Characteristic extraction device, characteristic extraction method, and characteristic extraction program

Abstract

To define a character for use in explanation of human characteristics in a specific scene.SOLUTION: A generation unit 15a generates sensor data reflecting human behaviors in a predetermined situation. An acquiring unit 15b acquires the generated sensor data. An extracting unit 15c separates and extracts a continuous variable representing human characteristics and a non-continuous variable representing individual difference regardless of the characteristics from the acquired sensor data.SELECTED DRAWING: Figure 3

Description

[Technical Field]

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

[0002] Conventionally, techniques for analyzing personality have been known. For example, in the Big Five theory, human personality is expressed as five latent factors consisting of openness, conscientiousness, structure, and neuroticism. In this Big Five theory, each individual's personality is expressed as a combination of specific characteristic values ​​(see Non-Patent Document 1).

[0003] Furthermore, personality traits such as "integrity" are defined using words that express personality traits found in dictionaries as factors.

[0004] When using such personality analysis to predict job aptitude, it may be necessary to define a personality that accurately describes a person's characteristics in a particular situation, such as the fact that clerical work requires "integrity" and especially "meticulousness." [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Xi Chen, et al., “InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets”, [online], 2016, [Retrieved March 25, 2022], Internet<URL:https: / / perceptual.mpi-inf.mpg.de / files / 2015 / 08 / Steil_Ubicomp15.pdf> Summary of the Invention [Problem to be solved by the invention]

[0006] However, with conventional technology, it is difficult to define a personality that describes a person's characteristics in a specific situation. In other words, while it has been possible to define universal personalities that apply to all situations using words from natural language dictionaries, etc., there is no dictionary that expresses only human personalities in specific situations, such as job selection, so it is difficult to define a personality that describes a person's characteristics in a specific situation.

[0007] The present invention has been made in view of the above, and aims to make it possible to define a personality that describes the characteristics of a person in a specific situation. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems and achieve the object, the characteristic extraction device of the present invention is characterized by having an acquisition unit that acquires sensor data that reflects human behavior in a predetermined state, and an extraction unit that separates and extracts continuous variables that represent human characteristics and discontinuous variables that represent individual differences that are not due to characteristics from the acquired sensor data. [Effects of the Invention]

[0009] The present invention makes it possible to define personalities that describe the characteristics of a person in a particular situation. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram for explaining an overview of the characteristic extraction device of this embodiment. [Figure 2] FIG. 2 is a diagram for explaining an outline of the characteristic extraction device of this embodiment. [Figure 3] FIG. 3 is a schematic diagram illustrating the general configuration of the characteristic extraction device of this embodiment. [Figure 4] FIG. 4 is a flowchart showing the procedure of the characteristic extraction process. [Figure 5] FIG. 5 is a diagram illustrating an example of a computer that executes a characteristic extraction program. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to this embodiment. In addition, in the description of the drawings, the same parts are designated by the same reference numerals.

[0012] [Overview of the characteristic extraction device] 1 and 2 are diagrams for explaining an overview of the characteristic extraction device of this embodiment. In this embodiment, the characteristic extraction device extracts factors expressing personality (traits) from human behavior sensor data acquired in a specific scene under an experimental environment, as shown in FIG.

[0013] The characteristic extraction device then identifies words that describe the extracted factors by arbitrarily changing their values, thereby enabling the definition of characteristics that are important in a particular situation.

[0014] Furthermore, when extracting factors that explain high-dimensional data such as behavioral sensor data, the characteristic extraction device separates behavioral features that vary depending on human characteristics from behavioral features that vary from individual to individual regardless of human characteristics.

[0015] Specifically, because human characteristics are defined by variables that take continuous values, differences in behavioral characteristics resulting from differences in human characteristics change continuously. In contrast, individual differences that are not due to human characteristics often take discontinuous values, and therefore differences in behavioral characteristics resulting from individual differences often manifest discontinuously as characteristics specific to each individual. Examples of individual differences include differences in noise caused by age, gender, and habit of wearing sensor devices.

