JP2012217518A - Human behavior analysis system and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate the degree of stress of the members constituting an organization and the whole organization by utilizing the characteristics of the stress in a group.SOLUTION: When human relation graph data for expressing a human relation based on data measured by a sensor or the like are given (101), an energy function describing the mutual action between human beings constituting a working place organization is defined to obtain such a macro index values (104 and 105) as the degree of stress of the organization by a process (103) minimizing the energy. When the index value of a node (for example, the degree of stress) of a part among the whole node constituting the human relation graph is found by questionnaire and other method for estimation, a processing part inputs them, and determines the other node condition by using the human relation graph data and by a minimizing processing of the energy function.

Description

  The present invention relates to a human behavior analysis system and method, and more particularly, to human relations and human beings measured by wearable sensors such as name tag type sensor nodes, wristwatch type sensor nodes, mobile phone usage logs, and other means. The present invention relates to a technology for analyzing a large amount of data reflecting various behaviors and optimizing organizational management and services, and a technology for estimating index values representing stress and other human internal states.

  Stress is one of the major social problems in modern society, and it is getting worse in various countries around the world. Several methods for measuring and estimating stress have been proposed. Individual features of voice features (Patent Document 1), various physiological information (Patent Document 2), skin temperature (Patent Document 3), etc. are used as indicators. There are methods to measure the degree of stress. In the field of psychology, research methods have been established for analyzing human internal conditions based on questionnaire surveys. For example, it is possible to measure a person's degree of stress by using an index called CES-D (Center for Epidemiological Studies Depression Scale) obtained by answering about 20 to 40 questions. CES-D is calculated as a score from 0 to 60, and a score of 16 or higher is considered to have a high risk of depression. By using this CES-D, it is possible to promote healthy organizational activities by looking at the stress state of the organization and applying appropriate measures. On the other hand, there is also research on the nature of stress in a group. For example, in Harvard University, it has been reported that stress propagates between best friends and neighbors (Non-patent Document 1). In this study, it was reported that those who are deeply connected with high-stress people are also prone to high stress.

  Stress has become a serious problem, especially in the workplace organization, and the importance of mental health care is screamed. Under such circumstances, human behavior can be measured quantitatively using a small sensor, and the relationship between the obtained behavior data and an index such as stress level can be analyzed, which can lead to organizational improvement. The nature is increasing. In recent years, wearable sensing devices have become smaller and lighter. For example, nameplate-type or wristwatch-type sensor nodes equipped with acceleration sensors, infrared sensors, small microphones, etc., are always worn by users without a heavy load. It has become possible. As a result, researchers reflect "people's communication status in the organization" and "behavior" and "life rhythm such as walking and sleep" from data such as face-to-face time, presence of conversation, and frequency of physical vibration It is now possible to acquire a large amount, various types of long-term time series data. If quantitative data concerning person-to-person communication can be obtained, a network structure between people in an organization can be obtained by using nodes as people and links as people-to-people communication. By analyzing the network structure and various indicators that represent this human relationship, the relationship between human views and human behavior in the organization that could not be elucidated by analyzing limited amounts or types of data, objectives, New knowledge about the relationship with the index to be used, such as organizational activity, productivity, or stress level, is gained, and the possibility of connecting it to services and management is expanding.

JP 2007-000366 A JP 08-252226 A JP 09-294724 A

J. et al. N. Rosenquist, J.M. H. Fowler, and N.W. A. Christakis, Social Network Determinants of Depression, Molecular Psychiatry, pp. 1-9, (2010).

  Non-Patent Document 1 shown above describes the knowledge that stress is transmitted between people who are particularly closely related to each other in the group, and may be applicable to stress care in the workplace organization. However, it does not show how to actually estimate the stress of the members of the organization.

  If CES-D, which is often used in the field of psychology, is measured for all the members constituting the organization, the degree of stress of the organization can be understood. However, when the scale of the organization is large, it is not easy to obtain answers from everyone using the questionnaire-based survey method. In addition, there is a problem that the answer standard of the questionnaire varies between individuals and the result is inaccurate. In addition, the timing of answering the questionnaire may also affect the answer results. For example, when there is a time margin compared with busy times, there may be a great difference in the answer results of the questionnaire. Furthermore, if the organization to which the organization belongs changes, the stress level of the individual will also change, and the stress level of the new organization will also change, but in order to measure them, a new questionnaire survey is conducted on all members of the new organization. Therefore, it is necessary to measure CES-D, which takes time and labor. Even in the same organization, changing the seat may change the physical distance from others and the stress level may change. In such cases, CES-D is again measured by questionnaire survey. The current situation is that the stress level cannot be evaluated without it.

By using the above-mentioned Patent Documents 1, 2, 3, etc., it is possible to measure an individual's stress value without conducting a questionnaire survey. However, even with these methods, the measurement method for all members of the organization is no different from the measurement method based on the questionnaire survey. In addition, as shown in Non-Patent Document 1, it is considered that stress has a great influence on human relations. However, the methods of Patent Documents 1, 2, and 3 are based on individual indicators such as voice feature values and skin temperature. It is a technique used to estimate the individual's stress level, and is not a stress estimation technique from the viewpoint of human relations in the workplace organization, how the state of others influences their own condition.
In view of the above points, an object of the present invention is to provide a human behavior analysis system and method capable of estimating the stress level of members constituting an organization or the entire organization using the characteristics of stress in a group. That is.

  Another object of the present invention is to solve these problems related to a method for investigating the degree of stress based on a questionnaire and a method for estimating an individual's stress level from an individual index. It is an object of the present invention to provide a human behavior analysis system and method capable of estimating an index value of a remaining person in an organization.

  Another object of the present invention is to define an energy function that determines the state of the system from observation data obtained by a sensor or the like, and by reducing or minimizing the energy function, an index value such as a stress degree is obtained. It is an object of the present invention to provide a human behavior analysis system and method capable of estimating

  Another object of the present invention is to provide a human behavior analysis system and method capable of predicting how an index value such as stress changes when the structure of a human relationship graph is changed. That is.

  In order to achieve the above object, the present invention defines an energy function that describes an interaction between persons constituting a workplace organization or the like when a human relation graph that is information representing human relations is given. By using a computer to minimize energy, it is possible to obtain a macro index of the system such as the degree of stress of the tissue. The present invention is based on a computer process for minimizing the energy function if the index values of some of the nodes constituting the human relationship graph, such as the degree of stress, are known by a questionnaire or other estimation method. The state of the node can be determined.

  The data targeted by the present invention, which is called human relationship graph data here, is general data representing the communication state between people. For example, quantitative communication between people can be obtained by wearing wearable sensors such as infrared sensors, acceleration sensors, and name tag type sensor nodes equipped with small microphones to members of the organization. This is based on the sensor data measured. From such data, a network structure having a person as a node can be obtained by establishing a link between persons with communication. In addition to data that is consciously worn and measured by such wearable sensors, the human relationship graph includes people and people who are unconsciously configured, such as mobile phone usage logs and e-mail transmission / reception relationships. It may be data reflecting the connection.

  In the present invention, an energy function based on a property on a human relationship graph of a state quantity to be measured, for example, stress, is defined in advance. This energy function is initially obtained by analyzing the relationship between the survey results based on the questionnaire and the network structure. For example, “a node connected to a high-stress person with one link is likely to be low-stress. In this case, the Ising model Hamiltonian describing the antiferromagnetic behavior is used as the energy function. be able to.

  Once such an energy function is defined for organizational stress, a questionnaire survey is conducted on all members of the organization when investigating the degree of stress at different times, such as after organizational reform. Even if it is not, if a questionnaire survey is conducted only for some members, for example, people of a certain position or higher, the stress state is estimated for other members by a process that minimizes network energy. Is possible.

  Furthermore, the present invention can change the number of links and nodes, change the link of the human relationship graph, or change the number of links and nodes in relation to personnel assignment changes and affiliation changes that cannot be implemented in practice or are difficult to try. Or, by changing the attribute of the node, by evaluating the macroscopic organization state obtained as a result of the energy function minimization process, for example, the degree of stress of the organization, the actual staff assignment change or affiliation change is performed. A simulator function is provided to estimate what will happen in the event of a failure.

According to the first solution of the present invention,
Using a human relationship graph in which the nodes of the network correspond to people and the links correspond to the relationships between people, an indicator that shows the reverse trend between adjacent people connected by links in the human relationship graph A human behavior analysis system for estimating an index value of another person based on a known index value of a part of a person,
In correspondence with the node identifier, a node attribute database in which a known index value for a node identifier whose index value is known, an estimated value of the index, and a link destination node identifier are stored correspondingly,
A processing unit that refers to the node attribute database and estimates a node index;
A process in which the processing unit determines an estimated value of the index of the node according to the known index value for at least a part of nodes for which a known index value is stored in the node attribute database; Processing for determining an initial value of the estimated value of the node index for the node of
The processing unit obtains the energy of the system configured by each node by using an Ising model based on the estimated value of the index of each node and the link information represented by the correspondence relationship between the node identifier and the link destination node identifier. When,
A process in which the processing unit repeats the calculation of the energy of the system again by changing the estimated value of the index of each node a predetermined number of times so that the energy of the system becomes smaller;
The processing unit determines an estimated value of the index of each node after a predetermined number of iterations or a statistical value of the estimated value of the index of each node in the repeated calculation process as an estimated value of the index and stores the estimated value in the node attribute database. The human behavior analysis system for performing processing is provided.

