WO2024105770A1

WO2024105770A1 - Factor behavior inference device, method, and program

Info

Publication number: WO2024105770A1
Application number: PCT/JP2022/042372
Authority: WO
Inventors: 登夢冨永; 修平山本; 健倉島
Original assignee: 日本電信電話株式会社
Priority date: 2022-11-15
Filing date: 2022-11-15
Publication date: 2024-05-23

Abstract

One embodiment of this invention acquires, for each predetermined constant period, data representing a health state, a plurality of lifestyle habits, and a work pattern of a target user, extracts a feature amount from each of the acquired data, uses the feature amount as a first variable, calculates a feature amount representative value per work pattern from the feature amount for each of the plurality of lifestyle habits, and uses the feature amount representative value as a second variable. By giving the first variable and the second variable to a model indicating a relationship between the health state per work pattern and the plurality of lifestyle habits, a parameter representing the degree of influence of the plurality of lifestyle habits on a change in the health state per work pattern is derived, and, on the basis of the derived parameter, information indicating the behavior that becomes a factor in the change in the health state per work pattern is generated and output.

Description

Factor behavior estimation device, method, and program

One aspect of the present invention relates to a factor behavior estimation device, method, and program used to estimate behavior that is a factor affecting a user's health condition for each work style, for example.

Changes in working patterns due to the novel coronavirus disease (COVID-19) are affecting people's lifestyles and health management behaviors. For example, Non-Patent Document 1 reports that a questionnaire survey of 13,989 Italians showed that people who started working from home due to COVID-19 had a lower level of severe sleep disorders than those who did not. In addition, Non-Patent Document 2 suggests that the results of 988 responses collected through an online survey system showed that people who started working from home due to COVID-19 exercised more, ate less junk food, and spent more time talking with colleagues compared to before they started working from home, which contributed to improving their physical and mental well-being.

These analysis results provide important insights into understanding the impact of COVID-19 on people's lives and health.

However, Non-Patent Documents 1 and 2 merely compare working styles and lifestyles between the period before and after the COVID-19 pandemic, and the analysis results do not allow us to know the impact of working from home or in the office on the relationship between daily lifestyle habits and health status. In other words, users cannot know what changes in their daily working styles will bring about in their lifestyle habits, or behavior, and what impact these changes in behavior will have on their health status.

This invention was made with the above in mind, and aims to provide technology that can estimate the behaviors that are factors that affect changes in a user's health condition for each work style.

In order to solve the above problem, one aspect of the factor behavior estimation device or factor behavior estimation method according to the present invention acquires data representing the health state, multiple lifestyle habits, and working style of a target user at predetermined fixed intervals, extracts features from each of the acquired data and sets these as first variables, and calculates a feature representative value for each working style from the features for each of the multiple lifestyle habits and sets this as a second variable. Then, by providing the first variable and the second variable to a model that represents the relationship between the health state and the multiple lifestyle habits for each working style, parameters that represent the degree of influence of the multiple lifestyle habits on changes in the health state for each working style are derived, and information that represents actions that are factors in changes in the health state for each working style is generated and output based on the derived parameters.

According to one aspect of the invention, the importance of each lifestyle habit that affects health changes is estimated for each work style, and the estimated results are presented to the user. This allows the user to identify lifestyle habits that should be taken into consideration in order to maintain their health, in other words, behaviors that are the cause of health changes, for each of their own work styles. Furthermore, data on health status, lifestyle habits, and work styles is obtained at regular intervals, and the estimated results of the importance of each lifestyle habit are updated based on this data each time, making it possible to dynamically re-present to the user lifestyle habits that should be taken into consideration in accordance with health changes at regular intervals.

In other words, according to one aspect of the present invention, it is possible to provide technology that can estimate the behaviors that are factors that affect changes in a user's health condition for each work style.