[0016] Therefore, the characteristic extraction device distinguishes between human characteristics as factors expressed as continuous variables and individual differences that are not related to characteristics as factors expressed as non-continuous variables, thereby enabling the characteristic extraction device to extract factors that represent human characteristics in specific situations.

[0017] Here, the horizontal axis of Figures 2(a) and (b) represents the characteristic value of a factor that represents a certain human characteristic, and the vertical axis represents the value for each individual. Figure 2(a) also illustrates a case in which behavioral features that vary due to human characteristics are not separated from behavioral features that vary for each individual regardless of human characteristics. In this case, the individual differences are too large, making it difficult to extract factors that represent human characteristics. In contrast, Figure 2(b) illustrates a case in which behavioral features that vary due to human characteristics are separated from behavioral features that vary for each individual regardless of human characteristics. In this case, factors that represent human characteristics are extracted by separating the differences in behavioral features caused by individual differences, as illustrated on the horizontal axis.

[0018] [Configuration of the characteristic extraction device] Fig. 3 is a schematic diagram illustrating the general configuration of a characteristic extraction device of this embodiment. As illustrated in Fig. 3, a characteristic extraction device 10 of this embodiment is realized by a general-purpose computer such as a personal computer, and includes an input unit 11, an output unit 12, a communication control unit 13, a storage unit 14, and a control unit 15.

[0019] The input unit 11 is realized using input devices such as a keyboard and a mouse, and inputs various instruction information such as a command to start processing to the control unit 15 in response to input operations by an operator. The output unit 12 is realized by a display device such as a liquid crystal display, a printing device such as a printer, etc. For example, the output unit 12 displays the results of the characteristic extraction processing described below.

[0020] The communication control unit 13 is realized by a NIC (Network Interface Card) or the like, and controls communication between the control unit 15 and an external device via a telecommunication line such as a LAN (Local Area Network) or the Internet. For example, the communication control unit 13 controls communication between the control unit 15 and a management device or the like that manages sensor data.

[0021] The storage unit 14 is realized by a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 14 stores in advance the processing program that operates the characteristic extraction device 10, data used during execution of the processing program, etc., or temporarily stores them each time processing is performed. The storage unit 14 may be configured to communicate with the control unit 15 via the communication control unit 13.

[0022] In this embodiment, the storage unit 14 stores the generator 14a and the discriminator 14b optimized by the learning unit 15d (described later). In this embodiment, the generator 14a and the discriminator 14b use a generative adversarial network, as described later, but are not limited to this and may be, for example, a variational autoencoder.

[0023] The control unit 15 is realized using a CPU (Central Processing Unit) or the like, and executes a processing program stored in a memory. As a result, the control unit 15 functions as a generating unit 15a, an acquiring unit 15b, an extracting unit 15c, a learning unit 15d, and an identifying unit 15e, as illustrated in FIG. 3, to execute the characteristic extraction process described below. Note that each or some of these functional units may be implemented in different hardware. For example, the identifying unit 15e may be implemented in hardware different from the other functional units. The control unit 15 may also include other functional units.

[0024] The generator 15a generates sensor data that reflects human behavior in a predetermined situation. Specifically, the generator 15a generates sensor data under an experimental environment using the optimized generator 14a. Here, the sensor data that reflects human behavior is, for example, data that represents head movement, eye movement, facial movement, or facial expression, and may be facial video data captured by a camera, a history of PC operation, or the like.

[0025] The generating unit 15a also assigns information for identifying individuals to non-continuous variables included in the generated sensor data. For example, the generating unit 15a explicitly assigns personal IDs as teacher labels to the non-continuous variables. This makes it easy to build a machine learning model by the identifying unit 15e, which will be described later.

[0026] The acquiring unit 15b acquires sensor data that reflects human behavior in a predetermined situation. Specifically, the acquiring unit 15b acquires sensor data that represents any one of head movement, eye movement, facial movement, and facial expression.