According to the second solution of the present invention,
Using a human relationship graph in which the nodes of the network correspond to people and the links correspond to the relationships between people, an indicator that shows the reverse trend between adjacent people connected by links in the human relationship graph A human behavior analysis method for estimating an index value of another person based on a known index value of a part of a person,
The processing unit is known in the node attribute database in which the known index value for the node identifier whose index value is known, the estimated value of the index, and the link destination node identifier are stored corresponding to the node identifier. A step of determining an estimated value of the index of the node according to the known index value and a processing unit for at least a part of the nodes storing the index value of the initial value of the estimated value of the node index for other nodes Steps to determine,
A processing unit that obtains an energy of a system constituted by each node by an Ising model based on an estimated value of an index of each node and link information represented by a correspondence relationship between the node identifier and the link destination node identifier; ,
A step of changing the estimated value of the index of each node and calculating the energy of the system again a predetermined number of times so that the energy of the system becomes smaller;
A step in which the processing unit determines an estimated value of the index of each node after it has been repeated a predetermined number of times or a statistical value of an estimated value of the index of each node in an iterative calculation process as an estimated value of the index and stores it in the node attribute database A method for analyzing human behavior is provided.

According to the present invention, it is possible to provide a human behavior analysis system and method capable of estimating the stress level of members constituting an organization or the entire organization using the characteristics of stress in a group.
In addition, the present invention solves these problems related to a method for investigating the degree of stress based on a questionnaire and a method for estimating the degree of individual stress from an individual index, and from the index values of some people, It is possible to estimate the index value.

In addition, according to the present invention, it is possible to estimate an index value such as the degree of stress by defining an energy function that determines the state of a system from observation data obtained by a sensor or the like, and making the energy function smaller or minimized. It is.
According to the present invention, it is possible to predict how an index value such as stress changes when the structure of the human relationship graph is changed.

It is a figure showing the whole human behavior analysis technique and system schematic diagram of the present invention. It is a figure which shows the relationship on the distance on a human relationship graph, ie, the number of paths, and an index value of this invention. It is a block diagram which shows the system configuration | structure of this invention. It is a figure which shows the example of the data set stored in the meeting information database of this invention. It is a figure which shows the example of the data set stored in the node attribute database of this invention. It is a flowchart which shows the process which estimates index value of this invention. It is a figure which shows the example (1) of simulation of this invention. It is a figure which shows the flowchart of simulation (1) of this invention. It is a figure which shows the experimental result of simulation (1) of this invention. It is a figure which shows the example (2) of the simulation which estimates the index value of a person with unknown index value of this invention. It is a figure which shows the flowchart of simulation (2) of this invention. It is a figure which shows the experimental result of simulation (2) of this invention. It is a figure which shows the example (3) of the simulation which estimates the whole index value at the time of changing the index value of a certain node of this invention. It is a figure which shows the experimental result of simulation (3) of this invention. It is a figure explaining correct | amending an estimated value using the index value of the person connected by two passes of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
1. Human Behavior Analysis System and Method FIG. 1 is a schematic diagram illustrating a human behavior analysis method and system processing flow in the present embodiment. Reference numeral 101 is input of human relation graph data representing human relations, reference numeral 102 is human relation graph analysis processing, reference numeral 103 is energy minimization calculation of human relation graph analysis processing, and reference numeral 104 is a human relation graph. The state determination process of each node in the analysis process, reference numeral 105 is a determination process of the macro state of the system in the analysis process of the human relation graph, and reference numeral 106 is a process of outputting data of the result of the analysis process of the human relation graph. is there.

  In FIG. 1, the human relationship graph data 101 includes, for example, face-to-face data obtained by wearing a wearable sensor such as an acceleration sensor, an infrared sensor, and a small microphone, for example, a name tag type sensor, and conversation with the face-to-face. Network information that reflects the combined communication data. The face-to-face data is, for example, face-to-face time obtained by detecting infrared rays from the infrared ray irradiating part of a person facing the face using an infrared ray sensor and an infrared sensor mounted on a name tag type sensor worn on a person. A person-to-person meeting is detected and converted into data. In addition to the above-mentioned face-to-face time obtained by, for example, an infrared sensor, the communication data combining the face-to-face and conversation may be data (indexed) taking into account the utterance time detected by a small microphone. I can do it. In addition to these, communication between people may be digitized by sensor data detected by an appropriate sensor.

  In this case, a node of the network corresponds to a person, and a link is drawn between nodes based on a rule that a link is established between nodes when communicating for a certain period of time or longer. For example, data relating the nodes is stored to create human relationship data. Even if it is not a human graph created by wearing wearable sensors like this, it reflects the connection between people that can be constructed from mobile phone usage logs, e-mail transmission / reception records, etc. Such information may be input as human relationship graph data to the human behavior analysis method and system of the present embodiment.

In the figure, the human relationship graph analysis processing 102 determines 104 the state of each node of the graph as a result of the energy minimization calculation 103 of the system from the human relationship graph data input at 101, and calculates the average value of all nodes. This is a process for determining 105 a macro state of the system by calculation or the like. The system here corresponds to an organization composed of a plurality of people corresponding to each node, for example.
In the figure, in the energy minimization calculation 103, the state of each node is set so that the energy of the entire system is reduced as much as possible while changing the state of each node (for example, the stress level of each node and the index value of stress) stochastically. The calculation is terminated when a certain period of time or a certain number of calculations have elapsed.

In the figure, the state of each node is determined 104 as a result of the energy minimization calculation 103.
In the figure, when the state of each node is determined 104, the macro state of the system, for example, the degree of stress of the organization, is determined 105 by, for example, calculating the average value of the index values of all nodes. In order to determine the macro state of the system, in addition to calculating the average value, the macro index value of the system can be calculated from the index value of each node by various formulas according to the definition of the macro index. It doesn't matter. For example, the specific heat, which is a macro quantity of the system, is calculated using the mean square of energy, the average of energy, and the temperature parameter, which are also macro quantities of the system.
In the same figure, as a result of the human relationship graph analysis process 102, when the state of each node (individual stress level) and the macro state of the system (organizational stress level) are determined, the human relationship graph reflecting the results Output data 106. For example, the results may be visually reflected and displayed on a network diagram, or macro values such as organizational stress may be output as numerical values or graphs, or output in the form of a matrix. May be.

FIG. 2 is a diagram for explaining an index (antiferromagnetic index) whose tendency is reversed between adjacent persons in the human relationship graph, which is an analysis target in the present embodiment.
In FIG. 2A, reference numeral 201 denotes a node number representing a person in the human relationship graph, reference numeral 202 denotes a CES-D value representing the stress level of the node itself, and reference numeral 203 denotes a node with node number i and one path, that is, one. The average value of the CES-D values of a plurality of nodes connected by the link of No. 204, the symbol 204 is the CES-D value of a plurality of nodes connected by following the two paths with the node of the node number i, ie, two links. The average value, 205, indicates an average value of CES-D values of a plurality of nodes connected to the node of node number i by r paths (r is a natural number), that is, r links. 2B plots a correlation coefficient R between the CES-D value of the node i and the average value of the CES-D of the node connected to the node i through the r path with respect to r. FIG. Reference numeral 207 in FIG. 2C shows an example of a human relationship graph. Reference numeral 208 denotes a focused node i, reference numeral 209 denotes a node connected to the node i through one path, and reference numeral 210 denotes a node connected to the node i through two paths.

FIG. 2 shows a case where the number of people constituting the human relationship graph, that is, the number of nodes is 100, but the number of nodes may be more or less.
In the figure, CES-D representing the degree of stress takes a value from 0 to 60, and the higher the value, the higher the degree of stress, and 16 or more are considered to have a higher risk of depression.
In the figure, when comparing the CES-D value 202 of the node number i with the average value 203 of the CES-D value of the node connected to the node of the node number i by one path, that is, one link, A tendency that the other is low when the other is high and the other is high when the other is low is observed. Thus, the CES-D value is a certain index (antiferromagnetic index) whose tendency is reversed between adjacent persons in the human relationship graph. When the two correlation coefficients (columns 202 and 203) are actually calculated, a negative value is obtained when r = 1 as shown in the graph 206. In the graph 206 showing the value of the correlation coefficient in the figure, when r = 0, the correlation coefficient R is naturally 1 because the correlation is between the CES-Ds of one's own.

