Anchoring personal characteristics estimation device

Abstract

To allow for estimating individual characteristics on anchoring by avoiding difficulty of experiments on cognitive bias called the anchoring effect that makes it impossible to guaranty independence if the same trial is repeated on the same experiment participant.SOLUTION: Answers received by a first reception unit and values indicative of anchors are applied to a first estimation mode, second estimation model, and third estimation model to estimate a first personal characteristic value and a second personal characteristic value.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an anchoring individual characteristic estimation device that predicts an individual characteristic related to anchoring. [Background technology]

[0002] (Prior art 1)

[0003] As shown in Non-Patent Document 1, the phenomenon in which information such as numerical values ​​given in advance strongly influences subsequent estimations and decision-making is widely known as a cognitive bias called the anchoring effect.

[0004] (Prior art 2)

[0005] Although many studies have been conducted to examine the anchoring effect under various conditions, there has been little research to quantitatively evaluate the mechanism of the anchoring effect.

[0006] Non-Patent Document 2 describes a model that uses Bayesian updating to explain the anchoring effect when estimating size. The technology described in Non-Patent Document 2 is applicable to cases where there is a relatively large amount of existing knowledge. However, there are problems in that it cannot be applied to cases where there is little knowledge, and there are problems in that the more the anchor differs from the prediction observed in actual experimental data, the greater the variance in the results, and the more inconsistent the results are.

[0007] (Prior art 3)

[0008] In order to solve the problems with the model described in Non-Patent Document 2, Non-Patent Document 3 describes an extended model that adds preprocessing using a logarithmic function and introduces degrees of freedom to the model described in Non-Patent Document 2. This model makes it possible to explain a large amount of experimental data on size estimation. However, it has the problem that it cannot be applied to estimation of proportions other than size estimation. Furthermore, it cannot explain the phenomenon where the effect disappears for anchors that deviate significantly from the prediction.

[0009] (Prior art 4)

[0010] Non-Patent Document 4 shows that in proportion estimation, the model described in Non-Patent Document 2 can be applied by using a logit function instead of a logarithmic function for preprocessing. [Prior art documents] [Non-patent literature]

[0011] [Non-Patent Document 1] Kahneman, D. and Tversky, A., Judgment under uncertainty: Heuristica and biases, Science, 185, 1124-1131 (1974) [Non-patent document 2] Turner, BB. and Schley, DR, The anchor integration model: A descriptive model of anchoring effects, Cognitive Psychology, 90, 1-47 (2016). [Non-patent document 3] Isao Ozawa and Takashi Takekawa, "Prior bias and anchoring effect in magnitude estimation - An example using proper nouns with little information content," IEICE Transactions on Information and Communication Engineers, J101-D, 405-413 (2018). [Non-patent document 4] Tomoaki Hamada and Takashi Takekawa, Analysis and Generalization of the Anchoring Effect in Proportion Estimation Using a Bayesian Updating Model, JSAI2021 (2021). Summary of the Invention [Problem to be solved by the invention]

[0012] The conventional models shown in Prior Art 1 to Prior Art 4 are models that interpret experimental results by assuming that the knowledge and behavioral patterns of experimental participants are constant. For this reason, they can handle only average behavior without considering individual differences, but they cannot predict individual behavior. This is due to the difficulty of conducting experiments on the anchoring effect, where independence cannot be guaranteed when the same trial is repeated with the same experimental participant.

[0013] The present invention aims to avoid the difficulty of conducting experiments on the anchoring effect, in which independence cannot be guaranteed when the same trial is repeatedly conducted on the same experimental participant, and to make it possible to estimate individual characteristics related to anchoring. [Means for solving the problem]

[0014] a first estimation model that updates a problem-specific knowledge and confidence parameter value for a problem before the anchor is presented to a problem-specific knowledge and confidence parameter value for a problem after the anchor is presented, in accordance with a first personal characteristic value related to the anchor value and the influence of the anchor; a second estimation model that predicts an individual's answer to a problem without an anchor in accordance with the problem-specific knowledge and confidence parameter value before the update by the first estimation model, and predicts an individual's answer to a problem with an anchor in accordance with the problem-specific knowledge and confidence parameter value updated by the first estimation model; and a third estimation model that predicts a problem-specific knowledge and confidence parameter value in accordance with a second personal characteristic value related to knowledge and confidence, and an analysis processing unit that applies values ​​indicating the answers and anchors received by the first reception unit to the first estimation model, the second estimation model, and the third estimation model to estimate the first personal characteristic value and the second personal characteristic value.

