Evaluation value estimation system, evaluation value estimation method, evaluation value estimation device, evaluation value estimation program, and storage medium on which said program is recorded

Abstract

This estimation system S comprises an emotion recognition unit 12 that recognizes an emotional valence intensity and arousal level of a user toward an object, and an evaluation value estimation unit 16 that estimates a user evaluation value using a learning model that receives input of the emotional valence intensity and arousal level of the user and outputs a user evaluation value.

Description

Rating value estimation system, rating value estimation method, rating value estimation device, rating value estimation program, and storage medium having the program recorded thereon

[0001] The present invention relates to a rating value estimation system, a rating value estimation method, a rating value estimation device, a rating value estimation program, and a storage medium on which the program is recorded, for estimating a user's willingness to a target object.

[0002] Generally, even when people feel the same way about something, the content and strength of their emotions vary from person to person. It is also difficult to accurately and quantitatively express such emotions, even when it comes to one's own feelings. Furthermore, since emotions toward a certain object change over time, it is desirable to express emotions quickly. Meanwhile, there is a demand for quantitatively grasping the emotions of many people toward a certain object in a short period of time.

[0003] To meet such demands, various methods for measuring emotions have been proposed. For example, the technology described in Patent Literature 1 uses a model with joints that mimic the human body, and grasps the emotions of a user based on how the user bends which joints.

[0004] Patent No. 6508988

[0005] However, when expressing a user's emotions through a model, as in the technology described in Patent Document 1, a certain amount of explanation is required for the method of expression. Furthermore, even if the expression through the model is similar, the emotions associated with that expression may differ for each user depending on cultural background, etc. Furthermore, in order to express a user's emotions through the model, it takes a certain amount of time to operate the model, and there is a risk that the emotions may change from the emotions felt when interacting with the object.

[0006] Therefore, emotion data collected using such models may have low versatility. For example, if an analysis is performed based on data collected from users living in a certain area, the results of that analysis may not necessarily be applicable to users living in other areas.

[0007] Furthermore, when using a model with such a complex structure, the input work for the user becomes cumbersome, which in turn makes it difficult to collect data from a large number of users.

[0008] The present invention has been made in consideration of the above points, and aims to provide a rating value estimation system, a rating value estimation method, a rating value estimation device, a rating value estimation program, and a storage medium on which the program is recorded, which are capable of quickly and easily collecting data related to a user's emotions in a highly versatile manner, and further capable of easily analyzing the collected data.

[0009] The rating value estimation system of the present invention is an rating value estimation system that estimates a user rating value, which is a value regarding a user's motivation for a target object, and is characterized by comprising an emotion recognition unit that recognizes the emotional valence intensity and arousal level of the user toward the target object, and an rating value estimation unit that estimates the user rating value using a learning model that receives the emotional valence intensity and arousal level of the user as input and outputs the user rating value.

[0010] Here, "valence" is a bipolar concept that indicates the qualitative difference in the emotion evoked (for example, positive valence such as attraction (pleasantness) toward the target object, or negative valence such as aversion (unpleasantness)). Furthermore, "valence intensity" refers to the strength of the emotion felt (Harabe Yoshihiro (2015). Measuring the intensity of general valence concepts using paired comparisons: A preliminary study using unpleasant emotions. Tezukayama Gakuin University Faculty of Human Sciences Annual Research Report, (17)).

[0011] Additionally, "arousal" here is a concept with two extremes: excitement (high arousal) and calmness (low arousal), and indicates the intensity of the emotion evoked (Bradley, MM, Greenwald, MK, Petry, MC, & Lang, PJ (1992). Remembering Pictures: Pleasure and Arousal in Memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 379-390.).

[0012] The term "evaluation value" used here refers to a value regarding the willingness to purchase a target item or a similar item. For example, the evaluation value may be expressed as a clear value using a point scale with an arbitrary value, such as past consumption frequency, the degree of recommendation to others, willingness to consume, or willingness to purchase.

[0013] As described above, the rating value estimation system of the present invention employs universal parameters, namely, valence and arousal. Therefore, the user rating value output by inputting these parameters is a highly versatile value that is not dependent on cultural background, etc. Furthermore, valence and arousal are parameters that can be easily obtained without using special equipment.

[0014] Furthermore, since these parameters are simplified parameters for expressing complex emotions, when analysis is performed using these parameters, the analysis process can be performed more easily and quantitatively than when using conventional parameters for expressing emotions (e.g., sentences written by the user).

[0015] Therefore, this system can quickly and easily collect data on users' emotions in a highly versatile manner, and can also easily analyze the collected data. This is particularly effective when the subject of evaluation is something like cooking, which involves complex factors such as personal physical condition and preferences, as well as cultural background.

[0016] Furthermore, in the evaluation value estimation system of the present invention, the emotion recognition unit has an emotion model presentation unit that presents to the user an emotion model that is a two-dimensional or three-dimensional model including at least a first axis indicating emotional valence intensity and a second axis indicating arousal level in a format that allows the user to select any coordinate in the emotion model, and a selected coordinate recognition unit that recognizes selected coordinates that are coordinates selected from the emotion model by the user when the target thing is presented, and it is preferable that the evaluation value estimation unit receives the selected coordinates as an input.

[0017] In this way, by adopting an emotion model, which is a planar or three-dimensional model including at least a first axis indicating valence intensity and a second axis indicating arousal level, as an interface for inputting emotions, users can easily and accurately express their emotions as clear values ​​called coordinates, which in turn makes it possible to easily and accurately collect data related to users' emotions.

[0018] Furthermore, since coordinate parameters are parameters that users can intuitively understand, there is no need to provide users with special explanations about how to express them when collecting data. This makes it possible to quickly collect parameters related to emotions that are prone to change before those changes actually occur. Furthermore, since coordinate parameters are simple and clear, using such coordinates as parameters for analysis makes the analysis process even easier.

[0019] In addition, when the emotion model is a three-dimensional model, the third axis may be, for example, an axis indicating the degree of arousal (Sato Toshio and Horigome Kazuya (2018). Examination of emotion scales using a three-dimensional model. The 82nd Annual Meeting of the Japanese Psychological Association, 721).

[0020] Furthermore, in the evaluation value estimation system of the present invention, it is preferable that the user evaluation value includes values ​​for a plurality of evaluation items, and the learning model takes one of the emotional valence intensity and the arousal level as input and outputs values ​​for the plurality of evaluation items.

[0021] Here, "evaluation items" are items based on the five human senses for the target object. For example, if the target object is food, the evaluation items would be the taste (gustation), aroma / flavor (olfaction), appearance (sight), texture (touch), and sound (hearing) of the food.

[0022] The user evaluation value does not necessarily have to be a value for one evaluation item, but may include values ​​for multiple evaluation items in this way. In addition, adopting multiple evaluation items in the user evaluation value in this way enables multifaceted evaluation, and the analysis results using the user evaluation value can be highly objective and accurate.

[0023] Furthermore, in the evaluation value estimation system of the present invention, when the evaluation value includes values ​​for multiple evaluation items, it is preferable to include an attractiveness value estimation unit that weights each of the values ​​for the multiple evaluation items output from the learning model by inputting one of the emotional valence intensity and the arousal level, and estimates the sum of the weighted values ​​as the attractiveness value.

