Behavioral analysis device, behavioral analysis method, and behavioral analysis program

Abstract

An action analysis device (10) comprises: an acquisition unit (15a) that acquires personality data including latent variables representing a psychological aspect of a user, and action data of the user; a calculation unit (15b) that calculates, on the basis of the personality data, an average value of the latent variables in an individual as individual personality data, and calculates purpose personality data by subtracting the individual personality data form the personality data; a learning unit (15c) that uses action data, individual personality data, and purpose personality data for a machine learning model (14d) to learn; and an estimation unit (15d) that uses the machine learning model (14d) to estimate a relationship between a personality characteristic indicated by the personality data and an action characteristic represented by action data.

Description

[Technical Field]

[0001] The present invention relates to a behavior analysis device, a behavior analysis method, and a behavior analysis program. [Background technology]

[0002] Linking and analyzing human behavior with values ​​representing psychological aspects can bring about various benefits. For example, the personality traits of a worker (or "user" as appropriate) can be used as values ​​representing psychological aspects, and behaviors characteristic of those values ​​(strength of personality traits) can be analyzed. This type of analysis is expected to clarify how behavior differs according to a worker's personality traits, leading to the realization of technologies such as designing work in line with predicted behavior simply by acquiring the worker's personality traits.

[0003] In the above-mentioned analysis, a method that treats values ​​representing psychological aspects as latent variables is effective. For example, one technique treats the strength of a worker's personality traits as one value of a latent variable and analyzes the results of generating typical behaviors while changing that value. This technique treats the value representing the strength of personality traits and the value representing individual differences, including measurement noise, as separate latent variables, making it possible to separate and extract continuous behavioral changes due to the strength of inconspicuous personality traits, which is the objective of the analysis, from discontinuous changes in behavioral sensor data due to significant individual differences, which become noise in the analysis. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Yamashita, Kanie, Takimoto, Koya, Kataoka, Oishi, Kumada, "Visualization of Head and Eye Movement Characteristics for Each Worker's Personality Using Generative Adversarial Networks," IEICE Transactions on Computer Science, Vol. J105-D, No.4, pp.1-12 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the above-mentioned conventional techniques make it difficult to effectively analyze human behavior. For example, the conventional techniques assume that the values ​​of multiple dimensions of the latent variables to be analyzed are uncorrelated, which can cause problems when analyzing personality traits or other traits that have strong correlations between each other. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems and achieve the objectives, the system is characterized by comprising an acquisition unit that acquires personality data including latent variables that represent the psychological aspects of a user and behavioral data of the user; a calculation unit that calculates an average value of the latent variables within an individual based on the personality data as personal personality data and calculates target personality data by dividing the personal personality data from the personality data; a learning unit that uses the behavioral data, the personal personality data, and the target personality data to train a machine learning model; and an estimation unit that uses the machine learning model to estimate the relationship between the personality characteristics indicated by the personality data and the behavioral features indicated by the behavioral data.

[0007] Furthermore, a behavioral analysis method according to the present invention is a behavioral analysis method executed by a behavioral analysis device, and is characterized by including: an acquisition step of acquiring personality data including latent variables representing psychological aspects of a user and behavioral data of the user; a calculation step of calculating an average value of the latent variables within an individual as personal personality data based on the personality data, and calculating target personality data by dividing the personal personality data from the personality data; a learning step of training a machine learning model using the behavioral data, the personal personality data, and the target personality data; and an estimation step of using the machine learning model to estimate the relationship between the personality characteristics indicated by the personality data and the behavioral features indicated by the behavioral data.

[0008] In addition, the behavioral analysis program of the present invention is characterized in that it causes a computer to execute the following steps: an acquisition procedure for acquiring personality data including latent variables that represent psychological aspects of a user and behavioral data of the user; a calculation procedure for calculating an average value of the latent variables within an individual as personal personality data based on the personality data and calculating target personality data by dividing the personal personality data from the personality data; a learning procedure for training a machine learning model using the behavioral data, the personal personality data, and the target personality data; and an estimation procedure for using the machine learning model to estimate the relationship between the personality characteristics indicated by the personality data and the behavioral features indicated by the behavioral data. [Effects of the Invention]

[0009] The present invention allows for efficient analysis of human behavior. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a behavior analysis system according to the first embodiment. [Figure 2] FIG. 2 is a block diagram illustrating an example of the configuration of the behavior analysis apparatus according to the first embodiment. [Figure 3] FIG. 3 is a flowchart showing an example of the flow of the behavior analysis process according to the first embodiment. [Figure 4] FIG. 4 is a flowchart showing an example of the flow of the data acquisition process according to the first embodiment. [Figure 5] FIG. 5 is a flowchart showing an example of the flow of the input data calculation process according to the first embodiment. [Figure 6] FIG. 6 is a flowchart showing an example of the flow of the learning process according to the first embodiment. [Figure 7] FIG. 7 is a flowchart showing an example of the flow of the data estimation process according to the first embodiment. [Figure 8] FIG. 8 is a diagram illustrating a computer that executes a program. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of a behavior analysis device, a behavior analysis method, and a behavior analysis program according to the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to the embodiments described below.