ここで、Ｓ_ｉ及びＳ_ｊはｉ番目、ｊ番目のスピンであり、＋１または−１をとる。Ｊはスピン間の交換相互作用であり、隣接するスピンｉとｊの間でのみ値をとる。Ｊ＞０の場合は隣り合うスピンを同じ向きにそろえるように（強磁性）作用し、Ｊ＜０の場合は逆向きにそろえるように（反強磁性）作用する。ｈは外から加えられる磁場である。 In the figure, as shown in the correlation coefficient graph 206 with respect to the number of paths (distance) r, the correlation coefficient gradually attenuates while alternately taking positive and negative values as the number of paths r increases. Such a tendency of the correlation coefficient is very similar to the behavior of the spin correlation function in a substance called antiferromagnetism in a magnetic material. This suggests the following.
The tendency for the stress level to be reversed between adjacent nodes is reminiscent of an antiferromagnetic Ising model that expresses an interaction that attempts to reverse the spin directions of adjacent electrons in a magnetic material. The Ising model has played an important role in magnetic material science as a basic model of magnetic materials. This is a model in which the spin configuration of an electron is determined so as to minimize an energy function composed of a term relating to an interaction between spins (exchange interaction) and a term relating to an external magnetic field. The Ising model Hamiltonian H is given by:

Here, S _i and S _j are the i-th and j-th spins and take +1 or −1. J is an exchange interaction between spins, and takes a value only between adjacent spins i and j. When J> 0, the adjacent spins work in the same direction (ferromagnetism), and when J <0, they work in the reverse direction (antiferromagnetism). h is a magnetic field applied from the outside.

  In the human behavior analysis method and system according to the present embodiment, the stress level exhibits an antiferromagnetic property as shown in a graph 206. Therefore, the spin direction is made to correspond to the level of stress of individuals constituting the tissue, the relationship with others is represented by a negative exchange interaction J <0 (for example, J = −1), and the external magnetic field h is the delivery date of the product. If it is a stressor (stressing factor, stimulus) such as customer response, the stress in the organization can be modeled by the Ising model. This is because, when an external magnetic field h is applied to the magnetic material, the spin direction is determined so that the energy is minimized by interaction between the spins, and the macroscopic magnetization M appears as an average. When a stressor h such as customer service is added to an organization, the degree of stress is determined as a macro characteristic of the organization as a result of the interaction between the organization members. Here, the spin representing the degree of stress takes a binary value of ± 1, and the direction of the stressor h is +1. The closer the average spin value is to 1, the higher the stress of the tissue.

Actually, spins are assigned to each node for the network structure of the organization constructed from communication data measured using wearable sensors such as name tag type sensors, and the spin of each node determined as a result of energy minimization calculation is determined. When the correlation coefficient between the value of the spin of the node number i and the average value of the spin values of the node connected to the node of the node number i by r paths, that is, r links, is calculated using the value A graph similar in shape to the graph shown in 206 can be drawn.
In the figure, reference numeral 207 denotes an example of a human relationship graph, which is a network diagram composed of nodes and links. Here, a high stress person having a CES-D of 30 or more and 60 or less is represented by a white circle, and a low stress person having a CES-D of 0 or more and less than 30 is represented by a black circle, and so on.

  In the same figure, when there is a CES-D correlation as indicated by reference numeral 206, the human relationship graph is in a state as shown in FIG. That is, when the node i 208 of interest is a high stress person, that is, a white circle, the node 209 connected to the node i by one path (and odd path) tends to be low stress, that is, a black circle. 2 nodes (and even paths) connected to each other tend to be highly stressed, that is, white circles. Although not shown in the figure, the case where black and white are reversed can also be established, and when the focused node i 208 is a low stress person, that is, a black circle, the node 209 connected to the node i by one path is There is a tendency to become high stress, that is, a white circle, and the node 210 connected to the node i through two passes tends to be low stress, that is, a black circle.

FIG. 3 is a diagram showing a configuration of the human behavior analysis system according to the present embodiment.
The human behavior analysis system includes, for example, a display device 303, an input device 304, a communication device 305, a CPU (processing unit) 306, a hard disk 307, and a memory 308. Note that both the input device 304 and the communication device 305 may be provided, or one of them may be provided. The human behavior analysis system may further include a data management server 302. Furthermore, you may provide the various sensors which measure the data which show the relationship between people, such as the above-mentioned wearable sensor.

  In FIG. 3, each reference numeral 301 indicates a human relationship graph data such as face-to-face time and a DVD or the like that stores some or all node index values (for example, CES-D values representing the degree of stress obtained by a method such as a questionnaire). Reference numeral 302 denotes a data management server that stores a human relationship graph and index values of some or all nodes. Reference numeral 303 denotes a display device, 304 denotes an input device, 305 denotes a communication device, 306 denotes a CPU, 307 denotes a hard disk, and 308 denotes a memory. Reference numeral 309 denotes a face-to-face information database that stores face-to-face information that is human relation graph data, 310 denotes a node attribute database that stores index values for each node, and 311 denotes a human relation graph analysis program. 312 is an energy minimization calculation process in the human relation graph analysis program 311, 313 is a state determination process of each node in the human relation graph analysis program 311, 314 is a macro state determination process of the system in the human relation graph analysis program 311 Reference numeral 315 denotes an Internet network.

In the figure, data representing a human relationship graph can be obtained from a wearable sensor such as a name tag type sensor, obtained from a use log of a mobile phone or an e-mail, or obtained by other methods. Face-to-face information such as “how many are facing” and a known index value obtained from a questionnaire or the like, for example, the degree of stress. These are recorded on a recording medium 301 such as a DVD, or a data management server 302 is stored.
In the figure, the face-to-face time data and the known index values of some or all nodes in the human relationship graph data to be analyzed are the communication among the hardware configurations that execute the human behavior analysis processing of the present embodiment. The apparatus 305 inputs from the data management server 302 via the Internet 315, or the input apparatus 304 inputs as media data 301 recorded on a recording medium such as a DVD.

In the figure, the face-to-face information of the human relationship graph data to be analyzed input via the input device 304 or the communication device 305 is temporarily stored in the face-to-face information database 309 in the hard disk 307. In the figure, even when index values of some or all nodes to be analyzed are input via the input device 304 or the communication device 305, they are temporarily stored in the node attribute database 310 in the hard disk 307.
In the figure, when analyzing the face-to-face data and the index values of some or all nodes in the human relationship graph data, these data stored in the hard disk 307 are read into the memory 308, and the CPU 306 executes the analysis program 311. By doing so, the analysis process is executed.
In the figure, in order to perform the energy minimization calculation process 312, an energy function of a system is defined using an Ising model, and a calculation for minimizing the energy function is performed. The energy minimization calculation will be described in detail using the metropolis method as an example.

(Metropolis method)
First, the flow of processing based on the Metropolis method will be described. These processes are executed by the CPU 306. More specific processing using hardware resources will be described later in detail with reference to a database and a flowchart.

The energy E is defined by the above formula (1). H in Eq. (1) becomes E. Here, S _i is an index value of each node, for example, a value representing the degree of stress. Here, description will be made assuming that S _i takes either +1 or −1.
Operation 1: First, the initial state energy E is calculated according to the equation (1).
Operation 2: Next, the state of the system is changed. Here, the change is, for example, an operation of inverting the index value S _i of a certain node i. If S _i is +1 in the initial state, the operation is changed to −1, and if S _i is −1 in the initial state, the operation is changed to +1. It is.
Operation 3: When the state is changed, the energy is calculated again and set to E ′.
Operation 4: The difference ΔE = E′−E between the energy E ′ after changing the state and the energy E before changing the state is calculated.
Operation 5: If ΔE <0, the new state with the changed state has lower energy, so accept the state and repeat steps 2, 3, 4, 5 or 6 a predetermined number of times with E = E ′. .
Operation 6: If ΔE ≧ 0, the new state with the changed state has higher energy or the same, but in this case, it is determined whether or not to accept the new state stochastically. Specifically, a random number rand from 0 to 1 is generated. If rand <p as compared with a certain probability p, a new state is accepted, E = E ′ is set, and operations 2, 3, 4, 5, or 6 are performed. Do. Otherwise, it returns to the state before changing the state. That is, S _i whose state has been changed is reversed again to return to the original state. Thus, even when ΔE ≧ 0, a new state is stochastically accepted and, for example, it is avoided to fall into a local minimum. Thereafter, the operation 2, 3, 4, 5 or 6 is repeated. Here, the probability p is, for example, p = exp (−ΔE / T) when the energy distribution of the system follows a canonical distribution, and T is a temperature parameter.
The above operations 1, 2, 3, 4, 5 or 6 are performed a predetermined number of times, and the calculation is terminated.
The energy minimization calculation may use a calculation method other than the metropolis method.

In the figure, when the energy minimization calculation process 312 is performed and the index values of all the nodes are unknown, the initial state in operation 1 is randomly determined. That is, the index value of each node is set to +1 or −1 using random numbers, and the energy minimization calculation is repeated a predetermined number of times. At this time, the index value is set in the data set stored in the node attribute database 310, and the updated value is reset every time the energy minimization calculation is performed (see the description of FIG. 5).
In this figure, when the energy minimization calculation is performed, if some or all of the index values of nodes are known by, for example, questionnaire surveys or other estimation methods, those node index values are binarized by a certain threshold value. And assigning a value of +1 or -1 and assigning +1 or -1 at random to a node whose index value is unknown is set as an initial state. Also at this time, the index value is set in the data set stored in the node attribute database 310, and the updated value is reset every time the energy minimization calculation is performed. Note that the above-described threshold value can be determined in advance.