[0024] The evaluation items that should be emphasized may differ depending on the object for which the evaluation value is to be estimated, the purpose for which the user evaluation value is to be used, etc. Therefore, when adding up the user evaluation values ​​to output a single attractiveness value, if different weighting is applied to each evaluation item according to the purpose of use, etc. before calculating the attractiveness value, the output attractiveness value, and ultimately the analysis results using that attractiveness value, can be made more accurate.

[0025] Furthermore, in the evaluation value estimation system of the present invention, when the user evaluation value includes values ​​for multiple evaluation items, it is preferable that the target object is a food product, and the multiple evaluation items include at least one of an evaluation item based on taste, an evaluation item based on smell, an evaluation item based on vision, and an evaluation item based on touch.

[0026] Generally, food evaluations are based on multiple criteria, such as taste (gustation), aroma / flavor (olfaction), appearance (sight), texture (touch), and the sound made when eating or cooking (hearing), and the evaluations that are output tend to be complex. Therefore, by using a system that can easily and instantly collect emotional data as input data for such food evaluations and that outputs user evaluation values ​​that are easy to handle during analysis, the evaluations can be performed easily and accurately.

[0027] Furthermore, in the evaluation value estimation system of the present invention, it is preferable that the user evaluation value includes values ​​for a plurality of evaluation criteria, and the learning model takes one of the emotional valence intensity and the arousal level as input and outputs a value based on the plurality of evaluation criteria.

[0028] Here, the "evaluation criteria" are the criteria for evaluating the willingness to purchase a target item or a similar item. Therefore, the "evaluation value" in this case is a value based on the evaluation criteria. Examples of the "evaluation criteria" that can be used include past consumption frequency, the degree of recommendation to others, willingness to consume, willingness to purchase, etc.

[0029] The user evaluation value does not necessarily have to be a single value, but may include values ​​for multiple evaluation criteria (i.e., multiple values) in this way. By adopting values ​​for multiple evaluation criteria in this way, it becomes possible to evaluate a single item from multiple perspectives, and the analysis results using the user evaluation value can be made highly objective and accurate.

[0030] Furthermore, in the evaluation value estimation system of the present invention, when the user evaluation value includes values ​​for multiple evaluation criteria, it is preferable to include an attractiveness value estimation unit that weights each of the values ​​for the multiple evaluation criteria output from the learning model by inputting one of the emotional valence intensity and the arousal level, and recognizes the sum of the weighted values ​​as the attractiveness value.

[0031] The evaluation criteria that should be emphasized may differ depending on the object for which the user evaluation value is to be estimated, the purpose for which the user evaluation value is to be used, etc. Therefore, when adding up the user evaluation values ​​to output a single attractiveness value, if different weighting is applied to each evaluation criterion according to the purpose of use, etc. before calculating the attractiveness value, the output attractiveness value, and ultimately the analysis results using the attractiveness value, can be made more accurate.

[0032] Furthermore, in the evaluation value estimation system of the present invention, when the evaluation value includes values ​​for multiple evaluation criteria, the values ​​for the multiple evaluation criteria may include at least one of a value for past consumption frequency, a value for the degree of recommendation to others, a value for the degree of willingness to consume, and a value for the degree of willingness to purchase.

[0033] Furthermore, in the evaluation value estimation system of the present invention, when an emotion model that is a two-dimensional model or a three-dimensional model including at least a first axis indicating emotional valence intensity and a second axis indicating arousal level is adopted as an interface for inputting emotions, the learning model may be correlation data that indicates the correlation between reference coordinates, which are coordinates selected from the emotion model, and a reference evaluation value, which is a value regarding motivation toward the target thing or a similar thing that has the same or similar attributes as the target thing.

[0034] Furthermore, in the evaluation value estimation system of the present invention, when correlation data is used as a learning model, the reference coordinates are coordinates selected from the emotion model for the target thing or the similar thing by a test user who has previously evaluated the target thing or the similar thing, and the correlation data may be data indicating the correlation between the reference coordinates and the reference evaluation value by the test user.

[0035] Furthermore, in the evaluation value estimation system of the present invention, when an emotion model that is a two-dimensional model or a three-dimensional model including at least a first axis indicating emotional valence intensity and a second axis indicating arousal level is adopted as an interface for inputting emotions, the learning model may be a prediction model that is generated by machine learning using, as training data, reference coordinates that are coordinates selected from the emotion model and reference evaluation values ​​that are values ​​regarding motivation for the target thing or similar things that have the same or similar attributes as the target thing, and that outputs the user evaluation value from the input selected coordinates.

[0036] Furthermore, in the evaluation value estimation system of the present invention, when a prediction model is used as a learning model, the reference coordinates are coordinates selected from the emotion model for the target thing or the similar thing by a test user who has previously evaluated the target thing or the similar thing, and the prediction model may be generated by machine learning using the reference coordinates and the reference evaluation value by the test user as training data.

[0037] Furthermore, the evaluation value estimation device of the present invention is an evaluation value estimation device that estimates a user evaluation value, which is a value regarding a user's motivation for a target thing, and is characterized by comprising an emotion recognition unit that recognizes the emotional valence intensity and arousal level of the user toward the target thing, and an evaluation value estimation unit that estimates the user evaluation value using a learning model that receives the emotional valence intensity and arousal level of the user as input and outputs the user evaluation value.

[0038] Furthermore, the evaluation value estimation method of the present invention is a method for estimating a user evaluation value, which is a value regarding a user's motivation for a target object, and is characterized by comprising: a step in which an emotion recognition unit recognizes the emotional valence intensity and arousal level of the user toward the target object; and a step in which an evaluation value estimation unit estimates the user evaluation value using a learning model that receives the emotional valence intensity and arousal level of the user as input and outputs the user evaluation value.

[0039] Furthermore, the evaluation value estimation program of the present invention is an evaluation value estimation program that causes a computer to execute an evaluation value estimation method for estimating a user evaluation value, which is a value regarding a user's motivation for a target object, and is characterized in that the computer is caused to execute the following steps: an emotion recognition unit recognizes the emotional valence intensity and arousal level of the user toward the target object; and an evaluation value estimation unit estimates the user evaluation value using a learning model that receives the emotional valence intensity and arousal level of the user as input and outputs the user evaluation value.

[0040] A recording medium according to the present invention stores the above-mentioned evaluation value estimation program, and is readable by the computer.

[0041] 1 is an explanatory diagram showing a schematic configuration of an estimation system according to an embodiment. A block diagram showing the configuration of a processing unit of the estimation system of FIG. 1. An image diagram showing an example of an image displayed on a user terminal when the estimation system of FIG. 1 recognizes user information. An image diagram showing an example of an image displayed on a user terminal when the estimation system of FIG. 1 recognizes a user's emotion. A schematic diagram showing an example of an emotion model used to recognize emotions in the estimation system of FIG. 1. A schematic diagram showing an example of an emotion model used to recognize emotions in an estimation system according to a modified example. An image diagram showing an example of an image displayed on a user terminal when the estimation system of FIG. 1 recognizes a reference rating value. An image diagram showing an example of correlation data recognized by the estimation system of FIG. 1. A flowchart showing processing performed by the estimation system of FIG. 1 when collecting correlation data. A flowchart showing processing performed by the estimation system of FIG. 1 when estimating an attractiveness value. An image diagram showing an example of an image displayed on a client terminal after the estimation system of FIG. 1 estimates an attractiveness value. A sample image used in a data collection experiment for generating a learning model for the estimation system of FIG. 1. Data related to a purchase probability score among experimental data for generating a learning model for the estimation system of FIG. 1. Data related to willingness to take home among experimental data for generating a learning model for the estimation system of FIG. 1. Among the experimental data used to generate the learning model of the estimation system in Figure 1, data related to the degree of recommendation to others (Net Promoter Score)

[0042] Hereinafter, a rating value estimation system according to an embodiment (hereinafter referred to as "estimation system S") and a rating value estimation method implemented using the same will be described with reference to the drawings.

