Personality estimation system

Abstract

To enable the estimation of a person's character with practical accuracy without using a language.SOLUTION: According to an embodiment, a character estimation system includes a measurement unit for recording measurement information corresponding to time series of a control input inputted from an input device to an inspection model for solving a problem by continuous or intermittent manual operations of the input device by a subject, and an estimation unit for calculating an index representing the character of the subject by inputting the measurement information to a neural network with the measurement information as input information and with the index representing the character of the subject as output information.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a personality estimation system, a personality estimation method, and a personality estimation program.

Background Art

[0002] It is known that a person's personality (individuality) correlates with work performance. In fact, many companies grasp a person's personality through personality tests and the like during the recruitment process, and use it for recruitment judgments and assignments.

[0003] Therefore, conventionally, a system has been proposed that obtains an index representing a person's personality based on the characteristics of the trial history of control inputs repeatedly given from a subject by operating an input device for an inspection model to which a problem is given so that personalities such as a person's character, ability, thinking method, and movement strategy can be evaluated and determined using an index (see, for example, Patent Document 1). In this system, since the use of language can be made unnecessary by using a mechanical model as an inspection model, it is said that a person's personality can be grasped more uniformly and objectively.

[0004] On the other hand, as a method of classifying and quantifying the characteristics of a person's personality, a method of representing a person's personality by five factors of openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism tendency is known (see, for example, Non-Patent Document 1). Note that conscientiousness may be translated as honesty, or the factor itself may be replaced by another factor. This classification method is called the Big Five or the Five-Factor Model (FFM), and typically, a person's personality is measured by a simple test in which the subject answers a 5-choice questionnaire test or a 10-item questionnaire called the TIPI (Ten Item Personality Inventory) on a 7-point scale for each.

[0005] Furthermore, as an advanced method of the Big Five personality traits, a method has been proposed that uses a neural network to determine five personality factors from the handwritten handwriting of subjects (see, for example, Non-Patent Document 2). [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2012-083426 [Non-patent literature]

[0007] [Non-Patent Document 1] Rachid Laajaj et al., “Challenges to capture the big five personality traits in non-WEIRD populations”, SCIENCE ADVANCES, 2019;5: eaaw5226, 10 July 2019, [Retrieved January 16, 2022], Internet<URL: https: / / www.science.org / doi / pdf / 10.1126 / sciadv.aaw5226> [Non-Patent Document 2] Mihai Gavrilescu et al., “Predicting the Big Five personality traits from handwriting”, Gavrilescu and Vizireanu EURASIP Journal on Image and Video Processing (2018) 2018:57, [Retrieved January 16, 2020], Internet <URL: https: / / jivp-eurasipjournals.springeropen.com / track / pdf / 10.1186 / s13640-018-0297-3.pdf > [Overview of the project] [Problems that the invention aims to solve]

[0008] However, the system described in Patent Document 1 requires repeated trials on a task and the characteristics of the trial history to determine an indicator representing a person's individuality, thus requiring effort and time from the subject.

[0009] Furthermore, conventional personality estimation methods based on the Big Five personality traits have a problem: since language ability is essential, estimation may not be possible at all if language ability is insufficient, or even if estimation is possible, the results may be unreliable. For example, in countries and regions with low literacy rates, it is difficult to conduct written tests or obtain handwriting samples. Also, when estimating the personality of young subjects, such as infants, whose language ability is uncertain, estimation may not be possible at all, or even if estimation is possible, the accuracy of the estimation may be questionable.

[0010] In fact, research has shown that conventional personality estimation methods based on the Big Five personality traits cannot estimate personality with practical accuracy unless the subjects have reached a certain level of education and acquired the necessary cognitive abilities.

[0011] Therefore, the present invention aims to enable the estimation of a person's personality with practical accuracy without using language. [Means for solving the problem]

[0012] The personality estimation system according to an embodiment of the present invention is a test model that solves a problem through continuous or intermittent manual operation of an input device by a subject. In one trial A measurement unit records measurement information corresponding to the time series of control inputs received from the input device, and a neural network takes the measurement information as input information and outputs an index representing the subject's personality as output information, Based on the measurement information in a single trial The system includes an estimation unit that calculates an index representing the personality of the subject.