Ｑは状態数であり、例えば所定計算回数を３００，０００回とし、このうち５回ごとにＡを計算するとすれば、Ｑ＝６０，０００である。この方法により各ノードの状態を決定３１３する場合には、Ｓ_ｉは＋１あるは−１の２値をとるが、（２）式のＡをＳ_ｉとした平均＜Ａ＞を計算するために、状態は−１以上１以下の実数値になる。 In the figure, as a result of the energy minimization calculation 312, the state of each node can be determined 313. As a method for determining the state of each node, the state of each node when the energy minimization calculation is repeated a predetermined number of times, that is, the stress value +1 or −1 of each node may be adopted, A value obtained by dividing the value of the state sampled during the energy minimization calculation of the number of times by the number of times of sampling, a so-called thermodynamic average value may be used. The thermodynamic mean <A> of a quantity A is calculated as follows:

Q is the number of states. For example, if the predetermined number of calculations is 300,000, and A is calculated every 5 times, Q = 60,000. When determining the state of each node 313 by this method, S _i takes a binary value of +1 or −1, but in order to calculate the average <A> where A in equation (2) is S _i The state is a real value between −1 and 1 inclusive.

となる。Ｎはノード数である。なお、上述のＭは、熱力学的な平均値をとっても良いし、通常の平均値でもよい。
同図において、分析プログラム３１１の計算結果は、その結果を反映した人間関係グラフデータをネットワーク図として視覚的に表示部３０３に表示されたり、マトリクス形式のデータとしてハードディスク３０７に格納される。 In the figure, if the state of each node is determined 313, the macro state of the system can be determined 314. For example, if the average stress level of organization and M, M is calculated as the average of the values of S _i of each node,

It becomes. N is the number of nodes. The above M may take a thermodynamic average value or a normal average value.
In the figure, the calculation result of the analysis program 311 is visually displayed on the display unit 303 as human relationship graph data reflecting the result, or stored in the hard disk 307 as matrix format data.

FIG. 4 is a diagram (a) showing an example of a data set related to face-to-face time information among human relationship graph data input from the outside in the human behavior analysis method and system of the present embodiment, and a drawing using the data set. It is a figure (b) which shows the example of a network diagram which can be performed. The data set relating to the meeting time information is stored in the meeting information database 309 in FIG.
In the figure, reference numerals 401, 402, 403, and 404 indicate user IDs arranged in the row direction, and reference numerals 405, 406, 407, and 408 indicate user IDs arranged in the column direction. Reference numerals 409, 410, 411, and 412 denote nodes in the network diagram in which users are drawn as nodes.
In the figure, a matrix element represents a meeting time between users constituting an organization, for example, a meeting time obtained by attaching a name tag type wearable sensor equipped with an infrared sensor to members of the organization. As described above, the face-to-face information database 309 stores face-to-face information (for example, face-to-face time) between users corresponding to combinations of user IDs. An appropriate format may be used instead of the matrix form as shown in the figure.

In the figure, the meeting time of the matrix element represents, for example, the average meeting time for each day in minutes. The method of measuring the meeting time may be measured using a wearable sensor as described above, or other methods may be used.
In the same figure, the same users, for example, User1 (401 and 405), User2 (402 and 406), User3 (403 and 407), and User100 (404 and 408) are facing each other, so the information is zero. Therefore, 0 is entered. In actual data, appropriate data other than 0 may be stored, or may be blank.
In the figure, as an example, the matrix elements of User1 (401) and User2 (406) are 13.55, which indicates that, for example, User1 and User2 face each other for an average of 13.55 a day.
If a data set relating to the meeting time as shown in the figure is used, information on who and who are facing each other in the organization can be obtained.

A network diagram representing the relationship between people in the organization can be drawn from the data indicating such a face-to-face relationship. In this case, if a user is a node and a rule such as “draw a link between nodes having a meeting time of 5 minutes or more” is defined, a network diagram composed of nodes and links can be drawn.
In the figure, when a network diagram is drawn using the rule of “draw a link between nodes having a meeting time of 5 minutes or more”, the matrix elements of User1 (401) and User2 (406) are 13.55. A link is drawn between 1 (409) and node 2 (410). Similarly, since the matrix elements of User1 (401) and User3 (407) are 15.7, a link is also drawn between node 1 (409) and node 3 (411). Since the matrix elements of User2 (402) and User3 (407) are 3.75, no link is drawn between node 2 (410) and node 3 (411). Similarly, links are established between the node 100 (412) and the node 1 (409), and between the node 100 (412) and the node 3 (411). In this way, the network diagram shown in FIG. 4B can be drawn. In addition to the threshold value of 5 minutes, an appropriate predetermined time may be set as the threshold value.

FIG. 5 is a diagram for explaining the configuration of the node attribute database 310 in the system configuration described in FIG.
The node attribute database 310 includes a node number (node identifier) indicated by 501 in the figure, a CES-D value (observed value, known index value) measured by a questionnaire survey indicated by 502, and an estimated value of an index indicated by 503. , 504, and corresponding link destination node numbers. In the figure, reference numeral 505 denotes a node for which an index value such as a stress level is obtained by a method such as a questionnaire, and 507 denotes a node for which no answer is obtained by the questionnaire survey and the CES-D value is unknown.
In the figure, regarding the CES-D value obtained by the questionnaire, the value (input value) is stored as it is in the column 502 of the data set for the node 505 where the value is obtained, and the value is not obtained. For the node 506, for example, a null character indicating “not answered” is set in the column 502.

In the data set shown in the drawing, for example, +1 or −1 is assigned to the estimated value column 503 as an initial value, and each time the energy minimization calculation 312 is repeated, the value of the column 503 is assigned to each node. Update.
In the same figure, when using the meeting information database 309 described in FIG. 3 and FIG. 4 and when a link is drawn when the meeting time is a certain threshold value or more, the number of the link destination node of each node is represented in the column of the data set. Stored in 504. For example, for one of the nodes at both ends of the link, the number of the other node is stored in the link destination node number, and vice versa. Thereby, a human relationship graph can be created. Note that the link destination node number is based on the input face-to-face information and is stored by the CPU 306. In addition, the link number between the node number 501 and the link destination node number 504 (link information) is determined in advance by an appropriate device such as the data management server 302. May be created and input from the data management server 302, the recording medium 301, or the like.

FIG. 6 is a diagram illustrating a flowchart of the process of estimating the stress level.
In the figure, a process 601 is a human relationship graph data input process, a process 602 is a calculation of energy E in an initial state, a process 603 is a calculation of energy E ′ when the state is changed, and a process 604 is an energy difference ΔE = E ′. -E calculation, process 605 is a process for determining whether ΔE is positive or negative, process 606 is a process for accepting a state change, and sets the energy E of the system to E ′, and process 607 is a process for determining whether the energy minimization calculation has been performed a predetermined number of times. A process 608 is a random number generation process of 0 to 1 and a process 609 is a process for determining whether the generated random number is larger or smaller than a certain value p. A process 610 is to return to the state before changing the system state. Processing 611 is processing for determining the index value of each node, and processing 612 is processing for determining the macro state of the system.

  First, in the human relationship graph data input processing 601, the CPU 306 inputs human relationship graph data from the recording medium 301 or the data management server 302 and stores it in the node attribute database 310 or the like. The human relationship graph data input in the human relationship graph data input process 601 includes link destination node information created from a face-to-face information data set 309 measured by a wearable sensor equipped with an acceleration sensor, an infrared sensor, a small microphone, etc., a questionnaire, etc. This is a node attribute data set 310 storing a known index value and the like for a certain node measured by means. If the link destination node information is not created, the CPU 306 inputs the face-to-face information data from the data management server 302 or the like, stores it in the face-to-face information database 309, and follows the rules for establishing the above-described link from the face-to-face information data. The link destination node information of the node attribute database 310 is created. For example, for each node ID in the input meeting information data, the meeting information data (meeting time) is compared with a threshold value, and if it is larger than the threshold value, there is a relationship, and the link destination node information corresponding to one node ID is Write the other node ID. The reverse is performed in the same manner for both node IDs.

  In the calculation 602 of the energy E in the initial state, the CPU 306 refers to the data set stored in the node attribute database 310, and the spin value of each node to which a value of +1 or −1 is allocated as an estimated value ( Index value) is obtained, and the energy function E is calculated according to the equation (1). The CPU 306 stores the estimated index value and the obtained energy function in the memory 308, for example. At this time, when the index values 502 of all the nodes are unknown, for example, the estimated value 503 of the index value of each node is determined to be +1 or −1 by using random numbers, and the energy E is calculated. When some or all of the node index values 502 are known by, for example, questionnaire surveys or other estimation methods, the node index values 502 are binarized with a predetermined threshold value, and the estimated value 503 is incremented by +1 or Assign a value of -1. For other nodes whose index value 502 is unknown, the estimated value 503 randomly assigned with +1 or -1 is set as the initial state, and the energy E is calculated.

Here, in the above formula (1), H is the energy of the system, S _i and S _j are estimated values of indices corresponding to the node identifiers i and j, and take +1 or −1, and J is between spins It is a parameter corresponding to the exchange interaction, and a negative value is used for the product of the estimated value of the index of the node identifier i in the node attribute database and the estimated value of the index of the link destination node identifier j corresponding to the node identifier i. In other cases, it is 0, and h is a predetermined value. The CPU 306 sets parameters J and h in equation (1) to J = −1 and h = 1. Here, with respect to the first term of the expression (1), the nodes i and j with links are in accordance with the node number 501 (corresponding to the node i) and the link destination node number 504 (corresponding to the node j) in the node attribute database 310. Calculated. 0 for nodes without links.