[0027] For example, the acquiring unit 15b acquires the generated sensor data. Specifically, the acquiring unit 15b acquires the sensor data generated by the generating unit 15a.

[0028] Furthermore, the acquiring unit 15b may acquire actual data reflecting human behavior in a predetermined situation from a management device or the like that manages sensor data output from the sensor via the input unit 11 or the communication control unit 13. The acquiring unit 15b may store the acquired actual data in the storage unit 14. In this case, the learning unit 15d and the identifying unit 15e, which will be described later, may acquire the sensor data to be processed from the storage unit 14.

[0029] The extraction unit 15c separates and extracts continuous variables representing human characteristics and discontinuous variables representing individual differences independent of characteristics from the acquired sensor data. Specifically, the extraction unit 15c separates and extracts continuous variables and discontinuous variables from the acquired sensor data using the optimized discriminator 14b.

[0030] The learning unit 15d optimizes the generator 14a and the discriminator 14b through learning. As described above, in this embodiment, the generator 14a and the discriminator 14b are constructed using a generative adversarial network. Specifically, the learning unit 15d first optimizes the discriminator 14b, and then optimizes the generator 14a.

[0031] In optimizing the discriminator 14b, the learning unit 15d acquires the sensor data generated by the generating unit 15a using the generator 14a. At this time, the generating unit 15a inputs several continuous variables randomly sampled from a normal distribution as factors consisting of continuous values ​​expected to express human characteristics, and randomly sampled discontinuous values ​​as discontinuous variables expressing individual differences independent of characteristics, to the generator 14a, and generates the sensor data.

[0032] The learning unit 15d inputs the generated sensor data to the discriminator 14b and performs error evaluation and backpropagation so that the discriminator 14b estimates the data as "generated data" in the path where the discriminator 14b distinguishes between "generated data" and "actual data."

[0033] Next, the learning unit 15d inputs the real data to the discriminator 14b, and performs error evaluation and backpropagation so that the discriminator 14b estimates the data as "real data" in the path where the discriminator 14b distinguishes between "generated data" and "real data."

[0034] In addition, in this embodiment, in addition to the paths normally provided in a generative adversarial network, the discriminator 14b has a path that separates and estimates whether the value is a continuous value expected to express characteristics or a discontinuous value of the personal ID that represents individual differences.

[0035] However, when real data is input, the learning unit 15d does not perform error evaluation and backpropagation in the path that estimates the continuous value expected to express the characteristic. On the other hand, the learning unit 15d performs error evaluation and backpropagation using the cross entropy function used for the non-continuous variable in the path that estimates the non-continuous value of the personal ID.

[0036] Next, the learning unit 15d optimizes the generator 14a. In optimizing the generator 14a, the learning unit 15d acquires the sensor data generated by the generation unit 15a using the generator 14a. At this time, similar to the case of optimizing the discriminator 14b, the generation unit 15a inputs several continuous variables randomly sampled from a normal distribution as factors consisting of continuous values ​​expected to express human characteristics, and randomly sampled discontinuous values ​​as discontinuous variables expressing individual differences independent of characteristics, to the generator 14a, and generates sensor data.

[0037] Then, the learning unit 15d inputs the generated sensor data to the discriminator 14b, and performs error evaluation and backpropagation so that the discriminator 14b estimates the data as "real data" in the path where the discriminator 14b distinguishes between "generated data" and "real data."

[0038] In the path to estimate continuous values ​​expected to express human characteristics, the learning unit 15d performs error evaluation and backpropagation using the negative log-likelihood function of the normal distribution used for continuous variables, so as to estimate continuous values ​​sampled from a normal distribution corresponding to the input data. As a result, the generator 14a generates data that makes it easier for the discriminator 14b to estimate differences in the continuous variables from which the data was generated. This makes it possible to extract factors.

