Reaction profile evaluation device, reaction profile evaluation method, and reaction profile evaluation program

Abstract

To provide a reaction profile evaluation device capable of quantitatively evaluating a reaction profile generated in rating scale value data.SOLUTION: A reaction profile evaluation device includes: a reception unit 51 for inputting rating scale value data 81 on a user with an answer tendency of the user as a reaction profile, and with the statistic distribution expressing the reaction profile of the user as a reaction profile distribution; a reaction profile distribution estimation unit 53 for estimating a plurality of reaction profile distribution candidates which are candidates of the reaction profile distribution, using a candidate reaction profile distribution template including types of the statistic distribution on the basis of the rating scale value data 81; and a reaction profile distribution evaluation unit 54 for selecting the reaction profile distribution suitable for the expression of the reaction profile of the user from the plurality of estimated reaction profile distribution candidates using the rating scale value data 81.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a response profile evaluation device, a response profile evaluation method, and a response profile evaluation program used when measuring and evaluating a user's subjective mental and physical state. [Background technology]

[0002] Psychological measurements using measurement scales are used as a means of easily and continuously acquiring subjective physical and mental states in daily life. For example, the Likert scale, which is an ordinal scale with multiple discrete responses, is commonly used as a measurement scale. In recent years, VAS measurements using the Visual Analogue Scale (VAS), which is an interval scale with continuous responses, as the measurement scale have been used for psychological measurements. When the VAS is used as the measurement scale, it is possible to obtain responses from subjects more easily and with a higher degree of granularity. The data of the subjects' (users') responses to the instruction sentences is called rating data.

[0003] In psychological measurements using a measurement scale, a consistent response tendency bias can occur for each subject that occurs regardless of the instruction text. This response tendency bias is included in the rating data and can have a negative effect on inter-subject analysis using psychological measurements in particular.

[0004] An example of a technique for removing such response tendency bias occurring in subjects is described in Patent Document 1. Patent Document 1 discloses a response style component removal device that targets response tendency biases that appear when evaluating subjective mental and physical states using a Likert scale and can remove response style (RS), which is a response tendency bias that is independent of the contents of a questionnaire. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent Publication No. 2021-026647 Summary of the Invention [Problem to be solved by the invention]

[0006] As mentioned above, in recent years, VAS measurement using VAS, which is a continuous response interval scale, as a measurement scale has been used. For this reason, even in VAS measurement, which uses a continuous measurement scale, there is a demand for technology that can quantitatively evaluate and remove response tendency bias. Hereinafter, response tendency bias is also called response profile (RP).

[0007] Patent Document 1 describes that the device described in Patent Document 1 can also be used when the measurement scale is a continuous value. However, since the nature of subjective evaluation responses on an interval scale (e.g., VAS) is different from that of subjective evaluation responses on an ordinal scale (e.g., Likert scale), it is generally difficult to apply a technique for removing a response profile when an ordinal scale is used, such as the technique described in Patent Document 1, to an interval scale such as VAS. In particular, VAS often includes different types of questions, such as unipolar VAS and bipolar VAS, and there is a problem that evaluation is difficult using a method based on a single mathematical model as in the conventional technique.

[0008] An object of the present invention is to provide a reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program capable of quantitatively evaluating a reaction profile occurring in evaluation value data. [Means for solving the problem]

[0009] The reaction profile evaluation device according to the present invention defines a user's response tendency as a reaction profile and a statistical distribution expressing the reaction profile of the user as a reaction profile distribution, and comprises a reception unit for inputting rating value data of the user, a reaction profile distribution estimation unit for estimating multiple reaction profile distribution candidates that are candidates for the reaction profile distribution based on the rating value data using a candidate reaction profile distribution template including a type of the statistical distribution, and a reaction profile distribution evaluation unit for selecting the reaction profile distribution suitable for expressing the reaction profile of the user from the multiple estimated reaction profile distribution candidates using the rating value data.

[0010] The reaction profile evaluation method according to the present invention is executed by a computer having a reception unit, a reaction profile distribution estimation unit, and a reaction profile distribution evaluation unit, and includes a step in which a user's response tendency is defined as a reaction profile, a statistical distribution representing the reaction profile of the user is defined as a reaction profile distribution, and the reception unit inputs rating value data of the user, a step in which the reaction profile distribution estimation unit estimates a plurality of reaction profile distribution candidates that are candidates for the reaction profile distribution based on the rating value data using a candidate reaction profile distribution template including a type of the statistical distribution, and a step in which the reaction profile distribution evaluation unit uses the rating value data to select the reaction profile distribution suitable for expressing the reaction profile of the user from the plurality of estimated reaction profile distribution candidates.

[0011] The reaction profile evaluation program according to the present invention is a program for causing a computer to realize the functions of the reception unit, the reaction profile distribution estimation unit, and the reaction profile distribution evaluation unit of the reaction profile evaluation apparatus according to the present invention. Effect of the Invention

[0012] According to the present invention, there is provided a reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program capable of quantitatively evaluating a reaction profile occurring in evaluation value data. [Brief description of the drawings]

[0013] [Figure 1] 1 is a block diagram showing an example of a main configuration of a reaction profile evaluation device according to a first embodiment of the present invention. [Diagram 2] 13 is a flowchart showing an example of a process in which the reaction profile evaluation device receives data from a user terminal. [Diagram 3] 1 is a flowchart showing an example of a reaction profile distribution (RP distribution) estimation process performed by the reaction profile evaluation device. [Figure 4] 13 is a flowchart showing an example of an evaluation process of an RP distribution performed by the reaction profile evaluation device. [Diagram 5] 13 shows experimental results of RP distribution estimation and RP distribution evaluation performed using the reaction profile evaluation device according to the present embodiment. [Figure 6] FIG. 13 is a diagram showing an example of RP distribution data of users displayed on a user terminal or an administrator terminal. [Figure 7] FIG. 13 is a diagram illustrating an example of the data structure of RP distribution data. [Figure 8] 11 is a flowchart showing an example of a series of processes for evaluating RP distribution with confidence intervals, which are performed by the reaction profile evaluation device in the second embodiment of the present invention. [Figure 9] FIG. 13 is a diagram showing an example of RP distribution data with confidence intervals of users, displayed on a user terminal or an administrator terminal. [Figure 10] FIG. 11 is a block diagram showing an example of a main configuration of a reaction profile evaluation device according to a third embodiment of the present invention. [Figure 11A] 13 is a flowchart showing an example of a learning process of a subjective state evaluation model performed by the reaction profile evaluation device. [Figure 11B] FIG. 11 is a diagram showing an example of a subjective state evaluation process performed by the response profile evaluation device. [Figure 12] FIG. 11 is a block diagram showing an example of the main configuration of a reaction profile evaluation device according to a fourth embodiment of the present invention. [Figure 13A] 13 is a flowchart showing an example of a process in which the reaction profile evaluation device receives data from a user terminal. [Figure 13B] 13 is a flowchart showing an example of a learning process of a biological state evaluation model performed by the reaction profile evaluation device. [Figure 13C] FIG. 2 is a diagram showing an example of a biological state evaluation process performed by the reaction profile evaluation device. [Figure 14A] 13 is a diagram showing an example of an evaluation result that the result display unit displays on the user terminal, and is an example of a display screen when requesting the user to input evaluation value data based on the evaluation result data. FIG. [Figure 14B] FIG. 13 is a diagram showing an example of an evaluation result displayed by the result display unit on the user terminal, and is an example of a display screen showing a plurality of examples of RP types A to D and the user's RP type to the user. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] According to the present invention, it is possible to quantitatively evaluate the response profile (user's response tendency) occurring in rating data, particularly rating data measured using a continuous response scale such as VAS. By obtaining the response profile, it is possible to eliminate the adverse effect of the response profile on inter-subject analysis. In the present invention, even for rating data measured using a continuous response scale such as VAS, which often includes questions with different measurement scales such as unipolar and bipolar scales, it is possible to quantitatively evaluate the response profile (RP) in a manner that allows comparison with other users. This leads to the realization of inter-subject analysis of psychological measurements that takes into account the effect of RP.

[0015] The reaction profile evaluation method according to the present invention can be executed by the reaction profile evaluation device according to the present invention. The reaction profile evaluation program according to the present invention is a program for causing a computer to realize the functions of each functional unit included in the reaction profile evaluation device according to the present invention.

[0016] Hereinafter, a reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program according to an embodiment of the present invention will be described with reference to the drawings. In the following description, a reaction profile, which is a response tendency of a user, may be referred to as "RP". EXAMPLES

[0017] A reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program according to a first embodiment of the present invention will be described.

[0018] 1 is a block diagram showing an example of the main configuration of a reaction profile evaluation device 1 according to this embodiment. The reaction profile evaluation device 1 according to this embodiment configures a reaction profile evaluation device system with one or more user terminals 7 and one or more administrator terminals 8.

[0019] The reaction profile evaluation device 1 is connected to a user terminal 7 and an administrator terminal 8 via a network 9. The reaction profile evaluation device 1 processes data received from the user terminal 7 and the administrator terminal 8, and transmits the processed data to the user terminal 7 and the administrator terminal 8.

[0020] The user terminal 7 is equipped with a biometric sensor 11 that measures the biological condition of the subject user, a biometric device 12 that controls the biometric sensor 11, an input / output device 13, a communication device 14, and a notification / alert device 15, and is used by the user.

[0021] The biometric sensor 11 includes a heart rate sensor 21 that detects the user's heart rate interval (RR Interval, RRI), an electrodermal activity sensor 22 that detects the user's sweat rate, and an acceleration sensor 23 that detects the user's movement. The heart rate sensor 21 may be, for example, a sensor that detects the heart rate based on an electrocardiogram, a pulse wave, a pressure change, or a heart sound.

[0022] The biometric sensor 11 is not limited to the above sensors, and may include sensors that detect body temperature, blinking, eye movement, electromyography, or brain waves, etc. The biometric sensor 11 may be a wearable device that can be worn by a user, or a device built into a smartphone that can be carried by a user.