In the calculation 603 of the energy E ′ when the state is changed, the CPU 306, for example, sets the index value S _i of a certain node i to −1 if it is +1 in the current state, and −1 in the current state. If so, the energy is changed to +1, and the energy is calculated in accordance with the equation (1) in this state (here, it is assumed as energy E ′). Until the predetermined number of calculations elapses 607, the estimated value of the index is stored in the memory. Here, the node whose value is changed can be selected at random.
The CPU 306 determines whether the obtained energy E ′ is smaller than the original energy E. For example, the CPU 306 calculates a difference ΔE = E′−E between the energy E in the initial state and the energy E ′ when the state is changed in the energy difference calculation 604, and whether ΔE is positive or negative in the determination process 605. judge.
If ΔE <0, the CPU 306 accepts the state because the new state with the changed state has lower energy, and sets E = E ′ (step 606).

On the other hand, if ΔE ≧ 0, the new state with the changed state has higher energy or the same energy. In this case, the CPU 306 determines whether or not to accept the new state stochastically. Specifically, for example, the CPU 306 generates a random number rand from 0 to 1 (process 608), compares the random number rand with a predetermined probability p (process 609), and if rand <p, a new state is obtained. If it is accepted, E = E ′ is set (process 606). Otherwise, the state is returned to the state before the change (process 610). That is, the CPU 306 inverts the estimated value S _i of the node index changed in the process 603 again, that is, multiplies S _i by −1 to restore the original state.
The CPU 306 determines whether or not the energy minimization calculation in the processes 603 to 606 and 608 to 610 has been performed a predetermined number of times (process 607). If the result of determination is that the predetermined number of times has not been reached, the CPU 306 returns to process 603 to calculate 603 for energy E ′ when the state is changed again, and thereafter processes 604, 605, 606, 609 and 610 described above. Execute. In the process 603, the nodes are randomly changed as described above. For example, the state of a node different from the node changed in the immediately preceding process may be changed.

On the other hand, if the CPU 306 reaches the predetermined number of times as a result of the determination process (607) on whether or not the energy minimization calculation has been performed a predetermined number of times, the state of each node is determined (process 611). The predetermined number of times described above can be determined in advance. As a method for determining the state of each node, the state S _i of each node when the energy minimization calculation is completed a predetermined number of times may be adopted, or sampling is performed in the middle of performing the energy minimization calculation a predetermined number of times. A value obtained by dividing the value of the state (the state sampled at each sampling interval) by the number of samplings, a so-called thermodynamic average value may be used. The thermodynamic average value of a quantity A can be calculated by equation (2) as described above. The sampling interval can be set in advance. The sampled state value (or its accumulated value) and the number of samplings can be stored in the memory 308.

The CPU 306 determines the macro state of the system based on the determined state of each node (process 612). For example, if the average stress level of the organization is M, M is defined as the average value of the index values S _i of the determined nodes, and the CPU 306 determines the average stress level M of the organization according to the above equation (3). calculate.
The CPU 306 stores the estimated value of the determined index value of each node (process 611) and the amount indicating the macro state of the system (process 612) in the hard disk 307 after a predetermined number of calculations have elapsed (607). Moreover, CPU306 displays on the display part 303 visually, for example as a network figure.
If the human behavior analysis method and system of this embodiment are used, the stress level of the person constituting the system can be estimated by a simulation by an operation of optimizing the energy of the system, It is possible to estimate a macro state, that is, the degree of stress of an organization.

An example of simulation by the human behavior analysis system will be shown below.

2.1 Simulation example 1
FIG. 7 is a diagram for explaining a simulation example (1).
In the figure, 701 is an actual human relationship graph and stress level, 702 is a human relationship graph and stress level when the number of links between nodes is reduced, and 703 is a human relationship graph when the number of links between nodes is increased. Stress level, 704 is a human relationship graph and stress level when the number of nodes and links are reduced, 705 is a human relationship graph and stress level when the number of nodes and links is increased, and 706 is a constant number of nodes The relationship graph and stress level when links are changed. 707 to 718 are nodes, 719 is a deleted node, 720 is a node, 721 is an added node, and 722 to 729 are nodes.

In the figure, the degree of stress is represented by a binary value of +1 or -1, high stress is represented by +1 and white circle, and low stress is represented by -1 and black circle.
In the same figure, the actual human relationship graph and the stress level 701 are those of the human relationship graph 207 shown in FIG. 2 when an index, for example, the stress level, whose magnitude tendency is reversed between adjacent nodes is targeted. Thus, if high stress is expressed by a white circle and low stress is expressed by a black circle, if a node of interest is a high stress person, that is, if it is a white circle, a node connected to that node in one pass is low stress, There is a tendency to become a black circle, and a node connected by two passes tends to be a high stress, that is, a white circle. Although not shown, a state in which black and white in the human relationship graph 701 is reversed can also be realized.
In the figure, the case of reducing links 702 and the case of increasing links 703 are applied to human relationships in an actual organization, for example, due to physical restrictions such as instructions from senior managers, organizational measures, and seat arrangement changes. , Corresponding to the disappearance of communication opportunities between specific persons among the members of the organization or communication between persons who have not communicated with each other.

  As shown in FIG. 7A, in the case of reducing the number of links 702, for example, the link between the node i707 and the node j708, which has been conventionally connected by one path, is broken. Nodes k709 and l710 connected with high stress, that is, white circle 714 and connected to node j with one path are both high stress, that is, white circles are low stress, ie, black circles 715 and 716. It happens.

As shown in FIG. 7B, in the case of increasing the number of links 703, for example, when a link is added to the node n711 or the node o713, a conventional high stress, that is, a white circle is a low stress state, that is, a black circle 717 and Change to 718 occurs.
In FIG. 7, the case of reducing the number of nodes 704 and the case of increasing the number of nodes 706 are applied to human relationships in an actual organization, for example, when there are no organizational members from a certain organization due to, for example, affiliation changes or personnel changes, or there are Each time a person joins an organization,
For example, the CPU 306 adds or deletes the link destination node identifier of the node attribute database 310 according to the input by the operator, and simulates a macro index of the system when the link is added or deleted.

  As shown in FIG. 7C, when the number of nodes is reduced 704, for example, when the node m 712 or 719 is deleted, a new link is generated between the node i 707 and the node n 711 that have not been directly connected to each other. As a result, the state of the node n changes from a high stress 711 to a low stress 720, that is, from a white circle to a black circle.

As shown in FIG. 7D, when the number of nodes is increased 705, for example, the node 721 is added to change the network structure, so that the node j708 which has been conventionally low stress, that is, a black circle, becomes high stress, that is, a white circle 722. Therefore, the node k709 and the node l710 connected to the node j by one path both have high stress, that is, what has been white circles is low stress, that is, 723 and 724 become black circles.
For example, the CPU 306 adds or deletes the node identifier in the node attribute database 310 according to the input by the operator, and simulates a macro index of the system when the node is added or deleted.

As shown in FIG. 7E, in the case where links are changed while keeping the number of nodes constant and the total number of links constant, 706 is applied to human relations in an actual organization. Due to physical constraints such as general measures and seat arrangement changes, the total communication volume and organizational members are the same, eliminating the opportunity for communication among specific members of the organizational members, Corresponding to creating and communicating time between people who were not communicating. Even in such a case, since the network structure changes, the state of each node changes. In the example shown in FIG. 706, the nodes k709, l710, and n711, which were conventionally high stress, become low stress. 726, 727, and 729, and the node j708 and the node m712, which have been conventionally low stress, become high stress 715 and 728.
For example, the CPU 306 deletes at least one of the link destination node identifiers corresponding to a predetermined node in the node attribute database 310 according to the input by the operator, adds another node identifier as the link destination node identifier, and switches the link. The system's macro index is simulated.

  As described above, the state of each node changes due to the increase / decrease of the link, and as a result, the stress level M of the entire organization calculated by, for example, the expression (3) also changes. In this way, it is possible to estimate the change in M due to various link increases and decreases by simulation, and to control workplace communication so that the stress level M of the entire organization is minimized.

2.2 Flowchart 1
FIG. 8 is a diagram illustrating a flowchart of 703 in the example (1) of the simulation described with reference to FIG. 7 when the number of links is increased while the number of nodes is constant.
The flowchart shown in FIG. 6 is basically the same as the flowchart shown in FIG. 6, but in FIG. 6, the CPU 306 determines the nodes from the data set stored in the face-to-face information database 309 in the human relationship graph data input 601. A data set stored in the attribute database 310 was created and input as human relationship graph data.

  When the number of links in FIG. 7 is increased and the simulation of 703 is performed, the CPU 306 has changed (added here) 801 the link destination node number 504 of the data set in FIG. 5 from the original data set. Is input 601 as human relationship graph data. The CPU 306 adds or deletes a node by adding or deleting the node number 501 already input and stored in the node attribute database 310, and adds or deletes a link by adding or deleting the link destination node number 504. Then, by deleting at least one of the link destination node numbers 504 for a certain node and adding the numbers of other nodes, the links are reconnected (process 801), and the new human relation graph data thus created is used. Then, each processing after the processing 602 may be executed.