[0012] [First embodiment] The configuration of the behavior analysis system according to the first embodiment, the configuration of the behavior analysis device, and the flow of the behavior analysis process will be described below in order, and finally the effects of the first embodiment will be described.

[0013] 1. Configuration of behavior analysis system 100 The configuration of a behavior analysis system 100 according to the first embodiment will be described in detail using Fig. 1. Fig. 1 is a diagram showing an example of the configuration of the behavior analysis system 100 according to the first embodiment. Below, an example of the overall configuration of the behavior analysis system 100 will be shown, and then each process will be described.

[0014] (1-1. Example of the overall configuration of the behavior analysis system 100) Behavior analysis system 100 includes behavior analysis device 10. Behavior analysis system 100 also includes response data A including personality data acquired by behavior analysis device 10, and users U who are operators who provide behavior data to behavior analysis device 10. Note that behavior analysis system 100 shown in FIG. 1 may include multiple behavior analysis devices 10. Response data A may be used to represent the response results of multiple users U. User U may be used to represent multiple operators.

[0015] (1-2. Processing of the behavior analysis system 100) Below, we will explain the personality data acquisition process, behavioral data acquisition process, input data calculation process, model learning process, and data estimation process as processes of the behavior analysis system 100. Note that the following processes can be executed in a different order. Also, some of the following processes may be omitted.

[0016] (1-2-1. Personality data acquisition process) Behavior analysis device 10 acquires personality data from response data A of user U (step S1). For example, behavior analysis device 10 acquires personality data from response data A regarding occupational aptitude answered by user U.

[0017] (1-2-2. Behavioral data acquisition process) Behavior analysis device 10 acquires behavioral data of user U from user U (step S2). For example, behavior analysis device 10 acquires behavioral data of head movement and behavioral data of eye movement from sensors installed on user U while he or she is working.

[0018] (1-2-3. Input data calculation process) Behavior analysis apparatus 10 calculates input data from the personality data and behavioral data (step S3). For example, behavior analysis apparatus 10 generates input data by dividing the behavioral data by a fixed time period. Behavior analysis apparatus 10 also generates input data by calculating an intra-personal average value from the personality data. Behavior analysis apparatus 10 also standardizes the personality data and generates input data by dividing the intra-personal average value.

[0019] (1-2-4. Model learning process) The behavior analysis apparatus 10 uses the input data to train a machine learning model (step S4). For example, the behavior analysis apparatus 10 uses a generative adversarial network (GAN) to optimize a discriminator and a generator.

[0020] (1-2-5. Data estimation processing) Behavior analysis apparatus 10 estimates the relationship between personality traits and behavioral features (step S5). For example, behavior analysis apparatus 10 generates sensor data of head movement and eye movement while arbitrarily changing the value of the target personality trait.

[0021] (1-3. Effects of the Behavior Analysis System 100) Below, the problems with conventional behavior analysis processing will be explained, and then the effects of the behavior analysis system 100 will be described in detail.

[0022] (1-3-1. Problems with conventional behavioral analysis processing) In conventional behavior analysis processing, the strength of a worker's personality trait is treated as one of the values ​​of a latent variable, and the results of generating typical behaviors are analyzed while changing this value. In this process, by using a value representing the strength of the personality trait and a value representing individual differences, including measurement noise, as separate latent variables, it becomes possible to separate and extract continuous behavioral changes due to the strength of inconspicuous personality traits, which is the objective of the analysis, from discontinuous changes in behavioral sensor data due to significant individual differences, which become noise in the analysis.

[0023] The latent variables analyzed by the above-mentioned methods may have multiple dimensions. For example, the Big Five personality traits, which are commonly used to describe human personality, are expressed as a combination of five traits, such as "a person's personality is strong in trait 1, average in trait 2, ..., and weak in trait 5." Conventional methods assume that the values ​​of multiple dimensions of such latent variables are uncorrelated. Therefore, problems arise when applying conventional techniques to personality traits that have a strong correlation between each trait.