[0013] Also, the personality estimation method according to an embodiment of the present invention is for an inspection model that solves problems by continuous or intermittent manual operations of an input device by a subject In one trial a step of recording measurement information corresponding to the time series of control inputs input from the input device, and inputting the measurement information into a neural network that uses the measurement information as input information and an index representing the personality of the subject as output information, Based on the measurement information in a single trial a step of calculating an index representing the personality of the subject The computer executes and is such.

[0014] Also, the personality estimation program according to an embodiment of the present invention causes a computer to perform, for an inspection model that solves problems by continuous or intermittent manual operations of an input device by a subject In one trial a step of recording measurement information corresponding to the time series of control inputs input from the input device and inputting the measurement information into a neural network that uses the measurement information as input information and an index representing the personality of the subject as output information, Based on the measurement information in a single trial a step of calculating an index representing the personality of the subject.

Brief Description of Drawings

[0015] [Figure 1] Configuration diagram of a personality estimation system according to an embodiment of the present invention. [Figure 2] Diagram showing a detailed configuration example of the neural network shown in FIG. 1. [Figure 3] Diagram showing the estimation accuracy of the personality of a subject by a neural network having the configuration shown in FIG. 2.

Modes for Carrying Out the Invention

[0016] A personality estimation system, a personality estimation method, and a personality estimation program according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0017] (Configuration and Function of Personality Estimation System) FIG. 1 is a configuration diagram of a personality estimation system according to an embodiment of the present invention.

[0018] The personality estimation system 1 is a system that obtains an index representing the personality of a subject without using language. The personality estimation system 1 includes a motion measurement system 2, a personality estimation system 3, an input device 4, and a display 5.

[0019] The motion measurement system 2 and the personality estimation system 3 can be constructed by causing a computer 6, which is a kind of electronic circuit, to load a personality estimation program. In other words, by causing the computer 6 to load the personality estimation program, the computer 6 can be made to function as the motion measurement system 2 and the personality estimation system 3. Of course, a plurality of computers, a plurality of displays, and / or a plurality of input devices may be used to constitute the personality estimation system 1.

[0020] The motion measurement system 2 measures a time series of index values representing the motion of a subject based on the time series of control inputs input from the input device 4 as operation information of the input device 4 referring to the test model 7 to which a task is given, and has a function as a measurement unit that records the measured time series of index values as measurement information 8. On the other hand, the personality estimation system 3 has a function as an estimation unit that calculates the value of an index representing the personality of the subject by information processing using a neural network 3A based on the measurement information 8 representing the motion of the subject recorded by the motion measurement system 2.

[0021] Therefore, it can be said that the personality estimation program is a program that causes the computer 6 to execute a process of recording measurement information 8 corresponding to the time series of control inputs input from the input device 4 by referring to the test model 7, and a process of calculating an index representing the personality of the subject by information processing based on the measurement information 8. Such a personality estimation program having such functions can also be distributed as a program product by recording it on an information recording medium or providing it via a network.

[0022] The personality estimation system 3 calculates an index value representing the subject's personality, which can be visually displayed on display 5 as a graph, such as the radar chart 9 exemplified in Figure 1. Of course, it is also possible to simply display a single index value representing the subject's personality, or to display multiple index values ​​in a table format on display 5.

[0023] The following describes the detailed functions of the exercise measurement system 2 and the personality estimation system 3.

[0024] The motion measurement system 2 displays a game-like test model 7 on the display 5, in which the subject solves a task by continuously or intermittently manually operating the input device 4. This encourages the subject to manually operate the input device 4 in order to solve the task while looking at the test model 7 displayed on the display 5, and the system has the function of recording measurement information 8 corresponding to the time series of control inputs received from the input device 4.

[0025] Therefore, if the inspection model 7 is a model that follows natural laws, such as a dynamics model, the motion measurement system 2 is configured to simulate the inspection model 7 in response to manual operation of the input device 4 and to prompt further manual operation of the input device 4 by displaying the simulation results on the display 5 in real time.

[0026] If the task of test model 7 is solved by continuous operation of input device 4, then input device 4 will be operated continuously by the subject. On the other hand, if the task of test model 7 is solved by intermittent operation of input device 4, then input device 4 will be operated intermittently by the subject.