In FIG. 8, the flow after processing 602 is the same as that shown in FIG.
In addition to the case where the number of links is increased, for each case shown in FIGS. 7A, 7 </ b> C to 7 </ b> E, the CPU 306 performs a process corresponding to each case described above in process 801.

2.3 Experimental result 1
FIG. 9 shows an example of the simulation described with reference to FIG. 7, in which the number of nodes is increased 703 when the number of links is increased and the number of links 703 is changed while the number of nodes and the number of links are fixed. FIG.
In the figure, reference numeral 901 indicates the correlation between the average stress level of the organization in 703 when the number of links is increased and the number of links is increased, that is, M calculated by the equation (3), and the number of links per node. Is a diagram showing the correlation between the average stress degree M of the organization and the energy of the system in 706 when the links are changed while keeping the number of nodes and the number of links constant. A p-value indicating significance is shown.

In the figure, when 35 networks (human relation graphs) with different numbers of links are created and each network node has an index value that takes a value of +1 or −1 corresponding to the degree of stress, an energy formula ( By calculating 1) and performing the calculation according to the flowchart shown in FIG. 8 so as to reduce the energy, FIG. 901 showing the correlation between the average stress degree M of the organization and the number of links per node can be obtained.
In the figure, 20 networks (human relationship graphs) are created by connecting links while keeping the number of nodes and the number of links constant, and each network node takes a value of +1 or -1 corresponding to the degree of stress. FIG. 902 showing the correlation between the average stress level of the tissue and the energy of the system by calculating the energy formula (1) when the index value is included and performing the calculation according to the flowchart shown in FIG. Can be obtained.

In FIG. 901, it can be seen from FIG. 901 that the average stress M of the organization decreases as the number of links increases. This indicates that, in an actual organization, the degree of stress M is lower when communicating with many people compared to an organization with less communication.
Referring to FIG. 902, it can be seen from FIG. 902 that a network of various energy states is constructed by changing the number of nodes and the number of links while keeping the number of links constant, and that the average stress M of the organization decreases as the energy becomes lower. This shows that, in an actual organization, even if the total communication amount is constant with the same members, the degree of stress of the organization as a whole can be lowered by changing the communication method.

3.1 Simulation example 2
FIG. 10 is an example (2) of the simulation for estimating the index values of other nodes from the index values of some nodes.
In the figure, 1001 is a human relationship graph when only the index values of some nodes are known, and 1002 is a human relationship graph in a state where the index values of all nodes are estimated. Reference numeral 1003 denotes a high stress node whose index value is known, and reference numerals 1004 to 1006 denote low stress nodes whose index value is known.

In the figure, before the simulation 1001, one node 1 (1003) is high stress, that is, a white circle, three nodes 2 (1004), node 3 (1005), and node 4 (1006) are low stress, That is, it is assumed that it is a black circle from, for example, a questionnaire survey for measuring CES-D performed on these four people. The index values of the four people may be obtained by a questionnaire as described above, or values estimated using other stress value estimation means may be used.
In the figure, before the simulation execution 1001, the stress level of the other nodes is unknown, and is represented by a gray circle in 1001 in the figure.
In the figure, when the simulation is executed using the human behavior analysis method and system of the present embodiment, the stress value of the node represented by the gray circle that was unknown before the simulation is determined by the procedure described below. Can do.

First, when the energy minimization calculation is performed according to the energy function (1) according to the flowchart shown in FIG. 6, the high stress node 1 always has a spin value S ₁ of +1, and the low stress nodes 2, 3, 4 Performs the calculation with the spin values S ₂ , S ₃ , and S ₄ always fixed to −1. That is, in the flowchart shown in FIG. 6, in the process 603 for changing the state, if the state of the node 1, 2, 3, or 4 is changed, the state is immediately restored and the index value is not known. The energy minimization calculation is performed by changing the state of other nodes that are not present.
In this figure, if the energy minimization calculation is executed a predetermined number of times, the index value of the node whose index value is unknown can be determined, and the human relationship graph 1002 reflecting the result can be output.
In this figure, the present embodiment is a method for estimating the index value of a node whose index value is unknown from a limited number of index values of known nodes by calculation that minimizes the energy of the system. The method used to measure the stress level by conducting a questionnaire survey on all members of the organization, but it is possible to estimate the overall stress level at a low cost. Become.

3.2 Flowchart 2
FIG. 11 is a flowchart of the simulation described with reference to FIG.
The flowchart shown in the figure is basically the same as the flowchart shown in FIG. 6, but in FIG. 6, from the data set 309 stored in the face-to-face information database in the human relationship graph data input 601 to the node attribute database. A stored data set 310 was created and input as human relationship graph data.

When the simulation described with reference to FIG. 10 is performed, a part of the estimated index 503 of the data set stored in the node attribute database in FIG. 5 that is set to a fixed value 1101 is input as human relationship graph data. 601.
In FIG. 11, the subsequent flow is the same as that shown in FIG. The process 603-2 is the same as the process 603 of FIG. 8, but as described in the description of FIG. 10 above, the estimated value is not changed for the node that fixes the estimated value of the index. Which node is fixed can be predetermined. For example, some or all of the nodes with known index values (observed values) can be used.

3.3 Experimental result 2
FIG. 12 is an example of a simulation experimental result for estimating index values of other nodes from index values of some nodes described in FIG. In the figure, (a) shows each value before simulation, and (b) shows each value after simulation.
In this figure, 1201 is a node number, 1202 is a CES-D value measured by a questionnaire, 1203 is an estimated value of an index, 1204 is a node number of a link destination, 1205 is an estimated value of an index value is −1 1206 is a node in which the estimated knowledge of the index value is unknown, 1207 is a node in which no answer is obtained by the questionnaire survey, and the CES-D value is unknown, 1208 is a CES-D value by the questionnaire survey The node being obtained.

An outline of the experiment will be described.
For an organization with 17 members, CES-D was measured by questionnaire survey and responses were obtained from 11 people. Among these, for the three people 1205 whose CES-D value is 16 or less and the stress level is low in the field of psychology, the estimated value of the index value in the simulation is fixed to −1, and other values are calculated by energy minimization calculation. A simulation experiment was conducted to estimate human values. Assume that the estimated value is obtained by simulation for the node 1206 in which the CES-D value is known by the questionnaire but is highly stressed and the node 1207 in which the answer is not obtained by the questionnaire and the value is unknown. Here, in the simulation, the calculation for minimizing the energy function of the equation (1) was performed by the metropolis method with the stress level of the node i, that is, the index value S _i being a binary value of +1 and −1. In addition, the calculation of the index value of each node after the calculation reaches a predetermined number of times and the calculation of the average stress level of the tissue were calculated by the thermodynamic average (Formula 3). Therefore, the estimated value takes a real value between −1 and +1.
In the experiment, an estimated value of a person 1206 whose answer is actually obtained by a questionnaire but whose CES-D is larger than 16 and an estimated value of a person 1207 whose answer is not obtained by a questionnaire were calculated by simulation.

As a result of the simulation using the method of the present embodiment, estimated values are obtained for all nodes as shown in FIG.
In this figure, when the correlation coefficient R between the CES-D value and the estimated value is calculated for a person 1208 whose CES-D value is known from the questionnaire, R = 0.687, and the p value indicating significance is 0. .019. From this, the distribution in the human relationship graph of the estimated stress level obtained as a result of the simulation shows the same tendency as the distribution of the human relationship graph in the CES-D value actually obtained in the questionnaire. Can be reproduced.
In the figure, an estimated value is obtained for the node 1207 in which no answer was obtained in the questionnaire survey and the CES-D value was unknown by simulation. This means that even if a questionnaire survey is not conducted for all the members of the organization, if only a part of the survey is conducted, the energy minimization calculation proposed in this embodiment can be used to calculate the It indicates that the index value can be estimated.

4.1 Simulation example 3
FIG. 13 is a diagram illustrating an example (3) of the simulation for estimating the index values of other nodes when the index value of a certain node is changed.
In the figure, reference numeral 1301 denotes a human relationship graph before simulation execution, and 1302 denotes a human relationship graph after simulation execution. Reference numeral 1303 denotes a node for changing the index value.
In the figure, it is assumed that the node i1303 is in a high stress state, that is, a white circle before the simulation is executed. In this situation, when the state of the node i 1303 changes to a low stress state, that is, a black circle for some reason, a simulation for estimating a change in the state of another node can be performed by the method of this embodiment. Here, for some reason, in an actual organization, for example, there is a change in the environment such as full concern and care around the high-stress person i or an increase in the reward for i. Or, instead of Mr. i, it may be a personnel reason in the organization that is very strong against stressors, so to put another Mr. i 'of the optimist type in Mr. i's position.
In the figure, the purpose of the simulation is to estimate which person constituting the organization can support the degree of stress M of the organization.

In the figure, the state of the node i is changed from the high stress state, that is, the white circle to the low stress state, that is, the black circle, and the index value (estimated value) of the node i is always kept at −1. If the energy minimization calculation is performed according to the flowchart shown in FIG. 6, a human relationship graph 1302 reflecting the result can be obtained, and the overall state accompanying the state change of the node i, that is, the change in the stress level M of the organization is estimated. Is possible.
For example, the CPU 306 selects a specific node by, for example, an input from the operator, and inverts and fixes the estimated value of the corresponding node number in the node attribute database 310. In this state, the process 602 and the subsequent steps in the flowchart are executed to obtain a macro second index of the system when the estimated value of the node index is inverted. Further, the CPU 306 may display the macro index of the system before the inversion of the estimated value, the obtained second index, the estimated value of the node index, and the like.