[0024] One situation in which correlations occur between multiple dimensions of latent variables is when values ​​representing psychological aspects are obtained from response data at different levels for each individual. For example, when obtaining a person's occupational aptitude as a personality trait, a list of occupations such as "auto repairman" and "insurance salesperson" is presented to the respondent, and the number of responses indicating "suitable" is tallied. Then, aptitudes such as "suitable for mechanic" and "suitable for salesperson" are obtained as personality traits based on the corresponding occupation names. The data obtained through such a procedure often results in response data with different levels of response for each individual.

[0025] Specifically, this includes not only intra-individual differences in responses that distinguish between occupational aptitudes, such as responding "suited" to occupations that contribute to "mechanic" aptitude, such as "auto repairman," but not to occupations that contribute to "salesperson" aptitude, such as "insurance salesperson." This often includes inter-individual differences in the level of "suitability" responses, such as when someone responds "suited" to "auto repairman," they are also more likely to respond "suited" to "insurance salesperson." If such personality traits were treated as latent variables, it would be difficult for the analytical model to distinguish whether differences in behavioral characteristics are due to differences in the strength of the "mechanic" trait or the "salesperson" trait. Furthermore, even if the model can appropriately capture differences in behavioral characteristics for each trait, differences due to the strength of the "salesperson" trait would be included in the analysis results when behaviors are generated by varying the strength of the "mechanic" trait, creating analytical problems.

[0026] The above problem cannot be solved by separating individual differences as discontinuous latent variables using the conventional behavioral analysis process described above. This is because the level of each individual's response is not discontinuous, but rather a continuous difference. Specifically, while discontinuous individual differences are defined by behavioral characteristics that allow a specific individual to be identified, the level of each individual's response is not linked to behavioral characteristics that allow a specific individual to be identified, such as when multiple individuals tend to respond that something is "suitable." Therefore, the problem cannot be solved by methods such as conventional behavioral analysis, which separate discontinuous individual differences from continuous latent variables that are the subject of analysis.

[0027] (1-3-2. Overview of the behavior analysis system 100) The behavior analysis system 100 solves the problem of taking values ​​of multiple dimensions as latent variables in a situation where values ​​representing psychological aspects are obtained from response data with different levels for each individual. Specifically, the behavior analysis system 100 takes values ​​representing the target psychological aspects (target personality data) and individual level values ​​(individual personality data) as continuous variables in separate dimensions. To achieve this, the behavior analysis system 100 defines the intra-individual average value for the response as the "level value of the response for each individual," and defines the value obtained by dividing the intra-individual average value by the original response data as the "value representing the psychological aspect." In other words, the behavior analysis system 100 can separate noise that continuously changes across multiple individuals and analyze values ​​such as personality traits.

[0028] (1-3-3. Effects of the Behavior Analysis System 100) As described above, although job aptitude data such as suitability for a list of occupations tends to contain noise that causes characteristic changes common to multiple people, behavior analysis system 100 can analyze the characteristics of behavior sensor data by treating such response data as personality traits that are uncorrelated with each other. In other words, behavior analysis system 100 can effectively analyze human behavior by estimating and visualizing the relationship between personality traits and behavioral features that are uncorrelated with each other.

[0029] 2. Configuration of Behavior Analysis Device 10 An example configuration of a behavior analysis apparatus 10 according to the first embodiment will be described in detail using Fig. 2. Fig. 2 is a block diagram showing an example configuration of a behavior analysis apparatus 10 according to the first embodiment. The behavior analysis apparatus 10 includes an input unit 11, an output unit 12, a communication unit 13, a storage unit 14, and a control unit 15.

[0030] (2-1. Input section 11) Input unit 11 controls input of various information to behavior analysis device 10. Input unit 11 is, for example, a mouse, a keyboard, or the like, and accepts input of setting information and the like to behavior analysis device 10.

[0031] (2-2. Output section 12) Output unit 12 controls the output of various information from behavior analysis apparatus 10. Output unit 12 is, for example, a display, and outputs setting information stored in behavior analysis apparatus 10, etc.

[0032] (2-3. Communications Department 13) The communication unit 13 controls data communication with other devices. For example, the communication unit 13 performs data communication with each communication device. The communication unit 13 can also perform data communication with an operator's terminal (not shown).

[0033] (2-4. Storage section 14) The storage unit 14 stores various pieces of information referenced by the control unit 15 when it operates, and various pieces of information acquired when the control unit 15 operates. The storage unit 14 includes a personality data storage unit 14a, a behavioral data storage unit 14b, an input data storage unit 14c, and a machine learning model 14d. Here, the storage unit 14 is, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. Note that, in the example of FIG. 2, the storage unit 14 is installed inside the behavior analysis device 10, but it may also be installed outside the behavior analysis device 10, or multiple storage units may be installed.

