommended to the customer. Furthermore, the expert or the customer may select beauty products included in the above list and/or beauty products not included in the above list based on the contribution of the customer's answers to each question to the score of the oiliness factor and the dryness factor. , determines that it should be recommended to the customer.

<4-2. Recommendation of beauty products based on factor scores and beauty product usage rate＞
(Introduction) In the above embodiment, the task of creating the list in advance, which links recommended beauty products to beauty needs codes or skin type codes, is done solely by experts etc. based on their own experience and knowledge. I will do it. In contrast, in this embodiment, the basic data includes information on whether or not beauty products are used in addition to the examinee's answers to questions about beauty, and a computer control program is executed to statistically analyze the basic data. By doing so, it is determined whether or not a beauty product should be recommended to the customer based on the customer's factor score.

(Factor analysis model [34]) Before explaining this embodiment in detail, a new factor analysis model (referred to as factor analysis model [34]) that includes information on whether or not beauty products are used will first be explained below. . Even if the factor scores calculated based on the above factor analysis model [13] and factor analysis model [21] are combined and analyzed, information on whether beauty products are used or not, beauty products can be used in almost the same way as in this embodiment. However, as will be explained later, the factor analysis model [34] that includes information on whether or not a beauty product is used is more accurate than the factor analysis model [13] or [ 21].

In the factor analysis model [34], the basic data includes information on whether or not beauty products are used, and from the basic data, there are two choices (0 or 1) for the 13 question items (hereinafter referred to as Q1 to Q13). ) and the 5 choices for the additional 8 aforementioned question items (hereinafter referred to as Q14 to Q21) (-2 (doesn't care), -1 (doesn't really care), 0 (usually), +1 (I'm a little concerned about it), +2 (I'm concerned about it)), as well as integer answers to the age question (hereinafter referred to as Q22) and the 12 answers shown in the table in Figure 18. Select a binary answer (1 (used), 0 (not used)) to the usage question, which is a question asking whether to use each beauty product (hereinafter referred to as G1 to G12). The answers to a total of 34 selected items will be analyzed using factor analysis. Note that the 12 beauty products G1 to G12 are beauty products whose usage rate in the basic data is 0.07 or more and 0.70 or less, but the beauty products of the present invention also have a usage rate of less than 0.07 or greater than 0.70. The determination method works effectively. Hereinafter, questions Q1 to Q22 and beauty products G1 to G12 will be referred to as items. Further, a value (binary value, 5 value, or integer value) representing the test subject's or customer's answer to each question or a value (binary value) representing whether or not each beauty product is used is referred to as the value of each item.

(Dividing basic data) Since we want to verify the accuracy of beauty product recommendation discrimination, we randomly shuffled the basic data and divided it into training data (11,868 items), cross-validation data (3,800 items), and test data (3,800 items). ), and first, we analyzed the training data using the factor analysis method.

(Results of factor analysis of training data) FIG. 9 is a table showing the factor loadings of the eight extracted factors (referred to as factors [1] to [8]). The number of factors to be extracted was set to 8 because the number of eigenvalues of the covariance matrix of the above responses standardized to an average of 0 and a standard deviation of 1 that is 1 or more is 8. The maximum likelihood method was used to estimate the factor loadings. The factor loadings shown in FIG. 9 are the factor loadings after nine rotations were performed until convergence using the varimax method without Kaiser's normalization. Further, FIG. 10 is a table showing factor score coefficients obtained by analysis using the above factor analysis method. Regression method was used to calculate factor scores. Furthermore, the following Table 4 is a table showing correlation coefficients between factor scores in the above factor analysis model [21] and factor analysis model [34]. Further, Table 5 below shows the sum of squares of factor loadings for each of the eight factors extracted in the factor analysis model [34] and the proportion of variance in answers explained by each factor.

As can be seen from Table 4 and FIG. 10, factor [1] is a factor related to aesthetics and can be almost identical to the EST factor (∵correlation coefficient is 0.988). Factor [2] is a factor related to oiliness and can be almost identical to factor O1 (∵correlation coefficient is 0.978). Factor [3] is a factor related to base makeup, and can be almost identical to the MB factor (∵correlation coefficient is 0.979). Factor [4] is a factor related to dryness and can be almost identical to factor D (∵correlation coefficient is 0.949). Factor [5] is a new factor and is related to awareness of moisturizing. Factor [6] is also a new factor and is related to awareness of aging care. Factor [7] is a factor related to oily-related skin dissatisfaction, and can be almost identical to the O2 factor (∵correlation coefficient is 0.984). Factor [8] is a new factor and is related to whitening awareness.

(Calculation of factor scores) As an example, based on the answers obtained by interviewing a certain customer regarding the above 34 items, this embodiment of the present invention determines whether beauty product G6 should be recommended to the customer. We will explain how to make a determination based on the following. Of course, recommendation determination can be made in the same manner for beauty products other than G6. First, using the above answer and equation (1), calculate the score f _i of each factor i of factors [1] to [8] for the customer. Here, a _ij is the factor score coefficient estimated by the factor analysis model [34] based on the training data, x _j is the value of the j-th item of the customer (answer value), and m _j and σ _j are the training values. These are the mean value and standard deviation of the value of the j-th item of the subject in the data. Σ means the sum of j. Here, the sum for j in equation (1) is taken over all items (that is, 33 items) other than the beauty product (here, G6) targeted for recommendation determination.

(Customer's Score) Next, it is determined to which of a plurality of predetermined score intervals for each factor i the calculated score f _i of each factor i of the customer belongs. A factor interval score is assigned to each interval in advance, and the value obtained by adding up the factor interval scores for all factors i in the interval to which the score of each factor i of the customer belongs is the score of the customer. Based on the score, it is determined whether beauty product G6 should be recommended to the customer.