[0023] The biomeasurement device 12 controls the biomeasurement sensor 11 and generates biomeasurement data from the biostatus measured by the biomeasurement sensor 11. The biomeasurement data is data on the biostatus of the user measured by the biomeasurement sensor 11, and is bioinformation of the user. The biomeasurement device 12 performs calculation processing and compression processing on the biostatus as necessary to generate the biomeasurement data.

[0024] The input / output device 13 is equipped with input devices such as a mouse, keyboard, touch panel, and microphone, and output devices such as a display and speaker, and is a device that performs input / output to and from the user terminal 7. The input device receives input from the user, for example, inputs rating value data 81 measured using a measurement scale to evaluate the subjective mental and physical state of the user. The output device displays on the display, for example, displays data acquired by the user terminal 7, and displays notifications to the user. The user terminal 7 displays on the display.

[0025] In the following, a measurement scale used to measure the rating value data 81 is mainly a continuous response type measurement scale such as a visual analogue scale (VAS). In a continuous response type measurement scale, the rating value data 81 is measured as an arbitrary number, for example, from 0 to 100. In this embodiment, an example is shown in which the VAS is measured electronically using a slider or the like displayed on the display of the user terminal 7. However, like VAS measurement using paper media that has been used in clinical settings, for example, the VAS measurement in this embodiment may be performed by a user marking a line segment drawn on a blank sheet of paper and inputting the value of the distance from one end of the line segment to the marked position using the input / output device 13.

[0026] This embodiment can also be applied to rating data 81 measured using a Likert scale, which is a multi-value discrete response type ordinal scale. In particular, when the number of options in an ordinal scale increases, it is known that the characteristics of the ordinal scale become closer to the characteristics of a continuous response type measurement scale, and this embodiment can be easily applied.

[0027] The subjective state of the mind and body of a user refers to the state of the user's body and mind as felt by the user, and is a state other than an objective state obtained by measuring, for example, blood pressure, pulse rate, or other values ​​using sensors. For example, the subjective state of the mind and body can be exemplified by measuring the degree of fatigue, stress, and pain at a certain point in time or over a certain period of time. For example, in the case of pain, the subjective state of the mind and body can be measured using a unipolar VAS, which is a VAS in which the left end (or the lower end, with a rating of 0) is instructed to be a state of no pain at all, and the right end (or the upper end, with a rating of 100) is instructed to be the worst pain imaginable to date. In addition, the subjective state of the mind and body can be exemplified by a slider (Affective Slider) for measuring the affective dimension consisting of arousal level (Arousal) and affective valence, which is a natural emotion. For example, when measuring affective valence with an Affective Slider, a bipolar VAS can be used, in which the left end (or bottom end, rated at 0) is represented by an emoticon that reflects a negative emotion, the middle (rated at 50) is represented by an emoticon that reflects a neutral emotion, and the right end (or top end, rated at 100) is represented by an emoticon that reflects a positive emotion.

[0028] The communication device 14 executes a process for enabling the user terminal 7 to communicate with the reaction profile evaluation device 1 and the manager terminal 8 via the network 9 .

[0029] The notification notification device 15 notifies the user in response to the output process of the notification from the reaction profile evaluation device 1. For example, the notification notification device 15 displays the notification on the display of the input / output device 13. The notification notification device 15 may output the notification to the user using, for example, vibration or sound. The notification notification device 15 may also notify the user when the biomeasurement device 12 detects any change in the biometric condition of the user.

[0030] In this embodiment, the user terminal 7 is configured as one device, but the user terminal 7 does not necessarily have to be configured as one device. For example, the input / output device 13 and the communication device 14 may be provided in a smartphone, and the biometric device 12, the biometric sensor 11, and the notification device 15 may be provided in a smartwatch, and the smartphone and the smartwatch may be considered to configure one user terminal 7.

[0031] The manager terminal 8 includes an input / output device 31, a communication device 32, and a notification device 33, and is used by the manager.

[0032] The input / output device 31 is equipped with input devices such as a mouse, keyboard, touch panel, and microphone, and output devices such as a display and speaker, and is a device that performs input and output to the administrator terminal 8. The input devices receive input from the administrator. The output devices display screens on the display, for example, displaying data acquired by the administrator terminal 8 and displaying notifications to the administrator. The administrator terminal 8 displays screens using a display.

[0033] The communication device 32 executes a process for the manager terminal 8 to communicate with the reaction profile evaluation device 1 and the user terminal 7 via the network 9 .

[0034] The notification device 33 notifies the administrator in response to the output process of the notification from the reaction profile evaluation device 1. For example, the notification device 33 displays the notification on the display of the input / output device 31. The notification device 33 may output the notification to the administrator using, for example, vibration or sound. The notification device 33 may also notify the administrator when the biomeasurement device 12 detects any change in the biometric condition of the user.

[0035] The reaction profile evaluation device 1 is composed of a computer (calculator) and includes a processor 2, a memory 3, a storage device 4, an input / output device 5, and a communication device 6. The reaction profile evaluation device 1 includes, as functional units, a reception unit 51, a preprocessing unit 52, an RP distribution estimation unit 53, an RP distribution evaluation unit 54, a result display unit 55, a data extraction unit 56, and an RP distribution evaluation unit with confidence interval 57. Details of these functional units will be described later. The data extraction unit 56 and the RP distribution evaluation unit with confidence interval 57 will be described in Example 2.

[0036] The memory 3 loads programs for implementing each functional unit of the reaction profile evaluation device 1. The programs for implementing these functional units are executed by the processor 2.

[0037] The processor 2 operates as a functional unit that provides a predetermined function by executing processing according to a program for realizing each functional unit. For example, the processor 2 functions as an RP distribution estimation unit 53 by executing an RP distribution estimation program. The same applies to the other functional units of the reaction profile evaluation device 1. Furthermore, the processor 2 also operates as a functional unit that provides each function of a plurality of processes executed by each program.

[0038] The storage device 4 stores data used by the above-mentioned functional units. For example, the storage device 4 stores rating value data 81, partial rating value data 82, candidate RP distribution data 83, RP distribution data 84, equalized rating value data 85, RP distribution data with confidence intervals 86, user characteristic data 87, RP distribution information 90, and distribution selection criteria information 91. These data will be described later. The equalized rating value data 85 and the RP distribution data with confidence intervals 86 will be described in the second embodiment.

[0039] The input / output device 5 includes an input device and an output device. Examples of the input device include a mouse, a keyboard, a touch panel, and a microphone. Examples of the output device include a display and a speaker.

[0040] The communication device 6 executes a process for the reaction profile evaluation device 1 to communicate with the user terminal 7 and the administrator terminal 8 via the network 9 .

[0041] In this embodiment, an example will be described in which the user always wears or carries the user terminal 7 in daily life, and the biometric sensor 11 is always operating. However, the reaction profile evaluation device 1 according to this embodiment can be used in ways other than those described above. For example, the user may operate the biometric sensor 11 using the user terminal 7 several times a day, such as when waking up or going to bed.

[0042] 2 is a flowchart showing an example of a process in which the reaction profile evaluation device 1 receives data from the user terminal 7. This process is executed by the reception unit 51 of the reaction profile evaluation device 1.

[0043] In the data reception process S21, when a connection with the user terminal 7 is established via the network 9, the reception unit 51 receives data from the user terminal 7. When the data reception process S21 starts, the reception unit 51 receives and stores the rating value data 81 from the input / output device 13 of the user terminal 7. The data reception process S21 continues until the connection between the reception unit 51 and the user terminal 7 is disconnected.

[0044] Furthermore, when user characteristic data 87 is input to the user terminal 7 via the input / output device 13, in the data reception process S21, the reception unit 51 also receives the user characteristic data 87. The user characteristic data 87 includes the age and sex of the user, a user ID, and the like.

[0045] In the following, as an example, we will consider an example in which arousal level (Arousal) and emotional valence (Valence) measured using a bipolar VAS at a specified time each day, and fatigue, stress, anxiety, depression, and deep sleep measured using a unipolar VAS are measured as evaluation value data 81.

[0046] In this embodiment, an example is shown in which the reaction profile evaluation device 1 continuously performs the data reception process S21 until the connection between the reception unit 51 and the user terminal 7 is cut off. The reaction profile evaluation device 1 does not necessarily have to continuously perform the data reception process S21. For example, the reception unit 51 may establish a connection with the user terminal 7 and perform the data reception process S21 only when the user terminal 7 has performed a data transmission process to the reaction profile evaluation device 1. In this case, the user terminal 7 may monitor the activity state of the user with the bio-measurement device 12, and when a predetermined event occurs, may notify the user using the notification device 15 and prompt the user to input the evaluation value data 81 (a response to the instruction sentence).

[0047] FIG. 3 is a flowchart showing an example of a process of estimating a reaction profile distribution (RP distribution) performed by the reaction profile evaluation device 1.

[0048] A reaction profile distribution is a statistical distribution that indicates a reaction profile (RP). A reaction profile distribution is expressed as a distribution consisting of one statistical distribution or a mixed distribution of multiple statistical distributions, and is expressed by a group of parameters that characterize this distribution (a group of features that determine the shape of the distribution).

[0049] In a data reading process S31, the reception unit 51 of the reaction profile evaluation device 1 inputs the rating value data 81 of the user.

[0050] In preprocessing S32, the preprocessing unit 52 of the reaction profile evaluation device 1 preprocesses the input rating value data 81. When subsequent processing is to be performed for each user, the preprocessing unit 52 first extracts user characteristic data 87 for only the user to be processed from the user characteristic data 87 (FIG. 2) received by the receiving unit 51.