[0034] (2-4-1. Personality Data Storage Unit 14a) The personality data storage unit 14a stores personality data acquired from the answer data A answered by the user U. For example, the personality data storage unit 14a stores answers to questions about occupational aptitude associated with the identification number of the user U. The personality data storage unit 14a may also store answers to questions about the Big Five (openness, conscientiousness, extroversion, agreeableness, and neuroticism) associated with the identification number of the user U.

[0035] (2-4-2. Behavioral Data Storage Unit 14b) The behavioral data storage unit 14b stores behavioral data acquired from the user U. For example, the behavioral data storage unit 14b stores time-series data of head movement and time-series data of eye movement associated with the user U's identification number.

[0036] (2-4-3. Input data storage unit 14c) The input data storage unit 14c stores input data to be input to the machine learning model 14d. For example, the input data storage unit 14c stores behavior data, individual personality data, and goal personality data divided into certain time periods.

[0037] (2-4-4. Machine Learning Model 14d) The machine learning model 14d is a trained model that outputs sensor data of head movement and eye movement when a value of a target personality trait is input. The machine learning model 14d is also a trained model that outputs a value of a target personality trait when sensor data of head movement and eye movement is input. For example, the machine learning model 14d is a trained model having a structure such as a Generative Adversarial Network (GAN) or a Variational AutoEncoder (VAE). The machine learning model 14d is also a trained model that is trained using input data calculated by a calculation unit 15b of the control unit 15, which will be described later.

[0038] (2-5. Control unit 15) Control unit 15 is responsible for overall control of behavior analysis device 10. Control unit 15 has acquisition unit 15a, calculation unit 15b, learning unit 15c, and estimation unit 15d. Here, control unit 15 is, for example, an electronic circuit such as a CPU (Central Processing Unit) or MPU (Micro Processing Unit), or an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).

[0039] (2-5-1. Acquisition part 15a) The acquiring unit 15a acquires personality data including latent variables representing psychological aspects of the user U, and behavioral data of the user U. For example, the acquiring unit 15a acquires answer data regarding the occupational aptitude of the user U as the personality data. At this time, the acquiring unit 15a acquires, as the personality data, values ​​of personality traits related to the occupational aptitude of the user U, such as "suited to mechanic," "suited to sales," and "suited to office work." The acquiring unit 15a may also acquire, as the personality data, values ​​of personality traits related to the Big Five of the user U, namely, "openness," "conscientiousness," "extroversion," "agreeableness," and "neuroticism."

[0040] On the other hand, the acquisition unit 15a acquires time-series data of head movement and eye movement of the user U as behavioral data. At this time, the acquisition unit 15a acquires time-series data of head movement about three spatial axes from an acceleration sensor and an angular velocity sensor installed on the user U. The acquisition unit 15a also acquires time-series data of eye movement about two axes, horizontal and vertical, from an electro-oculography sensor installed on the user U. The acquisition unit 15a may also acquire time-series data of head movement and eye movement of the user U from facial video data captured by a camera, a history of PC (Personal Computer) operation, or the like.

[0041] The acquiring unit 15a stores the acquired personality data in the personality data storage unit 14a, and stores the acquired behavioral data in the behavioral data storage unit 14b.

[0042] (2-5-2. Calculation section 15b) The calculation unit 15b calculates the average value of the intra-individual latent variables based on the personality data as individual personality data, and calculates the target personality data by dividing the individual personality data by the personality data. That is, first, the calculation unit 15b calculates the average value of the latent variables, which are the values ​​of the personality characteristics indicated by the acquired personality data, for each user U, and sets the average value as individual personality data (individual-level label data). Second, the calculation unit 15b standardizes all of the acquired personality data to convert it to a mean of 0 and a standard deviation of 1. At this time, the calculation unit 15b does not standardize in either the inter-individual or intra-individual directions, but simply standardizes all of the data to a mean of 0 and a standard deviation of 1. Third, the calculation unit 15b divides the standardized value of the personality data by the intra-individual average value, i.e., the individual personality data, and sets the target personality data (psychological aspect label data) that is the purpose of analysis. Meanwhile, the calculation unit 15b divides the acquired behavioral data using a moving window having a certain time width, and calculates input data related to the behavioral data.

[0043] The calculation unit 15b stores the calculated individual personality data and objective personality data in the input data storage unit 14c. The acquisition unit 15a also stores the divided behavioral data in the input data storage unit 14c.

[0044] (2-5-3. Learning Section 15c) The learning unit 15c uses the behavioral data, the individual personality data, and the objective personality data to train the machine learning model 14d. At this time, the learning unit 15c trains the machine learning model 14d using a generative adversarial network. Alternatively, the learning unit 15c may train the machine learning model 14d using a variational autoencoder.