In the example shown in Table 6, a certain customer's score for factor [1] belongs to the second interval of the factor (factor interval score is -0.013), and the score for factor [2] belongs to the first interval of the factor. (factor interval score is 0.359), (omitted), and the score for factor [8] also belongs to the first interval of the factor (factor interval score is 0.144), so the customer's score is based on these eight factors. The sum of the interval scores can be calculated as 1.868.

(Factor interval score) Next, the predetermined factor interval score will be explained. First, scores for each factor i are calculated for each subject belonging to the cross-validation data and/or training data in the same manner as for the customer described above. With reference to the calculated distribution of scores for factor i, the real axis (set of real numbers) is divided into a plurality of intervals without overlap or omission. In Table 6, the mean value and standard deviation of the scores for factor i are calculated for the cross-validation data, and using the method described in the factor analysis model [21] with reference to the quantile positions in the case of a normal distribution. Although the number of sections is divided into five sections, the number of sections may be other than five, and the sections may be divided so that each section includes the scores of factor i of approximately the same number of subjects. Next, for each interval, calculate the proportion r of subjects who use beauty product G6 among the subjects whose scores for factor i belong to the cross-validation data and/or training data belonging to that interval. Then, the difference (r−r ₀ ) between the ratio r and the usage rate r ₀ of the beauty product G6 that does not depend on the score of the factor i is defined as the factor interval score of the factor i of the beauty product G6.

(Discrimination based on threshold) If the customer's score for beauty product G6 calculated in this way is equal to or higher than the predetermined threshold h, beauty product G6 should be recommended to the customer. It is determined that there is, and if it is not, it is determined that it should not be recommended. The threshold value h is selected such that the proportion of subjects whose score for the beauty product G6 is h or more in the cross-validation data and/or the training data is approximately equal to the usage rate of the beauty product G6. Specifically, for example, prepare a decreasing sequence h ₁ , h ₂ , h ₃ , etc. of threshold candidates, increase the integer k as 1, 2, 3, etc., and use the cross-validation data and/or Alternatively, find the first k at which the percentage of subjects whose score for beauty product G6 is h _k or higher in the training data is greater than the usage rate of beauty product G6, and set h _k as the threshold value. Good.
In the example shown in Table 6, the threshold h for beauty product G6 determined using cross-validation data was h=0.610, and the customer's score was 1.868. In this example, since the customer's score is greater than the threshold h, it is determined that beauty product G6 should be recommended to the customer. Recommendation determination can be made in the same manner for beauty products other than G6. The determination results of the recommendation determination for the customer are shown in the column labeled "Discrimination" in the table of FIG.

The table in FIG. 11 shows an example of a display of the results of recommendation determination of beauty products according to the embodiment of the present invention. The "Usage" column in FIG. 11 displays whether or not each beauty product is used by the customer based on the customer's interview data. Similarly, in the "Usage Rate" column, the usage rate of each beauty product is displayed for the subjects in the test data. Similarly, the "precision rate" and "recall rate" columns display the precision rate and recall rate of discrimination by the recommended discrimination method of this embodiment. Preferably, in addition to numerical values, the precision is visually displayed using a bar graph or the like. Furthermore, it is preferable that the beauty products are sorted and displayed in descending order of relevance. For this customer, the conformity rate for beauty product G10 (cream) is 0.58, and the conformity rate for G6 (serum (moisturizing)) is 0.51, both exceeding 0.50, and the customer does not use these beauty products. I haven't. Therefore, it is considered desirable to first recommend these two beauty products to the customer.

Generally, the "accuracy rate" of discrimination refers to the proportion of subjects who actually use the beauty product among the subjects who have been determined to recommend it. Furthermore, the "recall rate" of discrimination refers to the proportion of subjects who are judged to be recommended among the subjects who actually use the beauty product. In FIG. 11, the "recall rate" was calculated using the entire test data. Regarding "precision rate," first, the score of each subject in the test data obtained in the same manner as above is converted using a sigmoid function to obtain a bounded score that is a value belonging to the [0, 1] interval. On the other hand, the [0, 1] interval is divided into 10 subintervals, e.g., [0, 0,1], (0.1, 0.2], ..., (0.9, 1.0), and the corresponding The usage rate of the beauty product among the test data subjects whose bounded scores belong to the same partial interval as the customer's bounded score belongs to was taken as the required precision rate.

(f1 score) The numerical values in the column labeled "8 factors" in Table 7 are based on the factor analysis model [34
] is the f1 score of the determination result obtained by performing the recommendation determination as described above. Here, the "f1 score" is the harmonic mean of the precision and recall of discrimination, as shown by the following equation.
f1 score = 2/(precision rate ^-1 + recall rate ^-1 ).
Note that the "precision rate" of the discrimination used to calculate the f1 score was calculated using the entire test data, similar to the "recall rate."

The numerical values in the column labeled "5 limiting factors" in Table 7 are calculated using the factor analysis model [34] in the same way as above, except for factor [1], factor [2], factor [3], factor This is the f1 score of the discrimination result when beauty product recommendations were judged using only the scores of five factors: [4] and factor [7]. In the case of "8 factors", the f1 score of the discrimination result is improved by 0.15 on average compared to the case of "limited 5 factors". Although not described in this specification, the recommendation of beauty products may be determined in the same way as above by combining the factor analysis model [21], which is a five-factor model, and information on whether or not beauty products are used. , the f1 score of the discrimination result is almost the same as the numerical value in the column labeled "5 Restricting Factors" in Table 7. In this way, when determining beauty and recommending beauty products using a factor analysis model [34], in addition to the customer's answers to questions about beauty, information on whether or not the customer uses beauty products can be used. By determining and understanding customers' beauty needs and skin type in more detail and elucidating their purchasing psychology, it will be possible to more accurately predict the items customers will purchase.