[0015] The second aspect is an anchoring personal characteristic estimation device in the first aspect, wherein the third estimation model is an estimation model that predicts a problem-specific knowledge / confidence parameter value according to the second personal characteristic value and the characteristic value of the problem, and the analysis processing unit applies values ​​indicating the answer and anchor received by the first receiving unit to the first estimation model, the second estimation model, and the third estimation model to estimate the first personal characteristic value, the second personal characteristic value, and the characteristic value of the problem.

[0016] A third aspect is an anchoring personal characteristic estimation device in which, in the first or second aspect, the second estimation model is configured with a probability density function that obtains individual responses according to a normal distribution with a mean that indicates the individual's average knowledge and a variance that is the inverse of a value that indicates the individual's confidence.

[0017] A fourth aspect is an anchoring personal characteristic estimation device according to any of the first to third aspects, wherein the first estimation model includes: a probability density function that acquires a value indicating the anchor's influence according to a gamma distribution with the anchor value and the first personal characteristic value as parameters; a knowledge average value calculation formula that acquires a value indicating the updated individual's knowledge average according to the value indicating the anchor's influence acquired by the probability density function and a value indicating the individual's knowledge average before the update; and a confidence value calculation formula that acquires a value indicating the updated individual's confidence according to the value indicating the anchor's influence acquired by the probability density function and the value indicating the individual's confidence before the update.

[0018] A fifth aspect is an anchoring personal characteristic estimation device according to any one of the first to fourth aspects, wherein the second personal characteristic value comprises a knowledge amount characteristic value indicating the amount of knowledge and a confidence characteristic value indicating the magnitude and variance of confidence, and the third estimation model comprises a probability density function for acquiring a knowledge average value, which acquires a value indicating the individual's knowledge average according to a normal distribution whose variance includes the inverse of the knowledge amount characteristic value, and a probability density function for acquiring a confidence value, which acquires a value indicating the individual's confidence according to a gamma distribution whose parameter includes the confidence characteristic value.

[0019] The sixth aspect is an anchoring personal characteristic estimation device in the fifth aspect, wherein the analysis processing unit estimates the first personal characteristic value and the second personal characteristic value using a probability density function for first personal characteristic value acquisition that acquires the first personal characteristic value according to a gamma distribution, a probability density function for knowledge amount characteristic value acquisition that acquires the knowledge amount characteristic value according to a gamma distribution, and a probability density function for confidence characteristic value acquisition that acquires the confidence characteristic value according to a gamma distribution.

[0020] A seventh aspect is a trained model trained using actual values ​​as training data, with values ​​indicating an individual's answer and an anchor as input data, and a first individual characteristic value related to the influence of the anchor and a second individual characteristic value related to knowledge and confidence as output data, and is an anchoring personal characteristic estimation device comprising: a personal characteristic estimation unit that inputs data indicating an individual's answer and data indicating the anchor and outputs the first personal characteristic value and the second personal characteristic value of the individual; a second reception unit that accepts individual's answers to questions without anchors and questions with anchors; and a processing execution unit that inputs the data indicating the individual's answer and the data indicating the anchor accepted by the second reception unit to the personal characteristic estimation unit and executes processing to cause the personal characteristic estimation unit to output the first personal characteristic value and the second personal characteristic value of the individual accepted by the second reception unit. [Effects of the Invention]

[0021] According to the first to sixth aspects, it is possible to avoid the difficulty of experiments on the anchoring effect, in that independence cannot be guaranteed when the same trial is repeatedly conducted on the same experimental participant, and to estimate individual characteristics related to anchoring.