[0051] The preprocessing unit 52 performs preprocessing according to the input rating value data 81. For example, when a VAS is recorded with a domain ranging from 0 to 100, normalization processing is performed on the rating value data 81 so that the domain of the VAS is in the range of 0 to 1. Also, for example, when a plurality of types of VAS are recorded and the domains of these VAS are different, normalization processing is performed on the rating value data 81 so that the domains are in the same range. Also, in the case of VAS, etc., an effort minimization bias may occur in which the person avoids making an effort to answer and answers appropriately without reading the question instruction very much. The preprocessing unit 52 may perform noise removal processing to detect answers in which this effort minimization bias is apparent from the answering time etc. included in the rating value data 81 and remove them from the processing target.

[0052] In the subsequent RP distribution candidate estimation process S34, the domain of the rating value data 81 may need to be greater than 0 and less than 1 in some cases. In this case, the preprocessing unit 52 may, for example, scale the extracted user rating value data 81(d) to rating value data 81(d') as a process for converting the domain from the range [0,1] of 0 to 1 to the range (0,1) of 0 to 1. This scaling is expressed by the formula d'={d(n-1)+0.5} / n, where n is the total number of data samples of the rating value data 81(d). Note that when referring to the domain, parentheses [] indicate a range that includes the value in the parentheses (closed interval), and parentheses () indicate a range that does not include the value in the parentheses (open interval).

[0053] In the partial data generation process S33, the preprocessing unit 52 divides the preprocessed rating value data 81 by data range to generate partial rating value data 82 for estimating main RP distribution candidates and secondary RP distribution candidates. The main RP distribution candidates and secondary RP distribution candidates will be described later.

[0054] First, division of the rating value data 81 will be described.

[0055] Research on response styles (answer tendencies) in the Likert scale, a multi-value discrete response measurement scale, has revealed a variety of typical response styles. Well-known response styles include the Acquiescence Response Style (ARS), which is biased toward the highest rating regardless of the item or instruction, the Dis-acquiescence Response Style (DRS), which is biased toward the lowest rating, and the Extreme Response Style (ERS), which is a tendency to give ratings at both extremes. In addition, the Mid-point Response Style (MRS), which is biased toward answers near the center, is often observed in Japanese people.

[0056] These response tendencies are also observed in a similar way when a continuous response type measurement scale such as VAS is used. Furthermore, a bimodal response style (Bimodal MRS, BiMRS) with two distribution peaks near the center is sometimes observed in MRS. BiMRS is often observed when a bipolar VAS is used, in which the center of the measurement scale is neutral and the two poles of this center have opposing concepts, or when a Likert scale is used with the Semantic Difference Method, in which adjective instructions showing opposite concepts are given at both poles as instructions. Below, we will show examples of response profiles (RPs) that occur in rating data 81, assuming ARS, DRS, ERS, MRS, and BiMRS.

[0057] Among the RPs, the ARS, DRS, and ERS, and the MRS and BiMRS have different ranges in which the rating value data 81 is likely to be localized and observed. Therefore, when an RP in which the rating value data 81 is localized at multiple positions is assumed, in the partial data generation process S33, the pre-processing unit 52 may divide the rating value data 81 by the data range of the domain to generate partial rating value data 82.

[0058] For example, in the case of rating value data 81 normalized to a range (0,1) of a domain greater than 0 and less than 1, it is considered that in MRS and BiMRS, the rating value data 81 appears near the center of the domain (i.e., near 0.5), while in ARS, DRS, and ERS, the rating value data 81 appears near the upper and lower limits of the domain (i.e., near 0 and 1). Therefore, the preprocessing unit 52 can set this vicinity to a range determined by a predetermined threshold th, and divide the rating value data 81 into main part rating value data, which is data near the center of the domain, and secondary part rating value data, which is data near the upper and lower limits of the domain. That is, the preprocessing unit 52 can use a predetermined threshold th to divide the rating value data 81 into partial rating value data 82 consisting of main partial rating value data whose domain is in the range [th, 1-th] and secondary partial rating value data whose domain is in the range (0, th) and the range (1-th, 1). Then, the subsequent RP distribution candidate estimation process S34 may be performed for each partial rating value data 82.

[0059] In the following, the vicinity is defined by a predetermined threshold th, but the method of defining the threshold is not limited to this example. For example, a range that is near the lower limit and the upper limit of the domain may be defined by using a threshold th1 related to the lower limit and a threshold th2 related to the upper limit. That is, in this case, the preprocessing unit 52 may use the predetermined thresholds th1 and th2 to divide the rating value data 81 into partial rating value data 82 consisting of main partial rating value data whose domain is in the range [th1, th2] and secondary partial rating value data whose domain is in the range (0, th1) and the range (th2, 1).

[0060] In the following, as an example, an example will be described in which rating value data 81 is divided into partial rating value data 82, and the partial rating value data 82 consists of main partial rating value data and secondary partial rating value data defined in the above-mentioned domain using a predetermined threshold value th.

[0061] The threshold value th can be set to any value. For example, in the rating value data 81 whose domain is normalized to the range [0,1] or the range (0,1), the threshold value th can be set to a value selected from the range of 0.05 to 0.30, for example, with reference to the experimental results described later.

[0062] In the RP distribution candidate estimation process S34, the RP distribution estimation unit 53 estimates a plurality of reaction profile distribution candidates (RP distribution candidates) based on each of the partial rating value data 82. The RP distribution candidates are candidates for the statistical distribution expressing the RP of the user, and can be estimated by fitting to the statistical distribution represented in the candidate RP distribution template.

[0063] In the RP distribution candidate estimation process S34, the RP distribution estimation unit 53 approximates the distribution shape of the partial rating value data 82 with a statistical distribution based on the candidate RP distribution template stored in the RP distribution information 90. The candidate RP distribution template includes the type of statistical distribution (e.g., normal distribution and beta distribution) that represents the distribution that can express RP (e.g., ERS and BiMRS) and the range of values ​​of the parameter θ that determines the shape of this distribution. The parameter θ is a feature that determines the shape of the distribution, and is, for example, the mean or standard deviation when the type of statistical distribution is normal distribution.

[0064] In the following, an example will be shown in which the RP distribution estimation unit 53 estimates an RP distribution candidate, which is a candidate for a statistical distribution that characterizes the RP of a user, through fitting to a parametric statistical distribution whose distribution shape is characterized by a parameter θ. In this fitting, the occurrence frequency distribution of the partial rating value data 82 is fitted to a probability density function of the statistical distribution. The RP distribution candidate may be estimated by fitting the cumulative occurrence frequency distribution of the partial rating value data 82 to a cumulative probability density function of the statistical distribution. In addition, when a multi-value discrete response type ordinal scale such as a Likert scale is used as the rating value data 81, the RP distribution candidate may be estimated by fitting to a probability mass function or cumulative probability mass function of an appropriate discrete statistical distribution instead of fitting to a probability density function of the statistical distribution.

[0065] In the following, the RP distribution candidate estimation process S34 will be explained separately as a main RP distribution candidate estimation process S34A and a secondary RP distribution candidate estimation process S34B. The RP distribution estimation unit 53 estimates a plurality of RP distribution candidates including at least one of a main RP distribution candidate expressing an RP appearing near the center of the rating value data 81 and a secondary RP distribution candidate expressing an RP appearing near at least one of the upper limit and lower limit of the rating value data 81.

[0066] First, the main RP distribution candidate estimation process S34A will be described. In the main RP distribution candidate estimation process S34A, the RP distribution estimation unit 53 estimates a response profile distribution candidate (main RP distribution candidate Main) for the main partial rating value data. The parameter θ that determines the distribution shape of the main RP distribution candidate Main is defined as the main RP feature amount θ. Main Let us assume that.

[0067] The RP distribution estimation unit 53 estimates the main RP feature amount θ of the main RP distribution candidate Main. Main is estimated. It is assumed that the RP distribution information 90 stores a candidate RP distribution template for MRS and a candidate RP distribution template for BiMRS as candidate RP distribution templates. It is further assumed that the RP distribution information 90 stores a candidate RP distribution template of a uniform distribution without a peak as the candidate RP distribution template. The MRS can be expressed as a unimodal continuous distribution. The BiMRS can be expressed as a mixed distribution (bimodal distribution) consisting of two continuous distributions.

[0068] The RP distribution estimation unit 53 estimates, for example, the main RP feature amount θ Main Normal distribution Normal(μ,σ) with mean μ and standard deviation σ, main RP feature θ Main The candidate RP distribution template is a beta distribution Beta(α,β) with parameters α and β, or a mixture distribution with two mixtures of these. The main RP distribution candidate Main is estimated by fitting to the candidate RP distribution template. For example, in the case of fitting to a mixture distribution of two beta distributions L and R, the main RP distribution candidate Main is estimated as BetaMixture(w L ,α L ,βL ,w R =1-w L ,α R ,β R ), where w is the mixture ratio of the two beta distributions, and the subscripts L and R indicate the parameters for each beta distribution. Parameter fitting can be performed using known methods. For example, in the case of MRS, maximum likelihood method or Bayesian estimation can be used, and in the case of BiMRS, the expectation-maximization algorithm or a similar algorithm can be used.

[0069] In this manner, the RP distribution estimation unit 53 estimates the main RP feature quantity θ Main and from this, the main RP distribution candidate Main is estimated.

[0070] Next, the secondary RP distribution candidate estimation process S34B will be described. In the secondary RP distribution candidate estimation process S34B, the RP distribution estimation unit 53 estimates a response profile distribution candidate (secondary RP distribution candidate DistSub) for the secondary partial rating value data. A parameter θ that determines the distribution shape of the secondary RP distribution candidate DistSub is set as the secondary RP feature amount θ DistSub Let us assume that.

[0071] The RP distribution estimation unit 53 estimates the secondary RP feature quantity θ of the secondary RP distribution candidate DistSub. DistSub For example, the RP distribution estimation unit 53 estimates the parameter α ADE , β ADE The beta distribution expressed using ADE ,β ADE ) is set as a candidate RP distribution template, and the secondary RP distribution candidate DistSub is estimated by fitting to the candidate RP distribution template. The RP distribution estimation unit 53 sets a constraint α for the secondary partial rating value data so that the distribution shapes of ARS, DRS, and ERS are guaranteed. ADE and β ADEBy parameter fitting imposed on, the secondary RP feature θ DistSub Based on this, the secondary RP distribution candidate DistSub is estimated.