In this embodiment, the customer's score is calculated as the sum of the factor interval scores for each factor, but the method of calculating the customer's score can be further generalized. For example, a customer's score may be calculated based on the proportion of subjects using beauty product G6 among subjects with the same beauty needs code and skin type code shown in Table 3. In a more general embodiment, where n is an integer greater than or equal to 1 and m is an integer greater than or equal to 2, the test subject's score on n factors T=(T ₁ , T ₂ , . . . , T _n ) is The n-dimensional number vector space R ⁿ is divided into m factor regions U ₁ , U ₂ , . . . , U _m that do not overlap with each other and have no omissions. ₁ , S ₂ , ..., S _n ) among the subjects related to the cross-validation data and/or training data having the factor score T included in the factor region U _k to which beauty products belong. Based on the proportion of subjects using G6, the factor domain score for the above division of the customer is calculated, and several divisions of the n-dimensional number vector space R ⁿ that do not overlap with each other and are not omitted are calculated. The customer's score is calculated by adding or averaging the factor domain scores of the customer related to the division, or by taking a weighted average, and the beauty product is selected based on the score and the threshold. G6 recommendation determination is performed. The above calculations and recommendation determination are performed by the usage rate standard recommendation determination means of the computer control program. From the viewpoint of increasing the accuracy of recommendation discrimination of beauty products, the number n of factors used for recommendation discrimination is desirably 2 or more, preferably 4 or more, and more preferably 8 or more. desirable. Of course, recommendation determination can be made in the same manner for beauty products other than beauty product G6.

<4-3. Recommendation of beauty products using machine learning (Part 1)>
Recommendation determination by the usage rate standard recommendation determination means described above is based on a certain rule based on the customer's n factor scores S = (S ₁ , S ₂ ,..., S _n ). This is a method in which a real value (predicted value) a called a customer's score is calculated, and based on whether or not the predicted value a is greater than or equal to a threshold value h, a beauty product g is recommended to the customer. I can do it. The above rules are based on the usage rate standard, but in the machine learning-based beauty product recommendation judgment of this embodiment, the assumption of the usage rate standard is removed, and the predicted value a is calculated from the customer's score S of n factors. The function used to calculate the is determined directly based on the data using machine learning. Now, a prediction function f whose input value is the customer's score S of n factors and whose output value is the predicted value a is expressed as a = f (S, Θ)
shall be. Here, Θ is a symbol collectively indicating several parameters that define the prediction function f. The scores of n factors for customer u are S(u), the output value of the prediction function for customer u is a(u) = f(S(u), Θ), and the presence or absence of use of beauty product g for customer u is Let the represented value (1 or 0) be y(u). For the subject u included in the training data and/or cross-validation data, the recommendation is made by optimizing the parameter Θ with the aim of minimizing the cost function J expressed by the following equation (4). A prediction function f useful for discrimination can be obtained.

In addition, when determining the recommendation of beauty product g, when calculating the score S of n factors of the customer using equation (1), a value (item item) indicating whether the customer uses beauty product g value) is not used. This point is similar to the case of recommendation determination by the usage rate-based recommendation determination means. The method of determining the threshold h is also the same as in the case of recommendation determination by the utilization rate standard recommendation determination means. Further, λ is a regularization parameter, and ||Θ|| ² means the sum of squares of several parameters that define the prediction function. Furthermore, in equation (4), the sum for customer u is taken for all subjects included in the training data of the basic data. m is the number of subjects such as training data.

The prediction function f can be constructed using a neural network. FIG. 12A shows an example of a forward propagation neural network that can be used to determine the recommendation of the beauty product G1. The input layer has n neurons corresponding to the number n of factors used for recommendation discrimination, and each neuron in the input layer is preferably standardized to the training data with an average of 0 and a standard deviation of 1. Output the score of each factor for customer u. The intermediate layer consists of one or more layers. Each neuron in each layer of the intermediate layer is composed of a biased sigmoid neuron, an unbiased sigmoid neuron, a hyperbolic tangent neuron, a normalized linear unit, and the like. In the embodiment shown in Figure (12A), the output layer is composed of one sigmoid neuron. Each neuron calculates the sum of the values obtained by multiplying the outputs of neurons in the previous layer by weights, adds a bias if necessary, and outputs a value obtained by applying an activation function to the sum. This output value becomes the input value of the neuron in the next layer. The above weights and biases constitute the parameter Θ of the prediction function f.