[0022] According to the seventh aspect, by inputting data indicating an individual's answers into the trained model, the anchoring characteristics of that individual can be estimated. [Brief explanation of the drawings]

[0023] [Figure 1] FIG. 1 is a block diagram showing the functional configuration of an anchoring personal characteristic estimation device according to an embodiment. [Figure 2] FIG. 2 is a diagram illustrating a network configuration of the system according to the embodiment, and is a network configuration diagram for realizing the functional configuration of FIG. [Figure 3] FIG. 3 is a diagram illustrating the hardware configuration of the administrator terminal, the experiment collaborator terminal, the general user terminal, and the server according to the embodiment, and is a hardware configuration diagram for realizing the functional configuration of FIG. [Figure 4A] FIG. 4A is a diagram illustrating an estimation result according to the embodiment. [Figure 4B] FIG. 4B is a diagram illustrating an estimation result according to the embodiment. [Figure 4C] FIG. 4C is a diagram illustrating an estimation result according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0024] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0025] FIG. 1 is a block diagram showing the functional configuration of an anchoring personal characteristic estimation device according to an embodiment.

[0026] The anchoring personal characteristic estimation device of the embodiment is configured to include a first reception unit 100, a question storage unit 150, an answer storage unit 200, an analysis processing unit 300, a second reception unit 400, a personal characteristic estimation unit 500, and a processing execution unit 600.

[0027] FIG. 2 is a diagram illustrating a network configuration of the system according to the embodiment, and is a network configuration diagram for realizing the functional configuration of FIG. As shown in Figure 2, the system of the embodiment is composed of an administrator terminal 10, an experimental collaborator terminal 20, a general user terminal 30, a server 40, and a network 50 connecting the terminals 10, 20, 30 and the server 40 so that data can be sent and received between the terminals 10, 20, 30 and the server 40.

[0028] The administrator terminal 10, the collaborator terminal 20, and the general user terminal 30 are information processing devices, such as personal computer terminals, smartphones, tablets, and other mobile device terminals. The network 50 is comprised of the Internet, an intranet, or the like. The server 40 is comprised of a server device or a virtual server built on a cloud computing service.

[0029] FIG. 3 is a diagram illustrating the hardware configuration of the administrator terminal 10, the experiment collaborator terminal 20, the general user terminal 30, and the server 40 in the embodiment, and is a hardware configuration diagram for realizing the functional configuration of FIG.

[0030] As shown in Figure 3, in the administrator terminal 10, the experimental collaborator terminal 20, the general user terminal 30 and the server 40, a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input device 16, a display device 17, a communication I / F 18 and an external storage device 19 are connected to each other so that they can communicate with each other via a system bus 15. The CPU 11 is a central processing unit that executes various programs and controls each device connected to the system bus 15. That is, the CPU 11 reads a program from the ROM 12 or the storage 14 and executes the program using the RAM 13 as a work area. The CPU 11 controls each device connected to the system bus 15 and performs various arithmetic processing in accordance with the program recorded in the ROM 12 or the storage 14. The ROM 12 or the storage 14 stores a BIOS (Basic Input / Output System) and an OS (Operating System), which are control programs executed by the CPU 11, various programs such as an analysis processing program that can be read and executed by a computer to realize this embodiment, and various necessary data.

[0031] The ROM 12 stores various control programs and various data. The RAM 13 functions as the main memory, work area, etc. of the CPU 11 and temporarily stores programs or data as a working area. The storage 14 is composed of an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various programs including the BIOS and OS, and various data.

[0032] The input device 16 includes a pointing device such as a mouse, a keyboard, a reading device such as a scanner, and is used to input various types of information.

[0033] The display device 17 is, for example, a liquid crystal display, and displays various information. The display device 17 may also function as the input device 16 by adopting a touch panel system.

[0034] The communication interface 18 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, Wi-Fi (registered trademark), etc. The communication interface 18 connects to a network and controls the transmission and reception of data.

[0035] The external storage device 19 is configured with various types of memory cards such as USB memory, HDD, SSD, or other external storage media that can be detachably connected.

[0036] (First Reception Section)

[0037] For example, 20 questions are prepared for participants to answer in the experiment. The questions include questions without anchors and questions paired with the presentation of anchors.