[0039] In the path for estimating a personal ID corresponding to input data, the learning unit 15d performs error evaluation and back propagation using a cross entropy function used for non-continuous variables.

[0040] By using the generator 14a trained in this way, it becomes possible to generate behavioral sensor data by specifying continuous values ​​expected to express characteristics and arbitrary personal ID values. Therefore, it becomes possible to generate sensor data that reflects human behavior for all individuals included in the sensor data to be analyzed by arbitrarily changing the values ​​of continuous variables expected to express human characteristics.

[0041] This allows the generation unit 15a and the extraction unit 15c to distinguish between continuous values ​​expected to represent characteristics and discontinuous values ​​representing personal IDs based on the qualitative difference between the continuity and discontinuity of the variables. Therefore, it is expected that individual differences not due to characteristics can be explained by the discontinuity. Additionally, by having the generation unit 15a explicitly provide personal IDs as teacher labels and allowing the identification unit 15e to perform learning, it is expected that the behavioral features can be explained by these variables while more clearly distinguishing between the two.

[0042] The identification unit 15e uses the sensor data generated by the generation unit 15a in accordance with changes in the value of the extracted continuous variable to identify information representing a characteristic corresponding to the continuous variable (a cluster of the continuous variable). Specifically, the identification unit 15e identifies words representing a characteristic corresponding to the extracted continuous variable. For example, the identification unit 15e expresses the sensor data that changes in accordance with changes in the value of the continuous variable using animation or the like to identify a cluster of the continuous variable (a factor representing a characteristic), and identifies a word that defines a character representing a characteristic for the cluster.

[0043] For example, the specification unit 15e implements on the 3D model behavior sensor data of the human head, eye movements, etc. for the length of the time window input to the learned model when the value of the factor representing the characteristic is low, medium, or high, thereby reproducing, as a moving image such as an animation, movements that affect the impression of the face, such as the degree to which the head is bowed or the way the eyes are winked.

[0044] The specification unit 15e then constructs a machine learning model that estimates characteristics from the face video data, thereby defining the personality represented by the factor when the value of the factor representing the characteristic is low, medium, or high. By using the machine learning model constructed for each case, it becomes possible to estimate the personality of an individual from the behavior sensor data of the individual in a specific situation and predict their suitability for work, for example.

[0045] The specification unit 15e may assign labels of the specified words (natural language) to the sensor data. The assigned word labels can be written on a questionnaire, for example, and can be used to predict job aptitude.

[0046] In this case, the identification unit 15e performs machine learning to estimate human characteristics in all general situations, for example, from facial video data via labels in natural language. That is, by obtaining output of characteristic labels that best represent the facial video data when the value of a factor representing the characteristic is low, medium, or high, the personality represented by the factor can be verbalized. Such a machine learning model can be constructed using human characteristics in all general situations.

[0047] Alternatively, the identification unit 15e may collect the results of defining characteristics by many people online without using a machine learning model. In this case, it becomes possible to identify the words that represent the characteristics of a given situation by, for example, majority vote using words that represent the characteristics of a given situation.

[0048] [Characteristics extraction processing] Next, the characteristic extraction process performed by the characteristic extraction device 10 according to this embodiment will be described with reference to Fig. 4. Fig. 4 is a flowchart showing the procedure of the characteristic extraction process. The flowchart in Fig. 4 starts, for example, when the user performs an operation input to instruct the start of the process.

[0049] First, the acquisition unit 15b acquires sensor data reflecting human behavior in a predetermined situation (step S1). For example, the acquisition unit 15b acquires sensor data representing head movement, eye movement, facial movement, or facial expression as sensor data for a specific scene generated by the generation unit 15a in an experimental environment. The generation unit 15a generates the sensor data using the optimized generator 14a.

[0050] Next, the extraction unit 15c separates and extracts continuous variables representing human characteristics and discontinuous variables representing individual differences not due to characteristics from the acquired sensor data (step S2). Specifically, the extraction unit 15c uses the optimized discriminator 14b to separate and extract continuous values ​​expected to represent human characteristics and discontinuous values ​​representing individual differences not due to characteristics from the acquired sensor data.

