JP6875305B2 - Relationship analysis device between olfactory preference and personality - Google Patents

Description

The present invention relates to a device for analyzing the relationship between a sense of smell and personality, and is particularly suitable for use in a device for analyzing the relationship between a person's taste for scent and personality.

Conventionally, a sensitivity estimation device for estimating the sensitivity during and after use of an object for sensitivity evaluation used by a user is known (see, for example, Patent Document 1). In the Kansei estimation device described in Patent Document 1, a preset hierarchical neural network skeleton learning method is used to classify each Kansei characteristic pattern for obtaining a predetermined comprehensive evaluation for a plurality of Kansei items. It makes it possible to estimate the user's sensibility with high accuracy and efficiency. For example, when the object to be evaluated is cosmetics, the sensitivity items at the time of application are "I like the color of the toner: visual", "I like the scent of the toner: odor", and "the scent of the toner is strong: the sense of smell" The user's sensibilities are estimated by using 18 items related to the five senses of human beings such as ... As input nodes of the neural network. In Patent Document 1, in addition to cosmetics, pharmaceuticals, foods and drinks, daily necessities, clothing, and the like are also described as sensitivity evaluation objects whose sensibilities can be estimated.

Further, Patent Document 2 clarifies the relationship between the material attributes possessed by the taste and smell material and the psychological factors of people regarding food, and is suitable for quantifying the sensitivity to food (food preference). The sensory sensitivity evaluation system is disclosed. In the sensory sensitivity evaluation system described in Patent Document 2, the material attribute data of the taste and odor material (component amount, physicochemical constant and physical amount obtained by instrumental analysis of the taste and odor material or detected through a sensor). (At least one) and biological information data (brain wave signal, brain wave power spectrum, brain wave intercorrelation coefficient between brain wave detection electrodes, self-phase relationship of brain wave at a specific electrode position) obtained from a subject who tastes the taste and smell material. Input numbers, estimated values of dipole position and moment, and at least one of the characteristic values of chaos analysis) in parallel, and output the output as food preference data (favorability for taste materials, arousal / sedation, desire, mood). , Relaxation, anger, and sadness) It has a hierarchical neural network, and the hierarchical neural network is trained by the back propagation method using food preference data as teacher data, and the trained hierarchy. Using a type neural network, the main factors of material attribute data and biometric data related to food preference data are obtained.

In addition, research has been conducted on the relationship between human taste preference and personality, and it has been reported that people who like bitterness have an antisocial personality, and people who like sweetness are prosocial. It has been reported that. However, the relationship between personality and olfactory preference, which seems to be more diverse than taste preference, has not yet been clarified. Even in the systems described in Patent Documents 1 and 2, the analysis of the relationship between the sense of smell preference and the personality is not disclosed.

Japanese Unexamined Patent Publication No. 2012-123794 Japanese Unexamined Patent Publication No. 2003-290179

The present invention has been made in view of such circumstances, and an object of the present invention is to provide an apparatus capable of analyzing the relationship between a person's sense of smell and personality.

In order to solve the above-mentioned problems, in the present invention, for each of a plurality of learning subjects, fragrance preference score data indicating the degree of personal preference defined in a plurality of stages for each of the plurality of scents, and predetermined psychology. Personality data indicating a plurality of individual personalities detected as a result of the test is input as learning data, and a plurality of learning explanatory variables are generated from the input scent preference score data for learning. Build a learning model to derive personality data from fragrance preference score data. Then, the personality of the prediction target person is predicted by applying the prediction scent preference score data input for the prediction target person to the learning model.

Further, in another aspect of the present invention, fragrance preference score data indicating the degree of personal preference defined in a plurality of stages for each of the plurality of scents for each of the plurality of learning subjects, and the result of a predetermined psychological test. Personality data indicating multiple personalities of the individual detected as is input as learning data, and after generating a plurality of learning explanatory variables from the input personality data for learning, fragrance preference is obtained from the personality data. Build a learning model to derive score data. Then, by applying the prediction personality data input for the prediction target person to the learning model, the scent preference of the prediction target person is predicted.

According to the present invention configured as described above, using a learning model constructed based on fragrance preference score data and personality data relating to a plurality of learning subjects, based on the fragrance preference score data relating to the prediction target subject. , The personality of the person to be predicted can be predicted. Further, according to the present invention, using a learning model constructed based on fragrance preference score data and personality data related to a plurality of learning target persons, the prediction target person is based on the personality data related to the prediction target person. Aroma preference can be predicted. As described above, according to the present invention, it is possible to analyze the relationship between a person's sense of smell and personality.

It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 1st Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 2nd Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 3rd Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 4th Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 5th Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 6th Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 7th Embodiment. It is a block diagram which shows the functional structure example of the relationship analysis apparatus by 8th Embodiment.

The device for analyzing the relationship between the olfactory preference and the personality according to the present embodiment analyzes the relationship between the human scent preference and the personality. As mentioned above, the relationship between taste preference and personality has already been studied and there are several reports. However, there are no research reports on the relationship between scent preference and personality.

Therefore, we conducted a questionnaire to multiple subjects, collected response information on 8 personalities and 41 food scent preferences, and based on the response information, factor analysis was performed on 6 factor food scent preferences (smelly, fragrant,). Sweet, pungent odor, liquor, fruit) was extracted. Next, a non-hierarchical cluster analysis using these factors was performed to divide the subject group into three groups and compared with an index (OAS: Odor Awareness Scale) indicating personality and fragrance recognizability. As a result, there were differences in some personalities and scent recognizability among groups with different scent preferences. From this, it was considered that there is a certain relationship between the scent preference and the personality, and further research was carried out, and the invention of the relationship analysis device between the olfactory preference and the personality described below was made.

The relationship analysis device between the olfactory preference and the personality according to the present embodiment (hereinafter, simply referred to as the relationship analysis device) predicts the personality from the preference for the human scent, or predicts the preference for the scent from the human personality. It made it possible. The first to fourth embodiments are devices for predicting personality from human scent preference, and the fifth to eighth embodiments are devices for predicting scent preference from human personality. Hereinafter, these embodiments will be described in order.

(First Embodiment)
Hereinafter, the first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a functional configuration example of the relationship analysis device 1A according to the first embodiment. As shown in FIG. 1, the relationship analysis device 1A according to the first embodiment includes a learning device 10A, a predictor 20A, and a learning model storage unit 30A. The learning device 10A includes a learning data input unit 11A, a learning variable generation unit 12A, and a model construction unit 13A as its functional configuration. The predictor 20A includes a prediction data input unit 21A, a prediction variable generation unit 22A, and a personality prediction unit 23A as its functional configuration.

Each of the above functional blocks can be configured by any of hardware, DSP (Digital Signal Processor), and software. For example, when configured by software, each of the above functional blocks is actually configured to include a computer CPU, RAM, ROM, etc., and a program stored in a recording medium such as RAM, ROM, hard disk, or semiconductor memory operates. It is realized by. This also applies to each functional block included in the relationship analysis apparatus according to the second to eighth embodiments described later.

The learning data input unit 11A is detected as a result of a predetermined psychological test and scent preference score data indicating the degree of personal preference defined in a plurality of stages for each of a plurality of learning subjects for each of the plurality of learning subjects. Personality data indicating a plurality of personalities of the individual is input as learning data. These data are collected as a result of questionnaires and psychological tests through, for example, paper media, websites, interviews, telephones, predetermined applications, etc., and are input by the learning data input unit 11A.

Here, as an example, the scent preference score data assumes that the degree of likes and dislikes for each of the scents of 41 types of foods is evaluated on a 9-point scale. In addition, the personality data, as an example, evaluates the degree of possibility of being applicable to each of a plurality of known items such as diplomacy, harmony, and integrity on a seven-point scale. The scent preference score data and personality data are input as a set for one learning target person. It should be noted that both the scent preference score data and the personality data shown here are examples, and the present invention is not limited thereto.