[0038] A specific example of a question paired with an anchor is "Is Tokugawa Ieyasu's height over 160cm? Or is it under 160cm?", and the question is "How tall is Tokugawa Ieyasu?". In this case, the number "160" is the anchor.

[0039] A specific example of an unanchored question is, "What percentage of women entered and left Japan in 2000?"

[0040] A plurality of participants, for example, 500 participants, answer the questions. Questions are created on the administrator terminal 10, and question data is stored in the question storage unit 150 of the server 40 in an accessible manner.

[0041] The first accepting unit 100 accepts individual answers to the above-mentioned unanchored and anchored questions. For example, the answers are accepted via the display screen of the experiment collaborator terminal 20.

[0042] The collaborator accesses the question storage unit 150 of the server 40 via the collaborator terminal 20, thereby reading out question data from the question storage unit 150 of the server 40. The collaborator operates an input device 16 such as a keyboard to input answers to the questions on the screen of the display device 17 of the collaborator terminal 20.

[0043] In the following, the suffix "i" is given to each participant in order to identify them individually, and the suffix "j" is given to each question in order to identify them individually.

[0044] Data indicating the answers input to the experiment participant terminal 20 is stored in the answer storage unit 200 of the server 40 via the network 50 .

[0045] One example is receiving answers via the display screen of the collaborator terminal 20. Answers may also be received via the display screen of the administrator terminal 10, or via the external storage device 19. For example, answer sheets with answers written on them may be collected from the collaborators, and the administrator may operate the input device 16, such as a keyboard, to receive the answers on the screen of the administrator terminal 10. Alternatively, answers may be received by connecting the external storage device 19, in which data indicating the answers is stored, to the administrator terminal 10.

[0046] (Answer storage section)

[0047] Data indicating the answer received by the first receiving unit 100 is stored in the answer storage unit 200 of the server 40.

[0048] (Analysis processing unit)

[0049] The analysis processing unit 300 executes an analysis process to find the characteristic values ​​of the individual experiment participants and the characteristic values ​​of the problem based on the data stored in the response storage unit 200. The administrator starts the analysis processing program stored in the ROM 12 or storage 14 via the administrator terminal 10 to execute the analysis processing.

[0050] In the following, the normal distribution and the gamma distribution are defined as in the following equations (20) and (21).

[0051] Normal distribution:

[0052]

number

[0053] Gamma distribution:

[0054]

number

[0055] Γ(α) is the gamma function.

[0056] The individual characteristic values ​​are divided into the first individual characteristic value ηi and the second individual characteristic value θi. The first individual characteristic value ηi indicates the individual's characteristics related to the anchor's influence. The second individual characteristic value θi indicates the individual's characteristics related to knowledge and confidence.

[0057] The first individual characteristic value ηi indicates the degree to which the anchor influences the individual. The greater the influence of the anchor, the greater the first individual characteristic value ηi.

[0058] The second individual characteristic value θi is composed of a knowledge amount characteristic value κi indicating the amount of knowledge, and confidence characteristic values ​​δi and ωi indicating the magnitude and variance of confidence. The confidence characteristic values ​​δi and ωi are composed of an individual difference δi in the magnitude of confidence, and an index ωi indicating whether confidence varies to the same extent for all questions.

[0059] The greater an individual's knowledge, the greater the value of the knowledge characteristic value κi. The greater an individual's confidence compared to other experimental participants, the greater the value of the individual difference in confidence δi. The smaller the deviation in confidence for each of the 20 questions, the greater the value of the index ωi, which shows whether confidence varies to the same extent for each question.

[0060] The first individual characteristic value ηi relating to the anchor's influence is introduced into equations (1), (2), and (3) which show the first estimation model below.

[0061]

number

[0062]

number

[0063]

number

[0064] The first estimation model shown in equations (1), (2), and (3) updates the problem-specific knowledge and confidence parameter values ​​μij, τij for the problem before the anchor was presented to the problem after the anchor was presented, depending on the anchor value aij and the first individual characteristic value ηi related to the anchor's influence. Here, the problem-specific knowledge and confidence parameter values ​​μij, τij consist of a value μij indicating the individual's average knowledge and a value τij indicating the individual's confidence.