[0051] Then, the identification unit 15e uses the sensor data generated by the generation unit 15a in accordance with changes in the value of the extracted continuous variable to identify information such as words that represent characteristics corresponding to the continuous variable (step S3). For example, the identification unit 15e expresses the sensor data that changes in accordance with changes in the value of the continuous variable using animation or the like to identify a cluster of the continuous variable (factor that represents the characteristic), and identifies a word that defines a character that represents the characteristic for the cluster. This completes the series of characteristic extraction processes.

[0052] [effect] As described above, in the characteristic extraction device 10 of this embodiment, the acquisition unit 15b acquires sensor data that reflects human behavior in a predetermined state. The extraction unit 15c separates and extracts continuous variables that represent human characteristics and non-continuous variables that represent individual differences that are not dependent on the characteristics from the acquired sensor data.

[0053] This makes it possible to distinguish between continuous values ​​expected to express characteristics and discontinuous values ​​expressing personal IDs based on the qualitative difference between continuity and discontinuity of variables. It is also expected that individual differences not attributable to characteristics can be explained by discontinuity. Therefore, the characteristic extraction device 10 can extract factors that represent human characteristics that are not individual-specific in a specific situation. Therefore, it becomes possible to define personality that explains human characteristics in a specific situation.

[0054] Furthermore, the generator 15a generates sensor data that reflects human behavior in a predetermined situation, and the acquirer 15b acquires the generated sensor data. This makes it possible to generate a large amount of sensor data under an experimental environment that assumes a specific situation. Therefore, it becomes possible to define information that accurately represents human characteristics.

[0055] Furthermore, the generating unit 15a assigns information for identifying individuals to the non-continuous variables included in the generated sensor data. As a result, the generating unit 15a performs learning by explicitly assigning the individual ID as a teacher label, and is thereby able to explain behavioral features using these variables while more clearly distinguishing between continuous variables and non-continuous variables.

[0056] Furthermore, the identification unit 15e uses the sensor data generated by the generation unit 15a in response to changes in the values ​​of the extracted continuous variables to identify information representing the characteristics corresponding to the continuous variables. Specifically, the identification unit 15e identifies words representing the characteristics corresponding to the extracted continuous variables. This makes it possible to define the characteristics using words that generally express human personality, for example.

[0057] The acquisition unit 15b also acquires sensor data representing head movement, eye movement, facial movement, or facial expression. This makes it possible to extract human characteristics from movements that affect the impression of the face, such as the degree to which the head is bowed or the way the eyes are winked. Furthermore, the extracted characteristics can be reproduced as animation on a 3D model, making it easy to identify information representing human characteristics.

[0058] [program] A program written in a computer-executable language may be created to execute the processing executed by the characteristic extraction device 10 according to the above embodiment. In one embodiment, the characteristic extraction device 10 can be implemented by installing a characteristic extraction program for executing the characteristic extraction processing described above as package software or online software on a desired computer. For example, by executing the characteristic extraction program on an information processing device, the information processing device can function as the characteristic extraction device 10. The information processing device referred to here includes desktop and notebook personal computers. Other examples of information processing devices include mobile communication terminals such as smartphones, mobile phones, and PHS (Personal Handyphone Systems), as well as slate terminals such as PDAs (Personal Digital Assistants). The functions of the characteristic extraction device 10 may also be implemented on a cloud server.

[0059] 5 is a diagram showing an example of a computer that executes a characteristic extraction program. The 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. These components are connected by a bus 1080.

[0060] The memory 1010 includes a ROM (Read Only Memory) 1011 and a RAM 1012. The ROM 1011 stores, for example, a boot program such as a BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to a hard disk drive 1031. The disk drive interface 1040 is connected to a disk drive 1041. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1041. The serial port interface 1050 is connected to, for example, a mouse 1051 and a keyboard 1052. The video adapter 1060 is connected to, for example, a display 1061.