[0027] However, in order to efficiently examine the subject's personality in the shortest possible time, it is preferable to collect and record as much useful data as possible per unit time. Therefore, it is advantageous to determine the test model 7 so that one or more variables that change continuously over time can be recorded as measurement information 8 by continuous manual operation of the input device 4.

[0028] In other words, if the inspection model 7 is a model in which the value of a variable is continuously determined at each time point by manual operation of the input device 4, rather than a model in which the value of a variable is determined by continuous or intermittent operation of the input device 4 over a certain period of time, then it becomes possible to collect the measurement information 8 necessary for inspection in a shorter time.

[0029] In this case, the values ​​of the variables determined at each time point by the manual operation of the input device 4 are recorded in the motion measurement system 2 as measurement information 8. In other words, the measurement information 8 can be recorded not as discrete time-series data, but as time-series data consisting of continuous variables of one or more dimensions.

[0030] In the example shown in Figure 1, a dynamic model in which one end of an inverted pendulum is rotatably connected to a cart that moves left and right by manual operation of the input device 4 is displayed on the display 5 as the inspection model 7. In other words, the inspection model 7 is a cart-type inverted pendulum in which the upper mass point is the free end and the lower end can be moved horizontally together with the cart by manual operation of the input device 4.

[0031] Therefore, by setting the task of keeping at least a part of the inverted pendulum, such as the free end, within a predetermined range, and performing a mechanical simulation based on the equations of motion, the simulation results of the test model 7 corresponding to the manual operation of the input device 4 can be continuously displayed on the display 5. This encourages the subject being tested for personality to manually operate the input device 4, and the value of the horizontal error Δx(t) from the top dead center of the inverted pendulum's point mass, which is determined at each time t by the manual operation of the input device 4, can be recorded as measurement information 8. That is, as illustrated in Figure 1, measurement information 8 representing the relationship between the error Δx(t) and time t can be acquired by the motion measurement system 2, and the acquired measurement information 8 can be provided to the personality estimation system 3.

[0032] Of course, any model can be adopted as the test model 7, not limited to the dynamic model exemplified in Figure 1. In that case, if a model that can collect continuous variables as measurement information 8 is adopted, it becomes possible to conduct an efficient personality test of the subject. In other words, it becomes possible to efficiently collect measurement information 8 in the motion measurement system 2.

[0033] The personality estimation system 3 is configured to calculate an index representing the subject's personality by inputting the measurement information 8, which is acquired by the motor measurement system 2, into a neural network 3A, which outputs an index representing the subject's personality.

[0034] Representative indicators of a subject's personality include the values ​​of the five factors used in the Big Five personality assessment. As mentioned above, the five factors are often defined as extraversion, agreeableness, conscientiousness, neuroticism, and openness, although conscientiousness is sometimes expressed as conscientiousness. Therefore, conscientiousness is essentially the same factor as conscientiousness. Furthermore, there are various variations, such as surgency being used instead of extraversion, dependability being used instead of conscientiousness or conscientiousness, emotional stability being used instead of neuroticism, or culture or intelligence being one of the five factors.

[0035] Therefore, when representing a subject's personality with five factors, the neural network 3A can calculate at least one value of the five factors as an index representing the subject's personality.

[0036] Figure 2 shows a detailed configuration example of the neural network 3A shown in Figure 1.

[0037] While the values ​​of the five factors representing a subject's personality can, in principle, be calculated using a simple neural network 3A, given that the measurement information 8 is time-series data and to obtain sufficient accuracy, it is more practical to use a multi-layer neural network that includes convolutional processing, and it is preferable to use a deep convolutional neural network (Deep-CNN).

[0038] A neural network 3A consists of an input layer, a hidden layer, and an output layer, and a neural network 3A with two or more hidden layers is often defined as a deep neural network. Furthermore, a convolutional neural network (CNN) is a neural network that includes convolutional layers that perform convolutional processing. Therefore, a deep-CNN can be defined as a neural network that includes convolutional layers and has two or more hidden layers.