[0043] In this embodiment, we will explain the case where the estimation system S is used in a service that estimates the attractiveness value of a target object (food in this embodiment) related to a user U and presents that attractiveness value to clients engaged in marketing, etc.

[0044] In the following description, an "evaluation value" refers to a value regarding willingness toward a target object or a similar object described below. Specifically, for example, past consumption frequency (e.g., Frequency of past consumption), degree of recommendation to others (e.g., Net promoter score), willingness to consume (e.g., Willingness to take home), willingness to buy (e.g., Purchase Probability Score, Willingness to buy, the Juster scale), or other values ​​expressed as clear values ​​using a point scale with an arbitrary value set may be used. The estimation system S uses these values.

[0045] Furthermore, the "reference evaluation value" is an evaluation value used to generate correlation data (learning model) described later, and is a value regarding the willingness of user U as a test user to the target object or a similar object.

[0046] In contrast, the "user evaluation value" is an evaluation value estimated by the evaluation value estimation unit 16 described later, and is a value regarding the motivation of user U as the person whose evaluation is estimated by the estimation system S (i.e., the target of marketing) toward the target object.

[0047] Here, the term "target object" refers to the object itself that is being evaluated (i.e., the user evaluation value is being estimated), and the term "similar object" refers to an object that has the same or similar attributes as the target object.

[0048] In this embodiment, the target thing is a dish. Therefore, similar things include dishes that have the same or similar taste as the target thing in at least one of the following: taste, aroma / flavor, appearance, texture, etc.; fragrances with the same or similar aroma / flavor as the target thing; photos of the target thing or similar dishes; dishes with the same or similar texture as the target thing; dishes that make the same or similar sound as the target thing when eaten or cooked; etc. In this embodiment, attributes of the target thing or similar things may include taste, aroma / flavor, appearance, texture, etc.

[0049] Generally, evaluations of dishes (and therefore foods) are based on multiple criteria, such as taste (gustation), aroma / flavor (olfaction), appearance (sight), texture (touch), and the sound made when eating or cooking (hearing), and the output evaluations tend to be complicated. Even for such items that tend to be evaluated in a complicated manner, the evaluation value estimation system of the present invention can be used to easily and accurately evaluate them.

[0050] However, the evaluation value estimation system of the present invention is not limited to such a configuration, and the estimated evaluation value cannot necessarily be used only to estimate an attractiveness value based on the user evaluation value for a dish related to the user.

[0051] Therefore, for example, the evaluation value estimation system of the present invention may estimate user evaluation values ​​from a large number of users for new products, whether they are food or non-food items (such as clothing), and refer to the user evaluation values ​​themselves for use in marketing the new products.

[0052] [System Overview] The overview of the configuration of the estimation system S will be described below with reference to FIGS. 1 and 2. FIG.

[0053] As shown in FIG. 1, the estimation system S is a computer system, and is configured by a server 1 owned by a provider of the service provided by the estimation system S.

[0054] The server 1 is configured to be able to communicate information with a user terminal 2 such as a smartphone or tablet used by a user U and a client terminal 3 such as a personal computer used by a client via the Internet network, public lines, etc.

[0055] It should be noted that the evaluation value estimation system of the present invention is not limited to being composed of a single server, but rather any of the terminals constituting the evaluation value estimation system may be configured to have the processing unit described below.

[0056] Therefore, for example, the rating value estimation system may be configured by a plurality of servers. Also, for example, at least one of the user terminal and the client terminal may be equipped with at least one of the processing units or at least part of the functions of that processing unit, and the rating value estimation system may be configured by cooperation between at least one of the user terminal and the client terminal and the server, or by only at least one of the user terminal and the client terminal.

[0057] Also, for example, functions corresponding to the user-side input unit 20 and user-side output unit 21 (see FIG. 2) provided in the user terminal 2 in this embodiment may be provided in a terminal having a processing unit, and configured as an independent rating value estimation device. Furthermore, the rating value estimation device may be provided with functions corresponding to the client-terminal-side input unit (not shown) and client-side output unit 30 (see FIG. 2) provided in the client terminal 3.

[0058] As shown in Figure 2, the user terminal 2 that can communicate with the server 1, which is the estimation system S, has a user-side input unit 20 and a user-side output unit 21 as functions (processing units) realized by at least one of the implemented hardware configuration and the program.

[0059] In this embodiment, the user input unit 20 and the user output unit 21 of the user terminal 2 are assumed to be touch panels (see FIG. 3, etc.).

[0060] However, the user terminal is not limited to such a configuration, and may be any terminal that can accept information input from a user and output information to present to the user. Therefore, for example, the user terminal may be configured to enable input and output using a keyboard, microphone, camera, speaker, etc. in addition to a touch panel.

[0061] In addition, the client terminal 3 capable of communicating with the server 1, which is the estimation system S, is equipped with a client-side input unit (not shown) and a client-side output unit 30 as functions (processing units) realized by at least one of the implemented hardware configuration and the program.

[0062] In this embodiment, the client-side input unit of the client terminal 3 is a keyboard and a touchpad, and the client-side output unit 30 is a monitor (see FIG. 10).

[0063] However, the client terminal is not limited to such a configuration, and may be any terminal capable of outputting information to present to the client. For example, the client terminal may be configured to be capable of outputting information using a speaker or the like.

[0064] Note that the user terminal and the client terminal are not limited to those configured with objects in real space, but may be configured with objects in virtual space, or may be configured with a combination of objects in real space and objects in virtual space. When configured using objects in virtual space, input and output are performed in the virtual space or a space that combines virtual space and real space.

[0065] [Configuration of Each Processing Unit] Next, the processing units that make up the estimation system S will be described with reference to FIGS.

[0066] As shown in Figure 2, the server 1 has the following functions (processing units) realized by at least one of the implemented hardware configuration and program: an attribute recognition unit 10, a target object presentation unit 11, an emotion recognition unit 12, a reference evaluation value recognition unit 13, a correlation data generation unit 14, a correlation data storage unit 15, an evaluation value estimation unit 16, and an attractiveness value estimation unit 17.

[0067] As shown in Figure 3, the attribute recognition unit 10 presents a questionnaire to the user U via the user terminal 2 in a format that can be answered to confirm the attributes of the user U, and recognizes the attributes of the user U based on the answers to the questionnaire.

[0068] Here, the "attributes" of a user refer to information that directly or indirectly influences the user's evaluation of an object. Examples include age, gender, whether or not the user has any food allergies, physical information such as height and weight, and place of origin (i.e., cultural background). The estimation system S uses age, gender, place of origin, and food allergies as attributes.