[0072] In the candidate distribution storage process S35, the RP distribution estimation unit 53 stores the candidate RP distribution data 83. The candidate RP distribution data 83 includes the main RP distribution candidate Main and the secondary RP distribution candidate DistSub obtained in the RP distribution candidate estimation process S34 (the main RP distribution candidate estimation process S34A and the secondary RP distribution candidate estimation process S34B), and the main RP feature amount θ Main and secondary RP feature θ DistSub In addition, when a uniform distribution without a peak is obtained as the main RP distribution candidate Main, the feature quantity that characterizes this uniform distribution is stored as the candidate RP distribution data 83.

[0073] The data localization ranges of the main RP distribution candidate Main and the secondary RP distribution candidate DistSub are unlikely to overlap with each other. Therefore, by dividing the rating value data 81 into partial rating value data 82 consisting of main partial rating value data and secondary partial rating value data, and obtaining candidate RP distribution data 83 using the partial rating value data 82, the accuracy with which the RP distribution estimation unit 53 estimates the RP distribution candidate can be improved.

[0074] In the above description, as an example, only specific user characteristic data 87 is extracted, and RP distribution candidates are estimated based on the occurrence frequency distribution of the rating data of only the specific user. The process of estimating RP distribution candidates is not limited to the process for a specific user. For example, when multiple users answer multiple types of questions and the number of times of answering is approximately equal among the multiple users, multiple item response theory models having RP as a latent variable may be provided as candidate RP distribution templates to estimate RP distribution candidates. In this case, if the rating data is a continuous response type measurement scale such as VAS, the RP distribution candidates and their RP feature amount θ may be estimated using a model that is expanded by introducing a parameter indicating RP into a beta response model.

[0075] 4 is a flowchart showing an example of an evaluation process of an RP distribution performed by the reaction profile evaluation device 1. In the evaluation process of an RP distribution, an RP distribution suitable for characterizing the RP of a user (i.e., an RP distribution suitable for the expression of RP of a user) is obtained using an evaluation index based on the rating value data 81 or partial rating value data 82 and a plurality of RP distribution candidates. In the following, as an example, an example of obtaining an RP distribution suitable for the expression of RP of a user using the partial rating value data 82 will be described.

[0076] In a data reading process S 41 , the reception unit 51 of the reaction profile evaluation device 1 inputs the partial rating value data 82 and the candidate RP distribution data 83 .

[0077] In the RP distribution evaluation process S42, the RP distribution evaluation unit 54 selects an RP distribution suitable for the user's expression of RP from among multiple RP distribution candidates, using the partial rating value data 82. The RP distribution evaluation unit 54 can select, for example, one RP distribution, multiple RP distributions, or a combination of multiple RP distributions as the RP distribution suitable for the user's expression of RP.

[0078] Specifically, the RP distribution evaluation unit 54 selects an evaluation index for determining an RP distribution suitable for characterizing the user's RP based on a combination of candidate RP distribution data 83. The evaluation index is an index for evaluating the degree of applicability of the distribution of the rating value data 81 (partial rating value data 82) to the RP distribution candidate, and is selected so that the evaluation index satisfies a condition determined by a predetermined threshold. The RP distribution evaluation unit 54 determines the distribution represented by the RP distribution candidates (main RP distribution candidate and secondary RP distribution candidate) corresponding to the selected evaluation index as the RP distribution suitable for characterizing the user's RP. In this embodiment, the RP distribution of the user is represented by a mixed distribution that mixes statistical distributions suitable for characterizing the user's RP.

[0079] In this embodiment, assuming a case where a main RP distribution candidate and a secondary RP distribution candidate are estimated, an example is shown in which the RP distribution evaluation process S42 is divided into a main RP distribution evaluation process S42A which is a process for the main RP distribution candidate and a secondary RP distribution evaluation process S42B which is a process for the secondary RP distribution candidate. In this embodiment, in particular, an example of hierarchical processing is shown in which, instead of exhaustively searching combinations of the main RP distribution candidate and the secondary RP distribution candidate, an appropriate RP distribution is selected from the main RP distribution candidates Main as the main RP distribution Main, and then a combination with the secondary RP distribution candidate is evaluated only for the selected main RP distribution Main. Note that an appropriate RP distribution may be selected in parallel for each of the main RP distribution candidate and the secondary RP distribution candidate, and only the combination of the obtained main RP distribution and the secondary RP distribution may be evaluated. By performing hierarchical processing or parallel processing instead of exhaustive searching, the evaluation space of the candidate RP distribution evaluation can be narrowed down, and the effect of reducing the amount of calculation and processing time is obtained.

[0080] As part of the RP distribution evaluation process S42, the RP distribution evaluation unit 54 first performs a main RP distribution evaluation process S42A. When MRS and BiMRS, or a uniform distribution without a peak is assumed as the main RP, the main RP distribution evaluation process S42A determines which type of statistical distribution is appropriate based on the evaluation index given in the distribution selection criteria information 91. The distribution selection criteria information 91 has items of evaluation index given in advance.

[0081] The evaluation index is the main RP feature θ Main The main RP distribution candidate characterized by the above can be compared with the occurrence frequency distribution of the main part rating value data from a necessary viewpoint to select the main RP distribution candidate. For example, a likelihood-related statistic or an information criterion can be adopted as an evaluation index to evaluate the fit of the occurrence frequency distribution of the main part rating value data to the main RP distribution candidate. For example, the likelihood, log-likelihood, likelihood ratio, and deviation can be used as the likelihood-related statistic. For example, Akaike's Information Criterion (AIC), Bayesian information criterion (BIC), Widely applicable information criterion (WAIC), and Widely applicable BIC (WBIC) can be used as the information criterion. In addition, the Kullback-Leibler divergence, which measures the distribution similarity, can be adopted as an evaluation index to evaluate the obtained distribution shape.

[0082] In addition, when MRS and BiMRS are assumed as the main RP and BiMRS is expressed by a mixture distribution, if the main RP distribution is evaluated simply by the information criterion, BiMRS, which has a complex RP distribution, tends to be selected even if the peak separation (bimodal separation) of the mixture distribution is not so large. It is desirable that BiMRS with a large number of parameters is selected only when the bimodal separation is high. For this reason, the bimodal separation may be used as an evaluation index, and the main RP distribution may be evaluated based on whether or not this evaluation index exceeds a predetermined minimum acceptable threshold accept_bidist, so that BiMRS is selected only when the bimodal separation of the BiMRS estimated as the main RP distribution is extremely high. For example, when BiMRS is approximated by a mixture beta distribution with a mixture number of 2, the bimodal separation can be calculated as the difference between the modes of two partial distributions that constitute the mixture distribution of BiMRS.

[0083] From the above, for example, only when the peak separation degree (bimodal separation degree) of the mixture distribution, which is the evaluation index, is greater than the minimum acceptable threshold accept_bidist and the AIC is small, BiMRS is selected as the main RP distribution. Otherwise, MRS or a uniform distribution without a peak (i.e., the main RP distribution does not exist) is selected as the main RP distribution Main and the main RP feature quantity θ characterizing the main RP distribution. Main Select .

[0084] After performing the main RP distribution evaluation process S42A, the RP distribution evaluation unit 54 performs the secondary RP distribution evaluation process S42B. In the secondary RP distribution evaluation process S42B, the main RP feature quantity θ Main The main RP distribution Main is characterized by the following, and each of the secondary RP features θ DistSub For the sub-RP distribution DistSub characterized by, the optimal mixture ratio w ADE Estimate the mixture proportion of both distributions, w ADE Calculate the evaluation index when the optimal mixture ratio w ADE can be estimated using the same method as the fitting method to the statistical distribution described in the RP distribution candidate estimation process S34. Then, each of the obtained evaluation indexes is compared with each of the evaluation indexes when the main RP distribution is selected as the RP distribution, and the condition with the best evaluation index is selected.

[0085] For example, when using the information criterion as an evaluation index, if the information criterion improves when a mixed distribution of the main RP distribution and the secondary RP distribution is adopted, this mixed distribution is selected as the RP distribution expressing the user's RP, and if it does not improve, the main RP distribution is selected as the RP distribution expressing the user's RP. If the information criterion is most improved when BiMRS is selected as the main RP distribution Main and ARS is selected as the partial distribution of the mixed distribution as the secondary RP distribution DistSub, then the RP distribution ResponseProfile(θ) is ResponseProfile(θ)=w ADE ×DistSub(θ DistSub )+(1-w ADE )×Main(θ Main ) and the RP distribution feature θ user θuser = {θ DistSub ,θ Main}={w ADE, α ARS ,β ARS ,w L ,α L ,β L ,w R ,α R ,β R The RP distribution evaluation unit 54 evaluates the obtained RP distribution feature amount θ user The information on the statistical distribution type is stored as RP distribution data 84.

[0086] 5 shows the results of an experiment on RP distribution estimation (FIG. 3) and RP distribution evaluation (FIG. 4) performed using the reaction profile evaluation device 1 according to this embodiment. This experiment was performed using pseudo data of the evaluation value data 81.

[0087] In this experiment, the RP distribution is defined as ResponseProfile(θ)=w ADE ×DistSub(θ DistSub )+(1-w ADE )×Main(θ Main ) and the RP distribution feature is expressed as θ user = {θ DistSub ,θ Main}={w ADE, α ARS ,β ARS ,w L ,α L ,β L ,w R ,α R ,β R}, the effect of the threshold value th used to generate the partial rating data 82 was evaluated using pseudo data of the rating data 81 generated from a statistical distribution with known true parameters.