[0065] The first estimation model is composed of a probability density function p(γ) in equation (1), a knowledge average value calculation equation in equation (2), and a confidence value calculation equation in equation (3).

[0066] The probability density function p(γ) in equation (1) obtains a value γ indicating the anchor's influence according to a gamma distribution whose parameters include the anchor value aij and the first personal characteristic value ηi. Equation (1) indicates that the value γ indicating the anchor's influence is determined probabilistically depending on the anchor value aij, the anchor's influence ηi held by the individual, the value μij indicating the individual's average knowledge, and the value τij indicating the individual's confidence. Note that in prior art 2, the value γ indicating the anchor's influence was a constant, but this embodiment indicates that the value γ indicating the anchor's influence changes depending on parameters including the anchor value aij and the first personal characteristic value ηi.

[0067] The formula for calculating the average knowledge value in equation (2) obtains the value μij indicating the updated average knowledge value of an individual according to the value γ indicating the anchor's influence obtained from the probability density function p(γ) in equation (1) and the value μij indicating the individual's average knowledge value before the update. Equation (2) shows that when an anchor aij is presented to an experimental participant, the value μij indicating the individual's average knowledge value is updated according to the anchor's direct influence γ in each problem.

[0068] The confidence value calculation formula (3) obtains the updated value τij of an individual's confidence according to the value γ indicating the anchor's influence obtained by the probability density function p(γ) in formula (1) and the value τij indicating the individual's confidence before the update. Formula (3) shows that the value τij indicating the individual's confidence is updated according to the anchor's direct influence γ in each problem.

[0069] The individual's response μ can be predicted using equation (4) below, which shows the second estimation model.

[0070]

number

[0071] The second estimation model in equation (4) is composed of a probability density function that obtains the individual's answer μ according to a normal distribution with the value μij indicating the individual's average knowledge as the "mean" and the inverse τ-1ij of the value τij indicating the individual's confidence as the "variance σ2".

[0072] Equation (4) means that when participant i answers the prediction μ for question j, he answers based on his knowledge p(μ) expressed as a probability distribution. Knowledge p(μ) expressed as a probability distribution is expressed using values ​​μij, which indicate the individual's average knowledge, and values ​​τij, which indicate the individual's confidence.

[0073] For questions without anchors, the individual's answer μ is predicted by applying the pre-update problem-specific knowledge and confidence parameter values ​​μij from the first estimation model (1), (2), and (3) to equation (4) which shows the second estimation model. In other words, when no anchor aij is given, the answer μ is output directly from the pre-update problem-specific knowledge and confidence parameter values ​​μij and τij.

[0074] For questions with anchors, the individual's answer μ is predicted by applying the problem-specific knowledge and confidence parameter values ​​μij and τij updated by the first estimation model (1), (2), and (3) to equation (4) showing the second estimation model. In other words, when an anchor aij is given, the answer μ is output according to the problem-specific knowledge and confidence parameter values ​​μij and τij updated under the influence of anchor aij.

[0075] The problem-specific knowledge and confidence parameter values ​​μij and τij can be predicted from equations (5) and (6) which show the third estimation model below.

[0076]

number

[0077]

number

[0078] The third estimation model shown in equations (5) and (6) predicts the problem-specific knowledge and confidence parameter values ​​μij and τij according to the second individual characteristic value θi (κi, δi, ωi) and the problem characteristic values ​​μj and τj. The problem characteristic values ​​μj and τj consist of the characteristic value μj indicating the correct answer to the problem and the characteristic value τj indicating the simplicity of the problem.

[0079] The third estimation model is composed of a probability density function p(μij) for acquiring the knowledge mean value shown in equation (5) and a probability density function p(τij) for acquiring the confidence value shown in equation (6).

[0080] The probability density function p(μij) for obtaining the average knowledge value shown in equation (5) uses the characteristic value μj indicating the correct answer to the problem as the “average” and obtains the value μij indicating the average knowledge value of an individual according to a normal distribution in which the inverse κ-1i of the knowledge amount characteristic value κi and the inverse τ-1j of the characteristic value τj indicating the simplicity of the problem are included in the “variance σ2”.