[0039] By inputting measurement information 8 consisting of time-series data into a Deep-CNN, it is possible to calculate indicators representing the subject's personality, such as the values ​​of five factors, with practical accuracy. A typical Deep-CNN has convolutional layers and pooling layers between the input layer and the output layer. However, when using measurement information 8 consisting of one-dimensional time-series data as input, as shown in Figure 2, a Deep-CNN is constructed with at least two convolutional layers C1 and C2 (first and second), a first pooling layer P1, four convolutional layers C3 to C6 (third to sixth), and a second pooling layer P2 from the input layer side, and the output layer is a softmax layer S. It was confirmed that when such a Deep-CNN is constructed for five factors and each factor is calculated using the corresponding Deep-CNN, an estimation accuracy of 90% or more can be achieved for all five factors.

[0040] Figure 3 shows the accuracy of personality estimation of a subject by neural network 3A, which has the configuration shown in Figure 2.

[0041] Figure 3(A) is a graph showing the relationship between the number of epochs of the Deep-CNN shown in Figure 2 and the estimation error of extroversion, one of the five factors, while Figure 3(B) is a graph showing the relationship between the number of epochs of the Deep-CNN shown in Figure 2 and the estimation accuracy of extroversion, one of the five factors. Specifically, the horizontal axis in Figure 3(A) and Figure 3(B) represents the number of epochs, the vertical axis in Figure 3(A) represents the estimation error of extroversion, and the vertical axis in Figure 3(B) represents the estimation accuracy (%) of extroversion.

[0042] The 5-factor scores were measured by having 100 subjects complete the TIPI-J test, a Japanese version of the TIPI test, which consists of 10 questions each answered on a 7-point scale. In the case of the TIPI-J test, the scores for each of the 5 factors are dimensionless discrete integer values ​​between 2 and 14.

[0043] Meanwhile, the 100 subjects also estimated scores for the five factors using Deep-CNN, as shown in Figure 2. The measurement information 8, which serves as input to Deep-CNN, was obtained using the inverted pendulum test model 7, as exemplified in Figure 1.

[0044] Furthermore, the first to third convolutional layers C1, C2, and C3, which constitute the Deep-CNN, were configured with 32 filters, a filter size of 6, and the ReLU (Rectified Linear Unit) activation function, while the fourth to sixth convolutional layers C4, C5, and C6 were configured with 64 filters, a filter size of 3, and the ReLU activation function.

[0045] Next, we calculated the estimation error between the measurements obtained from the TIPI-J test and the estimates obtained from Deep-CNN. The estimation error was expressed as the mean squared error between the measurements and estimates for a predetermined number of people. Since the 5-factor scores obtained from the TIPI-J test are dimensionless integers, the estimation error for the scores is also a dimensionless value.

[0046] Then, using the measured scores of the five factors from the TIPI-J test for some subjects as training data, a Deep-CNN was repeatedly trained to reduce the estimation error. In other words, the filter coefficients of the convolutional layers C1 to C6 that make up the Deep-CNN were optimized to reduce the estimation error between the measured values ​​from the TIPI-J test and the estimated values ​​from the Deep-CNN.

[0047] As a result, as shown in Figure 3(A), it was confirmed that the estimation error for extraversion, one of the five factors, decreased as the number of epochs increased, approaching zero. Conversely, as shown in Figure 3(B), the estimation accuracy of extraversion increased as the number of epochs increased, and it was confirmed that it exceeded 90% when the number of epochs reached 1000. The same was true for agreeableness, conscientiousness, neuroticism, and openness.

[0048] In addition, we also verified the estimation error and accuracy of the 5-factor scores by calculating the estimation error between the 5-factor scores of 20 participants (for whom the 5-factor scores measured by the TIPI-J test were not used as training data) and the estimated 5-factor scores by Deep-CNN, as well as the estimation accuracy of the 5-factor scores by Deep-CNN.

[0049] As a result, subjects who did not use the measured scores of the five factors as training data showed similar results to those of subjects who did use the measured scores as training data, as shown in the graph in Figure 3, confirming that the estimation accuracy for all five factors exceeded 90%. In other words, it was confirmed that the scores of the five factors can be estimated with an estimation accuracy of over 90% using the Deep-CNN configuration shown in Figure 2. Specifically, the estimation accuracy increased to 98.3% for extraversion, 98.2% for agreeableness, 98.2% for conscientiousness, 98.2% for neuroticism, and 97.5% for openness.

[0050] Based on these results, it is considered that if the Deep-CNN configuration has at least eight hidden layers as shown in Figure 2, and the input information of the Deep-CNN is measurement information 8 consisting of one-dimensional time-series data, while the output information of the Deep-CNN is a single index representing the subject's personality, the estimation accuracy of the index can be increased to over 90%. In other words, it becomes possible to estimate the subject's personality with practical accuracy without using language through a simple simulation using the test model 7.