[0088] Table 501 in Figure 5 shows an example of pseudo data used in the experiment, and shows the RP distribution and its RP distribution features for conditions numbered #11 to #36. Only 1000 samples of data were extracted from each beta mixture distribution defined by the RP distribution features shown in Table 501, and AIC and the degree of separation of the peaks of the mixture distribution (bimodal separation) were used as evaluation indices. The RP distribution feature θ when the minimum acceptable threshold accept_bidist for bimodal separation is set to 0.0 and the continuous distribution approximating the main RP distribution is set to a normal distribution or a beta distribution, user We evaluated the robustness of the estimates to the threshold th and the type of continuous distribution by evaluating

[0089] Graphs 502A and 502B show the RP distribution feature θ user The mixture ratio of the main RP distribution Main and the secondary RP distribution DistSub is w ADE The results are shown for the experiment. In this experiment, the continuous distributions (types of statistical distributions selected as candidate RP distribution templates) approximating the main RP distribution Main are normal distribution and beta distribution. Graph 502A is the case where the distribution type is normal distribution, and graph 502B is the case where the distribution type is beta distribution. In the experiment, the value of the threshold th was changed to 0.05, 0.15, 0.25, 0.1, 0.2, and 0.3.

[0090] From the graphs 502A and 502B, the RP distribution feature value θ user When evaluating, although there will be deviations from the true value under some conditions, as long as the threshold th is at least in the range of 0.05 to 0.3, regardless of whether the continuous distribution type is normal or beta, w ADE It can be confirmed that the estimation result of does not change significantly and is robust. In addition, the outline of the selected RP distribution reproduces the distribution of the pseudo data, including some conditions where deviation from the true value occurred. That is, in this embodiment, the RP distribution feature amount θ user It is less affected by changes in the response profile (RP) of each user, and stable results can be obtained.

[0091] In this way, according to this embodiment, even if the rating data 81 is measured using a measurement scale that is primarily a continuous response type, such as a VAS, the response profile (RP) that occurs in the rating data 81 can be quantitatively evaluated.

[0092] Fig. 6 is a diagram showing an example of user RP distribution data 84 displayed on the user terminal 7 or the administrator terminal 8. Graphs 601A and 601B shown in Fig. 6 are examples of RP distribution displays obtained by applying the method according to the present embodiment.

[0093] Graphs 601A and 601B are examples of probability density functions of the occurrence frequency distribution (actual distribution) and RP distribution of rating value data 81 displayed as histograms with a bin width of 0.05. Graph 601A shows the RP distribution obtained by assuming that the continuous distribution approximating the main RP distribution is a normal distribution. Graph 601B shows the RP distribution obtained by assuming that the continuous distribution approximating the main RP distribution is a beta distribution. In graphs 601A and 601B, the vertically shaded histograms are histograms showing actual rating value data consisting of multiple types of bipolar VAS and unipolar VAS for a certain user. The diagonally shaded histograms show the RP distribution (w ADE This is a histogram showing the probability density function of the RP distribution obtained by estimating that σ is 0.3.

[0094] In the example shown in Figure 6, a concentration of rating values ​​can be seen at both ends of the domain, as well as a bimodal concentration of rating values ​​at either side of the center. This means that this can be interpreted as an example in which the primary RP distribution is BiMRS and the secondary RP distribution is ERS.

[0095] As a result of applying this method, whether the continuous distribution is normal or beta, RP is w ADE It is estimated that the RP can be expressed by a mixture distribution consisting of ERS and BiMRS with a mean of 0.3, and it can be confirmed that an RP distribution that closely approximates the rating data 81 has been obtained. userThis provides the effect of enabling quantitative evaluation and comparison of RP occurring in rating value data 81 measured mainly using a continuous response type measurement scale such as VAS.

[0096] 7 is a diagram showing an example of the data structure of the RP distribution data 84. The RP distribution data 84 typically includes a user ID 7101, a processing date 7102, a processing ID 7103 for uniquely identifying the processing result, a start date and time 7104 and an end date and time 7105 of acquisition of the processed rating value data, an RP distribution information ID 7106 for identifying the RP distribution information 90 used in the processing, a distribution selection criteria information ID 7107 for identifying the distribution selection criteria information 91, and ADE The RP distribution data 84 may also have, as an index characterizing the RP distribution, DistSub 7109 indicating the type of secondary RP distribution, for example.

[0097] 7, the secondary RP distribution is shown, but the main RP distribution is not shown. The type of the main RP distribution (e.g., distribution shape) can be understood from the RP distribution information ID 7106 that identifies the RP distribution information 90 used in the processing.

[0098] The result display unit 55 can display the results calculated by the reaction profile evaluation device 1 on the display of the user terminal 7 or the administrator terminal 8. For example, the result display unit 55 can display at least one of the RP distribution (e.g., FIG. 6) and a group of parameters characterizing the RP distribution (e.g., RP distribution feature 7108 in FIG. 7) on the display.

[0099] As described above, the reaction profile evaluation device 1 according to this embodiment includes a receiving unit 51 that receives user rating value data 81, an RP distribution estimation unit 53 that estimates a plurality of RP distribution candidates expressing a reaction profile (RP), which is a response tendency of a user, based on the rating value data 81, and an RP distribution evaluation unit 54 that selects an RP distribution suitable for the expression of the user's RP from the plurality of RP distribution candidates, based on the rating value data 81 and the plurality of RP distribution candidates. The reaction profile evaluation device 1 according to this embodiment may include a result display unit 55 that displays at least one of the RP distribution and a group of parameters characterizing the RP distribution.

[0100] With this configuration, the response profile evaluation device 1 according to this embodiment has the effect of being able to quantitatively evaluate RP in a manner that allows comparison with other users, even for rating value data 81 measured using a continuous response measurement scale such as VAS, which often includes questions with different measurement scales, such as unipolar and bipolar scales. Furthermore, this effect has a secondary effect of contributing to the realization of inter-subject analysis of psychological measurements that takes into account the influence of RP. EXAMPLES

[0101] A reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program according to a second embodiment of the present invention will be described below. In the following, the differences between the second embodiment and the first embodiment will be mainly described.

[0102] When multiple types of VAS measurements are repeatedly taken over time in daily life, bias may occur in the number of responses and the type of responses depending on the type of VAS being measured. A response profile (RP) or response style is a bias in the tendency to respond consistently for each subject that occurs regardless of the instruction sentence (i.e., content-independent or task-independent). For this reason, in the configuration of the first embodiment, it is difficult to assume content independence when there is a bias in the number of responses and the type of responses, and there is a risk that an RP that is influenced by a type of VAS that has a large number of responses may be estimated. In consideration of this problem, the second embodiment is characterized by further comprising a process for pseudo-ensuring content independence.

[0103] The configuration of a reaction profile evaluation device 1 according to this embodiment is shown in FIG.

[0104] FIG. 8 is a flowchart showing an example of a series of processes for evaluating RP distribution with confidence intervals, which are performed by the reaction profile evaluation apparatus 1 in this embodiment.

[0105] In a data reading process S81, the reception unit 51 of the reaction profile evaluation device 1 inputs the rating value data 81.

[0106] Thereafter, each processing unit repeats a series of processes, which will be described later, a predetermined number of times B.

[0107] In the data extraction process S82, the data extraction unit 56 generates uniformed rating data 85, which is rating data 81 that pseudo-guarantees content independence. The data extraction unit 56 resamples the rating data 81 obtained from the user and consisting of multiple different answer items by restoring and extracting each answer item, thereby pseudo-eliminating the difference in the number of answers between answer items. The rating data 81 consisting of multiple different answer items includes, for example, rating data 81 for different types such as unipolar VAS and bipolar VAS, and rating data 81 for the same VAS type but with different instructions.

[0108] The data extraction unit 56 resamples the rating data 81 to generate uniformed rating data 85 in which the types of the rating data 81 or the number of the rating data 81 are uniformed. For example, when the measurement scale is composed of only bipolar or unipolar scales, a predetermined number of samples are extracted with replacement. When both bipolar and unipolar scales are mixed, the number of samples within each VAS type is uniformed by the first layer of extraction with replacement, and the difference in the number of samples between the bipolar and unipolar scales is uniformed by the second layer of extraction with replacement. This makes it possible to generate uniformed rating data 85 containing the same number of responses for each answer item. This process obtains a data set containing multiple answer items with the same number of responses, and it is possible to generate multiple rating data in which the influence of bias in the answer items and the number of responses is reduced, that is, rating data in which content independence is pseudo-guaranteed.

[0109] Next, the uniformized rating data 85 is used as input to perform the RP distribution estimation process S83 shown in Fig. 3 and the RP distribution evaluation process S84 shown in Fig. 4, and RP distribution data 84 is calculated for each uniformized rating data 85. These processes generate bootstrap samples by the bootstrap method for a predetermined number of iterations B, and calculate the RP distribution feature quantity θ user,i This is equivalent to constructing a bootstrap distribution of

[0110] Then, the RP distribution evaluation process with confidence interval S85 is performed. The RP distribution evaluation unit with confidence interval 57 evaluates the obtained RP distribution feature amount θ user,i The RP distribution data 84 expressing the bootstrap distribution of the above is subjected to a process S85 for evaluating the RP distribution with confidence intervals. In the process S85, the RP distribution evaluation unit 57 evaluates the RP distribution feature quantity θ user,i The RP distribution with confidence intervals is calculated by calculating the bootstrap statistics for the bootstrap distribution formed by the above. The RP distribution with confidence intervals is an RP distribution in which a plausible range (confidence interval) is also shown, and is represented by the RP distribution data with confidence intervals 86.

[0111] For example, by calculating the mean value and 95% confidence interval of the bootstrap distribution using the normal distribution approximation method, the RP distribution feature θ user,i can be calculated with confidence intervals. The RP distribution data 86 with confidence intervals can be calculated using known methods such as the percentile method, not limited to the normal distribution approximation method. This reduces the risk of being influenced by the rating values ​​of types with a large number of responses, and allows the user's RP to be estimated in a manner that takes into account the extent to which the estimated result may fluctuate.