FIG. 1 is a diagram showing an example of a system including a factorial behavior inference device according to an embodiment of the present invention. FIG. 2 is a block diagram showing an example of a hardware configuration of a factorial behavior inference device according to an embodiment of the present invention. FIG. 3 is a block diagram showing an example of a software configuration of the factor behavior inference device according to an embodiment of the present invention. FIG. 4 is a flowchart showing an example of a process procedure and process contents of the factor behavior inference process executed by the control unit of the factor behavior inference device shown in FIG. FIG. 5 is a flowchart showing an example of the process procedure and process contents of the model learning process from the process procedures shown in FIG. FIG. 6 is a flowchart showing an example of the procedure and contents of a process for generating and outputting information representing a result of an inference of a factor behavior, among the procedures shown in FIG. FIG. 7 is a diagram showing an example of information indicating the result of inference of the factor behavior generated by the process shown in FIG.

Below, an embodiment of the present invention will be described with reference to the drawings.

[One embodiment]
(Configuration example)
(1) System FIG. 1 is a diagram showing an example of a configuration of a factor behavior inference system according to an embodiment of the present invention.
The factor behavior inference system according to one embodiment of the present invention includes a factor behavior inference device SV, and enables transmission of information data via a network NW between this factor behavior inference device SV and user terminals UT1 to UTn used by a plurality of users who use the system.

User terminals UT1 to UTn are, for example, smartphones, tablet terminals or personal computers, and are equipped with communication tools such as a browser or mailer.

The user terminals UT1 to UTn each have a function according to an embodiment of the present invention to acquire data representing the user's health condition, lifestyle habits, and working patterns from sensors, and transmit each of the acquired data to the factor behavior estimation device SV via the network NW. The user terminals UT1 to UTn also have a function to receive, via the network NW, information representing the factor behavior estimation results transmitted from the factor behavior estimation device SV.

The data representing the health condition includes, for example, weight, blood pressure, blood sugar level, etc., and is measured, for example, by a vital sensor built into the user terminals UT1 to UTn, or a vital sensor provided on a wearable terminal worn by the user.

The data representing lifestyle habits may include, for example, the number of steps taken, hours of sleep, amount of food eaten, and mood, and may be measured, for example, by an application program provided in the user terminals UT1 to UTn.

Data representing working patterns includes, for example, whether or not a person works each day, the start and end times of work, and the place of work. This data is, for example, entered by the user and managed by a scheduler provided in the user terminals UT1 to UTn.

The network NW comprises, for example, a wide area network with the Internet at its core, and an access network for accessing this wide area network. Examples of the access network that can be used include a public communication network using wired or wireless connections, a Local Area Network (LAN) using wired or wireless connections, and a Cable Television (CATV) network. In addition, when the system is operated within a company or business establishment, for example, the network NW is composed of an in-house network such as a LAN or wireless LAN.

(2) Factor behavior estimation device SV
The factor behavior inference device SV is configured by, for example, a server computer located on the Web or on the cloud.

FIGS. 2 and 3 are block diagrams showing examples of the hardware and software configurations, respectively, of the factor behavior estimation device SV.

The causal behavior estimation device SV has a control unit 1 that uses a hardware processor such as a central processing unit (CPU). A storage unit having a program storage unit 2 and a data storage unit 3, and a communication interface (hereinafter the interface will be abbreviated as I/F) unit 4 are connected to this control unit 1 via a bus 5.

Under the control of the control unit 1, the communication I/F unit 4 transmits and receives information data to and from the user terminals UT1 to UTn in accordance with the communication protocol defined by the network NW.

The program storage unit 2 is configured by combining, for example, a non-volatile memory such as a solid state drive (SSD) as a storage medium that can be written to and read from at any time, and a non-volatile memory such as a read only memory (ROM), and stores application programs necessary for executing various control processes according to one embodiment, in addition to middleware such as an operating system (OS). Note that hereinafter, the OS and each application program will be collectively referred to as the program.

The data storage unit 3 is, for example, a combination of a non-volatile memory such as an SSD, which can be written to and read from at any time, and a volatile memory such as a RAM (Random Access Memory), as a storage medium, and the storage area includes a health condition storage unit 31, a lifestyle habit storage unit 32, a work pattern storage unit 33, a learning model storage unit 34, and a parameter storage unit 35, as the main storage units required to implement one embodiment of this invention.

The health condition memory unit 31, lifestyle habit memory unit 32, and work style memory unit 33 store data representing the health condition, lifestyle habit, and work style of each user acquired from the user terminals UT1 to UTn, in association with identification information of the user or the user terminals UT1 to UTn (hereinafter referred to as user ID).