[0081] Equation (5) is based on the idea that when there is average knowledge and the problem is simple, the output value approaches the correct answer μj.

[0082] The probability density function p(τij) for obtaining the confidence value shown in equation (6) obtains the value τij indicating the individual's confidence according to a gamma distribution whose parameters include the confidence characteristic values ​​δi, ωi, and the characteristic value μj indicating the correct answer to the problem.Equation (6) is based on the idea that the value τij indicating the individual's confidence will generally be large when a prediction is close to the correct answer.

[0083] The analysis processing unit 300 applies the answer μ received by the first receiving unit 100 and stored in the answer storage unit 200 and the value aij indicating the anchor of question j stored in the question storage unit 150 to the first estimation model shown in equations (1), (2), and (3), the second estimation model shown in equation (4), and the third estimation model shown in equations (5) and (6), to estimate the first individual characteristic value ηi and the second individual characteristic value θi (κi, δi, ωi) and the characteristic values ​​μj and τj of the questions.

[0084] In addition, preprocessing may be performed according to the nature of the experimental results before the analysis process. In a magnitude estimation experiment, preprocessing may be performed by applying a logarithmic function to the answer μ and the value aij indicating the anchor. In a proportion estimation experiment, preprocessing may be performed by applying a logit function to the answer μ and the value aij indicating the anchor.

[0085] If data on combinations of anchors aij and answers μ is collected from a combination of multiple experimental participants i (e.g., 500 people) and multiple experimental subjects j (e.g., 20 questions), it is possible to estimate the characteristic values ​​μj, τj of the problem and the individual characteristic values ​​ηi, θi (κi, δi, ωi).

[0086] Any analytical processing method can be used. For example, the analytical processing can be performed using the Markov Chain Monte Carlo (MCMC) method, making the assumptions shown in the following equations (7) to (12). In order to make accurate estimations, it is useful to make certain assumptions regarding the individual characteristic values ​​ηi, θi (κi, δi, ωi). As an example, assume that the individual characteristic values ​​ηi, θi (κi, δi, ωi) all follow a gamma distribution. The knowledge quantity characteristic value κi has an average of 1 by definition.

[0087]

number

[0088]

number

[0089]

number

[0090]

number

[0091] Equation (7) is a probability density function p(ηi) for acquiring the first personal characteristic value, and acquires the first personal characteristic value ηi according to the gamma distribution.

[0092] Equation (8) is a probability density function p(κi) for acquiring the knowledge quantity characteristic value, and acquires the knowledge quantity characteristic value κi according to the gamma distribution.

[0093] Equations (9) and (10) are probability density functions p(δi) and p(ωi) for obtaining the confidence characteristic value, respectively, and the confidence characteristic values ​​δi and ωi are obtained according to the gamma distribution.

[0094] In equations (7) to (10), αη, βη, ακ, αδ, βδ, αω, and βω are parameters that represent individual differences within a population.

[0095] In equations (7) to (10), a gamma distribution is assumed, but any distribution with a domain greater than or equal to 0 may be assumed. An appropriate distribution can be assumed depending on the experimental results.

[0096] The characteristic values ​​μj and τj of the problem are also assumed to follow a normal distribution and a gamma distribution, respectively, as shown in (11) and (12) below. For the characteristic values ​​μj and τj of the problem, it is appropriate to use the data μ*j and τ*j of the problem without anchors.

[0097]

number

[0098]

number

[0099] μ*j is the correct answer of the unanchored problem, and τ*j is the ease of the unanchored problem.

[0100] Bayesian estimation can be performed by setting prior distributions for the parameters αη, βη, ακ, αδ, βδ, αω, and βω, which represent individual differences within a population. The prior distribution does not have a significant effect if the data is sufficiently large, so for example, it can be set to an exponential distribution with a mean of 1.

[0101] As a result of the analysis by the analysis processing unit 300, the parameters αη, βη, ακ, αδ, βδ, αω, βω representing individual differences within the group, the first individual characteristic value ηi and the second individual characteristic value θi (κi, δi, ωi), and the characteristic values ​​μj and τj in question can be estimated.