[0045] The learning unit 15c adds random numbers to the individual personality data and the target personality data to bring them closer to a normal distribution, and then inputs values ​​randomly sampled from the individual personality data and the target personality data to which the random numbers have been added, thereby learning the machine learning model 14d. Details of the model learning process will be described later in (3-4. Flow of the model learning process).

[0046] (2-5-4. Estimation section 15d) The estimation unit 15d uses the machine learning model 14d to estimate the relationship between the personality trait indicated by the personality data and the behavioral feature indicated by the behavioral data. For example, the estimation unit 15d inputs the value of the target personality trait into the machine learning model 14d, and displays the sensor data of head movement and eye movement output from the machine learning model 14d as a graph on the output unit 12. The estimation unit 15d also inputs the sensor data of head movement and eye movement into the machine learning model 14d, and displays the value of the target personality trait as an estimated value on the output unit 12.

[0047] [3. Behavioral analysis processing flow] The flow of the behavior analysis process according to the first embodiment will be described in detail with reference to Figures 3 to 7. Below, the overall flow of the behavior analysis process will be described, and then the flow of each process will be described in the order of data acquisition process, input data calculation process, model learning process, and data estimation process.

[0048] (3-1. Overall flow of behavior analysis processing) The overall flow of the behavior analysis process will be described using Fig. 3. Fig. 3 is a flowchart showing the flow of the behavior analysis process according to the first embodiment. Note that the processes of steps S101 to S104 below can also be executed in a different order. Furthermore, some of the processes of steps S101 to S104 below may be omitted.

[0049] First, acquisition unit 15a of behavior analysis apparatus 10 executes a data acquisition process (step S101). Second, calculation unit 15b of behavior analysis apparatus 10 executes an input data calculation process (step S102). Third, learning unit 15c of behavior analysis apparatus 10 executes a model learning process (step S103). Fourth, estimation unit 15d of behavior analysis apparatus 10 executes a data estimation process (step S104), and then ends the process.

[0050] (3-2. Data acquisition process flow) The flow of the data acquisition process by the acquisition unit 15a will be described with reference to Fig. 4. Fig. 4 is a flowchart showing an example of the flow of the data acquisition process according to the first embodiment. Below, the personality data acquisition process (step S201) and the behavioral data acquisition process (steps S202 to S203) will be described in that order. Note that the processes of steps S201 to S203 below can also be executed in a different order. Also, some of the processes of steps S201 to S203 below may be omitted.

[0051] (3-2-1. Personality data acquisition process) The acquiring unit 15a acquires personality data from the answer data A (step S201). For example, the acquiring unit 15a acquires personality data from the answer data A regarding occupational aptitude answered by the user U.

[0052] (3-2-2. Behavioral data acquisition process) The acquiring unit 15a acquires time-series data of head movement as behavior data from the acceleration sensor and angular velocity sensor (step S202), and acquires time-series data of eye movement as behavior data from the electro-oculography sensor (step S203).

[0053] (3-3. Input data calculation process flow) The flow of the input data calculation process by the calculation unit 15b will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the flow of the input calculation process according to the first embodiment. Below, the behavioral data division process (step S301), the individual personality data calculation process (step S302), and the target personality data calculation process (steps S303 to S304) will be described in this order. Note that the processes of steps S301 to S304 below can also be executed in a different order. Also, some of the processes of steps S301 to S304 below may be omitted.

[0054] (3-3-1. Behavioral data division processing) The calculation unit 15b divides the behavioral data into time windows each having a certain time width, and generates input data (step S301).

[0055] (3-3-2. Personality data calculation process) The calculation unit 15b calculates the average value of the personality data within an individual as the individual personality data, and generates input data (step S302).

[0056] (3-3-3. Objective personality data calculation process) First, the calculation unit 15b standardizes all of the personality data (step S303). Second, the calculation unit 15b calculates target personality data by subtracting the individual personality data from the standardized personality data, and generates input data (step S304).

[0057] (3-3-4. Standardization of personality data) Standardization in the target personality data calculation process described above will now be described. The standardization performed in the process of step S303 above is neither inter-individual nor intra-individual standardization, but rather standardization for model learning of all variables so that they have an average of 0 and a standard deviation of 1. Note that if intra-individual standardization is performed in the process of step S303 above, the variables are converted into values ​​that only represent the relative magnitude of the variables within an individual, and information about differences in the level of personality traits between individuals is lost, and therefore this cannot be applied to the behavioral analysis process according to the first embodiment. Furthermore, if inter-individual standardization is performed in the process of step S303 above, information about the relative level of personality traits within individuals is lost, contrary to intra-individual standardization, and therefore this cannot be applied to the behavioral analysis process according to the first embodiment.