(Example Deep8) The first number in the column labeled "Deep8" in Table 7 standardizes the scores of eight factors, factor [1] to factor [8], calculated by the factor analysis model [34]. Then, it is used as the input layer in Figure (12A), and the middle layer is made up of three layers: a layer consisting of 24 unbiased sigmoid neurons, a layer consisting of 12 biased sigmoid neurons, and a layer consisting of 6 biased sigmoid neurons. It consists of a layer consisting of a certain sigmoid neuron, the output layer consists of one biased sigmoid neuron, the prediction function f is expressed by a fully connected neural network, and the regularization that appears in the cost function of equation (4) The parameters λ are λ=1×10 ^-2 , 3.2×10 ^-3 , 1×10 ^-3 , 3.2×10 ^-4 , 1×10 ^-4 , 3.2×10 ^-5 , 1×10 ^-5 , the parameter Θ is repeatedly updated 2000 times using the ADADELTA method in batch learning with the aim of minimizing the cost function using training data, and the optimum value of the parameter Θ is Θ = Θ (λ), and then calculate the f1 score of the discrimination result of beauty product g=G1 calculated using the cross-validation data when the parameter Θ is set as Θ=Θ(λ). The threshold value of is determined using training data, and among the values of λ above, λ that maximizes the f1 score is determined (in this example, λ _max = 1 × 10 ^-4 ), and the parameter The f1 score calculated for the test data with Θ set as Θ=Θ(λ _max ) is the numerical value shown for the beauty product G1. Recommendation was determined in the same manner for other beauty products, and f1 scores were calculated and shown in the corresponding column of Table 7.
The f1 scores in the column labeled "Deep8" in Table 7 are larger by 0.03 on average than the f1 scores in the column labeled "8 factors." In other words, recommendation discrimination based on machine learning using a neural network whose input layer outputs factor scores has higher discrimination accuracy than recommendation discrimination using the utilization rate-based recommendation discrimination means. In this embodiment, based on factor scores calculated from customer interview data, beauty product recommendations are determined using the above-mentioned optimized prediction function and threshold, and the determination result display means displays the determined results. The precision rate, recall rate, etc. of the discrimination can be displayed in a table similar to FIG. 11.

<4-4. Recommendation of beauty products using machine learning (Part 2)>
From the above-mentioned recommendation discrimination using machine learning using a neural network whose input layer outputs factor scores, we remove the assumption of factor scores and use customer item values (answers to questions and whether or not to use beauty products). In the embodiment described below, the predicted value a is directly calculated by machine learning. Now, the prediction function f whose input value is the value x = (x ₁ , x ₂ , ..., x _p ) of the customer's p items and whose output value is the predicted value a is expressed as a = f (x, Θ)
shall be. Here, Θ is a symbol collectively indicating several parameters that define the prediction function f. x(u) is the value of p items for customer u, a(u) = f(x(u), Θ) is the output value of the prediction function for customer, and represents whether or not beauty product g is used by customer u. Let the value (1 or 0) be y(u). For the test subject u included in the training data and/or the cross-validation data, the parameter Θ is optimized with the aim of minimizing the cost function J shown by equation (4). A useful prediction function f can be obtained. Note that when determining the recommendation of beauty product g, the value x(u) of p items of customer u does not include a value (item value) indicating whether the customer u uses beauty product g. . The method of determining the threshold h is the same as in the case of recommendation determination by the utilization rate standard recommendation determination means. Further, λ is a regularization parameter, and ||Θ|| ² means the sum of squares of several parameters that define the prediction function.

The prediction function f can be constructed using a neural network. FIG. 12B shows an example of a forward propagation neural network that can be used to determine the recommendation of the beauty product G1. The input layer has p neurons corresponding to the number p of items used for recommendation discrimination, and each neuron in the input layer is preferably standardized to the training data with an average of 0 and a standard deviation of 1. Output the value of each item for customer u. The intermediate layer consists of one or more layers. Each neuron in each layer of the intermediate layer is composed of a biased sigmoid neuron, an unbiased sigmoid neuron, a hyperbolic tangent neuron, a normalized linear unit, and the like. In the embodiment shown in Figure (12B), the output layer consists of one sigmoid neuron. Each neuron calculates the sum of the values obtained by multiplying the outputs of neurons in the previous layer by weights, adds a bias if necessary, and outputs a value obtained by applying an activation function to the sum. This output value becomes the input value of the neuron in the next layer. The above weights and biases constitute the parameter Θ of the prediction function f.

(Example DeepRaw) The first number in the column labeled "DeepRaw" in Table 7 is a number (1 or 0) representing the customer's answers to questions Q1 to Q22 and whether or not they use beauty products G2 to G12. It is standardized and used for the input layer in Figure (12B), with three intermediate layers: a layer consisting of 24 unbiased sigmoid neurons, a layer consisting of 12 biased sigmoid neurons, and a layer consisting of 6 biased sigmoid neurons. The output layer is composed of a sigmoid neuron with a certain bias, and the prediction function f is expressed by a fully connected neural network. λ = 1×10 ^-2 , 3.2×10 -3 , 1×10 ^-3 , 3.2×10 ^-4 , 1×10 ^-4 , 3.2× ^{10 -5 ,} 1× Using training data, the parameter Θ is set to either 10 ^-5 and set to 20 by the ADADELTA method in batch learning with the aim of minimizing the cost function.
00 times of repeated updates to find the preferred value Θ = Θ (λ) of the parameter Θ, and then the beauty product g = calculated using the cross-validation data when the parameter Θ is set as Θ = Θ (λ). The f1 score of the G1 discrimination result is determined. However, the threshold value at that time is determined using training data, and among the above λ values, the λ that maximizes the f1 score is determined (in this example, λ _max = 3.2 × 10 ^-4 ), parameter Θ is set as Θ = Θ (λ _max ), and the f1 score calculated for the test data is the numerical value shown for beauty product G1. . Recommendation determination can be performed in the same manner for other beauty products, and f1 scores can be calculated.
The f1 scores in the column labeled "DeepRaw" in Table 7 are on average 0.08 larger than the f1 scores in the column labeled "8 factors", and the f1 scores in the column labeled "Deep8" It is 0.05 larger on average. In other words, recommendation discrimination using machine learning generally has higher discrimination accuracy than recommendation discrimination using utilization rate-based recommendation discrimination means that uses factor scores. Recommendation discrimination using a neural network whose input layer outputs item values has higher discrimination accuracy than an output neural network.
In this embodiment, based on the customer's interview data that is the answer to the beauty-related question including the usage question, the recommended beauty product is determined using the above-mentioned optimized prediction function and threshold, and the determination result is displayed. The means can display the discrimination results and the precision rate, recall rate, etc. of the discrimination in a table similar to FIG. 11.