The learning variable generation unit 12A generates a plurality of learning explanatory variables from the learning scent preference score data input by the learning data input unit 11A. That is, the learning variable generation unit 12A synthesizes or converts the fragrance preference score data input by the learning data input unit 11A in order to increase the amount of information used for the analysis, and generates a new explanatory variable. Methods for creating new explanatory variables are, for example, standardization, factor analysis, simple regression, residual calculation by multiple regression, principal component analysis, four-rule calculation, mean, variance, standard deviation, number of missing values (questionnaire for food aroma preference, etc.) , The number of foods that answered that they do not know the scent). The newly generated explanatory variables belong to each learning target person.

In order to increase the amount of information used to generate new explanatory variables, the learning data input unit 11A uses at least one of the following correction data and preliminary data in addition to the fragrance preference score data and personality data. You may also enter more. The correction data is preferably acquired as much as possible, and the preliminary data is data to be used when it can be acquired. When these correction data and preliminary data are input by the learning data input unit 11A, the learning variable generation unit 12A also generates the learning explanatory variable using these data. For example, it is conceivable to predict the sleep time of the preliminary data from the age and gender of the correction data, and use the residual subtracted from the original sleep time as a variable.

<Correction data>
Age, gender, occupation, drinking, smoking <preliminary data>
Waking up, going to bed, other lifestyles, hobbies, subscription magazines, travel destinations and frequency, olfactory test, interest in and awareness of odors, amount of exercise, heart rate, respiratory rate, brain waves, cerebral blood flow, fMRI, brain shape, History of illness, height, weight, chest circumference, waist circumference, hip circumference, body fat ratio, muscle strength, sweat production and composition, bioimpedance, saliva metabolome, urinary exosome microRNA, fecal and skin surface metabolome and bacteria Flora analysis, genome of any cell in the body, methylome, proteome, metabolome and transcriptome, type and amount of neurotransmitters and cytokines in the body, wearable sensor, weight scale, height scale, body fat scale, scale, muscle strength Measuring instruments, sweat analyzers, bioimpedance analyzers, mass analyzers, various chromatographys, electrophoresis detection after PCR, real-time PCR equipments, next-generation DNA sequencers, microarrays, ELISA, and various chemical sensors, foods and luxury items Number of consumption, frequency and type of intake, impression of five senses on food and luxury items, preference for food and luxury items, degree of interest in nutrition and health, frequency of participation in volunteer activities, game process and results, record of interpersonal dialogue, Sensitivity information secondarily estimated from AI robot or chatbot exchanges, SNS, posted works (images and sentences), blogs and emails, purchase history of various products, application history to travel companies, bioelectric information

The model construction unit 13A uses the learning scent preference score data and personality data input by the learning data input unit 11A, and a plurality of learning explanatory variables generated by the learning variable generation unit 12A (hereinafter). , It may be called learning data including explanatory variables for learning), a learning model for deriving personality data from explanatory variables including fragrance preference score data is constructed, and the constructed learning model is stored in the learning model storage unit 30A. Let me. When one or both of the correction data and the preliminary data are input by the learning data input unit 11A, these may be added to the explanatory variables including the scent preference score data.

For example, the model building unit 13A builds a learning model by applying a known machine learning algorithm (for example, Random Forest) using the above-mentioned learning data. The learning model generated by this is composed of a large number of decision trees generated by the training data, and the explanatory variables including the fragrance preference score data are input, and the personality data is obtained by tracing a large number of decision trees. It is a learning model designed to output. The machine learning algorithm used in the model building unit 13A is not limited to Random Forest, and other algorithms may be applied to build the learning model.

The prediction data input unit 21A inputs scent preference score data as prediction data for the prediction target person. The scent preference score data input here is the same in content as that input by the learning data input unit 11A. The person to be predicted is a user whose personality is unknown because the scent preference score data has been obtained by a questionnaire or the like, but the personality data has not been obtained. The prediction data input unit 21A inputs the scent preference score data of the user (prediction target person) who wants to predict the personality. It is preferable to input the above-mentioned correction data and preliminary data as much as possible.

The prediction variable generation unit 22A generates a plurality of prediction explanatory variables from the prediction fragrance preference score data input by the prediction data input unit 21A. The method of generating the explanatory variables for prediction is the same as the method of generating the explanatory variables for learning by the learning variable generation unit 12A.

The personality prediction unit 23A stores the prediction fragrance preference score data input by the prediction data input unit 21A and a plurality of prediction explanatory variables generated by the prediction variable generation unit 22A in the learning model storage unit 30A. By applying it to the learning model, the personality of the person to be predicted is predicted.

As described above, in the first embodiment, the scent preference score data indicating the degree of personal preference defined in a plurality of stages for each of the plurality of scents for each of the plurality of learning subjects, and a predetermined psychological test. Personality data indicating a plurality of individual personalities detected as a result of the above are input as learning data, and a plurality of learning explanatory variables are generated from the input learning scent preference score data, and then the scent. Build a learning model to derive personality data from explanatory variables including preference score data. Then, the personality of the prediction target person is predicted by applying the prediction scent preference score data input for the prediction target person and the explanatory variables generated from the data to the learning model. When one or both of the correction data and the preliminary data are input by the prediction data input unit 21A, they may be added to the explanatory variables including the scent preference score data.

According to the first embodiment configured in this way, the scent preference score data regarding the prediction target person is also obtained by using the learning model constructed based on the scent preference score data and the personality data regarding the plurality of learning subjects. In addition, the personality of the person to be predicted can be predicted.

(Second Embodiment)
Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a block diagram showing a functional configuration example of the relationship analysis device 1B according to the second embodiment. In addition, in FIG. 2, those having the same reference numerals as those shown in FIG. 1 have the same functions, and therefore, duplicate description will be omitted here.

As shown in FIG. 2, the relationship analysis device 1B according to the second embodiment includes a learning device 10B, a predictor 20B, and a learning model storage unit 30B. The learner 10B further includes a learning target person classification unit 14B as its functional configuration, and also includes a model construction unit 13B in place of the model construction unit 13A. The predictor 20B further includes a prediction target person classification unit 24B as its functional configuration, and also includes a personality prediction unit 23B in place of the personality prediction unit 23A.

The learning target person classification unit 14B classifies a plurality of learning target persons by using a plurality of learning explanatory variables generated by the learning variable generation unit 12A. For example, the learning target person classification unit 14B uses a known k-nearest neighbor method (including the X-means method) as a method for classifying a group, and classifies a plurality of learning target persons into a plurality of groups. The classification method is not limited to this. For example, hierarchical cluster analysis, factor analysis, principal component analysis, self-organizing map, neural network, latent class analysis, latent profile analysis, or a combination of these methods may be applied.

If the number of people belonging to any of the groups is less than 20 by classifying the groups, the number of groups is reduced and the classification is performed again. That is, the group with less than 20 members is deleted, and the less than 20 learning subjects belonging to the group are reclassified into another group. Even if the reclassification is performed in this way, if the number of people belonging to each group is less than 20, the classification is not performed.

The model construction unit 13B constructs a learning model for each classification created by the learning target person classification unit 14B, and stores the constructed learning model in the learning model storage unit 30B. For example, when three classifications (groups) are generated by the learning target person classification unit 14B, the model building unit 13B uses the learning target person's fragrance preference score data, learning explanatory variables, and personality data belonging to the first classification. It is used to build a learning model for the first classification. The same applies to the other second classification and the third classification. It is the same as the model construction unit 13A described in the first embodiment except that the learning model is constructed for each classification.

The prediction target person classification unit 24B classifies the prediction target person using a plurality of prediction explanatory variables generated by the prediction variable generation unit 22A. Here, the prediction target person classification unit 24B determines to which of the groups created by the learning target person classification unit 14B the prediction target person belongs. For example, the prediction target person is classified using a known machine learning algorithm (for example, Random Forest). The machine learning algorithm used in the prediction target person classification unit 24B is not limited to Random Forest, and other algorithms such as SVM, discriminant analysis, and classification by decision tree may be applied.