[0058] (3-4. Model learning process flow) The flow of the model learning process by the learning unit 15c will be described with reference to FIG. 6. FIG. 6 is a flowchart showing an example of the flow of the model learning process according to the first embodiment. Below, the premise of the model learning process will be explained, followed by a description of the classifier optimization process (steps S401 to S405), the generator optimization process (steps S406 to S409), and the effect of the model learning process. Note that the processes of steps S401 to S409 below can also be executed in a different order. Also, some of the processes of steps S401 to S409 below may be omitted.

[0059] (3-4-1. Prerequisites for model learning process) The following describes a countermeasure for cases in which a commonly used normal distribution cannot be used as the probability distribution for sampling latent variables representing psychological aspects in the model learning process according to the first embodiment. First, the model learning process according to the first embodiment utilizes a technique known as "disentanglement" in the field of machine learning, which explains high-dimensional data such as behavioral sensor data using a small number of factors that are understandable to humans. Such a technique requires sampling latent variables from a certain probability distribution and learning to link the relationships between the latent variables and behavioral features. In this case, sampling is generally performed from a probability distribution such as a normal distribution, and for this purpose, inter-person standardization is often performed in advance on the values ​​representing psychological aspects. For example, the values ​​representing each psychological aspect are converted so that they have a mean of 0 and a standard deviation of 1.

[0060] However, the process of dividing the values ​​representing psychological aspects by the intra-individual mean value (performing a kind of inter-individual standardization) is incompatible with inter-individual standardization. For example, if such a process is performed, the values ​​representing each psychological aspect will no longer follow a mean of 0 and a standard deviation of 1. If this difference is very large, it may become difficult to sample latent variables from a pre-assumed normal distribution (e.g., a normal distribution with a mean of 0 and a standard deviation of 1). In such cases, the model learning process according to the first embodiment can perform sampling based on the distribution of values ​​representing each psychological aspect.

[0061] (3-4-2. Classifier optimization process) First, the learning unit 15c randomly samples continuous variables from the individual personality data and the objective personality data (step S401). Second, the learning unit 15c inputs the sampled continuous variables to a generator to generate behavioral data (step S402).

[0062] In the processing of steps S401 to S402, the learning unit 15c inputs randomly sampled continuous variables from the actual distribution of each personality trait value, etc. and the distribution of the average values ​​of the personality trait values, etc. within each individual to a generator as latent variables of continuous values ​​(target personality data) expected to represent personality traits, etc., and continuous values ​​(individual personality data) expected to represent the average values ​​of the personality trait values, etc. within each individual, to generate behavioral data. At this time, the learning unit 15c may perform processing to approximate the distribution to a normal distribution to facilitate learning by sampling from the original distribution to which a minute random value (random number) has been added.

[0063] Third, in the path for distinguishing between "generated data" and "real data," the learning unit 15c inputs the behavioral data generated by the generator to the classifier, and performs error evaluation and backpropagation so that the classifier distinguishes the data as "generated data" (step S403). Fourth, in the path for distinguishing between "generated data" and "real data," the learning unit 15c inputs the action data, which is input data, to the classifier, and performs error evaluation and backpropagation so that the classifier distinguishes the data as "real data" (step S404).

[0064] Fifth, the learning unit 15c performs error evaluation and backpropagation so that the classifier estimates personality data corresponding to the input behavioral data (step S405). That is, in the processing of step S405, in addition to the paths normally provided in the generative adversarial network, paths are added to the classifier that estimate continuous values ​​expected to represent personality traits, etc. and continuous values ​​expected to represent the average values ​​of personality trait values, etc. within each individual, and error evaluation and backpropagation are performed using a squared error function, etc.

[0065] (3-4-3. Generator optimization process) First, the learning unit 15c randomly samples continuous variables from the individual personality data and the objective personality data (step S406). Second, the learning unit 15c inputs the sampled continuous variables to a generator to generate behavioral data (step S407). Third, in the path for distinguishing between "generated data" and "real data," the learning unit 15c inputs the behavioral data generated by the generator to a classifier and performs error evaluation and backpropagation so that the classifier distinguishes the data as "real data" (step S408).

[0066] Fourth, the learning unit 15c performs error evaluation and backpropagation to estimate the personality data that is the source of the behavioral data generated by the generator (step S409). As in the process of step S409 above, in the path that estimates continuous values ​​expected to represent personality traits and the like and continuous values ​​expected to represent the average values ​​of personality trait values ​​and the like within each individual, error evaluation and backpropagation are performed using a squared error function or the like to estimate those values ​​sampled at the time of data generation.