<5. Flowchart of beauty judgment method＞
FIG. 13 is a flow diagram of a beauty determination method according to an embodiment of the present invention. In step S1, implementation of the beauty judgment method is started. Next, in step S2, it is determined whether an analyzed parameter is available and whether the analyzed parameter is to be used for beauty determination. If No, the process proceeds to step S3, where the basic data is analyzed by the factor analysis method to extract factors, the analyzed parameters are obtained, and the process proceeds to step S4. If Yes, the process advances to step S4. In step S4, it is determined whether the customer's interview data is available. If No, the process proceeds to step S5, where the customer is interviewed to obtain interview data, and the process proceeds to step S6. If Yes, the process advances to step S6. In step S6, the customer's factor scores are calculated based on the analyzed parameters and the interview data. Next, in step S7, the EST degree, the MB degree, preferably the oiliness degree and the dryness degree are calculated based on the analyzed parameters and the customer's factor scores, and the customer's beauty is judged. Next, in step S8, the judgment result of the customer's beauty judgment is displayed. Next, in step S9, the implementation of the beauty judgment method ends.

FIG. 14A is a detailed flowchart of step S7 of beauty determination. Beauty determination is started in step S70. Next, in step S71, the customer's EST degree and MB degree are calculated based on the analyzed parameters and the customer's factor scores. Next, in step S72, the customer's beauty needs code is determined based on the customer's EST degree and MB degree. Preferably, in a next step S73, the customer's oiliness and dryness are calculated based on the analyzed parameters and the customer's factor scores, and further in a next step S74, the customer's oiliness and dryness are calculated. , the customer's skin type code is determined. Finally, the beauty judgment ends in step S75. Note that step S7 of beauty determination preferably includes a step (not shown) of beauty product recommendation determination.

FIG. 14B is a detailed flowchart of step S8 of displaying the results of beauty evaluation. Display of the determination result is started in step S80. Next, in step S81, the determination result output means 53a plots the customer's EST degree and MB degree on the beauty needs identification screen of the output device 40, and preferably displays the customer's beauty needs code. Preferably, in the next step S82, the determination result output means 53a plots the customer's oiliness and dryness on the skin type identification surface of the output device 40, and preferably displays the customer's skin type code. Next, in step S83, the factor score answer contribution degree display means 51d displays the factor score answer contribution degree of the customer on the output device 40. Preferably, in the next step S84, the determination result output means 53b displays the determination result of the beauty product recommendation determination for the customer. Finally, in step S85, the step of displaying the determination result ends.

FIG. 14C is a detailed flowchart of step S84 for displaying the recommended beauty product determination results. In step S84a, display of the recommendation determination results is started. Next, in step S84b, the expert determines whether or not to make a recommendation determination. If Yes, the process proceeds to step S84c, where the determination result output means 53b displays the customer's beauty needs code and preferably the skin type code on the output device 40, preferably the beauty needs code and/or the skin type. A beauty product list indicating recommended beauty products linked in advance to the code is displayed, and the process proceeds to step S84e. In the case of No, the process advances to step S84d, where the determination result output means 53b displays the determination result of the recommendation determination for each beauty product and the degree of suitability of the recommendation classification on the output device 40, and the process advances to step S84e. Finally, in step S84e, the step of displaying the recommendation determination result ends.

<6. Configuration of beauty evaluation system and e-learning system>
<6-1. Independent system＞
FIG. 15 is a configuration diagram showing a system configuration in an embodiment of the present invention, which is composed of independent computers. The beauty evaluation system of this embodiment includes an independent computer 10 such as a personal computer, an input device 20 consisting of a keyboard, a mouse, a touch panel display, etc., an output device 40 consisting of a display, a printer, etc., and a register of a CPU. , internal storage means such as memory such as RAM and ROM, or removable storage means such as hard disk drive (HDD) or solid state drive (SSD), USB memory or SD card, or external storage means such as CD-R or DVD. and a computer-readable storage means 30 configured. This beauty evaluation system may be operated by a customer, or may be operated by a beauty advisor at a beauty supply store or the like while interviewing the customer. The input device 20 is used to input interview data of each customer. Here, the interview data for each customer includes answers to beauty-related questions, whether or not the customer uses beauty products, and other information. The computer 10 executes the computer control program 50, performs a beauty judgment based on the basic data stored in the storage means 30 and the interview data, and outputs the judgment result to the output device 40. Further, the computer 10 preferably executes the computer control program 50 to determine whether beauty products are recommended based on the basic data and the interview data, and outputs the determination results to the output device 40 . Preferably, the output device 40 outputs the discrimination accuracy such as precision and recall of the discrimination together with the discrimination results. In that case, the basic data includes training data, cross-validation data, and test data. It is desirable to divide it into data etc.

Computer 10 has a basic configuration 11. The basic configuration 11 includes an interview data input means 12 for inputting customer interview data, a computer-readable storage means 30 for storing basic data and temporarily storing the interview data, and a computer-readable storage means 30 for storing basic data and temporarily storing the interview data. A computer control program 50 for learning, outputting judgment results, etc., and a hard disk drive (HDD), solid state drive (SSD), removable storage device such as a USB memory or SD card, or external storage such as a CD-R or DVD. and a computer-readable storage medium 14 for storing analyzed parameters 15 and the like. When this beauty evaluation system constitutes a part of the e-learning system, the basic configuration 11 includes a customer evaluation result input means 16 for inputting the results of the beauty evaluation by the students themselves. The analyzed parameters 15 include standardized parameters such as the average and standard deviation of item values, factor loadings and factor score coefficients found in analysis using the factor analysis method, factor scores for each quantile, and recommended discrimination. , threshold values, neural network weights optimized by machine learning, learned parameters such as regularization parameters, etc.