[0102] For the analysis, the Markov Chain Monte Carlo (MCMC) method or the No-U-Turn Sampler (NUTS) algorithm may be used.

[0103] 4A, 4B, and 4C illustrate the estimation results.

[0104] Figure 4A shows the estimation results of the characteristic values ​​μj that indicate the correct answers to the questions. It can be seen that the characteristic values ​​μ1, μ2, μ3, μ4, and μ5 that indicate the correct answers to the questions 1, 2, 3, 4, and 5 are different.

[0105] Figure 4B shows the estimated results of the characteristic value τj, which indicates the simplicity of the problem. We can see that the characteristic values ​​τ1, τ2, τ3, τ4, and τ5, which indicate the simplicity of the problem, are different for problems 1, 2, 3, 4, and 5. The size of the range corresponds to the range of participants in the experiment.

[0106] Figure 4C shows the estimated results for the knowledge amount characteristic value κi. It can be seen that the knowledge amount characteristic values ​​κ1, κ2, κ3, κ4, and κ5 differ for each of the experimental participants 1, 2, 3, 4, and 5. Although the range between the knowledge amount characteristic values ​​κ1, κ2, κ3, κ4, and κ5 is large, there are individual differences in the peaks, and it is clear that it is fully possible to classify individuals into those with high knowledge and those with low knowledge.

[0107] (Estimating individual characteristics using a trained model)

[0108] The second accepting unit 400 accepts answers from individual general users to questions with and without anchors. Here, the general users are used to mean individuals in a population separate from the experimental participants.

[0109] The second accepting unit 400 accepts answers to questions from general users via, for example, the display screen of the general user terminal 30. It is desirable that the questions presented to the general users here are the same as the questions presented to the experimental participants.

[0110] A general user accesses the question storage unit 150 of the server 40 via the general user terminal 30, thereby reading out question data from the question storage unit 150 of the server 40. The general user operates an input device 16 such as a keyboard to input an answer to the question on the screen of the display device 17 of the general user terminal 30.

[0111] Data indicating the answers input to the general user terminal 30 is stored in the answer storage unit 200 of the server 40 via the network 50 .

[0112] (Answer storage section)

[0113] Data indicating the answer received by the second receiving unit 400 is stored in the answer storage unit 200 of the server 40.

[0114] (Personal characteristics estimation unit)

[0115] The individual characteristic estimation unit 500 receives data indicating the answer of an individual general user and data indicating an anchor as input, and outputs a first individual characteristic value ηi and a second individual characteristic value θi (κi, δi, ωi) of the individual general user.

[0116] The individual characteristic estimation unit 500 estimates the first individual characteristic value ηi and the second individual characteristic value θi (κi, δi, ωi) of an individual general user using the trained model.

[0117] The trained model is trained using actual values ​​as training data, with the input data being the individual response μ and the value aij indicating the anchor received by the first reception unit 100, and the output data being the first individual characteristic value ηi relating to the influence of the anchor estimated by the analysis processing unit 300 and the second individual characteristic value θi (κi, δi, ωi) relating to knowledge and confidence.

[0118] (Processing execution unit 600)

[0119] The processing execution unit 600 inputs the data μ indicating the answer of the individual general user and the data aij indicating the anchor received by the second receiving unit 400 into the personal characteristic estimation unit 500, and executes processing to cause the personal characteristic estimation unit 500 to output the first personal characteristic value ηi and the second personal characteristic value θi (κi, δi, ωi) of the individual general user received by the second receiving unit 400.

[0120] According to the embodiment, once the analysis process is performed with the cooperation of an experimental participant, the characteristics of that individual can be extracted by asking similar questions to another individual. This allows for simple and versatile use as a tool for analyzing personal characteristics. For example, by narrowing the scope of the problem to a specific target, highly accurate analysis is possible, making the system suitable for original marketing research and subsequent decision-making. The system is also suitable for corporate strategy applications, such as companies developing products and services by understanding the characteristics of individual consumers in advance. [Explanation of symbols]

[0121] 100 First Reception Section 300 Analysis processing section 400 Second Reception Section 500 Personal characteristics estimation unit 600 Processing execution unit