The learning model storage unit 34 stores the learned data that constitutes the learning model.

The parameter storage unit 35 stores the learned parameters generated by the model learning process by the control unit 1 as an estimated result of the degree of influence of lifestyle habits on the health condition.

The control unit 1 includes a data acquisition processing unit 11, a model configuration processing unit 12, a data extraction processing unit 13, a variable configuration processing unit 14, a representative value calculation processing unit 15, a model learning processing unit 16, and a factor behavior estimation information output processing unit 17 as processing functions necessary to implement one embodiment of this invention.

Each of the processing units 11 to 17 of the control unit 1 is realized by having the hardware processor of the control unit 1 execute application programs stored in the program storage unit 2. Note that some or all of the processing units 11 to 17 may be realized using hardware such as an LSI (Large Scale Integration) or an ASIC (Application Specific Integrated Circuit).

The data acquisition processing unit 11 receives data representing health conditions, lifestyle habits, and working patterns transmitted, for example, daily, from the user terminals UT1 to UTn via the communication I/F unit 4, and stores the received data representing health conditions, lifestyle habits, and working patterns in the health condition memory unit 31, lifestyle habit memory unit 32, and working pattern memory unit 33, respectively, in association with the user ID of the sender.

The model configuration processing unit 12 receives data representing the model structure of the learning model, for example, from an administrator terminal (not shown) used by a system administrator, via the communication I/F unit 4. Then, it provides the received data representing the model structure to the model learning processing unit 16.

The data extraction processing unit 13 reads data representing the health condition, lifestyle habits, and working style for each user from the health condition storage unit 31, lifestyle habits storage unit 32, and working style storage unit 33. Then, it extracts sets of data that correspond in time from the read data.

The variable configuration processing unit 14 calculates the feature quantities for each of the data representing the time-corresponding health condition, lifestyle habits, and working patterns extracted by the data extraction processing unit 13. At this time, if multiple types of lifestyle habits are acquired, the variable configuration processing unit 14 calculates the feature quantities for each of the lifestyle habits. The variable configuration processing unit 14 then passes the calculated feature quantities for the health condition, lifestyle habits, and working patterns to the model learning processing unit 16.

The representative value calculation processing unit 15 receives the feature quantities related to each lifestyle habit from among the feature quantities related to the health state, lifestyle habits, and working style calculated by the variable configuration processing unit 14. Then, it calculates a representative value by working style for each of the received feature quantities related to each lifestyle habit.

The model learning processing unit 16 performs learning processing by inputting the feature quantities related to the health condition, lifestyle, and work style calculated by the variable configuration processing unit 14 and the representative values of the feature quantities of each lifestyle habit for each work style calculated by the representative value calculation processing unit 15 into the model structure provided by the model configuration processing unit 12, thereby calculating parameters corresponding to each lifestyle habit that affects changes in the health condition. An example of the model learning processing will be explained in the operation example.

The model learning processing unit 16 then stores the calculated parameters corresponding to each of the lifestyle habits in the learning model storage unit 34 together with data representing the model structure when the parameters were obtained, each feature value related to the health condition, lifestyle habits, and working style, and a representative value of the feature value for each lifestyle habit. At the same time, the model learning processing unit 16 stores the parameters corresponding to each of the lifestyle habits in the parameter storage unit 35 as information representing an estimated result of the degree of influence of the lifestyle habits on the health condition.

The factor behavior estimation information output processing unit 17 reads parameters corresponding to each lifestyle habit from the parameter storage unit 35, generates display data indicating the importance of each lifestyle habit with respect to changes in health status based on the parameters corresponding to each lifestyle habit that have been read, and transmits the generated display data from the communication I/F unit 4 to the corresponding user terminals UT1 to UTn. An example of the factor behavior estimation process will also be described in the operation example.

(Example of operation)
Next, an example of the operation of the factor behavior inference device SV configured as above will be described.

(1) Acquisition of Data Related to a User The user terminals UT1 to UTn collect data on the user's daily health condition, lifestyle, and working style, for example, for each day, and obtain data representing the characteristics of each of the collected data, such as a representative value or average value, and transmit the obtained data together with the user ID to the factorial behavior inference device SV at a predetermined time every day.