[0051] The scores for the five factors corresponding to the TIPI-J test, estimated by Deep-CNN, are discrete integer values ​​ranging from 2 to 14. Therefore, displaying them in a radar chart (as exemplified in Figure 1) makes them easier to understand. Of course, the scores can also be displayed in a table format.

[0052] To this end, the personality estimation system 3 is equipped with a function to display indicator values ​​such as the five factors representing the subject's personality on the display 5, as described above. Since the observer of the indicator values ​​representing the subject's personality may be different from the subject themselves, the display for showing the test model 7 and the display for showing the indicator values ​​representing the subject's personality may be separate.

[0053] Incidentally, it has been reported that the values ​​of the five factors representing a person's personality vary depending on nationality and race. Therefore, it is possible to configure a neural network 3A, such as Deep-CNN, to output different information depending on at least one of nationality and race. In that case, the neural network 3A can be trained using measurement values ​​taken in tests on multiple subjects of different nationalities or races as training data. In this case, the neural network 3A for determining the five factors representing the subject's personality will be prepared separately for each nationality and / or race, making it possible to calculate a more accurate index representing the subject's personality according to at least one of nationality and race.

[0054] (Personality estimation method) Next, we will explain the flow of the personality estimation method using personality estimation system 1.

[0055] When the personality estimation system 1 estimates a subject's personality, the motion measurement system 2 displays a test model 7 on the display 5, which has a task such as an unstable inverted pendulum on a cart that must remain within a predetermined range, as exemplified in Figure 1. Then, when the subject operates the input device 4 to solve the task while referring to the test model 7 displayed on the display 5, the motion measurement system 2 performs a simulation based on the control input from the input device 4 and updates the state of the test model 7 in real time.

[0056] As a result, the subject manually operates the input device 4 continuously or intermittently on the test model 7 to which they are to solve a task, and the motion measurement system 2 collects a time series of control inputs from the input device 4. The motion measurement system 2 then updates the parameter values ​​for outputting the test model 7 to the display 5 in real time according to the time series of control inputs from the input device 4, and records the time series of variables corresponding to the time series of control inputs from the input device 4 as motion measurement information 8 of the subject.

[0057] Once measurement information 8 of a sufficiently long duration is recorded, the measurement information 8, consisting of a time series of variables, is provided from the motion measurement system 2 to the personality estimation system 3. The personality estimation system 3 then inputs the measurement information 8, consisting of a time series of variables, into a neural network 3A such as a Deep-CNN. The neural network 3A then calculates an index representing the subject's personality.

[0058] As a concrete example, the position error Δx(t) of the inverted pendulum shown in Figure 1 can be input as measurement information 8 into a Deep-CNN having the configuration shown in Figure 2, and the five Big Five factors can be determined. The values ​​of the five factors obtained can be displayed on the display 5 in the form of a radar chart 9 as exemplified in Figure 1. This allows for an understanding of the subject's personality.

[0059] (effect) The personality estimation system 1, personality estimation method, and personality estimation program described above record the subject's movements as measurement information 8 represented as time-series data by operating an input device 4 that references an examination model 7 such as a dynamic model for which a task has been set, and estimate an index value representing the subject's personality using a neural network 3A that takes the measurement information 8 as input data.

[0060] Therefore, the personality estimation system 1, personality estimation method, and personality estimation program can estimate a person's personality without using language. As a result, it is possible to estimate a subject's personality with good accuracy without being affected by the subject's educational level or cognitive ability. Specifically, it is possible to estimate the personality with good accuracy not only for young subjects whose reading and writing abilities are not necessarily sufficient, but also for subjects from countries and regions with low literacy rates, and for subjects who speak different languages.

[0061] Furthermore, the automation and online implementation of personality tests will become easier. For example, it will be possible to estimate the personality of subjects using the internet.

[0062] (Other embodiments) Although specific embodiments have been described above, these embodiments are merely examples and do not limit the scope of the invention. The novel methods and apparatus described herein can be embodied in various other forms. Furthermore, various omissions, substitutions, and modifications can be made in the forms of methods and apparatus described herein, without departing from the spirit of the invention. The attached claims and equivalents include such various forms and modifications as being encompassed within the scope and spirit of the invention.