Claims

1. An anchoring individual characteristic estimation device that predicts an individual characteristic related to anchoring, a first reception unit that receives individual answers to unanchored questions and anchored questions; a first estimation model that updates a problem-specific knowledge / confidence parameter value related to a question before the anchor is presented to a problem-specific knowledge / confidence parameter value related to a question after the anchor is presented, according to a first individual characteristic value related to the anchor value and the influence of the anchor; a second estimation model that predicts an individual's answer to questions without an anchor in accordance with the problem-specific knowledge and confidence parameter values ​​before the update by the first estimation model, and that predicts an individual's answer to questions with an anchor in accordance with the problem-specific knowledge and confidence parameter values ​​after the update by the first estimation model; a third estimation model for predicting a question-specific knowledge / confidence parameter value according to a second individual characteristic value related to knowledge / confidence; an analysis processing unit that applies values ​​indicating the answer and the anchor received by the first reception unit to the first estimation model, the second estimation model, and a third estimation model to estimate the first individual characteristic value and the second individual characteristic value; An anchoring personal characteristic estimation device comprising:

2. the third estimation model is an estimation model that predicts a question-specific knowledge / confidence parameter value according to the second individual characteristic value and a characteristic value of the question, the analysis processing unit applies values ​​indicating the answer and the anchor received by the first receiving unit to the first estimation model, the second estimation model, and the third estimation model to estimate the first individual characteristic value, the second individual characteristic value, and the characteristic value of the question. The anchoring personal characteristic estimation device according to claim 1 .

3. The second estimation model is configured with a probability density function that obtains the individual's answer according to a normal distribution having a mean value indicating the individual's average knowledge and a variance equal to the inverse of a value indicating the individual's confidence. The anchoring personal characteristic estimation device according to claim 1 or 2.

4. The first estimation model is a probability density function that acquires a value indicating the influence of the anchor according to a gamma distribution with the anchor value and the first personal characteristic value as parameters; a knowledge average value calculation formula for obtaining a value indicating an individual's knowledge average after updating according to a value indicating the influence of the anchor obtained by the probability density function and a value indicating the individual's knowledge average before updating; a confidence value calculation formula for obtaining a value indicating the confidence of an individual after updating according to a value indicating the influence of the anchor obtained by the probability density function and a value indicating the confidence of the individual before updating; comprising: The device for estimating personal characteristics of anchoring according to any one of claims 1 to 3.

5. the second individual characteristic value comprises a knowledge amount characteristic value indicating an amount of knowledge and a confidence characteristic value indicating a magnitude and a variance of confidence, The third estimation model is a probability density function for acquiring a knowledge average value, which acquires a value indicating an individual's knowledge average according to a normal distribution in which the reciprocal of the knowledge amount characteristic value is included in the variance; a probability density function for acquiring a confidence value, which acquires a value indicating the confidence of an individual according to a gamma distribution including the confidence characteristic value as a parameter; comprising: The anchoring personal characteristic estimation device according to claim 1 .

6. The analysis processing unit a probability density function for acquiring a first personal characteristic value that acquires the first personal characteristic value according to a gamma distribution; a probability density function for acquiring the knowledge quantity characteristic value according to a gamma distribution; a probability density function for obtaining the confidence characteristic value according to a gamma distribution; and estimating the first individual characteristic value and the second individual characteristic value using The anchoring personal characteristic estimation device according to claim 5 .

7. a trained model trained using as training data actual values ​​in which values ​​indicating an individual's answer and an anchor are used as input data and a first individual characteristic value relating to the influence of the anchor and a second individual characteristic value relating to knowledge and confidence are used as output data, the trained model comprising: a personal characteristic estimation unit that receives as input data indicative of an individual's answer and data indicative of the anchor, and outputs the first individual characteristic value and the second individual characteristic value of the individual; a second reception unit that receives individual answers to unanchored questions and anchored questions; a processing execution unit that executes processing to input data indicating the answer of the individual and data indicating the anchor received by the second reception unit into the personal characteristic estimation unit, and to output the first personal characteristic value and the second personal characteristic value of the individual received by the second reception unit from the personal characteristic estimation unit; An anchoring personal characteristic estimation device comprising:

Images