In response to this, the control unit 1 of the factor behavior inference device SV, under the control of the data acquisition processing unit 11, receives the data on the health condition, lifestyle habits, and working style transmitted from the user terminals UT1 to UTn via the communication I/F unit 4. The received data on the health condition, lifestyle habits, and working style is then stored sequentially by day in the health condition storage unit 31, lifestyle habit storage unit 32, and working style storage unit 33, respectively, in association with the user ID.

(2) Estimation of Factor Behavior Each time new daily data is stored in the health condition storage unit 31, the lifestyle habit storage unit 32, and the work pattern storage unit 33 for each user, the control unit 1 of the factor behavior estimation device SV executes a process of estimating factor behaviors that affect changes in the health condition of that user as follows.

FIG. 4 is a flowchart showing an example of the process steps and process contents of the factor behavior inference process executed by the control unit 1 of the factor behavior inference device SV.

(2-1) Data Extraction Processing First, in step S10, the control unit 1 of the factor behavior inference device SV reads data representing the health condition, lifestyle habits, and working style for each user from the health condition storage unit 31, lifestyle habit storage unit 32, and working style storage unit 33 under the control of the data extraction processing unit 13. Then, the control unit 1 organizes the temporal correspondence of each of the read data, and extracts a set of temporally corresponding data as an analysis target.

For example, the data extraction processing unit 13 first reads data y _i,t representing health condition, data x _i,s,k representing lifestyle, and data w _i,(us,ue) representing work style from the health condition memory unit 31, lifestyle habit memory unit 32, and work style memory unit 33, respectively.

Here, the acquisition times of data representing the health condition of a certain user i are t=t ₀ , t ₁ , . . . , t _T .

Then, the analysis unit D _i,j is

where j=0,...,T-1.

This represents a procedure for associating a set of data A ij representing lifestyle habits and a set of data B _ij representing working patterns acquired between acquisition times t _j and t _j+1 _of two consecutive health conditions of a user i, and constructing them as an analysis unit D _ij .

The data extraction processing unit 13 extracts a _set D

In step S11, it is determined whether or not a set D of temporally corresponding data has been constructed. If a set D of temporally corresponding data has been constructed, the data extraction processing unit 13 passes the constructed set D to the variable construction processing unit 14 as preprocessed data in step S12.

On the other hand, if D = φ, that is, if a temporally corresponding data set D cannot be constructed, the data extraction processing unit 13 proceeds to output processing of the factor behavior estimation result in step S17.

(2-2) Variable Construction When the data set D is received from the data extraction processing unit 13, the variable construction processing unit 14 constructs, in step S13, variables to be trained into the learning model for each of health condition, lifestyle, and working style, for example, as follows:

(2-2-1) Variables related to health conditions The variable construction processing unit 14 first calculates the difference between data y _i,tj and y _i,tj+1 representing two health conditions included in the analysis unit D _ij.

The calculated result y _i,j is set as the feature of the data representing the health condition. However, a scale other than the above may be used as the feature as long as it represents the difference between y _i,tj and y _i,tj+1 , which are data representing two health conditions.

(2-2-2) Variables Related to Lifestyle Habits The variable construction processing unit 14 next constructs A _i,j = {x _i,s,k |t _j <s<t _j+1 , k ∈ K} corresponding to the analysis unit D _ij .
For a certain user i, a measurement value related to a certain lifestyle habit per unit time acquired at time t _j <s <t _j+1 is expressed as

The calculated x _i,j,k is used as the feature quantity of lifestyle habit k.

However, a scale other than the above may be used for the feature amount of the data representing the lifestyle habits, as long as the scale expresses a representative value of the measurement values of a certain lifestyle habit k acquired at time _tj < s < _tj+1 of a certain user i, such as an average value of the measurement values. Also, a different scale may be used for each of a plurality of lifestyle habits.

(2-2-3) Variables Related to Working Styles Finally, the variable configuration processing unit 14 calculates the feature quantities of the data related to working styles as follows.
For example, suppose there are c types of work styles (w = { _w1 , ..., _wc }). In this case, the variable construction processing unit 14 counts the work hours for each of the c types of work styles in the work style data included in _Bij corresponding to the analysis unit _Dij , and determines the work style that has the longest time as the work style of _Dij .