[0069] The attribute recognition unit 10 may be configured to recognize the attributes of the user U without referring to the questionnaire and its responses, but by referring to information previously obtained about the user U. Here, the previously obtained information may be, for example, information entered by the user U in a service provided by another system provided by the provider of the service provided by the estimation system S.

[0070] The target object presentation unit 11 presents, via the user terminal 2, information about the target object for which the evaluation value of the user U is to be recognized. Specifically, in this embodiment, the target object is a dish, and so the target object presentation unit 11 presents an image of the dish as well as information about places where the dish can be eaten.

[0071] Although the estimation system S presents information about the target object in this manner, in the present invention, the method of presenting the target object to the user may be changed as appropriate as long as the method allows the user to recognize the target object.

[0072] Therefore, for example, as described later in this embodiment, the target object itself (more specifically, the dish itself) may be presented to the test user, and only an image of the target object may be presented to the user whose attractiveness value is to be estimated.

[0073] However, in the evaluation value estimation system of the present invention, it is preferable to present the same information to both the test user and the user whose attractiveness value is to be estimated, and it is also preferable to present the information in the same environment (for example, whether or not the target object itself is presented). Also, in the evaluation value estimation system of the present invention, if the target object is not presented on the system (for example, if only the target object itself is presented without presenting any information), the target object presentation unit may be omitted.

[0074] The emotion recognition unit 12 recognizes the emotion of the user U when using the estimation system S. The emotion recognition unit 12 includes an emotion model presentation unit 12a and a coordinate recognition unit 12b (selected coordinate recognition unit).

[0075] As shown in Fig. 4 , the emotion model presenting unit 12a presents to the user U, via the user terminal 2, an emotion model that is a planar model including a first axis indicating valence intensity and a second axis indicating arousal level, in a format that allows the user U to select any coordinate in the emotion model. The estimation system S employs EmojiGrid (registered trademark) shown in Fig. 5A as the emotion model to be presented.

[0076] Here, "valence" is a bipolar concept that indicates the qualitative difference in the emotion evoked (for example, positive valence such as attraction (pleasantness) toward the target object, or negative valence such as aversion (unpleasantness)). Furthermore, "valence intensity" refers to the strength of the emotion felt (Harabe Yoshihiro (2015). Measuring the intensity of general valence concepts using paired comparisons: A preliminary study using unpleasant emotions. Tezukayama Gakuin University Faculty of Human Sciences Annual Research Report, (17)).

[0077] Additionally, "arousal" here is a concept with two extremes: excitement (high arousal) and calmness (low arousal), and indicates the intensity of the emotion evoked (Bradley, MM, Greenwald, MK, Petry, MC, & Lang, PJ (1992). Remembering Pictures: Pleasure and Arousal in Memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 379-390.).

[0078] The emotion model in the present invention is not limited to EmojiGrid, but may be any two-dimensional or three-dimensional model including a first axis indicating emotional valence intensity and a second axis indicating arousal level.

[0079] For this reason, for example, the Russell emotional cycle model may be adopted as the emotional model, as in the modified example shown in Fig. 5B. Furthermore, a model that uses the emotional cycle model and arranges colors or pictograms instead of letters may also be adopted.

[0080] In addition, in the present embodiment and the modified examples, a case has been described in which a planar model is used as the emotion model. However, the emotion model of the present invention is not limited to such a configuration, and may be a three-dimensional model of such a planar model.

[0081] In addition, when the emotion model is a three-dimensional model, the third axis may be, for example, an axis indicating the degree of arousal (Sato Toshio and Horigome Kazuya (2018). Examination of emotion scales using a three-dimensional model. The 82nd Annual Meeting of the Japanese Psychological Association, 721).

[0082] The coordinate recognition unit 12b recognizes coordinates (selected coordinates or reference coordinates) in an emotion model selected by the user U via the user terminal 2. In the estimation system S, the coordinate recognition unit 12b recognizes coordinates designated by the user U by touching an emotion model displayed on a touch panel that is the user-side input unit 20 and user-side output unit 21 of the user terminal 2.

[0083] Then, the coordinate recognition unit 12b recognizes the recognized coordinates themselves as the emotion of the user U at that time. That is, the coordinate recognition unit 12b does not particularly identify the emotion based on the coordinates.

[0084] As described above, in the estimation system S, the emotion recognition unit 12 employs an emotion model, which is a planar model including at least a first axis indicating emotional valence intensity and a second axis indicating arousal level, as an interface for the user U to input emotions.

[0085] As a result, in this estimation system S, the user U can easily and accurately express his / her own emotions as clear values ​​called coordinates, and as a result, data on the emotions of the user U can be easily and accurately collected.

[0086] However, the emotion recognition unit in the evaluation value estimation system of the present invention is not limited to such a configuration, and may be any unit that recognizes the emotional valence intensity and arousal level of the user toward the target object.

[0087] Therefore, for example, the emotion recognition unit may recognize at least one of the emotional valence intensity and the arousal level as a numerical value. Furthermore, the recognition method does not necessarily have to be based on an input by the user, and may be based on estimation or calculation from the user's facial expression, movement, etc.

[0088] The reference evaluation value recognition unit 13 recognizes the reference evaluation value for the object presented when the user U uses the estimation system S.

[0089] Specifically, as shown in Figure 6, the reference evaluation value recognition unit 13 presents a questionnaire set up for each specified evaluation criterion to the user U in a format that the user can answer, via the user terminal 2, and recognizes the answers as reference evaluation values.

[0090] Here, the "evaluation criterion" is a criterion for evaluation regarding willingness to purchase a target item or a similar item. Therefore, the "evaluation value" in this case is a value in the evaluation criterion. Note that examples of evaluation criteria that can be used include past consumption frequency (e.g., Frequency of past consumption), recommendation level to others (e.g., Net promoter score), willingness to consume (e.g., Willingness to take home), and willingness to buy (e.g., Purchase Probability Score, Willingness to buy, the Juster scale).

[0091] The estimation system S is configured to ask the user U for at least one of a value regarding past consumption frequency, a value regarding the degree of recommendation to others, a value regarding willingness to consume, and a value regarding willingness to purchase for each of the specified evaluation items (in this embodiment, taste, aroma / flavor, and appearance).

[0092] Here, "evaluation items" are items based on the five human senses for the target object. For example, if the target object is food, the evaluation items would be the taste (gustation), aroma / flavor (olfaction), appearance (sight), texture (touch), and sound (hearing) of the food.

[0093] The correlation data generation unit 14 generates correlation data (learning model) based on the emotions (reference coordinates) of user U as a test user recognized by the emotion recognition unit 12 and the reference evaluation value of user U recognized by the reference evaluation value recognition unit 13.

[0094] In the estimation system S, correlation data is generated as two-dimensional data indicating the correlation between coordinates selected from the emotion model and specific values ​​of the evaluation values, as shown in Fig. 7. This correlation data is generated for each of a plurality of evaluation criteria for each of predetermined evaluation items (in this embodiment, taste, aroma / flavor, and appearance) for each user attribute.

[0095] The correlation data generated in this manner takes as input the selected coordinates selected from the emotion model by the user U, who is the user whose attractiveness value is to be estimated, and outputs a user evaluation value.

[0096] It should be noted that the learning model in the evaluation value estimation system of the present invention is not limited to such correlation data, but may be any model that takes valence intensity and arousal level as input and outputs a user evaluation value.