The personality prediction unit 23B uses the model construction unit 13B to classify a plurality of prediction fragrance preference score data input by the prediction data input unit 21A and a plurality of prediction explanatory variables generated by the prediction variable generation unit 22A. Of the plurality of learning models constructed and stored in the learning model storage unit 30B, the personality of the predicted target person is applied to the learning model corresponding to the classification to which the predicted target person is assigned by the predicted target person classification unit 24B. Predict.

As described above, in the second embodiment, a plurality of learning subjects are classified into a plurality of groups, and a learning model is constructed for each classification. Then, the personality is predicted by using the learning model of the classification to which the prediction target belongs. By doing so, it is possible to improve the accuracy of the learning model and improve the prediction accuracy of the personality.

(Third Embodiment)
Next, a third embodiment of the present invention will be described with reference to the drawings. FIG. 3 is a block diagram showing a functional configuration example of the relationship analysis device 1C according to the third embodiment. In addition, in FIG. 3, those having the same reference numerals as those shown in FIG. 1 have the same functions, and therefore, duplicate description will be omitted here.

As shown in FIG. 3, the relationship analysis device 1C according to the third embodiment includes a learning device 10C, a predictor 20C, and a learning model storage unit 30A. The learning device 10C further includes a learning missing value complementing unit 15C as its functional configuration, and also includes a learning variable generation unit 12C and a model construction unit 13C in place of the learning variable generation unit 12A and the model construction unit 13A. There is. The predictor 20C further includes a missing value complementing unit 25C for prediction as its functional configuration, and also includes a variable generation unit 22C for prediction and a personality prediction unit 23C in place of the variable generation unit 22A for prediction and the personality prediction unit 23A. There is.

The learning missing value complementing unit 15C is based on other score data existing when at least one scent-related score data is missing from the learning scent preference score data input by the learning data input unit 11A. , Estimate and supplement the score data of the missing scent. That is, when the scent preference score data has a missing value, the learning missing value complementing unit 15C complements the missing value by estimating by machine learning using the scent preference score data in which the value exists. The machine learning to be applied is, for example, Random Forest (including missing value complementation by Miss Forest).

Here, the missing value of the aroma preference score data can occur when there is no experience of smelling aroma among 41 kinds of foods. That is, personality data based on the results of psychological tests exists, and scent preference scores for one or more foods are entered, but the learnees answered that they did not answer or did not know about scent preference for other scents. In this case, the scent preference score data is in a state where the score data is missing for at least one scent.

Further, the scent preference score data having a value is scent preference score data relating to other learning subjects. For complementing the missing value, it is preferable to use scent preference score data of a predetermined number of people (for example, 20 people) or more.

Here, Random Forest is used for complementing missing values, but the applicable machine learning algorithm is not limited to this. That is, any algorithm that can predict other variables based on one or more variables can be applied. For example, k-neighborhood method, SVM, principal component analysis, factor analysis, multiple substitution method, multiple regression, principal component regression, PLS, robust regression, (multi-term) logistic regression, laser regression, Ridge regression, elastic-net, quantification. Class I, neural network, decision tree, regression tree, Markov Monte Carlo method, Bayesian network and the like may be applied. The method that is most convenient to use should be selected according to the difference in the quantity, quality, and tendency of the data used.

When correction data or preliminary data is input by the learning data input unit 11A, these data may also be used for complementing the missing value. Further, the learning missing value complementing unit 15C may complement the missing values when the correction data or the preliminary data input by the learning data input unit 11A has missing values. Similarly, when the personality data input by the learning data input unit 11A has a missing value (due to forgetting to answer the psychological test or completing the psychological test in the middle, at least one personality data related to the personality is lost. If so, the missing value may be complemented by other personality data. However, the model construction unit 13C does not complement by mixing the data divided into the explanatory variable and the objective variable.

The learning variable generation unit 12C generates a plurality of learning explanatory variables from the learning scent preference score data input by the learning data input unit 11A and the scent preference score data complemented by the learning missing value complementing unit 15C. To do. With the learning variable generation unit 12A described in the first embodiment, except that the scent preference score data complemented by the learning missing value complementing unit 15C is added as the data to be input to the learning variable generation unit 12C. The same is true.

The model building unit 13C is input by the learning data input unit 11A, and the missing value is input by the learning fragrance preference score data complemented by the learning missing value complementing unit 15C and the learning data input unit 11A, and is defective. Using the learning personality data whose values are complemented by the learning missing value complementing unit 15C and the plurality of learning explanatory variables generated by the learning variable generation unit 12C, from the explanatory variables including the fragrance preference score data. Build a learning model to derive personality data. It is the same as the model building unit 13A described in the first embodiment except that the scent preference score data complemented by the learning missing value complementing unit 15C is added as the data to be input to the model building unit 13C.

The prediction missing value complementing unit 25C is based on other score data existing when at least one scent-related score data is missing from the prediction scent preference score data input by the prediction data input unit 21A. , Estimate and supplement the score data of the missing scent. In this case, the "other existing score data" is the fragrance preference score data without defects, which is input by the learning data input unit 11A and complemented by the learning defect value complement unit 15C. When correction data or preliminary data is input by the prediction data input unit 21A, these data may also be used for complementing the missing value.

That is, when there is a defect in the fragrance preference score data input by the prediction data input unit 21A, the prediction missing value complement unit 25C is the defect acquired by the learning data input unit 11A and the learning missing value complement unit 15C. Missing values are estimated and complemented by machine learning by adding fragrance preference score data without scent. For machine learning, it is preferable to use the same method as that used in the learning missing value complementing unit 15C.

The prediction variable generation unit 22C generates a plurality of prediction explanatory variables from the prediction fragrance preference score data input by the prediction data input unit 21A and the fragrance preference score data complemented by the prediction missing value complement unit 25C. To do. With the prediction variable generation unit 22A described in the first embodiment, except that the scent preference score data complemented by the prediction missing value complementing unit 25C is added as the data to be input to the prediction variable generation unit 22C. The same is true.

The personality prediction unit 23C is a plurality of prediction fragrance preference score data input by the prediction data input unit 21A and complemented by the prediction missing value complement unit 25C and a plurality of generations generated by the prediction variable generation unit 22C. By applying the explanatory variables for prediction to the learning model stored in the learning model storage unit 30A, the personality of the person to be predicted is predicted. It is the same as the personality prediction unit 23A described in the first embodiment, except that the scent preference score data complemented by the prediction missing value complementing unit 25C is added as the data to be input to the personality prediction unit 23C.

As described above, in the third embodiment, if there is a missing value in the scent preference score data, it is complemented, and then explanatory variables are generated, a learning model is constructed, and personality is predicted. .. By doing so, it is possible to improve the accuracy of the built learning model and the prediction data, and improve the prediction accuracy of the personality.

(Fourth Embodiment)
Next, a fourth embodiment of the present invention will be described with reference to the drawings. FIG. 4 is a block diagram showing a functional configuration example of the relationship analysis device 1D according to the fourth embodiment. In addition, in FIG. 4, those having the same reference numerals as those shown in FIGS. 1 to 3 have the same functions, and therefore, duplicate description will be omitted here.

As shown in FIG. 4, the relationship analysis device 1D according to the fourth embodiment includes a learning device 10D, a predictor 20D, and a learning model storage unit 30B. The learning device 10D further includes a learning missing value recomplementing unit 16D and a learning variable regeneration unit 17D as its functional configuration, and also includes a model building unit 13D in place of the model building unit 13B. The predictor 20D further includes a missing value replenishment unit 26D for prediction and a variable regeneration unit 27D for prediction as its functional configuration, and also includes a personality prediction unit 23D in place of the personality prediction unit 23B.

The learning missing value re-complementing unit 16D deletes the fragrance preference score data complemented by the learning missing value complementing unit 15C after the classification is performed by the learning target person classification unit 14B, and is created by the learning target person classification unit 14B. For each classification, the score data of the missing scent is estimated and re-complemented from the other scent preference score data existing. In this case, the "existing other scent preference score data" is the scent preference score data relating to other learning subjects in the same classification as the classification to which the learning subject belongs corresponding to the deficient scent preference score data. .. When the correction data and the preliminary data are complemented by the learning missing value complementing unit 15C, the correction data and the preliminary data are also deleted and re-complemented.