From the work style data included in B _ij , the total measurement time WTc for each work style is calculated as follows:

(7)
where c=1, . . . , C.

Based on this equation (7), the working style of D _ij is as follows:

(8)

The structure is as follows. However,

Here, w _ij is a one-hot vector in which the element w _i,j,c of c for which WTc is the longest is 1 and the other elements are 0.

The variable configuration processing unit 14 then passes the calculated feature amount y _i,j related to the health state, the feature amount x _i,j,k related to each lifestyle habit k, and the feature amount w _ij related to the working style to the model learning processing unit 16 .

(2-3) Calculation of Representative Value of Feature Amount Related to Each Lifestyle Habit Subsequently, in step S14, the control unit 1 of the factor behavior inference device SV receives the feature amount x _i,j,k calculated for each lifestyle habit from the variable configuration processing unit 14 under the control of the representative value calculation processing unit 15. Then, for each of the received feature amounts x _i,j,k related to each lifestyle habit, a representative value by work type is calculated as follows.

In other words, since the distribution of the feature quantities of the lifestyle habit data differs for each work style, it is necessary to control the difference during model learning. Therefore, the representative value calculation processing unit 15 calculates the average value of the feature quantities x _i,j,k of the lifestyle habit data k of a certain user i in a certain work style W c as follows:

It is calculated as follows.

However, the method for finding a representative value of the feature quantity of lifestyle habit data k for a certain working style wc of a certain user i is not limited to the above formula. In addition, different methods may be used to calculate representative values for multiple lifestyle habits.

The representative value calculation processing unit 15 passes the calculated representative values of the features of each of the lifestyle habit data to the model learning processing unit 16.

(2-4) Learning Using the Learning Model Next, in step S16, the control unit 1 of the factorial behavior inference device SV executes a process of learning parameters corresponding to each lifestyle habit using the learning model under the control of the model learning processing unit 16 as follows.

FIG. 5 is a flowchart showing an example of the processing procedure and processing content executed by the model learning processing unit 16.

(2-4-1) Configuration of Model Structure Data representing the model structure is designated, for example, by a system administrator and transmitted from an administrator terminal. In response to this, the control unit 1 of the factorial behavior inference device SV receives the data representing the model structure from the model configuration processing unit 12 via the communication I/F unit 4, and passes it to the model learning processing unit 16.

First, in step S161, the model learning processing unit 16 receives data representing the model structure from the model configuration processing unit 12.

As the model structure, the following model is used to learn the relationship between health status and lifestyle habits for each work style. In this example, the relationship between the feature value y _i,j related to the health status and the feature value x _i,j,k related to each lifestyle habit k is expressed as follows using a parameter θ:

This is expressed as a regression model shown below.

Here, it is assumed that the parameter θ _k has a different value θ _k,c depending on the work style w c. Considering that the feature quantity x _i,j,k related to the lifestyle with the work style w _i,j,c =1 is associated with this θ _k,c ,

This can be expressed as follows.

Furthermore, in order to control the difference in distribution of the lifestyle feature x _i,j,k for each work style, the representative value of the lifestyle feature x _i,j,k is used.

Let us assume that.

Therefore, the regression model representing the relationship between the feature value y _i,j related to the health state and the feature value x _i,j,k related to each lifestyle habit k is as follows:

It can be expressed as follows.

In this example, we have explained the case where a linear regression model is used as the mathematical model with the highest interpretability, but any other mathematical model can be used as long as it is a machine learning model that can express to the user the importance of each lifestyle habit as a feature for each working style.

(2-4-2) Parameter Learning Next, in step S162, the model learning processing unit 16 obtains the feature amount y _i,j related to the health state, the feature amount x _i,j,k related to each lifestyle habit k, and the feature amount w _ij related to the working style from the variable configuration processing unit 14, and further obtains the representative value of the feature amount of each of the lifestyle habit data from the representative value calculation processing unit 15 in step S163.

Then, in step S164, the model learning processing unit 16 performs a learning process by inputting, as variables, each feature value related to the health condition, lifestyle, and work style obtained from the variable configuration processing unit 14 and the representative value of the feature value of each lifestyle for each work style obtained from the representative value calculation processing unit 15 into the model structure obtained from the model configuration processing unit 12, and thereby calculates a parameter θ that indicates the degree of influence of each lifestyle on changes in the health condition as follows:

Here, we will use an example of learning using the least squares method, which is the most common method.