[0097] Therefore, for example, the learning model may be a prediction model that is generated by machine learning using, as training data, coordinates selected from the emotion model and reference evaluation values ​​that are values ​​regarding motivation for the target object or similar objects, and that outputs evaluation values ​​from input coordinates. Furthermore, the input is not limited to coordinates, and may be the values ​​of emotional valence intensity and arousal level themselves.

[0098] Furthermore, although not adopted in the estimation system S, when a process for estimating a user evaluation value for a target object of a user as a marketing target is performed as described below, the user evaluation value estimated in the process may be used to update the generated learning model.

[0099] Furthermore, in the correlation data (learning model) of this embodiment, values ​​based on multiple evaluation criteria are output for each of multiple evaluation items for one input data (coordinates in this embodiment, and ultimately valence intensity and arousal level). However, the learning model in an evaluation value estimation system or the like is not limited to this configuration, and may be any model that takes valence intensity and arousal level as input and outputs a user evaluation value.

[0100] Therefore, for example, the learning model may output one value for one input data, may output one value for multiple input data, or may output multiple values ​​for multiple input data.

[0101] The correlation data storage unit 15 classifies and stores the correlation data generated by the correlation data generation unit 14 according to the attributes of the user U. Specifically, the correlation data storage unit 15 stores the correlation data in association with the attributes of the user U who input the reference coordinates and the reference evaluation value when generating the correlation data.

[0102] The evaluation value estimation unit 16 estimates a user evaluation value of the target object by the user U based on the coordinates selected from the emotion model and the correlation data acquired from the correlation data storage unit 15 .

[0103] In the estimation system S, the evaluation value estimation unit 16 estimates, as user evaluation values, values ​​for past consumption frequency, values ​​for the degree of recommendation to others, values ​​for the willingness to consume, and values ​​for the willingness to purchase for each of the specified evaluation items (in this embodiment, taste, aroma / flavor, and appearance).

[0104] This is because by recognizing values ​​for multiple evaluation criteria for each of multiple evaluation items related to a single target object, it becomes possible to make a multifaceted evaluation, and the analysis results using the user evaluation values ​​will be highly objective and accurate.

[0105] However, the rating value estimation system of the present invention is not limited to such a configuration, and therefore, for example, the rating value estimation system may recognize only one rating item or only one rating criterion.

[0106] The evaluation value estimation unit 16 acquires and recognizes correlation data (learning model) corresponding to the attributes of the user U from the correlation data storage unit 15, and estimates the user evaluation value using the correlation data. This is because the accuracy of estimating the user evaluation value is further improved when the attributes of the users are identical or similar.

[0107] However, the rating value estimation system of the present invention is not limited to such a configuration, and may be configured to estimate a user rating value using a learning model. Therefore, for example, the rating value estimation system may be configured to use the same learning model for different users when estimating a user rating value, without changing the learning model depending on the user's attributes.

[0108] The attractiveness value estimation unit 17 estimates an attractiveness value based on the user evaluation value estimated by the evaluation value estimation unit 16 .

[0109] Specifically, first, the attractiveness value estimation unit 17 weights each value (such as a value for past consumption frequency) in multiple evaluation criteria included in each of the values ​​for multiple evaluation items (in this embodiment, taste, aroma / flavor, appearance, and texture) estimated by the evaluation value estimation unit 16 based on a predetermined rule in accordance with the attributes of user U (for example, different weighting for each evaluation criterion), and estimates the sum of each of these weighted values ​​as the attractiveness value for each of the multiple evaluation items for the target object (food in this embodiment).

[0110] The attractiveness value estimation unit 17 then further weights the attractiveness values ​​for each of the multiple evaluation items estimated in this manner based on predetermined rules (for example, different weighting for each evaluation item) in accordance with the attributes of user U, and estimates the sum of these weighted values ​​as the attractiveness value for the target object itself.

[0111] This is because which evaluation criteria or which evaluation items should be emphasized may differ depending on the object for which the user evaluation value is to be estimated and the purpose for which the user evaluation value is to be used.

[0112] Therefore, when summing user evaluation values ​​to output a single attractiveness value, weighting the values ​​according to the intended use before calculating the attractiveness value can make the output attractiveness value, and ultimately the analysis results using the attractiveness value, more accurate. This also applies when a predictive model is used as the learning model instead of correlation data.

[0113] However, the evaluation value estimation system of the present invention is not limited to such a configuration, and the method for calculating the attractiveness value may be set as appropriate.

[0114] Therefore, for example, weighting may be performed according to the type of target object, or according to rules based on other criteria, rather than the user's attributes as in the present embodiment, or a combination of multiple rules. Furthermore, weighting may not be performed, and the attractiveness value may simply be the sum of the user evaluation values. Furthermore, for example, weighting may be performed only for a predetermined evaluation criterion, or only for a predetermined evaluation item.

[0115] Furthermore, instead of adding up multiple user evaluation values, one user evaluation value that should be emphasized based on the type of target object may be determined, and that value may be used as the attractiveness value. Alternatively, one or multiple user evaluation values ​​may be presented as is without estimating the attractiveness value, and in such cases, the attractiveness value estimation unit may be omitted.

[0116] The weighting rules may be, for example, rules established based on the results of a survey conducted to determine which evaluation items are emphasized based on at least one of the user's attributes and the type of target object. Alternatively, rules may be established based on the results of a survey conducted to determine which evaluation criteria are emphasized based on at least one of the user's attributes and the type of target object. Furthermore, a combination of these rules may be used.

[0117] [Processing performed by each processing unit when generating a learning model] Next, with reference to Figures 2 to 4, 6, and 8, we will explain the processing performed by each processing unit when the estimation system S generates correlation data, which is a learning model.

[0118] The user U in this process is a test user who evaluates the target subject in advance in order to collect data for generating correlation data, which is a learning model.

[0119] As will be described below, the estimation system S presents the target object itself to the user U, who is a test user, and performs processing to generate correlation data, which is a learning model. However, similar objects may be presented instead of the target object, and processing may be performed. Furthermore, if there are multiple test users, the target object may be presented to some of the multiple test users, and similar objects may be presented to other test users, and processing may be performed.

[0120] In this process, first, the attribute recognition unit 10 presents a questionnaire to the user U via the user terminal 2 in a format that the user can answer in order to confirm the attributes of the user U, as shown in FIG. 3 (FIG. 8 / STEP 100).

[0121] Next, the attribute recognition unit 10 recognizes the attributes of the user U based on the answers to the questionnaire (FIG. 8 / STEP 101).

[0122] Next, the target object presentation unit 11 presents information (image, name, ingredients, etc.) about the target object, ie, the food, to the user U via the user terminal 2 (FIG. 8 / STEP 102).

[0123] In this process in the present embodiment, information about the target object, that is, the food, is presented, and the food itself is provided to the user U.

[0124] Next, the emotion model presentation unit 12a of the emotion recognition unit 12 presents the emotion model, which is a planar model including a first axis indicating emotional valence intensity and a second axis indicating arousal level, to the user U via the user terminal 2 in a format in which any of the coordinates in the emotion model can be selected (FIG. 8 / STEP 103).

[0125] Next, the coordinate recognition unit 12b of the emotion recognition unit 12 recognizes the coordinates selected from the emotion model by the user U via the user terminal 2 as reference coordinates (FIG. 8 / STEP 104).