[0067] By executing the above generator optimization process, when the generator of the machine learning model 14d generates data while changing the values ​​of the latent variables, it is possible to explain the differences in behavioral characteristics linked to values ​​representing personality traits, etc., or values ​​representing the average values ​​of personality trait values, etc. within each individual, i.e., to reproduce the differences in behavioral characteristics in the generated data.

[0068] (3-4-4. Effect of model learning process) By using a generator trained by the above-described model learning process, it becomes possible to generate behavioral data by arbitrarily specifying two continuous values: one representing a personality trait, etc., and the other representing the average value of the personality trait value within each individual. Furthermore, by using the generator to generate behavioral sensor data of head movement and eye movement while arbitrarily changing the value of the continuous value representing the target personality trait, etc., it becomes possible to analyze the relationship between the target personality trait value and behavioral features without being affected by differences in the level of responses between individuals.

[0069] (3-5. Data estimation process flow) The flow of the data estimation process by the estimation unit 15d will be described with reference to FIG. 7. FIG. 7 is a flowchart showing an example of the flow of the data estimation process according to the first embodiment. Below, the behavioral data estimation process (steps S501 to S502), the personality data estimation process (steps S503 to S504), and the effect of the data estimation process will be described in that order. Note that the processes of steps S501 to S504 below can also be executed in a different order. Also, some of the processes of steps S501 to S504 below may be omitted.

[0070] (3-5-1. Behavioral data estimation processing) First, the estimation unit 15d inputs personality data to the trained machine learning model 14d (step S501). Second, the estimation unit 15d displays the behavioral data output from the trained machine learning model 14d as an estimation result (step S502).

[0071] (3-5-2. Personality data estimation processing) First, the estimation unit 15d inputs the behavioral data into the trained machine learning model 14d (step S503). Second, the estimation unit 15d displays the personality data output from the trained machine learning model 14d as an estimation result (step S504).

[0072] (3-5-3. Effect of data estimation processing) The data estimation process described above not only makes it possible to estimate the strength of personality traits from behavioral data such as sensor data, but also to generate behavioral data such as sensor data while varying the strength of the personality traits. The data generated in this way makes it possible to visualize the relationship between personality traits and behavioral features.

[0073] 4. Effects of the First Embodiment Finally, the effects of the first embodiment will be described below: Effects 1 to 5 corresponding to the processing according to the first embodiment will be described below.

[0074] (4-1. Effect 1) In the first embodiment described above, personality data including latent variables representing psychological aspects of user U and behavioral data of user U are acquired, the average value of the latent variables within an individual is calculated as personal personality data based on the personality data, the target personality data is calculated by dividing the personal personality data from the personality data, the behavioral data, the personal personality data, and the target personality data are used to train machine learning model 14d, and the relationship between the personality traits indicated by the personality data and the behavioral features indicated by the behavioral data is estimated using machine learning model 14d. Therefore, in the first embodiment, analysis of human behavior can be performed effectively.

[0075] (4-2. Effect 2) In the first embodiment described above, the machine learning model 14d is trained using a generative adversarial network. Therefore, in the first embodiment, the relationship between personality traits and behavioral features can be estimated with higher accuracy, thereby enabling effective analysis of human behavior.

[0076] (4-3. Effect 3) In the first embodiment described above, random numbers are added to the individual personality data and the target personality data to bring them closer to a normal distribution, and values ​​randomly sampled from the individual personality data and the target personality data to which the random numbers have been added are input to train the machine learning model 14d. Therefore, in the first embodiment, sampling from the distribution of real data makes it possible to effectively analyze human behavior.

[0077] (4-4. Effect 4) In the first embodiment described above, answer data regarding the occupational aptitude of the user U is acquired as the personality data. Therefore, in the first embodiment, it is possible to effectively perform an analysis of human behavior regarding occupational aptitude.

[0078] (4-5. Effect 5) In the above-described first embodiment, time-series data of head movement and eye movement of the user U is acquired as behavioral data. Therefore, in the first embodiment, analysis of human behavior such as office work can be effectively performed.

[0079] [System configuration, etc.] The components of each device shown in the drawings according to the above embodiments are conceptual functional units and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of each device can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc. Furthermore, all or any part of the processing functions performed by each device can be realized by a CPU and a program analyzed and executed by the CPU, or can be realized as hardware using wired logic.

[0080] Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method.In addition, the information including the processing procedures, control procedures, specific names, various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified.