The complement of the missing value by the learning missing value complementing unit 15C is the complementing before the classification by the learning target person classification unit 14B is performed, and is the same classification as the learning target person corresponding to the scent preference score data in which the defect occurs. Regardless of whether or not it belongs to, it is a complement using scent preference score data for other learning subjects. On the other hand, the complementation by the learning deficiency value replenishing unit 16D used only the scent preference score data related to other learning subjects belonging to the same classification as the learning subject corresponding to the scent preference score data in which the defect occurred. Since it is a complement, it is possible to improve the accuracy of the complemented (estimated) scent preference score data.

In order to improve the accuracy of complementation, the index value obtained as data indicating the certainty of estimation when complementing the missing value by the learning missing value complementing unit 15C and the missing value by the learning missing value recomplementing unit 16D are used. When complementing, the index value obtained as data indicating the accuracy of estimation may be compared, and the one with the larger accuracy of estimation may be adopted. For example, when Random Forest is used as an estimation machine learning algorithm, a variable called an OOB (Out-of-bag) error is included in the data output at the time of estimation. It is determined that the complement with the lower OOB error value is correct, and whether the complemented scent preference score data before or after classification is used is determined.

The learning variable generation unit 17D was generated by the learning variable generation unit 12C from the fragrance preference score data re-complemented by the learning missing value replenishment unit 16D after the classification was performed by the learning target person classification unit 14B. Delete the learning explanatory variables. Then, for each classification created by the learning target person classification unit 14B, the learning scent preference score data input by the learning data input unit 11A and the scent re-complemented by the learning missing value replenishment unit 16D. Multiple learning explanatory variables are regenerated from the preference score data.

The generation of the explanatory variables for learning by the learning variable generation unit 12C is the generation before the classification by the learning target person classification unit 14B is performed, and the fragrance complemented by the learning missing value complementing unit 15C regardless of the classification. It is the generation of explanatory variables for learning using the preference score data. On the other hand, the generation of the explanatory variables for learning by the learning variable regeneration unit 17D is the scent preference score data for other learning subjects belonging to the same classification as the learning subject corresponding to the fragrance preference score data in which the defect occurs. This is the generation of explanatory variables for learning using the fragrance preference score data with high estimation accuracy, which is complemented by the missing value replenishment unit 16D for learning. Therefore, it is possible to improve the accuracy of the learning explanatory variables generated by the learning variable regeneration unit 17D.

The model building unit 13D includes learning scent preference score data and personality data input by the learning data input unit 11A, scent preference score data re-complemented by the learning missing value replenishing unit 16D, and learning variables. A learning model for deriving personality data from explanatory variables including fragrance preference score data is constructed using a plurality of learning explanatory variables regenerated by the regeneration unit 17D, and the constructed learning model is stored in the learning model storage unit. Store in 30B. As the data to be input to the model construction unit 13D, the fragrance preference score data re-complemented by the learning missing value replenishment unit 16D is added, and the learning explanatory variable generated by the learning variable generation unit 12A is used for learning variable reproduction. It is the same as the model building unit 13B described in the second embodiment except that it is replaced with the learning explanatory variable regenerated by the nurturing unit 17D.

The prediction target person classification unit 26D deletes the fragrance preference score data complemented by the prediction target person classification unit 25C after the classification is performed by the prediction target person classification unit 24B, and the prediction target person classification unit 24B predicts. For the classification assigned to the subject, the missing scent score data is estimated and re-complemented from other existing score data. In this case, the "other existing scent preference score data" is the scent preference score data related to the learning target person in the same classification as the classification to which the prediction target person corresponding to the deficient scent preference score data belongs, and is learned. It means the fragrance preference score data without defects, which is input by the data input unit 11A and complemented by the learning defect value replenishment unit 16D. When the correction data and the preliminary data are complemented by the prediction missing value complementing unit 25C, the correction data and the preliminary data are also deleted and re-complemented.

The missing value complementation by the prediction target person complementing unit 25C is the complementation before the classification by the prediction target person classification unit 24B is performed, and is the same classification as the prediction target person corresponding to the fragrance preference score data in which the defect occurs. Regardless of whether or not it belongs to, it is a complement using scent preference score data for general learning subjects. On the other hand, the complementation by the missing value replenishment unit 26D for prediction is a complementation using only the scent preference score data for the learning target person belonging to the same classification as the prediction target person corresponding to the scent preference score data in which the defect occurs. Therefore, it is possible to improve the accuracy of the complemented (estimated) scent preference score data.

In order to improve the accuracy of complementation, the index value obtained as data indicating the certainty of estimation when complementing the missing value by the prediction missing value complementing unit 25C and the missing value by the predicting missing value recomplementing unit 26D are used. At the time of complementation, the index value obtained as data indicating the certainty of estimation may be compared, and the one with the larger certainty of estimation may be adopted.

The prediction variable generation unit 27D was generated by the prediction variable generation unit 22C from the fragrance preference score data re-complemented by the prediction missing value replenishment unit 26D after the classification by the prediction target person classification unit 24B was performed. Delete the predictive explanatory variables. Then, the classification to which the prediction target person is assigned by the prediction target person classification unit 24B is re-complemented by the prediction fragrance preference score data input by the prediction data input unit 21A and the prediction missing value replenishment unit 26D. A plurality of predictive explanatory variables are regenerated from the aroma preference score data obtained.

The generation of the explanatory variables for prediction by the variable generation unit 22C for prediction is the generation before the classification by the prediction target person classification unit 24B is performed, and the fragrance complemented by the missing value complementation unit 25C for prediction regardless of the classification. It is the generation of explanatory variables for prediction using preference score data. On the other hand, the prediction variable generation unit 27D generates the prediction explanatory variable only for the scent preference score data related to the learning target person who belongs to the same classification as the prediction target person corresponding to the fragrance preference score data in which the defect occurs. It is the generation of the explanatory variable for prediction using the fragrance preference score data with high estimation accuracy, which is complemented by the missing value replenishment unit 26D for prediction. Therefore, it is possible to improve the accuracy of the predictive explanatory variable generated by the predictive variable regeneration unit 27D.

The personality prediction unit 23D is provided by the prediction fragrance preference score data input by the prediction data input unit 21A, the fragrance preference score data re-complemented by the prediction missing value replenishment unit 26D, and the prediction variable generation unit 27D. A plurality of regenerated explanatory variables for prediction are used as a learning model corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit 24B among the training models for each classification stored in the learning model storage unit 30B. By applying, the personality of the person to be predicted is predicted. As the data to be input to the personality prediction unit 23D, the fragrance preference score data re-complemented by the prediction missing value replenishment unit 26D is added, and the prediction explanatory variable generated by the prediction variable generation unit 22A is used for prediction variable reproduction. It is the same as the personality prediction unit 23B described in the second embodiment except that it is replaced with the prediction explanatory variable regenerated by the generation unit 27D.

As described above, in the fourth embodiment, a plurality of learning subjects are classified into a plurality of groups, and a learning model is constructed for each classification. Then, the personality is predicted by using the learning model of the classification to which the prediction target belongs. By doing so, it is possible to improve the accuracy of the learning model and improve the prediction accuracy of the personality.

Further, in the fourth embodiment, if there is a missing value in the scent preference score data, it is complemented, and then explanatory variables are generated, a learning model is constructed, and personality is predicted. By doing so, it is possible to improve the accuracy of the fragrance preference score data and the explanatory variables used for the constructed learning model and prediction, and to improve the prediction accuracy of the personality.

Further, in the fourth embodiment, variables are generated from the scent preference score data in which the missing values are complemented using the general scent preference score data, added to the explanatory variables, and the classification is created based on the explanatory variables. Later, for each classification, the missing value complementation and the generation of explanatory variables are re-executed using the scent preference score data belonging to the classification. By doing so, it is possible to further improve the accuracy of the fragrance preference score data and the explanatory variables used for the constructed learning model and prediction, and further improve the prediction accuracy of the personality.