[0126] 4, first, the emotion model presenting unit 12a displays an emotion model (EmojiGrid) on the touch panel of the user terminal 2. Thereafter, when the user U touches any point on this emotion model, the coordinate recognizing unit 12b determines that the coordinates of that point have been selected, and recognizes the coordinates as reference coordinates.

[0127] Next, the reference evaluation value recognition unit 13 presents a questionnaire regarding the reference evaluation value by the user U to the user U in a format that the user can answer, via the user terminal 2 (FIG. 8 / STEP 105).

[0128] Next, the reference evaluation value recognition unit 13 recognizes the reference evaluation value based on the answers to the questionnaire (FIG. 8 / STEP 106).

[0129] 6 , first, the reference evaluation value recognition unit 13 displays a plurality of questions for inquiring about values ​​regarding past consumption frequency, etc., on the touch panel of the user terminal 2. Thereafter, when the user U selects one of the answers to the questions, the reference evaluation value recognition unit 13 recognizes the selected answer as a reference evaluation value in the evaluation criteria corresponding to that question.

[0130] In the estimation system S, the questionnaire conducted by the reference evaluation value recognition unit 13 is conducted sequentially for each of the predetermined evaluation items (in this embodiment, taste, aroma / flavor, appearance, and texture).

[0131] Next, the correlation data generation unit 14 generates correlation data (learning model) based on the emotion (reference coordinates) of user U recognized by the emotion recognition unit 12 and the reference evaluation value of user U recognized by the reference evaluation value recognition unit 13 (Figure 8 / STEP 107).

[0132] In the estimation system S, this correlation data is generated for each of the predetermined evaluation items (in this embodiment, taste, aroma / flavor, appearance, and texture) in the same number as the number of types of reference evaluation values ​​(i.e., evaluation criteria).

[0133] Next, the correlation data generating unit 14 classifies the generated correlation data according to the attributes recognized by the attribute recognizing unit 10, and stores the data in the correlation data storage unit 15 (FIG. 8 / STEP 108), thereby terminating this processing.

[0134] In this embodiment, information (image, name, ingredients, etc.) about the target object, which is a dish, is presented to user U, the test user, and the above processing is performed when user U eats the dish itself.

[0135] However, the process for generating a learning model in the rating value estimation system of the present invention does not necessarily have to be performed at such a timing, and may be performed before the process for estimating a user rating value, which will be described later. Therefore, for example, the process for generating the learning model may be performed when the target object or a similar object is presented, or when the target object or a similar object is acquired (e.g., when purchased).

[0136] Furthermore, the process of recognizing the attributes of user U (STEP 100 and STEP 101) may be performed at any time before the process of storing correlation data (STEP 108). Therefore, for example, it may be performed in advance independently of the series of processes, or it may be performed after the correlation data is generated (after STEP 107).

[0137] In addition, in the present embodiment, the case where the learning model is correlation data showing the correlation between the reference coordinates and the reference evaluation value by the test user, user U, has been described. However, after performing the same processes as in STEP 102 to STEP 106, machine learning may be performed using the reference coordinates obtained by the processes and the reference evaluation value by the test user, user U, as training data to generate a prediction model, and the prediction model may be used as the learning model.

[0138] In this embodiment, the estimation system S generates a learning model by performing the processes described above in steps 100 to 108. However, the rating value estimation system of the present invention is not limited to this configuration, and may be configured to use a separately generated learning model.

[0139] [Processing performed by each processing unit when estimating user evaluation value and attractiveness value] Next, with reference to Figures 2 to 4, 9, and 10, we will explain the processing performed by each processing unit when the estimation system S estimates user evaluation value and attractiveness value.

[0140] In this process, user U is the target of marketing for the target thing. The attractiveness value estimated for user U is presented to the client conducting the marketing via client terminal 3. Note that in the following process, processing may be performed using a similar thing instead of the target thing, and the processing results may be used to estimate at least one of the user evaluation value and attractiveness value of the target thing.

[0141] In this process, first, the attribute recognition unit 10 presents a questionnaire to the user U via the user terminal 2 in a format that the user can answer in order to confirm the attributes of the user U, as shown in FIG. 3 (FIG. 9 / STEP 200).

[0142] Next, the attribute recognition unit 10 recognizes the attributes of the user U based on the answers to the questionnaire (FIG. 9 / STEP 201).

[0143] Next, the target object presentation unit 11 presents information (image, name, ingredients, etc.) about the target object, ie, the food, to the user U via the user terminal 2 (FIG. 9 / STEP 202).

[0144] In this process in the present embodiment, only information about the target object, the food, is presented, and the food itself is not provided to user U. However, when the correlation data (learning model) used in this process is generated, it is preferable to present it to user U, the target of marketing, in the same way as it was presented to the test user.

[0145] Next, the emotion model presentation unit 12a of the emotion recognition unit 12 presents the emotion model, which is a planar model including a first axis indicating emotional valence intensity and a second axis indicating arousal level, to the user U via the user terminal 2 in a format in which any of the coordinates in the emotion model can be selected (FIG. 9 / STEP 203).

[0146] Next, the coordinate recognition unit 12b of the emotion recognition unit 12 recognizes the coordinates selected from the emotion model by the user U via the user terminal 2 as selected coordinates (FIG. 9 / STEP 204).

[0147] 4, first, the emotion model presenting unit 12a displays an emotion model (EmojiGrid) on the touch panel of the user terminal 2. Thereafter, when the user U touches any point on this emotion model, the coordinate recognizing unit 12b determines that the coordinates of that point have been selected, and recognizes the coordinates as selected coordinates.

[0148] Next, the evaluation value estimation unit 16 acquires correlation data corresponding to the attribute recognized by the attribute recognition unit 10 from the correlation data storage unit 15 and recognizes it (FIG. 9 / STEP 205).

[0149] Specifically, the evaluation value estimation unit 16 acquires and recognizes the correlation data from among the multiple correlation data stored in the correlation data storage unit 15, which has at least one attribute that matches or is similar to the attribute recognized by the attribute recognition unit 10 and has the largest number of matching or similar items.

[0150] Next, the evaluation value estimation unit 16 recognizes a user evaluation value based on the recognized correlation data and the emotion (selected coordinates) of the user U recognized by the emotion recognition unit 12 (FIG. 9 / STEP 206).

[0151] Specifically, the evaluation value estimation unit 16 takes the recognized selected coordinates as input and outputs values ​​for each of a plurality of evaluation criteria for each of the predetermined evaluation items (in this embodiment, taste, aroma / flavor, appearance, and texture) from the recognized correlation data.

[0152] Next, the evaluation value estimation unit 16 weights each of the recognized user evaluation values ​​based on the attributes recognized by the attribute recognition unit 10 (FIG. 9 / STEP 207).

[0153] Specifically, for example, the evaluation value estimation unit 16 recognizes a weighting rule that is predetermined for each attribute, and weights each of the values ​​in the multiple evaluation criteria included in each user evaluation value in accordance with that rule.

[0154] Next, the attractiveness value estimation unit 17 calculates an attractiveness value based on the weighted user evaluation value (FIG. 9 / STEP 208).

[0155] Specifically, the attractiveness value estimation unit 17 adds up the values ​​of multiple evaluation criteria included in the user evaluation value for each evaluation item, and estimates the attractiveness value for each evaluation item for the target thing.