[0061] Here, the hard disk drive 1031 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. Each piece of information described in the above embodiment is stored in the hard disk drive 1031 or memory 1010, for example.

[0062] The characteristic extraction program is stored in the hard disk drive 1031 as a program module 1093 in which instructions to be executed by the computer 1000 are written. Specifically, the program module 1093 in which each process executed by the characteristic extraction device 10 described in the above embodiment is written is stored in the hard disk drive 1031.

[0063] Furthermore, data used for information processing by the characteristic extraction program is stored as program data 1094, for example, in the hard disk drive 1031. Then, the CPU 1020 reads the program module 1093 and the program data 1094 stored in the hard disk drive 1031 into the RAM 1012 as necessary, and executes each of the above-described procedures.

[0064] The program module 1093 and program data 1094 related to the characteristic extraction program are not limited to being stored in the hard disk drive 1031, but may be stored in a removable storage medium, for example, and read by the CPU 1020 via the disk drive 1041. Alternatively, the program module 1093 and program data 1094 related to the characteristic extraction program may be stored in another computer connected via a network such as a LAN or a WAN (Wide Area Network), and read by the CPU 1020 via the network interface 1070.

[0065] Although the present invention has been described above as an embodiment, the present invention is not limited to the description and drawings that form part of the disclosure of the present invention. In other words, other embodiments, examples, and operational techniques that can be made by those skilled in the art based on the present invention are all included in the scope of the present invention. [Explanation of symbols]

[0066] 10 Characteristic extraction device 11 Input section 12 Output section 13 Communication control section 14 Storage section 15 Control Unit 15a Generator 15b Acquisition part 15c Extraction part 15d Learning Department 15e Specific part

Claims

1. an acquisition unit that acquires sensor data reflecting human behavior in a predetermined situation; an extraction unit that uses a discriminator to separate and extract continuous variables that are behavioral features that vary depending on human characteristics and non-continuous variables that represent individual differences that are not dependent on characteristics from the acquired sensor data; an identification unit that uses sensor data acquired using the extracted values ​​of the continuous variables to represent the sensor data as animation and identify information representing characteristics corresponding to the continuous variables; A characteristic extraction device comprising:

2. a generating unit configured to generate sensor data reflecting human behavior in a predetermined situation using a generator; The characteristic extraction device according to claim 1 , wherein the acquisition unit acquires the sensor data generated by the generation unit as the sensor data reflecting the human behavior.

3. The characteristic extraction device according to claim 2 , wherein the generating unit assigns information for identifying an individual to the non-continuous variables.

4. The characteristic extraction device according to claim 1 , wherein the specifying unit specifies a word that represents a characteristic corresponding to the extracted continuous variable.

5. The characteristic extraction device according to claim 1 , wherein the acquisition unit acquires the sensor data representing any one of head movement, eye movement, facial movement, and facial expression.

6. A characteristic extraction method executed by a characteristic extraction device, an acquisition step of acquiring sensor data reflecting human behavior in a given situation; an extraction step of separating and extracting, from the acquired sensor data, continuous variables that are behavioral characteristics that vary depending on human characteristics and non-continuous variables that represent individual differences that are not dependent on characteristics, using a discriminator; an identifying step of identifying information representing characteristics corresponding to the continuous variables by using sensor data acquired using the extracted values ​​of the continuous variables and representing the sensor data as animation; A feature extraction method comprising:

7. an acquisition step of acquiring sensor data reflecting human behavior in a given situation; an extraction step of separating and extracting, from the acquired sensor data, continuous variables that are behavioral characteristics that vary depending on human characteristics and non-continuous variables that represent individual differences that are not dependent on characteristics using a discriminator; an identifying step of identifying information representing characteristics corresponding to the continuous variables by using sensor data acquired using the extracted values ​​of the continuous variables and representing the sensor data as animation; A characteristic extraction program for causing a computer to execute the above.

Images