The first to fourth embodiments for predicting personality from a person's taste for scent have been described above. Next, the fifth to eighth embodiments for predicting the preference for fragrance from the personality of a person will be described. In the fifth to eighth embodiments, the prediction direction is only reversed from that of the first to fourth embodiments, and the processing logic itself executed by each functional block is the first to fourth embodiments. Since it is not much different from the above, it will be briefly explained below.

(Fifth Embodiment)
Next, a fifth embodiment of the present invention will be described with reference to the drawings. FIG. 5 is a block diagram showing a functional configuration example of the relationship analysis device 2A according to the fifth embodiment. As shown in FIG. 5, the relationship analysis device 2A according to the fifth embodiment includes a learning device 50A, a predictor 60A, and a learning model storage unit 70A. The learning device 50A includes a learning data input unit 51A, a learning variable generation unit 52A, and a model construction unit 53A as its functional configuration. The predictor 60A includes a prediction data input unit 61A, a prediction variable generation unit 62A, and a scent preference prediction unit 63A as its functional configuration.

The learning data input unit 51A is detected as a result of a predetermined psychological test and scent preference score data indicating the degree of personal preference defined in a plurality of stages for each of a plurality of learning subjects for each of the plurality of learning subjects. Personality data indicating a plurality of personalities of the individual is input as learning data.

The learning variable generation unit 52A generates a plurality of learning explanatory variables from the learning personality data input by the learning data input unit 51A. In order to increase the amount of information used to generate new explanatory variables, the learning data input unit 51A is requested to further input at least one of the above-mentioned correction data and preliminary data in addition to the fragrance preference score data and personality data. You may do it.

The model building unit 53A uses the learning scent preference score data and personality data input by the learning data input unit 51A and a plurality of learning explanatory variables generated by the learning variable generation unit 52A to perform personality. A learning model for deriving fragrance preference score data from explanatory variables including data is constructed, and the constructed learning model is stored in the learning model storage unit 70A.

The prediction data input unit 61A inputs personality data as prediction data for the prediction target person.

The prediction variable generation unit 62A generates a plurality of prediction explanatory variables from the prediction personality data input by the prediction data input unit 61A.

The fragrance preference prediction unit 63A stores the prediction personality data input by the prediction data input unit 61A and a plurality of prediction explanatory variables generated by the prediction variable generation unit 62A in the learning model storage unit 70A. By applying it to the learning model, the scent preference of the prediction target person is predicted.

As described above, in the fifth embodiment, the scent preference score data indicating the degree of personal preference defined in a plurality of stages for each of the plurality of scents for each of the plurality of learning subjects, and a predetermined psychological test. Personality data indicating a plurality of personalities of an individual detected as a result of is input as learning data, and after generating a plurality of explanatory variables for learning from the input personality data for learning, the personality data is input. Build a learning model to derive fragrance preference score data from the included explanatory variables. Then, by applying the personality data for prediction input for the prediction target person and the explanatory variables generated from the personality data for prediction to the training model, the fragrance preference of the prediction target person is predicted.

According to the fifth embodiment configured in this way, the learning model constructed based on the fragrance preference score data and the personality data regarding a plurality of learning subjects is used, and based on the personality data regarding the prediction target person. , The fragrance preference of the person to be predicted can be predicted.

(Sixth Embodiment)
Next, a sixth embodiment of the present invention will be described with reference to the drawings. FIG. 6 is a block diagram showing a functional configuration example of the relationship analysis device 2B according to the sixth embodiment. Note that, in FIG. 6, those having the same reference numerals as those shown in FIG. 5 have the same functions, and therefore, duplicate description will be omitted here.

As shown in FIG. 6, the relationship analysis device 2B according to the sixth embodiment includes a learning device 50B, a predictor 60B, and a learning model storage unit 70B. The learner 50B further includes a learning target person classification unit 54B as its functional configuration, and also includes a model construction unit 53B in place of the model construction unit 53A. The predictor 60B further includes a prediction target person classification unit 64B as its functional configuration, and also includes a scent preference prediction unit 63B in place of the scent preference prediction unit 63A.

The learning target person classification unit 54B classifies a plurality of learning target persons by using a plurality of learning explanatory variables generated by the learning variable generation unit 52A.

The model construction unit 53B constructs a learning model for each classification created by the learning target person classification unit 54B, and stores the constructed learning model in the learning model storage unit 70B.

The prediction target person classification unit 64B classifies the prediction target person using a plurality of prediction explanatory variables generated by the prediction variable generation unit 62A.

The fragrance preference prediction unit 63B classifies the prediction personality data input by the prediction data input unit 61A and a plurality of prediction explanatory variables generated by the prediction variable generation unit 62A into the learning model storage unit 70B for each classification. Among the stored learning models, the prediction target person's fragrance preference is predicted by applying it to the learning model corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit 64B.

As described above, in the sixth embodiment, a plurality of learning subjects are classified into a plurality of groups, and a learning model is constructed for each classification. Then, the scent preference is predicted using the learning model of the classification to which the prediction target belongs. By doing so, it is possible to improve the accuracy of the learning model and improve the prediction accuracy of the scent preference.

(7th Embodiment)
Next, a seventh embodiment of the present invention will be described with reference to the drawings. FIG. 7 is a block diagram showing a functional configuration example of the relationship analysis device 2C according to the seventh embodiment. Note that, in FIG. 7, those having the same reference numerals as those shown in FIG. 5 have the same functions, and therefore, duplicate description will be omitted here.

As shown in FIG. 7, the relationship analysis device 2C according to the seventh embodiment includes a learning device 50C, a predictor 60C, and a learning model storage unit 70A. The learning device 50C further includes a learning missing value complementing unit 55C as its functional configuration, and also includes a learning variable generation unit 52C and a model construction unit 53C in place of the learning variable generation unit 52A and the model construction unit 53A. There is. The predictor 60C further includes a missing value complementing unit 65C for prediction as its functional configuration, and replaces the variable generation unit 62A for prediction and the scent preference prediction unit 63A with the variable generation unit 62C for prediction and the scent preference prediction unit 63C. I have.

The learning missing value complementing unit 55C lacks the learning personality data input by the learning data input unit 51A from other existing personality data when at least one personality data related to the personality is missing. Estimate and supplement the personality data of the personality you are doing.

When correction data or preliminary data is input by the learning data input unit 51A, these data may also be used for complementing the missing value. Further, the learning missing value complementing unit 55C may complement the missing values when the correction data or the preliminary data input by the learning data input unit 51A has missing values. Similarly, when a missing value occurs in the scent preference score data input by the learning data input unit 51A, the missing value may be supplemented with other scent preference score data. However, the model construction unit 53C does not complement by mixing the data divided into the explanatory variable and the objective variable.

The learning variable generation unit 52C generates a plurality of learning explanatory variables from the learning personality data input by the learning data input unit 51A and the personality data complemented by the learning missing value complementing unit 55C.

The model construction unit 53C is generated by the learning fragrance preference score data and personality data input by the learning data input unit 51A, the personality data complemented by the missing value complementing unit 55C, and the learning variable generation unit 52C. A learning model for deriving fragrance preference score data from explanatory variables including personality data is constructed by using a plurality of explanatory variables for learning.

The prediction missing value complementing unit 65C is missing from other existing personality data when at least one personality data related to the personality is missing from the prediction personality data input by the prediction data input unit 61A. Estimate and supplement the personality data of the personality you are doing. When correction data or preliminary data is input by the prediction data input unit 61A, these data may also be used to supplement the missing value.

The prediction variable generation unit 62C generates a plurality of prediction explanatory variables from the prediction personality data input by the prediction data input unit 61A and the personality data complemented by the prediction missing value complement unit 65C.

The fragrance preference prediction unit 63C inputs the missing value by the prediction data input unit 61A, and learns the personality data complemented by the prediction missing value complement unit 65C and the prediction explanatory variable generated by the prediction variable generation unit 62C. By applying it to the learning model stored in the model storage unit 70A, the fragrance preference of the prediction target person is predicted.