[0112] Fig. 9 is a diagram showing an example of RP distribution data 86 with confidence intervals for users, displayed on the user terminal 7 or the administrator terminal 8. The example shown in Fig. 9 shows an example in which the occurrence frequency distribution 901A (actual distribution) of each equalized rating value data 85 and the corresponding probability density function 901B of the RP distribution are displayed as a histogram with a bin width of 0.05 when the number of repetitions B is 1000, the continuous distribution type is normal distribution, and the method described in the second embodiment is applied.

[0113] In the histogram of Fig. 9, the RP distribution feature θ of the number of repetitions (i.e., B number of repetitions) user,i The median RP distribution feature θ user,i The envelope drawn by is represented by the thick line at the top of the region indicating 25-50% (thick line indicating 50%), and the sum of the multiple regions filled with hatching is the RP distribution feature θ user,i The figure shows the range of the RP distribution, which is the 95% range (the plausible range of the RP distribution).

[0114] By repeating sampling with replacement, the occurrence frequency of the uniformed rating value data 85 fluctuates as shown in occurrence frequency distribution 901A, and reflecting this fluctuation, the RP distribution shape estimated as an RP distribution with confidence interval also fluctuates as shown in probability density function 901B. According to the method of this embodiment, even from rating value data 81 that includes bias in the number of responses, etc., it is possible to estimate and display the RP of a user with a confidence interval.

[0115] As described above, the reaction profile evaluation device 1 according to this embodiment includes a data extraction unit 56 that resamples rating value data 81 to generate homogenized rating value data 85 in which the type of rating value data or the number of rating value data items is pseudo-homogenized, and an RP distribution evaluation unit 57 that estimates an RP distribution based on the generated multiple homogenized rating value data 85 and calculates RP distribution data 86 with confidence intervals for the RP distribution data group obtained from each homogenized rating value data 85.

[0116] As a result, the reaction profile evaluation device 1 according to this embodiment can pseudo-guarantee content independence and estimate the RP of the user more plausibly even if there is a bias in the number of answers or the type of answers. Furthermore, by estimating the RP distribution for a plurality of uniformed rating value data 85, it is possible to estimate the confidence interval of the degree of fluctuation that may occur in the RP distribution of the user, and it is possible to take into account the reliability of the estimated RP distribution. EXAMPLES

[0117] A reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program according to a third embodiment of the present invention will be described below. In the following, the differences between the present embodiment and the first and second embodiments will be mainly described.

[0118] Example 3 is an example in which the estimated RP can be used to take into account RP in inter-subject analysis using psychological measurements. In this example, an example is described in which rating data 81 obtained by measuring the subjective mental and physical state of the user as a psychological measurement is used as an input to detect changes in the subjective mental and physical state of the user.

[0119] 10 is a block diagram showing an example of the main components of a reaction profile evaluation device 1 according to the present embodiment. The reaction profile evaluation device 1 according to the present embodiment includes a subjective state evaluation unit 58 and a subjective state evaluation model 92 in the reaction profile evaluation device according to embodiments 1 and 2 (FIG. 1), and stores evaluation result data 93.

[0120] In the following, as an example of detecting changes in the user's subjective mental and physical state, an example of anomaly detection is shown in which rating value data 81 measured in the past is compared to determine whether or not there is an abnormality in the trend of the rating value data 81 for the most recent specified period.

[0121] Fig. 11A is a flowchart showing an example of a learning process of the subjective state evaluation model 92, which is performed by the response profile evaluation device 1. The subjective state evaluation model 92 learns in accordance with the flowchart shown in Fig. 11A.

[0122] In the data reading process S111, the reception unit 51 inputs the rating value data 81, the user characteristic data 87, and the RP distribution data with confidence intervals 86. In this embodiment, as an example, the RP distribution data with confidence intervals 86 is input, but the RP distribution data 84 may be input instead of the RP distribution data with confidence intervals 86.

[0123] In preprocessing S112, the preprocessing unit 52 performs preprocessing on the data input in the data reading process S111.

[0124] First, the preprocessing unit 52 performs processing on the rating value data 81, which corresponds to the preprocessing shown in FIG.

[0125] Next, the preprocessing unit 52 creates a learning data set from the data input in the data reading process S111. For example, if the most recent predetermined period is one week, and daily anxiety is measured as the user's subjective mental and physical state as the rating data 81, a data set consisting of the one-week anxiety rating data 81, the user characteristic data 87, and the RP distribution data 86 with confidence intervals for a period during which no abnormality occurs in anxiety is extracted with a moving window of one day to create a learning data set. At this time, since the RP of the user is a relatively stable response tendency over time regardless of instruction, the previously evaluated RP distribution data 86 with confidence intervals may be used, considering it to be a static characteristic. On the other hand, if the situation requires consideration of the characteristics of the time change of the RP itself, for example, the daily RP distribution data 86 with confidence intervals may be calculated, and the one-week RP distribution data 86 with confidence intervals may be extracted with a moving window of one day to be added to the data set, similar to the anxiety rating data 81.

[0126] In the subjective state assessment model learning process S113, the subjective state assessment unit 58 uses the created learning data set to create a subjective state assessment model 92. The subjective state assessment model 92 is a model for inputting a data set consisting of the input one-week anxiety rating value data 81, user characteristic data 87, and RP distribution data with confidence intervals 86, and evaluating the degree to which these data sets deviate from normal data as an anomaly.

[0127] The subjective state assessment model 92 is typically an unsupervised model based on a known anomaly detection algorithm, and can be configured as a known statistical model or machine learning model. As the statistical model, for example, a statistical model that outputs a z-score using the average value or standard deviation for the data distribution of the data set, a statistical model that outputs a deviation value, and a statistical model that outputs a quantile on the data distribution can be used. In addition, as the machine learning model, for example, a machine learning model that outputs a distance based on a cluster center of the data distribution estimated nonparametrically from the data set can be used.

[0128] When using an unsupervised model based on an anomaly detection algorithm as the user's subjective state assessment model 92, a model may simply be created for each user. In this case, as an example of using the RP of this method in intra-subject analysis, by using RP distribution data 86 with confidence intervals as input in addition to rating value data 81, it is possible to assess the subjective state taking into account the time-varying characteristics of RP, and it is possible to distinguish between changes in RP and changes in the target rating value data 81 and detect changes in the subjective state of mind and body with high detection accuracy.

[0129] On the other hand, when creating a model for each user, a large amount of rating data 81 of a specific user is required, but the rating data 81 requires the user's answers, so the measurement cost is high. In addition, when a subjective state assessment model 92 is created by simply grouping multiple users together, a large amount of rating data 81 can be used as a learning data set, but since the RP is different for each user, excessive abnormalities are likely to be detected for users whose RP is different from other users. In this method, if the RP distribution data 86 with confidence intervals is used as an input, the RP characteristics can also be taken into account as an input data set, as an example of using the RP of this method in inter-subject analysis. Therefore, by creating a subjective state assessment model 92 by grouping multiple users together, a large amount of learning data can be used, and the RP that differs between users can be taken into account, making it possible to detect changes in the subjective state of the mind and body with high detection accuracy.

[0130] The series of learning processes shown in Fig. 11A is executed at least once before executing the process of subjective state evaluation (Fig. 11B) described later. In addition, by executing this learning process at regular intervals as the rating value data 81 increases, the subjective state evaluation model 92 can be re-trained. As a result, a subjective state evaluation model 92 with higher evaluation accuracy can be generated.

[0131] 11B is a diagram showing an example of a subjective state evaluation process performed by the response profile evaluation device 1. The response profile evaluation device 1 uses a subjective state evaluation model 92 to evaluate the mental and physical subjective states of the user.

[0132] In the data reading process S111, the reception unit 51 inputs the rating value data 81, the user characteristic data 87, and the RP distribution data 86 with confidence intervals.

[0133] In the pre-processing S112, the pre-processing unit 52 generates an input data set for evaluating the subjective state. For example, the pre-processing unit 52 generates a data set including the user's anxiety rating data 81 for the most recent week, the user characteristic data 87, and the RP distribution data 86 with confidence intervals as the input data set.

[0134] In subjective state evaluation S114, the subjective state evaluation unit 58 evaluates the user's subjective mental and physical state using the input data set generated in preprocessing S112 and the subjective state evaluation model 92, and determines the degree of abnormality indicating whether the subjective mental and physical state has changed significantly from normal. The subjective state evaluation unit 58 outputs the determined degree of abnormality as a continuous value, for example, from 0 to 1, and stores it in evaluation result data 93.

[0135] As a result, the reaction profile evaluation device 1 according to this embodiment has the effect of being able to detect changes in the subjective mental and physical state of the user with high accuracy, taking into account the RP that differs for each user. EXAMPLES

[0136] A reaction profile evaluation device, a reaction profile evaluation method, and a reaction profile evaluation program according to a fourth embodiment of the present invention will be described below. In the following, the differences between the first and second embodiments will be mainly described.

[0137] In the third embodiment, the rating data 81 itself is input as an analysis target, and changes in the user's subjective mental and physical state are detected in consideration of the user's RP.

[0138] In the fourth embodiment, biometric data obtained by the biometric sensor 11 measuring a living body (user) is input, and the user's RP is taken into consideration to estimate the subjective mental and physical state of the user when estimating the rating data 81. In this embodiment, as an example, an example will be described in which the subjective mental and physical state to be estimated is a natural emotion in daily life, and a natural emotion measured in an emotion dimension consisting of arousal (Arousal) and emotional valence (Valence) is estimated from the biometric data, and if the emotion does not occur often in daily life, an answer is obtained as the rating data 81.