[0156] In addition, the attractiveness value estimation unit 17 weights the attractiveness values ​​for each of the estimated multiple evaluation items according to the weighting rules acquired in STEP 207, and then adds up the values ​​to estimate the attractiveness value of the target object itself.

[0157] Next, as shown in Figure 10, the attractiveness value estimation unit 17 presents the calculated attractiveness value to the client via the client terminal 3 as the attractiveness value of the user U for the target object (Figure 9 / STEP 209), and then terminates this processing.

[0158] As described above, the estimation system S and the rating value estimation method executed using the estimation system S employ universal parameters, namely, valence and arousal. Therefore, the user rating value output by inputting these parameters is also a highly versatile value that is not dependent on cultural background, etc. Furthermore, valence and arousal are parameters that can be easily acquired without using special equipment.

[0159] Furthermore, since these parameters are simplified parameters for expressing complex emotions, when analysis is performed using these parameters, the analysis process can be performed more easily and quantitatively than when using conventional parameters for expressing emotions (e.g., sentences written by the user).

[0160] Therefore, according to this system and method, data regarding the emotions of the user U can be collected quickly and easily in a highly versatile manner, and the collected data can be easily analyzed.

[0161] In the evaluation system S, the user evaluation value is a value for a plurality of evaluation criteria for each of a plurality of evaluation items, and the evaluation system S estimates all of these values.

[0162] However, the rating value estimation system of the present invention is not limited to such a configuration, and at least one of the multiple user rating values ​​may be recognized (specifically, estimated, acquired, etc.) using another system. Specifically, for example, if the target object is a dish, the rating value regarding taste may be estimated by the rating value estimation system, and the rating value regarding appearance may be recognized based on actual sales.

[0163] [Experimental Example] Next, an experiment for collecting actual data related to the evaluation criteria and the experimental data obtained by the experiment will be described with reference to FIGS.

[0164] This experiment corresponds to STEP 100 to STEP 106 of the above-mentioned [Processing Executed by Each Processing Unit When Generating a Learning Model] described with reference to Figures 7 and 8. Therefore, the experimental data obtained by this experiment is used to create correlation data showing the correlation between the reference coordinates and the reference evaluation value by user U, who is a test user (see Figure 8 / STEP 107 to STEP 108), and to use as training data for generating a prediction model.

[0165] The subjects of this experiment were 156 men and women aged 18 or older who belonged to a specific region (specifically, had nationality in a specific country). The gender ratio of the subjects was 54.5% male and 45.5% female. The average age of the subjects was 40.2 ± SD 9.5 years.

[0166] In this experiment, subjects were presented with six different food images (i.e., objects similar to the target dish) as shown in Figure 11, and then asked to enter their emotions (emotional valence intensity and arousal level) using EmojiGrid (see Figure 4) and to fill out a questionnaire regarding the evaluation criteria (see Figure 6). Note that Figure 11 is an illustration, and in the actual experiment, photographs of food were used.

[0167] In other words, the evaluation item used in this experiment was "appearance."The evaluation criteria used in this experiment were also the degree of willingness to purchase (Purchase Probability Score), the degree of willingness to consume (Willingness to take home), and the degree of recommendation to others (Net Promoter Score).

[0168] In the "Purchase Probability Score" shown in FIG. 12, the questionnaire used the question "Are you likely to purchase this food item?", and the evaluation value obtained from the questionnaire was an 11-point scale, with "I will definitely not buy it" being 0 points and "I will definitely buy it" being 10 points.

[0169] Furthermore, for the "Willingness to take home" shown in FIG. 13, the questionnaire included the question "How much do you want to take this food home?", and the evaluation value obtained from the questionnaire was on an 11-point scale, with "not at all" being 0 points and "very much so" being 10 points.

[0170] Furthermore, for the "Recommendability to others (Net Promoter Score)" shown in Figure 14, the questionnaire used the question "To what extent would you like to recommend this food to your close friends and family?", and the evaluation value obtained from the questionnaire was an 11-point scale, with "Not at all like that" being 0 points and "Strongly like that" being 10 points.

[0171] 12 to 14 show the collected experimental data. Of the data in each figure, the data in the upper left plots emotions (coordinates selected in EmojiGrid) input for all rating values. The data in the upper right plots emotions input for rating values ​​of 0 and 1. The data in the middle left plots emotions input for rating values ​​of 2, 3, and 4. The data in the middle right plots emotions input for rating values ​​of 5 and 6. The data in the lower left plots emotions input for rating values ​​of 7 and 8. The data in the lower right plots emotions input for rating values ​​of 9 and 10.

[0172] 12 to 14, it was found that there was a clear and significant correlation between the evaluation values ​​and emotions. Therefore, it was thought that a significant learning model could be generated by using this data.

[0173] As described above, the estimation system S employs an emotion model (EmojiGrid), which is a planar model including at least a first axis indicating emotional valence intensity and a second axis indicating arousal level, as an interface for the user U to input emotions. This allows the user U to easily and accurately express his or her own emotions as clear values ​​known as coordinates.

[0174] Furthermore, since the coordinate parameter is a parameter that can be intuitively understood by the user U, there is no need to provide the user with a special explanation of how to express the parameter when collecting data. This allows the estimation system S to quickly collect emotion-related parameters that are prone to change before those changes occur.

[0175] Furthermore, since coordinates are simple and clear parameters, analysis can be performed even more easily by using such coordinates as parameters.

[0176] Other Embodiments Although the illustrated embodiments have been described above, the present invention is not limited to such embodiments.

[0177] For example, in the above embodiment, the estimation system S is a single computer system. However, the present invention also includes a rating value estimation program for causing any one or more computers to execute the above-described rating value estimation method, and a recording medium on which the program is recorded and which is readable by a computer used by a user or the like.

[0178] In the above embodiment, the attractiveness value of the target object of user U as the target of marketing is presented to a client who is conducting marketing, etc. However, the evaluation value estimation system of the present invention is not limited to this configuration.

[0179] Therefore, for example, the user evaluation value itself estimated using the evaluation value estimation system of the present invention, or the user evaluation value and attractiveness value, may be presented to the user himself / herself. Furthermore, at least one of the estimated user evaluation value and attractiveness value may be used as is as data for machine learning in marketing, etc., rather than being presented to someone.

[0180] 1...server, 2...user terminal, 3...client terminal, 10...attribute recognition unit, 11...target object presentation unit, 12...emotion recognition unit, 12a...emotion model presentation unit, 12b...coordinate recognition unit (selected coordinate recognition unit), 13...reference evaluation value recognition unit, 14...correlation data generation unit, 15...correlation data storage unit, 16...evaluation value estimation unit, 17...attractiveness value estimation unit, 20...user side input unit, 21...user side output unit, 30...client side output unit, S...estimation system (evaluation value estimation system), U...user

Claims

1. An evaluation value estimation system for estimating a user evaluation value, which is a value regarding a user's motivation for a target object, comprising: An emotion recognition unit that recognizes the intensity of emotional value and arousal level of the target object by the user; An evaluation value estimation unit that estimates the user evaluation value using a learning model that takes the intensity of emotional value and the arousal level by the user as inputs and outputs the user evaluation value; The user evaluation value includes values for a plurality of evaluation items; The learning model takes one of the intensity of emotional value and the arousal level as an input and outputs values for the plurality of evaluation items, characterized by the evaluation value estimation system.