As described above, in the seventh embodiment, if there is a missing value in the personality data, the missing value is complemented, and then the explanatory variables are generated, the learning model is constructed, and the scent preference is predicted. By doing so, it is possible to improve the accuracy of the built learning model and the prediction data, and improve the prediction accuracy of the scent preference.

(8th Embodiment)
Next, an eighth embodiment of the present invention will be described with reference to the drawings. FIG. 8 is a block diagram showing a functional configuration example of the relationship analysis device 2D according to the eighth embodiment. Note that, in FIG. 8, those having the same reference numerals as those shown in FIGS. 5 to 7 have the same functions, and therefore, duplicate description will be omitted here.

As shown in FIG. 8, the relationship analysis device 2D according to the eighth embodiment includes a learning device 50D, a predictor 60D, and a learning model storage unit 70B. The learning device 50D further includes a learning missing value replenishment unit 56D and a learning variable generation unit 57D as its functional configuration, and also includes a model construction unit 53D in place of the model construction unit 53B. The predictor 60D further includes a missing value replenishment unit 66D for prediction and a variable regeneration unit 67D for prediction as its functional configuration, and also includes a scent preference prediction unit 63D in place of the scent preference prediction unit 63B.

The learning missing value re-complementing unit 56D was created by the learning target person classification unit 54B by deleting the personality data complemented by the learning target person classification unit 55C after the classification was performed by the learning target person classification unit 54B. For each classification, the personality data of the missing personality is estimated and re-complemented from other existing personality data. When the correction data and the preliminary data are complemented by the learning missing value complementing unit 55C, the correction data and the preliminary data are also deleted and re-complemented.

The learning variable generation unit 57D is for learning that was generated by the learning variable generation unit 52C from the personality data re-complemented by the learning missing value replenishment unit 56D after the classification by the learning target person classification unit 54B was performed. Delete the explanatory variable. Then, for each classification created by the learning target person classification unit 54B, from the learning personality data input by the learning data input unit 51A and the personality data re-complemented by the learning missing value replenishment unit 56D. Regenerate multiple learning explanatory variables.

The model construction unit 53D regenerates the learning fragrance preference score data and personality data input by the learning data input unit 51A, the personality data re-complemented by the learning missing value replenishing unit 56D, and the learning variables. A learning model for deriving fragrance preference score data from explanatory variables including personality data is constructed using a plurality of learning explanatory variables regenerated by the unit 57D, and the constructed learning model is stored in the learning model storage unit 70B. Remember.

The prediction target person classification unit 66D deletes the personality data supplemented by the prediction target person classification unit 65C after the classification is performed by the prediction target person classification unit 64B, and the prediction target person classification unit 64B deletes the personality data complemented by the prediction target person classification unit 64B. For the classification to which is assigned, the personality data of the missing personality is estimated and re-complemented from other existing personality data. When the correction data and the preliminary data are complemented by the prediction missing value complementing unit 65C, the correction data and the preliminary data are also deleted and re-complemented.

The prediction variable generation unit 67D is for prediction that was generated by the prediction variable generation unit 62C from the personality data re-complemented by the prediction missing value replenishment unit 66D after the classification by the prediction target person classification unit 64B was performed. Delete the explanatory variable. Then, the classification to which the prediction target person was assigned by the prediction target person classification unit 64B was re-complemented by the prediction personality data input by the prediction data input unit 61A and the prediction missing value replenishment unit 66D. Regenerate multiple predictive explanatory variables from personality data.

The fragrance preference prediction unit 63D provides personality data input by the prediction data input unit 61A and re-complemented by the prediction missing value replenishment unit 66D and prediction explanatory variables regenerated by the prediction variable regeneration unit 67D. Among the learning models stored in the learning model storage unit 70B, the fragrance preference prediction unit of the prediction target person is predicted by applying it to the learning model corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit 64B. To do.

As described above, in the eighth embodiment, a plurality of learning subjects are classified into a plurality of groups, and a learning model is constructed for each classification. Then, the scent preference is predicted using the learning model of the classification to which the prediction target belongs. By doing so, it is possible to improve the accuracy of the learning model and improve the prediction accuracy of the scent preference.

Further, in the eighth embodiment, if there is a missing value in the personality data, it is complemented, and then an explanatory variable is generated, a learning model is constructed, and a scent preference is predicted. By doing so, it is possible to improve the accuracy of the built learning model, the personality data used for the prediction, and the explanatory variables, and to improve the prediction accuracy of the fragrance preference.

Further, in the eighth embodiment, an explanatory variable is generated from the personality data in which the missing value is complemented by using the general personality data, the explanatory variable is added to the explanatory variable, and a classification is created based on the explanatory variable. For each classification, missing value completion and explanatory variable generation are re-executed using the personality data belonging to the classification. By doing so, it is possible to further improve the accuracy of the built learning model, the personality data used for the prediction, and the explanatory variables, and further improve the prediction accuracy of the fragrance preference.

It should be noted that the first to eighth embodiments are merely examples of embodiment of the present invention, and the technical scope of the present invention should not be interpreted in a limited manner by these. It is something that does not become. That is, the present invention can be implemented in various forms without departing from its gist or its main features.

For example, in the above embodiment, an example of analyzing the relationship between the scent of food and the personality has been described, but it is also possible to analyze the relationship between the scent other than food and the personality.

Further, in the first to fourth embodiments, the learning scent preference score data input by the learning data input unit 11A (in the case of the third and fourth embodiments, the missing values are complemented). Including. The same applies hereinafter) and a plurality of learning explanatory variables generated by the learning variable generation units 12A and 12C (in the case of the fourth embodiment, those regenerated by the learning variable generation unit 17D are also included. An example of constructing a learning model using both of the same) has been described below, but the learning model is constructed using either the input scent preference score data for learning or the generated explanatory variables for learning. You may try to build it. Similarly, in the fifth to eighth embodiments, the learning personality data input by the learning data input unit 51A (in the case of the seventh and eighth embodiments, the missing values are complemented) are also included. Hereinafter, the same applies) or a plurality of learning explanatory variables generated by the learning variable generation units 52A and 52C (in the case of the fourth embodiment, those regenerated by the learning variable generation unit 57D are also included. The learning model may be constructed using either one of the same).

In this case, the personality prediction units 23A to 23D in the first to fourth embodiments are missing the prediction scent preference score data (in the case of the third and fourth embodiments, they are missing) input by the prediction data input unit 21A. Including those whose values are complemented. The same shall apply hereinafter) or a plurality of predictive explanatory variables generated by the predictive variable generation units 22A and 22C (in the case of the fourth embodiment, reproduced by the predictive variable generation unit 27D). The personality of the person to be predicted is predicted by applying one of the following (the same applies hereinafter) to the learning model. Similarly, the scent preference prediction units 63A to 63D in the fifth to eighth embodiments are prediction personality data (missing values in the case of the seventh and eighth embodiments) input by the prediction data input unit 61A. (The same shall apply hereinafter) or a plurality of predictive explanatory variables generated by the predictive variable generators 62A and 62C (in the case of the fourth embodiment, the predictive variable generator 67D regenerates the data). By applying either one of the above to the learning model, the scent preference of the person to be predicted is predicted.

Further, in the second and fourth embodiments, an example of classifying the prediction target person using a plurality of learning explanatory variables generated by the learning variable generation units 12A and 12C has been described, but the learning data has been described. The classification may be performed using the learning scent preference score data (including the one in which the missing value is complemented in the case of the fourth embodiment. The same shall apply hereinafter) input by the input unit 11A, or the input may be performed. The classification may be performed using both the scent preference score data for learning and the generated explanatory variables for learning. Similarly, in the sixth and eighth embodiments, in addition to the example of classifying the prediction target person using a plurality of learning explanatory variables generated by the learning variable generation units 52A and 52C, the learning data input unit. Classification may be performed using the learning personality data input by 51A (including the data in which missing values are complemented in the case of the eighth embodiment. The same shall apply hereinafter), or the input learning personality data may be used. The classification may be performed using both the personality data of the above and a plurality of generated explanatory variables for learning.