[0139] Fig. 12 is a block diagram showing an example of the main components of a reaction profile evaluation device 1 according to this embodiment. The reaction profile evaluation device 1 according to this embodiment includes a biological state evaluation unit 59, biological measurement data 88, and a biological state evaluation model 94 in the reaction profile evaluation device according to embodiments 1 and 2 (Fig. 1), and stores evaluation result data 93. The biological state evaluation model 94 is a model for estimating the subjective mental and physical state of a user from the biological state of the user.

[0140] 13A is a flowchart showing an example of processing by the reaction profile evaluation device 1 to receive data from the user terminal 7. In data reception processing S131, the reception unit 51 of the reaction profile evaluation device 1 receives the rating value data 81 and the user characteristic data 87 from the user terminal 7, as in the data reception processing S21 shown in FIG. 2 (Example 1), and further receives the biomeasurement data 88 from the user terminal 7.

[0141] Fig. 13B is a flowchart showing an example of the learning process of the biological state evaluating model 94, performed by the reaction profile evaluating device 1. The biological state evaluating model 94 learns according to the flowchart shown in Fig. 13B.

[0142] In the data reading process S132, the reception unit 51 inputs the rating value data 81, the user characteristic data 87, the biometric data 88, and the RP distribution data with confidence intervals 86. In this embodiment, as an example, the RP distribution data with confidence intervals 86 is input, but the RP distribution data 84 may be input instead of the RP distribution data with confidence intervals 86.

[0143] In the following, unless otherwise specified, a set of biomeasurement data 88 is composed of 30 minutes of heartbeat interval data, electrodermal activity data, and acceleration data, a set of rating value data 81 is the intensity of emotional valence and arousal for this 30 minutes, and an example will be described in which a biological state assessment model 94 is trained by using a plurality of these sets. Therefore, the rating value data 81 corresponds to the correct answer data for estimating the subjective mental and physical state.

[0144] In preprocessing S133, the preprocessing unit 52 performs preprocessing on the input data. First, the preprocessing unit 52 performs processing equivalent to preprocessing S32 in Fig. 3 (Example 1) on the rating value data 81, which is the correct answer data for estimating the subjective mental and physical state.

[0145] In addition to this process, the preprocessing unit 52 may convert the rating value data 81 into quantiles for each user, for example, based on the RP distribution data with confidence intervals 86. This quantile conversion is a process of converting certain data into occurrence probability, etc., based on the cumulative density function of its occurrence frequency. For example, when a normal distribution approximation method is used for the RP distribution data with confidence intervals 86, the cumulative density function of the RP distribution may be obtained using the average value of the bootstrap distribution, and the relative occurrence probability (a value between 0 and 1) of the rating value data 81 for each user may be used as the correct answer data.

[0146] Furthermore, when classifying natural emotions as a subjective mental and physical state instead of estimating a continuous value of the natural emotion, for example, a margin ε may be used to binarize the data, excluding data in the range of 0.5±ε, which is the central value between 0 and 1, with 0 to 0.5−ε set as negative example 0 and 0.5+ε to 1 set as positive example 1.

[0147] Furthermore, as preprocessing S133 for the biometric data 88, the preprocessing unit 52 performs at least one of a correction process for individual differences contained in the biometric data 88, a feature extraction process from the biometric data 88, and a compression process (dimension reduction) of the features extracted from the biometric data 88. The preprocessing unit 52 may explicitly execute multiple processes (subtasks) in sequence as a pipeline for these processes, or may execute them end-to-end (all at once without dividing them into subtasks).

[0148] For example, the series of pre-processing steps S133 may be configured as a correction process for individual differences by performing data normalization processing based on a series of biometric data 88 previously measured from the user, and by performing feature compression processing using principal component analysis with the biometric data 88 normalized for each user as input, without performing feature extraction processing.

[0149] In addition, in the case of end-to-end processing, the pre-processing S133 may be configured to set a process of removing measurement noise from the biomeasurement data 88, and the feature extraction process and feature compression process may be configured to use a latent vector of a Variational Auto Encoder (VAE), which is a deep generative model that inputs the noise-removed biomeasurement data 88. In this case, the VAE model is learned from the biomeasurement data 88 by unsupervised learning.

[0150] For the purpose of correcting individual differences and removing noise, normalization processing of the signal scale to be used or noise removal processing may be performed. For example, the normalization processing may be min-max normalization, which normalizes the signal by the maximum and minimum values, z-score normalization, which normalizes the signal by the average and standard deviation, or quantile normalization, which normalizes the signal by the quantile of the signal intensity distribution, for each user, each measurement day, each measurement day of a user, or among all users. In addition, if it is known that the signal intensity varies due to aging, etc., deviation value processing, which normalizes the signal intensity for each age group, may be performed in consideration of the user's age information included in the user characteristic data 87. Furthermore, in the noise removal processing, clipping processing or winsorization processing, which removes abnormal values ​​of the signal and places them within a certain range, or moving average processing, which suppresses and smoothes sudden fluctuations at one time, or zeroth order differentiation processing using a Savitzky-Golay filter, may be performed.

[0151] In the process of extracting features from the biomeasurement data 88, features may be extracted according to the biosignals of the biomeasurement data 88 used. For example, for the heartbeat interval data acquired by the heartbeat sensor 21, the average heart rate, the low frequency component (LF) or high frequency component (HF) obtained by frequency domain analysis and known to mainly reflect sympathetic nerve activity or parasympathetic nerve activity, respectively, the feature amount using the SDNN, RMSSD, and NN50 used in the time domain analysis, the Lorenz plot used in the nonlinear domain analysis, the feature amount obtained by the detrended fluctuation analysis, and the feature amount obtained by the complex demodulation method may be used. For the electrodermal activity data acquired by the electrodermal activity sensor 22, the skin conductance level (SCL) or the skin conductance response (SCR) may be used. In addition, for the three-axis acceleration data obtained from the acceleration sensor 23, the acceleration norm or the number of zero crossings, which is the number of times that a signal processed by a band-pass filter for the acceleration norm passes through a threshold of ±0.01 G when the gravitational acceleration is 1 G, may be used.

[0152] The feature compression process can use a known algorithm, such as the above-mentioned principal component analysis, an autoencoder (AE), or uniform manifold approximation and projection (UMAP).

[0153] Thereafter, the preprocessing unit 52 creates a learning dataset using the data that has been subjected to various conversion processes. For example, the learning dataset may be created by using the converted biometric data 88, the RP distribution data with confidence intervals 86, and the user characteristic data 87 as the input X of the biological state evaluation model 94, and the rating value data 81 as the output y of the biological state evaluation model 94. Alternatively, the learning dataset may be created by using the converted biometric data 88 and the user characteristic data 87 as the input X, and using the rating value data 81 that has been quantile-transformed by the RP distribution data with confidence intervals 86 as the output y. Furthermore, the learning dataset may be created by combining these processes.

[0154] After the series of pre-processing steps S133 are completed, a biological state evaluation model learning step S134 is performed.

[0155] In the biological state evaluation model learning process S134, the biological state evaluating section 59 creates a biological state evaluation model 94 using the created learning data set.

[0156] For example, when the biological state evaluation model 94 is a model for estimating natural emotions in the user's daily life, the biological state evaluation model 94 may learn an emotion dimension estimation model that estimates "the intensity of valence and arousal for a certain 30 minutes" as natural emotions through supervised learning. The biological state evaluation model 94 may be created using a known algorithm. For example, a logistic regression model, a decision tree, a Random Forest, a Support Vector Machine, a neural network, a Bayesian neural network, and a deep learning model may be used as a machine learning algorithm. As for the algorithm, a classification algorithm or a regression algorithm may be used depending on the subjective state of the mind and body to be estimated. For example, when the intensity of valence and arousal is estimated as a value from 0 to 1, a regression algorithm may be used, and when the intensity of valence and arousal is estimated as a high or low, a classification algorithm may be used instead of a regression algorithm.

[0157] Furthermore, the biological state evaluation model 94 may be composed of a plurality of models that take into account individual differences. In this case, it becomes possible to estimate the subjective mental and physical state that is suited to each user, and this has the effect of improving suitability to the user and the acceptability of the subjective value estimation result from the user.

[0158] When the biological state evaluation model 94 is configured with a plurality of models considering individual differences, a dedicated process may be added to the biological state evaluation model 94, or the biological state evaluation model 94 itself may be configured as end-to-end. For example, a case where the assessment value data 81 quantile-converted by the RP distribution data 86 with confidence intervals is used as the output y corresponds to an example of adding a dedicated process. In addition, when configured as end-to-end, the biological state evaluation model 94 may be configured as a neural network, and a layer close to the final layer may be configured by multitask learning that divides the layer for each user or into user groups with similar RP distributions, so that estimation according to the RP of the user can be realized. Alternatively, a biological state evaluation model 94 common to users may be created, and the biological state evaluation model 94 may be fine-tuned for each user so that the biological state evaluation model 94 is adapted to a user with a specific RP.

[0159] The series of learning processes shown in Fig. 13B is executed at least once before executing the biological state evaluation process (Fig. 13C) described below. Furthermore, by executing this learning process at regular intervals as the assessment value data 81 and biomeasurement data 88 increase, the biological state evaluation model 94 can be re-trained. As a result, a biological state evaluation model 94 with even higher evaluation accuracy can be generated.

[0160] 13C is a diagram showing an example of a biological state evaluation process performed by the reaction profile evaluation device 1. The reaction profile evaluation device 1 uses a biological state evaluation model 94 to evaluate the subjective mental and physical state of the user from the biological state of the user.

[0161] In the data reading process S132, the reception unit 51 inputs the user characteristic data 87, the biometric data 88, and the RP distribution data 86 with confidence intervals.

[0162] In preprocessing S133, the preprocessing unit 52 generates an input data set for evaluating the user's subjective mental and physical state from the user's biological state. For example, the preprocessing unit 52 generates a data set consisting of user characteristic data 87, biological measurement data 88, and RP distribution data with confidence intervals 86 as the input data set.