FIG. 16 is a configuration diagram of the computer control program 50. The computer control program 50 outputs a beauty determination program 51 for performing beauty determination, a recommendation determination program 52 for recommending beauty products, and a determination result of beauty determination and a determination result of recommendation determination of beauty products. It includes a result output program 53 for.

The beauty evaluation program 51 includes a factor extraction means 51a for extracting factors by analyzing basic data using a factor analysis method, and a score of the customer's EST factor based on the extracted factors and the customer's interview data. A beauty needs determining means 51b for obtaining the scores of the MB factors and determining the beauty needs of the customer based on the scores of these factors, and preferably a score of the customer's O1 factor based on the extracted factors and the customer's interview data. and a skin type determining means 51c for determining the customer's skin type based on the scores of these factors, and preferably determining the contribution of each answer of the interview data to the customer's score of each factor. factor score answer contribution display means 51d for displaying on the output device 40 a factor score answer contribution degree which is a quantity representing the factor score answer contribution degree.

A recommendation discrimination program 52 for recommending and discriminating beauty products recommends the beauty product to the customer based on the usage rate of the beauty product by the basic data test subject who has a factor score close to the customer's factor score. The output value of the usage rate standard recommendation determination means 52a for performing recommendation determination, or the prediction function for performing recommendation determination of the beauty product by machine learning based on customer factor scores or customer answers in interview data. It has prediction function output value calculation means 52b for calculating. Here, the prediction function is defined by parameters, the parameters are optimized by machine learning using basic data, the output value of the prediction function preferably takes a real value between 0 and 1, and the output value is If the output value is above the threshold, it is determined that the beauty product should be recommended to the customer, and if the output value is less than the threshold, it is determined that the beauty product should not be recommended to the customer. Discern.

The result output program 53 includes a determination result output means 53a for outputting the determination result of the beauty determination, and a determination result output means 53b for outputting the determination result of the beauty product recommendation determination. The determination result output means 53a plots and displays the customer's EST degree and MB degree on the beauty needs identification surface, preferably plots and displays the customer's oiliness degree and dryness degree on the skin type identification surface, and preferably displays the factor score response contribution in a bar graph, etc. The determination result output means 53b displays the recommendation determination result indicating whether or not the beauty product should be recommended to the customer, preferably displays the degree of suitability of the determination as a numerical value, a bar graph, etc., and preferably also displays the recall rate of the determination. When this beauty evaluation system constitutes a part of the e-learning system, the result output program 53 performs a judgment for comparing the judgment results of the beauty judgment by the student himself and the judgment results of the beauty judgment by the present beauty judgment system. It has a result comparing means 53c, and preferably a determination result comparing means 53d for comparing the determination result of the beauty product recommendation determination by the student himself and the determination result of the beauty product recommendation determination by the present beauty evaluation system.

<6-2. A system that performs beauty judgment on the server＞
Figure (17A) is a configuration diagram showing a system configuration in another embodiment of the present invention. In this embodiment, in principle, the server 10a performs beauty judgment and beauty product recommendation determination, and input and output is performed by a terminal 10b close to the customer. The beauty evaluation system of this embodiment includes a server 10a and a terminal 10b that can communicate bidirectionally via a network 10c such as the Internet. The terminal 10b includes a computer such as a personal computer, a smartphone, or a tablet, an input device 20 including a keyboard, a mouse, a touch panel display, and the like, and an output device 40 including a display, a printer, and the like. The server 10a is a computer such as a personal computer or a server machine, and includes the basic configuration 11 described above, and optionally includes an input device including a keyboard, a mouse, a touch panel display, etc., and an output device including a display, a printer, etc. It may have. The terminal 10b in this embodiment may be operated by a customer, or may be operated by a beauty advisor at a beauty supply store or the like while interviewing the customer. The input device 20 is used to input customer interview data. Terminal 10b
has an interview data transmitting means 54a for transmitting interview data input using the input device 20 to the server 10a. The server 10a receives the inquiry data transmitted by the inquiry data transmitting means 54a of the terminal 10b, using the inquiry data receiving means 54b. The server 10a executes the computer control program 50 to make a beauty judgment of the customer based on the interview data, preferably also makes a recommendation judgment of beauty products, and sends the judgment result, preferably the judgment result, to the terminal 10b. It has a determination result transmitting means 55a for transmitting to. The terminal 10b has a determination result receiving means 55b for receiving the determination result transmitted by the determination result transmitting means 55a of the server 10a, and preferably the determination result. The terminal 10b includes a determination result output means 53a for outputting the determination result received by the determination result receiving means 55b, and preferably the determination result, to the output device 40.

In this embodiment, the server 10a stores the customer interview data received by the interview data receiving means 54b in the storage means 30, and when the number of stored interview data satisfies a certain standard, All or part of the interview data is added to basic data or training data, the updated basic data or training data is stored in the storage means 30, and factor analysis or analysis is performed based on the basic data or training data. The analyzed parameters 15 may be updated by performing machine learning, and the updated analyzed parameters 15 may be stored in the storage medium 14 or the storage means 30.