2. In the evaluation value estimation system according to Claim 1, A charm value estimation unit that weights each of the values for the plurality of evaluation items output from the learning model by inputting one of the intensity of emotional value and the arousal level, and estimates the sum of the weighted values as a charm value, characterized by the evaluation value estimation system.

3. In the evaluation value estimation system according to Claim 1, The target object is food; The plurality of evaluation items include at least one of an evaluation item based on taste, an evaluation item based on smell, an evaluation item based on vision, and an evaluation item based on touch, characterized by the evaluation value estimation system.

4. An evaluation value estimation system for estimating a user evaluation value, which is a value regarding a user's motivation for a target object, comprising: An emotion recognition unit that recognizes the intensity of emotional value and arousal level of the target object by the user; An evaluation value estimation unit that estimates the user evaluation value using a learning model that takes the intensity of emotional value and the arousal level by the user as inputs and outputs the user evaluation value; The user evaluation value includes values for a plurality of evaluation criteria; The learning model takes one of the intensity of emotional value and the arousal level as an input and outputs values based on the plurality of evaluation criteria, characterized by the evaluation value estimation system.

5. In the evaluation value estimation system according to Claim 4, A charm value estimation unit that weights each of the values for the plurality of evaluation criteria output from the learning model by inputting one of the intensity of emotional value and the arousal level, and recognizes the sum of the weighted values as a charm value, characterized by the evaluation value estimation system. **Claim 6**: In the evaluation value estimation system according to claim 4, the values for the plurality of evaluation criteria include at least one of the values for past consumption frequency, the value for the degree of recommendation to others, the value for the degree of consumption desire, and the value for the degree of purchase desire. An evaluation value estimation system characterized by this. **Claim 7**: In the evaluation value estimation system according to claim 1 or claim 4, the emotion recognition unit includes an emotion model which is a planar model or a three-dimensional model including at least a first axis indicating the intensity of emotional value and a second axis indicating the arousal level, and presents the emotion model to the user in a form where any coordinate in the emotion model can be selected. It has an emotion model presentation unit and a selected coordinate recognition unit that recognizes the selected coordinates which are the coordinates selected from the emotion model when the user is presented with the target thing. The evaluation value estimation unit takes the selected coordinates as an input. An evaluation value estimation system characterized by this. **Claim 8**: In the evaluation value estimation system according to claim 7, the learning model is correlation data showing the correlation between a reference coordinate which is a coordinate selected from the emotion model and a reference evaluation value which is a value for the desire for the target thing or a similar thing having the same or similar attributes as the target thing. An evaluation value estimation system characterized by this. **Claim 9**: In the evaluation value estimation system according to claim 8, the reference coordinate is the coordinate selected from the emotion model by a test user who preliminarily evaluates the target thing or the similar thing with respect to the target thing or the similar thing. The correlation data is data showing the correlation between the reference coordinate and the reference evaluation value by the test user. An evaluation value estimation system characterized by this. **Claim 10**: In the evaluation value estimation system according to claim 7, a prediction model that is generated by machine learning using, as teacher data, a reference coordinate which is a coordinate selected from the emotion model and a reference evaluation value which is a value for the desire for the target thing or a similar thing having the same or similar attributes as the target thing, and outputs the user evaluation value from the input selected coordinates. An evaluation value estimation system characterized by this. **Claim 11**: In the evaluation value estimation system according to claim 10, the reference coordinate is the coordinate selected from the emotion model by a test user who preliminarily evaluates the target thing or the similar thing with respect to the target thing or the similar thing. The evaluation value estimation system is characterized in that the prediction model is generated by machine learning using the reference coordinates and the reference evaluation value by the test user as teacher data.

12. An evaluation value estimation device for estimating a user evaluation value, which is a value regarding a user's motivation for a target thing, a sentiment recognition unit that recognizes the sentiment value intensity and arousal level of the target thing by the user, and an evaluation value estimation unit that estimates the user evaluation value using a learning model that takes the sentiment value intensity and the arousal level by the user as inputs and outputs the user evaluation value. The user evaluation value includes values for a plurality of evaluation items. The learning model is characterized in that it takes one of the sentiment value intensity and the arousal level as an input and outputs values for the plurality of evaluation items.

13. An evaluation value estimation device for estimating a user evaluation value, which is a value regarding a user's motivation for a target thing, a sentiment recognition unit that recognizes the sentiment value intensity and arousal level of the target thing by the user, and an evaluation value estimation unit that estimates the user evaluation value using a learning model that takes the sentiment value intensity and the arousal level by the user as inputs and outputs the user evaluation value. The user evaluation value includes values for a plurality of evaluation criteria. The learning model is characterized in that it takes one of the sentiment value intensity and the arousal level as an input and outputs values based on the plurality of evaluation criteria.

14. An evaluation value estimation method for estimating a user evaluation value, which is a value regarding a user's motivation for a target thing, a step in which a sentiment recognition unit recognizes the sentiment value intensity and arousal level of the target thing by the user, and a step in which an evaluation value estimation unit estimates the user evaluation value using a learning model that takes the sentiment value intensity and the arousal level by the user as inputs and outputs the user evaluation value. The user evaluation value includes values for a plurality of evaluation items. The learning model is characterized in that it takes one of the sentiment value intensity and the arousal level as an input and outputs values for the plurality of evaluation items.

15. An evaluation value estimation method for estimating a user evaluation value, which is a value regarding a user's motivation for a target thing, a step in which a sentiment recognition unit recognizes the sentiment value intensity and arousal level of the target thing by the user, The evaluation value estimation unit estimates the user evaluation value by using a learning model that takes the emotional value intensity and the arousal level input by the user as inputs and outputs the user evaluation value. The user evaluation value includes values for a plurality of evaluation criteria. The learning model is characterized in that it takes one of the emotional value intensity and the arousal level as an input and outputs values based on the plurality of evaluation criteria. An evaluation value estimation method.

16. In an evaluation value estimation program for causing a computer to execute an evaluation value estimation method for estimating a user evaluation value, which is a value regarding a user's intention toward a target object, the computer is caused to: a step in which an emotion recognition unit recognizes the emotional value intensity and the arousal level of the user toward the target object; a step in which an evaluation value estimation unit estimates the user evaluation value by using a learning model that takes the emotional value intensity and the arousal level input by the user as inputs and outputs the user evaluation value; The user evaluation value includes values for a plurality of evaluation items. The learning model is characterized in that it takes one of the emotional value intensity and the arousal level as an input and outputs values for the plurality of evaluation items. An evaluation value estimation program.

17. In an evaluation value estimation program for causing a computer to execute an evaluation value estimation method for estimating a user evaluation value, which is a value regarding a user's intention toward a target object, the computer is caused to: a step in which an emotion recognition unit recognizes the emotional value intensity and the arousal level of the user toward the target object; a step in which an evaluation value estimation unit estimates the user evaluation value by using a learning model that takes the emotional value intensity and the arousal level input by the user as inputs and outputs the user evaluation value; The user evaluation value includes values for a plurality of evaluation criteria. The learning model is characterized in that it takes one of the emotional value intensity and the arousal level as an input and outputs values based on the plurality of evaluation criteria. An evaluation value estimation program.

18. A recording medium that records the evaluation value estimation program according to Claim 16 or Claim 17, and is readable by the computer.