[0163] In the biological state evaluation process S135, the biological state evaluation unit 59 estimates the user's subjective mental and physical state using the input data set generated in the preprocessing S133 and the biological state evaluation model 94. For example, when estimating "the intensity of emotional valence (Valence) and arousal (Arousal) for a certain 30 minutes", the estimated intensities of arousal (Arousal) and valence (Valence) are estimated as continuous values ​​from 0 to 1. The biological state evaluation unit 59 outputs the estimated subjective mental and physical state and stores it in the evaluation result data 93.

[0164] 14A and 14B are diagrams showing examples of evaluation results that are output and displayed by the result display unit 55 on the display of the input / output device 13 of the user terminal 7. As an example, Fig. 14A and Fig. 14B show a case where a smartphone is used as the input / output device 13. In this embodiment, an example is shown where the evaluation result is notified to the user in the form of a message to a notification application.

[0165] A screen 1301 in FIG. 14A is an example of a display screen when the user is requested to input evaluation value data 81 (answer to the instruction sentence) based on evaluation result data 93.

[0166] In the conventional technology, a notification application displays notification 1301A to a user at irregular intervals, for example, as shown in Signal #1 on screen 1301 in Fig. 14A, and attempts to collect rating value data 81 as correct answer data for emotions corresponding to a certain moment. However, since emotions such as strong joy or sadness do not occur frequently in daily life, simple irregular notifications may take a long time to collect correct answer data, or the specified period may end without acquiring a sufficient number of correct answer data.

[0167] In this embodiment, since the evaluation result data 93 is obtained taking into account the RP of the user, it is possible to evaluate whether the user is in an unusual situation or not taking into account the RP. For this reason, for example, a situation in which the estimated strength of the arousal level Arousal and the emotional valence Valence are close to 0 or 1 is considered to be a strong emotion, and the timing when the user is in a subjective state (emotional state) that is different from the usual can be displayed on the user terminal 7 as a notification 1301B in a format like Event #1 shown on the screen 1301. This makes it possible to prompt the user to respond when the emotion experienced by the user is estimated to be an emotion that does not occur often in daily life, which ultimately contributes to the collection of the evaluation value data 81.

[0168] In this embodiment, the user's response based on the notification 1301B can be obtained as the rating value data 81 in the data reception process S131 of Fig. 13A in this way. Therefore, in this embodiment, compared to the conventional technology in which the notification 1301A such as Signal #1 is irregularly displayed to the user to obtain rating value data, and this rating value data is used as correct answer data for emotions, it is possible to obtain the effect that the rating value data 81 corresponding to a rare emotional state can be obtained with high efficiency.

[0169] 14B is an example of a display screen that shows a user examples of multiple RP types A to D and the user's RP type. Screen 1302 shows the user's RP type when multiple models that take into account individual differences for each RP type A to D (similar user groups A to D) are created using RP distribution data with confidence intervals 86. A display screen such as screen 1302 allows a user to compare their own answer characteristics (i.e., RP) with other users or similar user groups different from themselves and judge what characteristics they have.

[0170] As a result, the reaction profile evaluation device 1 according to this embodiment has the effect of being able to estimate the subjective mental and physical state of the user with high accuracy from the biomeasurement data 88, taking into account the RP that differs for each user.

[0171] The present invention is not limited to the above-described embodiments, and various modifications are possible. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to an embodiment having all of the configurations described. It is also possible to replace a part of the configuration of one embodiment with the configuration of another embodiment. It is also possible to add the configuration of another embodiment to the configuration of one embodiment. It is also possible to delete a part of the configuration of each embodiment, or to add or replace another configuration.

[0172] Further, each component of the device according to the present invention may be realized in hardware, for example, by designing a part or all of them as an integrated circuit. Further, each component may be realized in software, for example, by having the processor 2 interpret and execute a program that realizes each function. Information such as a program, table, file, measurement information, and calculation information that realizes each function can be recorded in a recording device such as a memory 3, a hard disk drive, and an SSD (Solid State Drive), or a computer-readable recording medium such as an IC card, an SD card, and a DVD. Thus, each component of the device according to the present invention can realize each function as a processing unit, a processing unit, a program module, and the like.

[0173] In addition, in each drawing, the control lines and information lines are shown as those considered necessary for the explanation, and not all the control lines and information lines necessary for the product are shown. In an actual product, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0174] 1...reaction profile evaluation device, 2...processor, 3...memory, 4...storage device, 5...input / output device, 6...communication device, 7...user terminal, 8...administrator terminal, 9...network, 11...biometric sensor, 12...biometric device, 13...input / output device, 14...communication device, 15...notification / alarm device, 21...heart rate sensor, 22...electrodermal activity sensor, 23...acceleration sensor, 31...input / output device, 32...communication device, 33...notification / alarm device, 51...reception unit, 52...preprocessing unit, 53...RP distribution estimation unit, 54...RP distribution evaluation unit, 55...result display unit, 56...data extraction unit, 57...RP distribution evaluation unit with confidence interval, 58...subjective state evaluation unit, 59...biometric state evaluation unit, 81...rating data, 82...partial rating data, 83...candidate RP distribution data, 84...RP Distribution data, 85...uniformized rating value data, 86...RP distribution data with confidence interval, 87...user characteristic data, 88...biometric data, 90...RP distribution information, 91...distribution selection criterion information, 92...subjective state assessment model, 93...assessment result data, 94...biometric state assessment model, 501...table, 502A, 502B...graph, 601A, 601B...graph, 901A...occurrence frequency distribution, 901B...probability density function, 1301...screen, 1301A...notification, 1301B...notification, 1302...screen, ID7101...user, 7102...processing date, 7103...processing ID, 7104 data start date and time, 7105...data end date and time, 7106...RP distribution information ID, 7107...distribution selection criterion information ID, 7108...RP distribution feature, 7109...DistSub.

Claims

1. The user's response tendency is made into a response profile, A statistical distribution representing the reaction profile of the user is defined as a reaction profile distribution; A reception unit for inputting the user's rating data; a reaction profile distribution estimation unit that estimates a plurality of reaction profile distribution candidates, which are candidates for the reaction profile distribution, based on the evaluation value data, by using a candidate reaction profile distribution template including a type of the statistical distribution; a reaction profile distribution evaluation unit that selects a reaction profile distribution suitable for expressing the reaction profile of the user from among a plurality of reaction profile distribution candidates estimated using the evaluation value data; A reaction profile evaluation device comprising:

2. The reaction profile distribution estimation unit estimates a plurality of reaction profile distribution candidates including at least one of a main reaction profile distribution candidate expressing the reaction profile appearing near the center of the domain of the rating value data and a secondary reaction profile distribution candidate expressing the reaction profile appearing near at least one of an upper limit vicinity and a lower limit vicinity of the domain of the rating value data; the reaction profile distribution evaluation unit selects one reaction profile distribution, a plurality of reaction profile distributions, or a combination of a plurality of reaction profile distributions, suitable for expressing the reaction profile of the user, from among the plurality of estimated reaction profile distribution candidates; The reaction profile evaluation apparatus according to claim 1 .

3. a data extraction unit that resamples the rating value data to generate uniformed rating value data in which the types of the rating value data or the number of the rating value data are uniformed in a pseudo manner; A reaction profile distribution evaluation unit with confidence intervals that calculates a reaction profile distribution with confidence intervals, which is the reaction profile distribution with a plausible range being also shown; Equipped with The reaction profile distribution estimation unit estimates the reaction profile distribution candidate based on a plurality of the uniformized rating value data, The reaction profile distribution evaluation unit selects the reaction profile distribution using a plurality of the homogenized rating value data, The reaction profile distribution with confidence interval evaluation unit calculates the reaction profile distribution with confidence interval from the selected reaction profile distribution. The reaction profile evaluation apparatus according to claim 2 .

4. a result display unit that displays at least one of the reaction profile distribution and a group of parameters that characterize the reaction profile distribution; The reaction profile evaluation apparatus according to claim 1 .

5. a pre-processing unit that divides the rating data by a data range of the domain and generates partial rating data for estimating the main reaction profile distribution candidate and the sub-reaction profile distribution candidate; The reaction profile evaluation apparatus according to claim 2 .

6. The reaction profile distribution is represented by one of the statistical distributions or a distribution consisting of a mixture distribution of a plurality of the statistical distributions, and is represented by a group of parameters characterizing the distribution. The reaction profile evaluation apparatus according to claim 1 .

7. The reaction profile distribution evaluation unit selects the reaction profile distribution using at least one of a likelihood-related statistic, an information criterion, a distribution similarity, and a degree of separation of peaks of a mixture distribution as an evaluation index. The reaction profile evaluation apparatus according to claim 1 .

8. a condition evaluation unit that evaluates a condition of the user using at least one of the evaluation value data and biomeasurement data obtained by measuring a biological condition of the user and the reaction profile distribution; The reaction profile evaluation apparatus according to claim 1 .

9. The rating data is data measured using a continuous response measurement scale. The reaction profile evaluation apparatus according to claim 1 .

10. The method is executed by a computer including a reception unit, a reaction profile distribution estimation unit, and a reaction profile distribution evaluation unit, The user's response tendency is made into a response profile, A statistical distribution representing the reaction profile of the user is defined as a reaction profile distribution; the receiving unit inputs rating value data of the user; the reaction profile distribution estimation unit estimates a plurality of reaction profile distribution candidates, which are candidates for the reaction profile distribution, based on the evaluation value data, by using a candidate reaction profile distribution template including a type of the statistical distribution; a step in which the reaction profile distribution evaluation unit selects a reaction profile distribution suitable for expressing the reaction profile of the user from among a plurality of reaction profile distribution candidates estimated using the rating value data; A reaction profile evaluation method comprising:

11. A reaction profile evaluation program for causing a computer to realize the functions of the reception unit, the reaction profile distribution estimation unit, and the reaction profile distribution evaluation unit of the reaction profile evaluation apparatus according to claim 1.

Images