1A, 1B, 1C, 1D relationship analyzer (for personality prediction)
2A, 2B, 2C, 2D relationship analyzer (for scent preference prediction)
10A, 10B, 10C, 10D Learner 11A Learning data input unit 12A, 12C Learning variable generation unit 13A, 13B, 13C, 13D Model construction unit 14B Learning target person classification unit 15C Learning missing value complementing unit 16D Learning defect Value recompletion unit 17D Learning variable generation unit 20A, 20B, 20C, 20D Predictor 21A Prediction data input unit 22A, 22C Prediction variable generation unit 23A, 23B, 23C, 23D Personality prediction unit 24B Prediction target person classification unit 25C Missing value complement for prediction 26D Missing value replenishment for prediction 27D Variable regeneration for prediction 30A, 30B Learning model storage 50A, 50B, 50C, 50D Learner 51A Data input for learning 52A, 52C Learning variable Generation unit 53A, 53B, 53C, 53D Model construction unit 54B Learning target person classification unit 55C Learning missing value complementing unit 56D Learning missing value recomplementing unit 57D Learning variable generation unit 60A, 60B, 60C, 60D Predictor 61A Prediction data input section 62A, 62C Prediction variable generation section 63A, 63B, 63C, 63D Fragrance preference prediction section 64B Prediction target person classification section 65C Prediction missing value complement section 66D Prediction missing value complement section 67D Prediction variable playback Narube 70A, 70B Learning model storage unit

Claims (14)