[0063] For example, in the embodiment described above, the case where the inspection model 7 for acquiring measurement information 8 by operating the input device 4 is a virtual model displayed on the display 5 was explained, but an actual model may also be used as the inspection model 7. In that case, the input device 4 is configured to be able to operate the actual model in order to solve the set task, and the motion measurement system 2 is configured to acquire the time series of control inputs input from the input device 4 to the inspection model 7. As a specific example, an inverted pendulum that actually falls over on a cart may be manufactured and used as the inspection model 7. If the inspection model 7 is constructed in hardware, the creation of a program for performing simulations, etc., can be omitted.

[0064] Furthermore, while the Big Five personality traits were used as a specific example of an indicator representing a person's personality in the embodiments described above, other indicators may also be used to represent a person's personality. For example, dementia such as Alzheimer's disease could be considered a type of personality, and an indicator representing the type and degree of dementia could be obtained using a non-verbal testing model and a neural network such as Deep-CNN. Similarly, developmental disorders such as Asperger's syndrome and other autism spectrum disorders (ASD), learning disorders (LD), and attention deficit hyperactivity disorder (ADHD) could be considered a type of personality, and an indicator representing the type and degree of developmental disorder could be obtained using a non-verbal testing model and a neural network such as Deep-CNN. [Explanation of Symbols]

[0065] 1. Personality Estimation System 2. Motion Measurement System 3. Personality Estimation System 3A Neural Network 4 Input devices 5 displays 6 Computers 7. Test Model 8. Measurement Information 9 Radar Chart C1, C2, C3, C4, C5, C6 Convolutional Layers P1, P2 pooling layers S Softmax Layer

Claims

1. A measurement unit records measurement information corresponding to the time series of control inputs received from an input device during a single trial, for an examination model in which a subject solves a problem by continuously or intermittently manually operating an input device. An estimation unit calculates an index representing the subject's personality based on the measurement information in a single trial by inputting the measurement information into a neural network that takes the measurement information as input information and outputs an index representing the subject's personality as output information. A personality estimation system having the following characteristics.

2. The personality estimation system according to claim 1, wherein the estimation unit is configured to input the measurement information into the neural network to calculate at least one value of the plurality of factors when the personality of the subject is represented by a plurality of factors as the index.

3. The personality estimation system according to claim 1 or 2, wherein the measurement unit is configured to prompt manual operation of the input device by displaying the simulation results of the inspection model corresponding to the manual operation of the input device on a display, and to record the values ​​of variables determined at each time by the manual operation of the input device as measurement information.

4. The personality estimation system according to claim 3, wherein the measuring unit is configured to prompt manual operation of the input device by continuously displaying on a display the simulation results corresponding to the manual operation of the input device of an inspection model having the problem of keeping at least a part of the inverted pendulum, in which the upper end mass point is a free end and the lower end can be moved horizontally by manual operation of the input device, within a predetermined range, and to record the value of the error from the top dead center of the inverted pendulum at each time determined by the manual operation of the input device at each time as measurement information.

5. The personality estimation system according to any one of claims 1 to 4, wherein the estimation unit is configured to calculate an index representing the personality of the subject according to at least one of the nationality and race, using a plurality of neural networks consisting of at least one of a plurality of neural networks prepared separately for nationality and a plurality of neural networks prepared separately for race, so as to output different output information according to at least one of the nationality and race.

6. A test model in which a subject solves a task by continuously or intermittently manually operating an input device, comprising the steps of recording measurement information corresponding to the time series of control inputs received from the input device during a single trial, The steps include: inputting the measurement information into a neural network that takes the measurement information as input information and outputs an index representing the subject's personality, thereby calculating an index representing the subject's personality based on the measurement information in one trial; A personality estimation method performed by a computer.

7. On the computer, A test model in which a subject solves a task by continuously or intermittently manually operating an input device, comprising the steps of recording measurement information corresponding to the time series of control inputs received from the input device during a single trial, and A step of inputting the measurement information into a neural network that takes the measurement information as input information and outputs an index representing the subject's personality, thereby calculating an index representing the subject's personality based on the measurement information in one trial. A personality estimation program that performs the following actions.

Images