[0081] 〔program〕 It is also possible to create a program written in a computer-executable language that executes the processing performed by behavior analysis apparatus 10 described in the above embodiment. In this case, the same effects as those of the above embodiment can be achieved by having a computer execute the program. Furthermore, such a program may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read and executed by a computer to achieve processing similar to that of the above embodiment.

[0082] 8 is a diagram showing a computer that executes a program. As shown in the example of FIG. 8, a computer 1000 includes, for example, a memory 1010, a CPU 1020, a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070, and these components are connected by a bus 1080.

[0083] As shown in FIG. 8, the memory 1010 includes a ROM (Read Only Memory) 1011 and a RAM 1012. The ROM 1011 stores a boot program such as a BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to a hard disk drive 1090 as shown in FIG. 8. The disk drive interface 1040 is connected to a disk drive 1100 as shown in FIG. 8. A removable storage medium such as a magnetic disk or an optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to a mouse 1110 and a keyboard 1120 as shown in FIG. 8. The video adapter 1060 is connected to a display 1130 as shown in FIG. 8.

[0084] 8, the hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, the above programs are stored, for example, on the hard disk drive 1090 as program modules in which instructions to be executed by the computer 1000 are written.

[0085] The various data described in the above embodiment are stored as program data, for example, in the memory 1010 or the hard disk drive 1090. The CPU 1020 then reads the program module 1093 and the program data 1094 stored in the memory 1010 or the hard disk drive 1090 into the RAM 1012 as needed, and executes various processing procedures.

[0086] Note that the program module 1093 and program data 1094 related to the program are not limited to being stored in the hard disk drive 1090, and may be stored in, for example, a removable storage medium and read by the CPU 1020 via a disk drive or the like. Alternatively, the program module 1093 and program data 1094 related to the program may be stored in another computer connected via a network (such as a LAN (Local Area Network) or WAN (Wide Area Network)) and read by the CPU 1020 via the network interface 1070.

[0087] The above-described embodiments and their modifications are included in the technology disclosed in this application, as well as in the scope of the invention described in the claims and their equivalents. [Explanation of symbols]

[0088] 10 Behavior analysis device 11 Input section 12 Output section 13 Communications Department 14 Storage section 14a Personality data storage section 14b Behavioral data storage unit 14c Input data storage section 14d Machine Learning Model 15 Control Unit 15a Acquisition part 15b Calculation part 15c Learning Department 15d Estimation part 100 Behavioral Analysis System

Claims

1. an acquisition unit that acquires personality data including latent variables representing psychological aspects of a user and behavioral data of the user; a calculation unit that calculates an average value of the latent variables within an individual as individual personality data based on the personality data, and calculates target personality data by dividing the individual personality data by the personality data; a learning unit that uses the behavioral data, the individual personality data, and the objective personality data to learn a machine learning model; an estimation unit that estimates a relationship between a personality trait indicated by the personality data and a behavioral feature indicated by the behavioral data using the machine learning model; A behavior analysis device comprising:

2. The learning unit training the machine learning model using a generative adversarial network; The behavior analysis device according to claim 1 .

3. The learning unit adding random numbers to the individual personality data and the target personality data to bring them closer to a normal distribution, and training the machine learning model by inputting values ​​randomly sampled from the individual personality data and the target personality data to which the random numbers have been added; The behavior analysis device according to claim 1 .

4. The acquisition unit As the personality data, response data regarding the user's occupational aptitude is acquired. The behavior analysis device according to claim 1 .

5. The acquisition unit acquiring time-series data of the user's head movement and eye movement as the behavioral data; The behavior analysis device according to claim 1 .

6. A behavior analysis method executed by a behavior analysis device, comprising: an acquisition step of acquiring personality data including latent variables representing psychological aspects of a user and behavioral data of the user; a calculation step of calculating an average value of the latent variables within an individual as personal personality data based on the personality data, and calculating target personality data by dividing the personal personality data by the personality data; a learning process of learning a machine learning model using the behavioral data, the individual personality data, and the objective personality data; an estimation step of estimating a relationship between a personality trait indicated by the personality data and a behavioral feature indicated by the behavioral data using the machine learning model; A behavioral analysis method comprising:

7. an acquisition step of acquiring personality data including latent variables representing psychological aspects of a user and behavioral data of the user; a calculation step of calculating an average value of the latent variables within an individual as personal personality data based on the personality data, and calculating target personality data by dividing the personal personality data by the personality data; a learning procedure for learning a machine learning model using the behavioral data, the individual personality data, and the objective personality data; an estimation step of estimating a relationship between a personality trait indicated by the personality data and a behavioral feature indicated by the behavioral data using the machine learning model; A behavioral analysis program characterized by causing a computer to execute the above.

Images