For each of multiple learning subjects, scent preference score data showing the degree of personal preference defined in multiple stages for each scent, and multiple personalities of individuals detected as a result of a predetermined psychological test. A learning data input unit that inputs the indicated personality data as learning data, respectively.
Using the learning scent preference score data and the personality data input by the learning data input unit, a model construction for constructing a learning model for deriving the personality data from the explanatory variables including the scent preference score data. Department and
For the prediction target person, a prediction data input unit that inputs the above scent preference score data as prediction data, and
By applying the prediction scent preference score data input by the prediction data input unit to the learning model constructed by the model construction unit, the personality prediction unit that predicts the personality of the prediction target person and the personality prediction unit. A device for analyzing the relationship between olfactory taste and personality. A learning variable generation unit that generates a plurality of learning explanatory variables from the learning fragrance preference score data input by the learning data input unit, and a learning variable generation unit.
It further includes a prediction variable generation unit that generates a plurality of prediction explanatory variables from the prediction fragrance preference score data input by the prediction data input unit.
The model construction unit includes both or one of the fragrance preference score data for learning input by the learning data input unit and the plurality of learning explanatory variables generated by the learning variable generation unit. The learning model is constructed by using the personality data for learning input by the data input unit for learning.
The personality prediction unit uses both or one of the above-mentioned fragrance preference score data for prediction input by the above-mentioned data input unit for prediction and the plurality of explanatory variables for prediction generated by the above-mentioned variable generation unit for prediction. The relationship analysis device between olfactory preference and personality according to claim 1, wherein the personality of the prediction target person is predicted by applying it to the learning model constructed by the model construction unit. A learning variable generation unit that generates a plurality of learning explanatory variables from the learning fragrance preference score data input by the learning data input unit, and a learning variable generation unit.
Using both or one of the fragrance preference score data for learning input by the learning data input unit and the plurality of learning explanatory variables generated by the learning variable generation unit, the plurality of learnings are performed. The learning target person classification department that classifies the target person, and
A prediction variable generation unit that generates a plurality of prediction explanatory variables from the prediction fragrance preference score data input by the prediction data input unit, and a prediction variable generation unit.
Using both or one of the above-mentioned fragrance preference score data for prediction input by the above-mentioned data input unit for prediction and the above-mentioned explanatory variables for prediction generated by the above-mentioned variable generation unit for prediction, the person to be predicted. It is further equipped with a forecast target person classification unit that classifies
The model building unit builds the learning model for each classification created by the learning target person classification unit.
The personality prediction unit may generate both or one of the fragrance preference score data for prediction input by the data input unit for prediction and the plurality of explanatory variables for prediction generated by the variable generation unit for prediction. Among the learning models constructed for each classification by the model construction unit, the personality of the prediction target person is applied to the learning model corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit. The relationship analysis device between olfactory preference and personality according to claim 1, wherein the data is predicted. When the scent preference score data for learning input by the learning data input unit is missing from the other scent preference score data existing, when the scent preference score data for at least one scent is missing, it is missing. It is further equipped with a missing value complementation unit for learning that estimates and complements the scent preference score data of the existing scent.
The learning variable generation unit generates the plurality of learning explanatory variables from the fragrance preference score data that is input by the learning data input unit and the missing values are complemented by the learning missing value complementing unit.
The model construction unit regenerates the missing value by inputting it by the learning data input unit and complementing the missing value by the learning missing value complementing unit with the learning fragrance preference score data and the learning variable generation unit. The second aspect of the present invention is characterized in that the learning model is constructed by using both or one of the plurality of learning explanatory variables and the personality data input by the learning data input unit. A device for analyzing the relationship between the described olfactory preference and personality. When the scent preference score data for learning input by the learning data input unit is missing from the other scent preference score data existing, when the scent preference score data for at least one scent is missing, the scent preference score data is missing. A learning deficiency value complement part that estimates and complements the scent preference score data of the existing scent,
After the classification by the learning target classification unit is performed, the fragrance preference score data complemented by the learning missing value complementing unit is deleted, and other classifications existing for each classification created by the learning target classification unit are present. From the scent preference score data, the learning deficiency value replenishment unit that estimates and recomplements the scent preference score data of the above-mentioned missing scent, and
After the classification by the learning target person classification unit is performed, the learning explanatory variables generated by the learning variable generation unit are deleted from the fragrance preference score data re-complemented by the learning missing value replenishment unit, and the above For each classification created by the learning target person classification unit, from the scent preference score data for learning input by the learning data input unit and the fragrance preference score data re-complemented by the missing value replenishment unit for learning. It is further provided with a learning variable generation unit that regenerates the above-mentioned plurality of learning explanatory variables.
The learning variable generation unit generates the plurality of learning explanatory variables from the fragrance preference score data that is input by the learning data input unit and the missing values are complemented by the learning missing value complementing unit.
The model construction unit regenerates the missing value by the learning data input unit and the learning variable generation unit and the learning fragrance preference score data re-complemented by the learning missing value replenishment unit. The third aspect of claim 3, wherein the learning model is constructed by using both or one of the plurality of learning explanatory variables and the personality data input by the learning data input unit. A device for analyzing the relationship between the described olfactory preference and personality. When the scent preference score data for prediction input by the prediction data input unit is missing from the other scent preference score data existing, when the scent preference score data for at least one scent is missing, it is missing. It also has a predictive missing value complement section that estimates and complements the scent preference score data of the existing scent.
The prediction variable generation unit generates the plurality of prediction explanatory variables from the fragrance preference score data input by the prediction data input unit and the missing values complemented by the prediction missing value complementing unit.
The personality prediction unit inputs the missing value by the prediction data input unit and complements the missing value by the prediction missing value complementing unit, and the fragrance preference score data and the plurality of prediction explanations generated by the prediction variable generation unit. The olfactory preference according to claim 2 or 4, wherein the personality of the prediction target person is predicted by applying both or one of the variables to the learning model constructed by the model construction unit. Relationship analysis device between and personality. When the scent preference score data for prediction input by the prediction data input unit is missing from the other scent preference score data existing, when the scent preference score data for at least one scent is missing, it is missing. Predictive missing value complementation unit that estimates and complements the scent preference score data of the existing scent,
After the classification by the prediction target person classification unit is performed, the scent preference score data complemented by the prediction target person complementing unit is deleted, and the classification to which the prediction target person is assigned by the prediction target person classification unit is performed. From the other scent preference score data that exists, the scent preference score data of the above-mentioned deficient scent is estimated and re-complemented.
After the classification by the prediction target person classification unit is performed, the prediction explanatory variable generated by the prediction variable generation unit is deleted from the fragrance preference score data re-complemented by the prediction missing value replenishment unit, and the above For the classification to which the prediction target person is assigned by the prediction target person classification unit, the plurality of fragrance preference score data input by the prediction data input unit and the missing value is re-complemented by the prediction target person replenishment unit. It also has a predictive variable regeneration unit that regenerates the predictive explanatory variables.
The prediction variable generation unit generates the plurality of prediction explanatory variables from the fragrance preference score data input by the prediction data input unit and the missing values complemented by the prediction missing value complementing unit.
The personality prediction unit is the fragrance preference score data input by the prediction data input unit and the missing values are re-complemented by the prediction missing value replenishment unit, and the plurality of items regenerated by the prediction variable regeneration unit. Of the learning models constructed for each classification by the model construction unit, learning corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit. The relationship analysis device between olfactory preference and personality according to claim 3 or 5, wherein the personality of the person to be predicted is predicted by applying it to a model. For each of multiple learning subjects, scent preference score data showing the degree of personal preference defined in multiple stages for each scent, and multiple personalities of individuals detected as a result of a predetermined psychological test. A learning data input unit that inputs the indicated personality data as learning data, respectively.
Model construction for constructing a learning model for deriving the fragrance preference score data from the explanatory variables including the personality data using the fragrance preference score data for learning and the personality data input by the learning data input unit. Department and
Forecast data input unit that inputs the above personality data as forecast data for the forecast target person,
By applying the prediction personality data input by the prediction data input unit to the learning model constructed by the model construction unit, the scent preference prediction unit predicts the scent preference of the prediction target person. A device for analyzing the relationship between odor preference and personality. A learning variable generation unit that generates a plurality of learning explanatory variables from the learning personality data input by the learning data input unit, and a learning variable generation unit.
It further includes a prediction variable generation unit that generates a plurality of prediction explanatory variables from the prediction personality data input by the prediction data input unit.
The model construction unit includes both or one of the personality data for learning input by the data input unit for learning and the plurality of explanatory variables for learning generated by the variable generation unit for learning, and the learning. Using the above-mentioned fragrance preference score data for learning input by the data input unit for learning, the above-mentioned learning model is constructed.
The fragrance preference prediction unit uses both or one of the prediction personality data input by the prediction data input unit and the plurality of prediction explanatory variables generated by the prediction variable generation unit. The relationship analysis device between olfactory preference and personality according to claim 8, wherein the fragrance preference of the prediction target person is predicted by applying it to the learning model constructed by the model construction unit. A learning variable generation unit that generates a plurality of learning explanatory variables from the learning personality data input by the learning data input unit, and a learning variable generation unit.
Using both or one of the personality data for learning input by the data input unit for learning and the explanatory variables for learning generated by the variable generation unit for learning, the plurality of learning subjects. And the learning target classification department that classifies
A prediction variable generation unit that generates a plurality of prediction explanatory variables from the prediction personality data input by the prediction data input unit, and a prediction variable generation unit.
Classification of the prediction target person using both or one of the prediction personality data input by the prediction data input unit and the plurality of prediction explanatory variables generated by the prediction variable generation unit. Further equipped with a forecast target person classification department to perform
The model building unit builds the learning model for each classification created by the learning target person classification unit.
The fragrance preference prediction unit uses both or one of the prediction personality data input by the prediction data input unit and the plurality of prediction explanatory variables generated by the prediction variable generation unit. Among the learning models constructed for each classification by the model construction unit, the fragrance preference of the prediction target person is applied to the learning model corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit. The relationship analysis device between olfactory preference and personality according to claim 9, wherein the data is predicted. When the personality data related to at least one personality is missing for the personality data for learning input by the learning data input unit, the personality data of the missing personality is obtained from the other existing personality data. It is further equipped with a missing value complement for learning that estimates and complements.
The learning variable generation unit generates the plurality of learning explanatory variables from the personality data that is input by the learning data input unit and the missing values are complemented by the learning missing value complementing unit.
The model building unit is the personality data for learning that is input by the learning data input unit and the missing values are complemented by the learning missing value complementing unit, and the above that is regenerated by the learning variable generation unit. A claim characterized in that the learning model is constructed by using both or one of a plurality of explanatory variables for learning and the fragrance preference score data for learning input by the learning data input unit. 9. The relationship analysis device between olfactory preference and personality according to 9. When the personality data related to at least one personality is missing for the personality data for learning input by the learning data input unit, the personality data of the missing personality is obtained from the other existing personality data. A learning missing value complement that estimates and complements,
After the classification by the learning target person classification unit is performed, the personality data complemented by the learning missing value complementing unit is deleted, and other personality data existing for each classification created by the learning target person classification unit is deleted. From the learning missing value re-complementing unit that estimates and re-complements the personality data of the above-mentioned missing personality,
After the classification by the learning target person classification unit is performed, the learning explanatory variables generated by the learning variable generation unit are deleted from the personality data re-complemented by the learning missing value replenishment unit, and the learning target is deleted. Each of the above-mentioned plurality of learning explanations from the above-mentioned personality data for learning input by the above-mentioned learning data input unit and the personality data re-complemented by the above-mentioned learning missing value replenishment unit for each classification created by the person classification unit. It also has a learning variable generator that regenerates variables.
The learning variable generation unit generates the plurality of learning explanatory variables from the personality data that is input by the learning data input unit and the missing values are complemented by the learning missing value complementing unit.
The model construction unit was regenerated by the personality data for learning and the variable generation unit for learning, in which the missing values were input by the data input unit for learning and the missing values were re-complemented by the missing value replenishment unit for learning. A claim characterized in that the learning model is constructed by using both or one of the plurality of explanatory variables for learning and the fragrance preference score data for learning input by the learning data input unit. Item 10. The relationship analysis device between olfactory preference and personality according to Item 10. For the prediction personality data input by the prediction data input unit, when the personality data relating to at least one personality is missing, the personality data of the missing personality is obtained from the other existing personality data. It is further equipped with a missing value complement for prediction that estimates and complements.
The prediction variable generation unit generates the plurality of prediction explanatory variables from the personality data that is input by the prediction data input unit and the missing values are complemented by the prediction missing value complementing unit.
The fragrance preference prediction unit is the personality data input by the prediction data input unit and the missing values complemented by the prediction missing value complementing unit, and the plurality of prediction explanatory variables generated by the prediction variable generation unit. The olfactory preference according to claim 9 or 11, wherein the scent preference of the prediction target person is predicted by applying both or either of the above to the learning model constructed by the model construction unit. Relationship analysis device between and personality. For the prediction personality data input by the prediction data input unit, when the personality data relating to at least one personality is missing, the personality data of the missing personality is obtained from the other existing personality data. A predictive missing value complement that estimates and complements,
After the classification by the prediction target person classification unit is performed, the personality data complemented by the prediction target person complementing unit is deleted, and the classification to which the prediction target person is assigned by the prediction target person classification unit exists. A predictive missing value re-complementing unit that estimates and re-complements the personality data of the above-mentioned missing personality from other personality data,
After the classification by the prediction target person classification unit is performed, the prediction explanatory variable generated by the prediction variable generation unit is deleted from the personality data re-complemented by the prediction missing value replenishment unit, and the prediction target is deleted. Regarding the classification to which the prediction target person is assigned by the person classification unit, the plurality of personality data for prediction input by the data input unit for prediction and the personality data re-complemented by the missing value replenishment unit for prediction are used. It also has a predictive variable regeneration unit that regenerates the predictive explanatory variables of.
The prediction variable generation unit generates the plurality of prediction explanatory variables from the personality data that is input by the prediction data input unit and the missing values are complemented by the prediction missing value complementing unit.
The fragrance preference prediction unit is the personality data input by the prediction data input unit and the missing values re-complemented by the prediction missing value replenishment unit, and the plurality of personality data regenerated by the prediction variable regeneration unit. Of the training models constructed for each classification by the model construction unit, both or one of the explanatory variables for prediction is a learning model corresponding to the classification to which the prediction target person is assigned by the prediction target person classification unit. The relationship analysis device between olfactory preference and personality according to claim 10 or 12, wherein the fragrance preference prediction unit of the prediction target person is predicted by applying to the above.

Images