In addition, when the beauty evaluation system of this embodiment constitutes a part of the e-learning system, the terminal 10b further includes a customer evaluation result input means 16 for inputting the results of the beauty evaluation by the student himself/herself. , has a customer determination result transmitting means 56b for transmitting the student's determination result input by the customer determination result input means 16 to the server 10a. Further, the server 10a further includes a customer determination result receiving means 56c for receiving the student determination result transmitted by the customer determination result transmitting means 56b of the terminal 10b, and a student determination result receiving means 56c included in the basic configuration 11. It has a determination result comparing means 53c for comparing the determination result of the beauty determination and the determination result of the beauty determination by the beauty determination system.
Preferably, an e-learning system for determining recommendation of beauty products is similarly configured. That is, when the beauty judgment system of this embodiment constitutes a part of the e-learning system, the terminal 10b is preferably used to input the customer judgment results for inputting the judgment results of the beauty product recommendation judgment by the student himself/herself. It has an input means and a customer determination result transmitting means for transmitting the student determination results input by the customer determination result input means to the server 10a. Further, the server 10a preferably includes a customer determination result receiving means for receiving the student's determination result transmitted by the customer determination result transmitting means of the terminal 10b, and a student's hairdresser included in the basic configuration 11. It has a determination result comparing means 53d for comparing the determination result of the determination and the determination result of the beauty determination by the beauty determination system.

<6-3. A system that performs beauty judgment on a terminal>
Figure (17B) is a configuration diagram showing the configuration of a beauty evaluation system in yet another embodiment of the present invention. In general, factor analysis and machine learning require high-load calculations, but if the analyzed parameters 15 of the model are known, the calculations of factor scores, prediction function output values, and recommendation discrimination operations themselves are generally light. It is a load, and can be fully performed on a terminal such as a mobile terminal. Therefore, in this embodiment, the terminal 10b performs the calculation of factor scores, the calculation of the output value of the prediction function, and the calculation of recommendation determination.

In the beauty evaluation system of this embodiment, in addition to inputting medical interview data and outputting the results of beauty evaluation, the terminal 10b performs calculations necessary for beauty evaluation and preferably for recommending beauty products. The terminal 10b in this embodiment includes the basic configuration 11 described above, and includes an input device 20, an output device 40, and an interview data transmission means for transmitting interview data input using the input device 20 to the server 10a. 54a. The server 10a includes the basic configuration 11, includes an interview data receiving means 54b, and may optionally include an input device and an output device. The server 10a has an analyzed parameter transmitting means 57a for transmitting the analyzed parameter 15 to the terminal 10b. The terminal 10b has an analyzed parameter receiving means 57b for receiving the analyzed parameter 15 transmitted by the analyzed parameter transmitting means 57a of the server 10a. The terminal 10b executes the computer control program 50 based on the analyzed parameters 15 received by the analyzed parameter receiving means 57b and the interview data inputted using the input device 20 to judge the beauty of the customer. also performs recommendation discrimination of beauty products, and includes, as part of the basic configuration 11, a judgment result output means 53a for outputting the judgment result of the beauty judgment and preferably the judgment result of the recommendation judgment to the output device 40. have

Also in this embodiment, the server 10a adds interview data to basic data or training data based on certain criteria, and performs factor analysis or machine learning based on the additionally updated basic data or training data to perform analysis. The analyzed parameters 15 may be updated and the updated analyzed parameters 15 may be stored in the storage medium 14 or the storage means 30.

In addition, when the beauty judgment system of this embodiment constitutes a part of the e-learning system, the terminal 10b further includes the judgment result of the beauty judgment by the student himself, which is included as a part of the basic configuration 11. It has customer judgment result input means 16 for inputting customer judgment result input means 16, and customer judgment result transmission means 56b for transmitting judgment results by students input by customer judgment result input means 16 to server 10a. Further, the server 10a further includes a customer determination result receiving means 56c for receiving the student determination result transmitted by the customer determination result transmitting means 56b of the terminal 10b, and a student determination result receiving means 56c included in the basic configuration 11. It has a determination result comparing means 53c for comparing the determination result of the beauty determination and the determination result of the beauty determination by the beauty determination system.
Preferably, an e-learning system for determining recommendation of beauty products is similarly configured. That is, when the beauty evaluation system of this embodiment constitutes a part of the e-learning system, the terminal 10b is preferably used to recommend beauty products by the students themselves, which is included as part of the basic configuration 11. It has customer discrimination result input means for inputting the discrimination result of discrimination, and customer discrimination result transmission means for transmitting the discrimination result by the student inputted by the customer discrimination result input means to the server 10a. Further, the server 10a preferably includes a customer determination result receiving means for receiving the student's determination result transmitted by the customer determination result transmitting means of the terminal 10b, and a student's recommendation included in the basic configuration 11. It has a determination result comparing means 53d for comparing the determination result of the determination and the determination result of the recommendation determination by the beauty determination system.

It goes without saying that the present invention is not limited to the above-described embodiments, but includes various modifications and design changes within the scope of the present invention without departing from the technical idea of the present invention. .

According to the present invention, a beauty evaluation method, a sensory evaluation method, which can objectively and accurately determine a customer's skin type as well as beauty needs based on interview data using a factor analysis method using big data; and a beauty evaluation system. According to the present invention, recommendations for cosmetics and beauty services, beauty education, training of beauty advisors, sensory evaluation of beauty products, etc. in online stores or physical stores such as beauty salons, cosmetic stores, clinics, drug stores, and aesthetic salons, etc. can be performed more accurately than before. The present invention has wide industrial applicability.