4.2 Experimental result 3
FIG. 14 is a diagram showing the experimental results of the simulation described in FIG.
In the figure, 1401 is a stress level of an organization calculated by binarizing the stress level CES-D obtained by the questionnaire survey to ± 1, 1402 is a node having the maximum CES-D value in the same organization, that is, a high stress person, When the simulation is performed with the estimated value of the index value fixed at −1, the degree of stress of the organization is 1403, the index value of one node having the CES-D value of 16 or less, that is, the low-stress person in the same organization The degree of stress of the tissue when the simulation is performed with the estimated value of -1 being fixed to -1.

The contents of the experiment will be described.
First, the stress level CES-D of each member of the organization consisting of 17 people was measured by questionnaire, and the stress level of the organization was set to -1 for those with a value of 16 or less and +1 for those with a value greater than 16. M was calculated according to equation (3). Specifically, N = 17 and S _i take +1 or −1 in the equation (3), and in this case, it is not necessary to take a thermodynamic average. As a result of this calculation, M = 0.545545.

Next, we simulated what happens when a person who has the highest CES-D value and the highest stress in the same organization is rescued by some kind of support and measures are taken to reduce the stress in the real world. For this purpose, the estimated value of the node with the highest CES-D value (CES-D = 46) is fixed to −1, and energy minimization calculation is performed using the method of the present embodiment according to the flowchart shown in FIG. The index value S _i of each node was obtained, and the stress level M of the tissue was calculated according to the equation (3). This resulted in M = 0.447 1402.
In addition, we simulated what happens when measures are taken in the real world for people with low CES-D values and low stress in the same organization, with some kind of further support. For this purpose, the estimated value of the node of CES-D = 14 is fixed to −1, and the index value S _i of each node is calculated by energy minimization calculation using the method of the present embodiment according to the flowchart shown in FIG. The degree of stress M of the tissue was calculated according to the equation (3). As a result, M = 0.462 was 1403.

From the results shown in the figure, it can be seen that in this experiment, the stress level of the entire organization can be lowered by providing some support to a highly stressed person. On the other hand, the results show that the stress level of the entire organization increases if people with low stress are supported.
From the results shown in the figure, it can be seen that in this experiment, the overall stress level can be adjusted by controlling the stress of the specific person 1303 as described in FIG. Therefore, it is shown that stress can be controlled by using a simulator based on the method of the present embodiment, for example, for manager operation support of a manager.

5. Method for Using Values Related to Two Passes for Correction FIG. 15 is a diagram illustrating an example (modification) of a correction method in the case of estimating the stress value of the members constituting the tissue.
In the figure, 1501 is a target node i for which an index value is estimated, 1502 is a node j connected to the node i by one path, 1503 is a node j connected to the node i by one path, and node i is connected to the node i by two paths. k.
In the figure, there are one or more nodes j connected to the node i by one path and one node k connected to the node i by two paths.
In the figure, the method of the present embodiment that has been described so far utilizes the fact that the index value has a reverse tendency between node i and node j connected to node i through one path, In the energy function defined in (1), the energy minimization calculation was performed on the assumption that J representing the exchange interaction was negative.

As shown in the graph 206 in FIG. 2, the node i and the node k connected to the node i through two paths have a positive correlation coefficient of the magnitude of the index value, and 2 if the stress degree of the node i is high. The stress level of the node k connected by the path is high, and conversely, if the stress level of the node i is low, the stress level of the node k tends to be low.
Therefore, in order to improve the accuracy of the estimation of the stress level of the node i, using the index value of the node k connected to the node i through two paths is one of the methods for correcting the estimated value (in the present embodiment). As a variation).

ここで、Ｈは系のエネルギー、Ｓ_ｉ、Ｓ_ｊ、Ｓ_ｋは（１）式と同様±１の２値をとり、ストレス度の高低を表す。この場合、第１項は１パスで繋がったノードｉとｊの相互作用を表し、Ｊが負（例えばＪ＝−１）で、互いにＳ_ｉ、Ｓ_ｊを逆向きに揃わせようとする効果がある。第２項が補正項に相当する項であるが、２パス離れたノードｉとｋの相互作用を表し、Ｊ’は正（例えばＪ’＝１）で、互いにＳ_ｉ、Ｓ_ｋを同じ向きに揃わせようとする効果がある。２パスのリンクないノード間については、Ｊ’＝０である。ｈは（１）式と同様である。
このエネルギー関数を用いることにより、指標値を推定する対象ノードｉの指標値、例えばストレス度を、１パスで繋がったノードｊのストレス度と、２パスで繋がったノードｋのストレス度から推定することが可能になる。 For example, instead of the energy function (1), the following energy function to which the effect of the node k is added is used.

Here, H is the energy of the system, and S _i , S _j , and S _k take a binary value of ± 1 as in equation (1), and represent the level of stress. In this case, the first term represents the interaction between nodes i and j connected by one path, J is negative (for example, J = −1), and S _i and S _j are arranged to be aligned in opposite directions. There is. The second term is a term corresponding to the correction term, but represents the interaction between nodes i and k separated by two paths, J ′ is positive (eg, J ′ = 1), and S _i and S _k are in the same direction. There is an effect of trying to align. J ′ = 0 between nodes with no two-path link. h is the same as in equation (1).
By using this energy function, the index value of the target node i for which the index value is estimated, for example, the stress level, is estimated from the stress level of the node j connected by 1 path and the stress level of the node k connected by 2 paths. It becomes possible.

The second term is calculated for nodes connected by two paths, but the nodes connected by two paths by referring to the link destination node number 504 corresponding to the node number 501 of the node attribute database 310 twice. Can be judged. For example, referring to the node attribute database 310, a link destination node number 504 (for example, node j) corresponding to the node number 501 (for example, node i) of the target node is acquired (1 path), and each node of the acquired 1 path The node number 501 in the node attribute database 310 is searched for the number (node j), and the corresponding link destination node number (for example, node k) is acquired (two paths).
In addition to the two paths, the index value of each node may be estimated in consideration of the index value of a person connected through a finite number of paths greater than that.

6). Other According to the human behavior analysis method and system, the human relationship graph is input from the outside, and in some index (antiferromagnetic index, for example, the degree of stress) in which the tendency is reversed between adjacent people in the graph, The index value of another person can be estimated from the index value of that person. Traditionally, questionnaire surveys have mainly been used to measure human internal conditions such as the degree of stress. However, this method minimizes the energy of the system to minimize the elements of the system, that is, the members of the organization. Since index values can be estimated and index values representing the macro state of the system can be determined from the results, questionnaire results or estimated values of some people can be obtained without conducting a questionnaire survey for all members The value of another person's index can be estimated from Furthermore, by calculating the stable state of human relationships by minimizing energy, not only the indicators of neighboring people in the human relationship graph but also the estimated values of the indicators of sub-networks connected by two or more finite paths can be obtained. Can be used to estimate an individual's index.

Also, if the technology of the present invention is implemented as a basic algorithm, it can be used in the analysis of index values that are antiferromagnetic and have opposite tendencies between adjacent nodes in a human relationship graph. It can be applied in a wide range of fields such as psychology, service, and management.
Further, the human behavior analysis method and system of the present invention includes a human behavior analysis program for causing a computer to execute each procedure, a computer-readable recording medium recording the human behavior analysis program, and a human behavior analysis program. It can be provided by a program product that can be loaded into the internal memory, a computer such as a server including the program, and the like.

  The present invention is generally applicable to a large amount of data reflecting human relations and various human behaviors measured by wearable sensors such as name tag type sensor nodes, wristwatch type sensor nodes, mobile phone usage logs, and other means. It can be used for a system that analyzes the computer.