10 Computer (PC) 10a Server 10b Terminal 10c Network 11 Basic configuration 12 Interview data input means 14 Storage medium 15 Analyzed parameters 16 Customer judgment result input means 20 Input device 30 Storage means 40 Output device 50 Computer control program 51 Beauty judgment program 51a Factor extraction means 51b Beauty needs determination means 51c Skin type determination means 51d Factor score answer contribution display means 52 Recommendation discrimination program 52a Usage rate standard recommendation determination means 52b Prediction function output value calculation means 53 Result output program 53a Judgment result output means 53b Discrimination Result output means 53c Judgment result comparison means 53d Judgment result comparison means 54a Interview data transmission means 54b Interview data reception means 55a Judgment result transmission means 55b Judgment result reception means 56b Customer judgment result transmission means 56c Customer judgment result reception means 57a Analyzed parameter transmission Means 57b Analyzed parameter receiving means G1 to G12 Beauty products Q1 to Q22 Questions about beauty

Claims (11)

Basic data that is the answers of many test subjects to three or more skin-related questions,
Questionnaire data that is the answer to the customer's question,
A beauty determination method for determining the beauty of the customer by processing by executing a computer-controlled program,
The computer control program includes:
a factor extraction procedure for extracting at least an EST factor related to aesthetics and an MB factor related to base makeup that is orthogonal to the EST factor from the basic data by a factor analysis method ;
Based on the interview data, the EST factor score S _EST and the MB factor score S _MB of the customer are calculated, and a point (S _EST , S _MB ) on the beauty needs identification plane with XY coordinate axes is calculated. a beauty needs determination procedure for expressing and determining the beauty needs of the customer based on the position of the customer;
A beauty evaluation method characterized by having the following. The above question includes at least one of the items related to firmness, elasticity, brightness and transparency, skin texture, and youthfulness, which are closely related to esthetics, and is also closely related to base makeup, such as makeup smearing and makeup. 2. The beauty evaluation method according to claim 1, including at least one of the items related to poor texture and rough texture. In addition to the EST factor and the MB factor, the factor extraction procedure extracts at least an O factor related to excessive sebum and a D factor related to skin dryness,
The computer control program calculates the customer's O factor score S _O and D factor score S _D based on the interview data, and calculates a point ( _3. The beauty judgment method according to claim 1, further comprising a skin type judgment step of _expressing and judging the skin type of the customer based on the position of the skin. The computer control program has a factor score answer contribution degree display procedure for displaying the degree of contribution of the answer to each question of the customer to the EST factor score S _EST and/or the MB factor score S _MB of the customer. The beauty evaluation method according to any one of claims 1 to 3. In addition to determining the customer's beauty needs and/or skin type, the method also determines whether at least one beauty product g should be recommended to the customer,
The questions include a question asking whether the subject uses beauty product g,
The computer control program is based on at least one utilization rate calculated from the customer's n factor scores S=(S ₁ , S ₂ , . . . , S _n ), where n is an integer of 1 or more. , a usage rate standard recommendation determination procedure for determining whether beauty product g should be recommended to the customer;
The utilization rate is calculated based on the number of subjects whose factor scores are included in a certain region of an n-dimensional vector space R ⁿ that includes the customer's n factor scores S. The beauty evaluation method according to any one of claims 1 to 4, wherein the calculation is based on the proportion of subjects using the beauty product g. In addition to determining the customer's beauty needs and/or skin type, the method also uses machine learning to determine whether at least one beauty product g should be recommended to the customer.
The questions include a usage question that asks whether the subject uses beauty product g;
The computer control program takes as an input value the test subject's or the customer's factor score or a numerical value representing the answer to the question, and uses as an output value a predicted value of the numerical value representing the answer to the use/absence question. , has a prediction function output value calculation procedure for calculating an output value of the prediction function,
5. The beauty evaluation method according to claim 1, wherein the prediction function is optimized by machine learning using basic data. The computer control program according to any one of claims 1 to 6. A computer-readable storage medium recording the computer control program according to any one of claims 1 to 6. 4. A sensory evaluation method for a beauty product using the beauty evaluation method according to claim 3 , wherein the cooperator who cooperates with the evaluation uses the beauty product for a certain period of time, and the cooperator uses the beauty product for a certain period of time. By executing the computer control program, one or more of the EST factor, MB factor, O factor , and D factor is determined based on the interview data that is the answer of the cooperator to the question. A sensory evaluation method, characterized in that the scores of the factors are calculated both before and after the period, and the improvement effect is measured from the amount of change in the scores during the period. Consists of independent computers or servers and terminals that can communicate bidirectionally with each other,
Basic data that is the answers of many test subjects to three or more skin-related questions,
Questionnaire data that is the answer to the customer's question,
A beauty judgment system that processes the above by executing a computer control program to judge the beauty of the customer,
The computer control program includes:
a factor extraction procedure for extracting at least an EST factor related to aesthetics and an MB factor related to base makeup that is orthogonal to the EST factor from the basic data by a factor analysis method ;
Based on the interview data, the EST factor score S _EST and the MB factor score S _MB of the customer are calculated, and a point (S _EST , S _MB ) on the beauty needs identification plane with XY coordinate axes is calculated. a beauty needs determination procedure for expressing and determining the beauty needs of the customer based on the position of the customer;
A beauty evaluation system characterized by having the following. It is an e-learning system where students can learn about determining customers' beauty needs.
It consists of an independent computer or a server and a terminal that can communicate with each other bidirectionally, and is configured to be able to execute the computer control program according to claim 10,
Judgment result display means for displaying the judgment result of the beauty judgment by the beauty judgment system according to claim 10 regarding the interview data of the real or virtual customer, or the judgment result and the judgment result by the student himself/herself. An e-learning system characterized by having a determination result comparison means for comparing.

Images