101 Human relationship graph data input 102 Human relationship graph analysis processing 103 Energy minimization calculation 104 State determination of each node 105 Macro state determination of system 106 Result data output 201 Node number 202 Own CES-D value 203 Connected by one path Average CES-D value of existing nodes
204 Average CES-D value of nodes connected by two paths 205 Average CES-D value of nodes connected by r paths 206 Own CES-D value and average CES-D values of nodes connected by r paths Graph 207 in which correlation coefficients are plotted 207 Example of human relationship graph 208 Node 209 of interest Low stress node 210 High stress node 301 Recording medium 302 Data management server 303 Display device 304 Input device 305 Communication device 306 CPU
307 Hard disk 308 Memory 309 Face-to-face information database 310 Node attribute database 311 Analysis unit program 312 Energy minimization calculation 313 Index value determination 314 for each node 315 Macro index value calculation 315 Internet network 401 to 408 User ID
409 to 412 Node number 502 CES-D value obtained by questionnaire survey 503 Estimated value of index value 504 Node number of link destination 505 Node 506 obtained CES-D value by questionnaire survey No answer is obtained by questionnaire survey CES-D value unknown node 601 Human relationship graph data input 602 Initial state energy E calculation 603 Energy E ′ calculation 604 when changing the state 604 Energy difference ΔE = E′−E calculation 605 ΔE positive / negative determination 606 E = E ′ 607 Predetermined number of calculation progress determination 608 Random number generation between 0 and 1 609 Random number and probability p magnitude comparison 610 Processing to return the state to the state before changing the state 611 Index value determination 612 of each node Macro index value calculation 701 Current human relationship graph 702 The number of links is constant with the number of nodes Human relationship graph 703 when the number of links is constant and the number of links is fixed Human relationship graph 704 When the number of nodes is decreased Human relationship graph 705 When the number of nodes is increased Human relationship graph 706 Number of nodes and links Human relation graph 707 to 718 when link is changed at a fixed level Node 719 Deleted node 720 Node 721 Added node 722 to 729 Node 801 Change the node number of the link destination node of the data set of the node attribute database 901 Number of nodes Experimental result 902 when the number of links is increased at a constant value Experimental result 902 when the number of nodes and the number of links are fixed and links are changed 1001 Human relationship graph 1002 where the index values of some nodes are known Index values of all nodes Human relationship graph 1003 with an estimated index value and high stress Nodes 1004 to 1006 Nodes with known index values and low stress 1101 Estimated values of node attribute database data set 1201 Node number 1202 CES-D value 1203 obtained by questionnaire survey 1204 Estimated values of index values 1204 Link destination Node number 1205 of node 1206 for which simulation is performed assuming that the estimated value is known Node 1206 for which estimation is performed by simulation assuming that the index value is unknown Node 1208 for which an answer is not obtained by a questionnaire survey and CES-D value is unknown A node 1301 whose answer is obtained by a questionnaire and the CES-D value is known A human relation graph 1302 before a simulation A human relation graph 1303 after a simulation by changing an index value of a node A node 1401 which changes an index value Result of calculating the stress level of the entire organization from the stress level of the organizational members obtained from the Kate result 1402 Result of simulation with the estimated stress level of the high stress person fixed to -1 1403 Stress level of the low stress person Simulation result with the estimated value fixed at -1 1501 Node 150 of interest Node 1502 connected to the node of interest through one path 1503 Node connected to the node of interest through two paths

Claims (19)

Using a human relationship graph in which the nodes of the network correspond to people and the links correspond to the relationships between people, an indicator that shows the reverse trend between adjacent people connected by links in the human relationship graph A human behavior analysis system for estimating an index value of another person based on a known index value of a part of a person,
In correspondence with the node identifier, a node attribute database in which a known index value for a node identifier whose index value is known, an estimated value of the index, and a link destination node identifier are stored correspondingly,
A processing unit that refers to the node attribute database and estimates a node index;
A process in which the processing unit determines an estimated value of the index of the node according to the known index value for at least a part of nodes for which a known index value is stored in the node attribute database; Processing for determining an initial value of the estimated value of the node index for the node of
The processing unit obtains the energy of the system configured by each node by using an Ising model based on the estimated value of the index of each node and the link information represented by the correspondence relationship between the node identifier and the link destination node identifier. When,
A process in which the processing unit repeats the calculation of the energy of the system again by changing the estimated value of the index of each node a predetermined number of times so that the energy of the system becomes smaller;
The processing unit determines an estimated value of the index of each node after a predetermined number of iterations or a statistical value of the estimated value of the index of each node in the repeated calculation process as an estimated value of the index and stores the estimated value in the node attribute database. The human behavior analysis system for executing processing.   The human behavior analysis system according to claim 1, wherein the index is an index representing a stress of a person corresponding to a node.   The human behavior analysis system according to claim 1, wherein the processing unit obtains a macro index of the system by obtaining an average of estimated values of indices of the obtained nodes.   The human behavior analysis system according to claim 3, wherein the macro index of the system is an index representing the stress of the entire organization composed of persons corresponding to the node.   The human behavior analysis system according to claim 1, wherein the statistical value of the estimated value of the index of each node in the iterative calculation process is a thermodynamic average of the estimated value of the index of each node. The human behavior analysis system according to claim 5, wherein the Ising model is represented by the following expression.

Here, H is the energy of the system, S _i and S _j are estimated values of indices corresponding to the node identifiers i and j, take +1 or −1, and J is a parameter corresponding to the exchange interaction between spins. Yes, in the node attribute database, the product of the estimated value of the index of the node identifier i and the estimated value of the index of the link destination node identifier j corresponding to the node identifier i takes a negative value, otherwise 0. And h is a predetermined value.   2. The human behavior analysis system according to claim 1, wherein the estimated value of the index of each node is changed to minimize the energy of the system or to minimize it in a predetermined number of calculations. Face-to-face information in which communication including at least one of face-to-face and conversation between people is represented by a wearable sensor attached to each person corresponding to each node corresponds to each node identifier combination. Further comprising a face-to-face information database stored;
The processing unit compares the face-to-face information corresponding to each node identifier combination stored in the face-to-face information database with a predetermined threshold to determine whether to link between the node combinations; The human behavior analysis according to claim 1, wherein when a link is established, the other node identifier of the combination is stored in the corresponding link destination node identifier of the node attribute database for each node identifier of the combination of the node identifiers. system.   The human behavior analysis system according to claim 8, wherein the processing unit inputs a combination of the node identifiers and face-to-face information from a data management server or a recording medium, and stores the face-to-face information database. The human behavior analysis system according to claim 8, further comprising the sensor that is attached to each person corresponding to each node and quantifies communication including at least one of a person-to-person meeting and a conversation.   2. The human behavior analysis system according to claim 1, wherein the processing unit inputs the link information corresponding to a node identifier and a link destination node identifier from a data management server or a recording medium, and stores the link information in the node attribute database. The data management server further comprising:
The data management server
Face-to-face information in which communication including at least one of face-to-face and conversation between people is represented by a wearable sensor attached to each person corresponding to each node corresponds to each node identifier combination. A second face-to-face information database stored;
A second processing unit,
The second processing unit compares the face-to-face information corresponding to each combination of node identifiers stored in the face-to-face information database with a predetermined threshold value, and determines whether to link between the combinations of the nodes. When establishing a link, for each node identifier of the combination of node identifiers, the other node identifier of the combination is stored in the corresponding link destination node identifier of the second node attribute database,
The human behavior analysis system according to claim 11, wherein the processing unit inputs the link information corresponding to a node identifier and a link destination node identifier from the data management server. The processor is
According to the input by the operator, add or delete the link destination node identifier of the node attribute database,
The human behavior analysis system according to claim 1, which simulates a macro index of a system when a link is added or deleted. The processor is
According to the input by the operator, the node identifier of the node attribute database is added or deleted,
The human behavior analysis system according to claim 1, which simulates a macro index of a system when a node is added or deleted. The processor is
According to the input by the operator, at least one link destination node identifier corresponding to the predetermined node of the node attribute database is deleted and another node identifier is added as a link destination node identifier,
The human behavior analysis system according to claim 1, which simulates a macro index of a system when a link is switched. In the process of reducing the energy of the system, the processing unit,
The human behavior analysis according to claim 1, wherein the estimated value of an index corresponding to the known index value is not changed while the estimated value of the index is fixed and the estimated value of the index of another node is changed. system. The processing unit reverses and fixes the estimated value of the index of a predetermined node after the macro index of the system is obtained, and executes the above steps again to invert the estimated value of the node index The second macro index of the system of
The human behavior analysis system according to claim 1, wherein a macro index of the system and a macro second index of the system are displayed. The human behavior analysis system according to claim 1, wherein the Ising model uses the following expression with a correction term added.

here,
H is the energy of the system,
S _i , S _j and S _k are estimated values of indices corresponding to the node identifiers i, j and k, and take +1 or −1,
J is a parameter corresponding to the exchange interaction between spins, and the product of the estimated value of the index of the node identifier i and the estimated value of the index of the link destination node identifier j corresponding to the node identifier i in the node attribute database Takes a negative value, otherwise it is 0,
J ′ represents an estimated value of an index of a link destination node identifier k further corresponding to the node identifier j, an estimate of the node identifier i, and the index for the link destination node identifier j corresponding to the node identifier i in the node attribute database. Takes a positive value for the product with the value, 0 otherwise,
h is a predetermined value. Using a human relationship graph in which the nodes of the network correspond to people and the links correspond to the relationships between people, an indicator that shows the reverse trend between adjacent people connected by links in the human relationship graph A human behavior analysis method for estimating an index value of another person based on a known index value of a part of a person,
The processing unit is known in the node attribute database in which the known index value for the node identifier whose index value is known, the estimated value of the index, and the link destination node identifier are stored corresponding to the node identifier. A step of determining an estimated value of the index of the node according to the known index value and a processing unit for at least a part of the nodes storing the index value of the initial value of the estimated value of the node index for other nodes Steps to determine,
A processing unit that obtains an energy of a system constituted by each node by an Ising model based on an estimated value of an index of each node and link information represented by a correspondence relationship between the node identifier and the link destination node identifier; ,
A step of changing the estimated value of the index of each node and calculating the energy of the system again a predetermined number of times so that the energy of the system becomes smaller;
A step in which the processing unit determines an estimated value of the index of each node after it has been repeated a predetermined number of times or a statistical value of an estimated value of the index of each node in an iterative calculation process as an estimated value of the index and stores it in the node attribute database And a human behavior analysis method.

Images