First, a vector a=(a ₀ ,...,a _n ) having elements a ₀ ,...,a _n is expressed as {a _i } _i . Then, the model structure shown in equation (14) is expressed as follows:

It is transformed as follows.

here,

Then, equation (15) becomes ｙ＝Ｘ ^T Θ＋ｅ
This is expressed as:

In the least squares method, a parameter Θ that minimizes the error term e in the above equation is calculated.
Here, if the estimated value of y is y^=X ^T Θ, the error term e is e=y-y^=y-X ^T Θ (20)
Then, the following optimization problem is

(twenty one)
The desired parameters can be obtained by solving

This can be done by setting the loss function as L(Θ)＝(y−X ^T Θ) ^T (y−X ^T Θ) and searching for the point where the gradient of this loss function with respect to Θ is 0.

Therefore, the optimal parameter Θ is

(twenty five)
can be obtained as:

In step S165, the model learning processing unit 16 stores the optimal parameters Θ corresponding to each of the lifestyle habits in the learning model storage unit 34 together with data representing the model structure when the parameters Θ were obtained, each feature value related to the health state, lifestyle habits, and working style, and a representative value of the feature value for each lifestyle habit.

In addition, in step S166, the model learning processing unit 16 stores the calculated optimal parameters Θ corresponding to each of the lifestyle habits in the parameter storage unit 35 as estimated results of the degree of influence of each lifestyle habit on changes in health status by work style.

(3) Output of inference results Finally, in step S17, the control unit 1 of the factor behavior inference device SV generates display data indicating the importance of each lifestyle habit for each work style under the control of the factor behavior inference information output processing unit 17, and transmits the generated display data to the user terminals UT1 to UTn.

FIG. 6 is a flowchart showing an example of the processing procedure and processing contents of the factor behavior inference information output processing unit 17.

In other words, first, in step S171, the factor behavior inference information output processing unit 17 reads the learned parameters Θ corresponding to each of the lifestyle habits from the parameter storage unit 35. Then, in step S172, the factor behavior inference information output processing unit 17 calculates the importance of each lifestyle habit with respect to the health status by work type based on the learned parameters Θ corresponding to each of the lifestyle habits that have been read.

For example, in a certain work pattern w _c , the ratio of the absolute value of a parameter of a certain lifestyle habit k to the sum of the absolute values of the parameters corresponding to each lifestyle habit is defined as the importance of that lifestyle habit π _k,c .

(26)
It is calculated as follows.

A specific example is given below. Let us assume that the health condition is a daily weight change, and that the lifestyle habits are exercise, diet, sleep, and mood. In addition, two work styles are assumed: "at home" and "going to work." In this case, as a result of the above calculation by the factor behavior estimation information output processing unit 17, the importance π _k,c of each lifestyle habit by work style is, for example, π (exercise, at home) = 0.50, π (diet, at home) = 0.30, π (sleep, at home) = 0.15, π (mood, at home) = 0.05.
π(exercise, going to work) = 0.15, π(meal, going to work) = 0.30, π(sleep, going to work) = 0.30, π(mood, going to work) = 0.25
It is calculated as follows:

The factor behavior inference information output processing unit 17 generates display data in step S173 in order to present the calculation result of the importance π _k,c of each lifestyle habit by each work style to the user.
FIG. 7 shows an example of display data generated for presenting the calculation results of the importance π _k,c of each lifestyle habit for each work style to the user.

Finally, in step S174, the factor behavior estimation information output processing unit 17 transmits the generated display data from the communication I/F unit 4 via the network NW to the user terminals UT1 to UTn of the target users. Based on the display data displayed on the user terminals UT1 to UTn, the target users can identify which of the four lifestyle habits above is the lifestyle habit that affects their own weight change, that is, the behavior that is the cause of health changes, based on its importance.

Note that, as long as the purpose of presenting to the user important lifestyle habits that affect changes in health status according to work style can be achieved, another method may be used, such as showing the importance level π _k,c by a bar graph.

In the above example, the factor behavior inference information output processing unit 17 transmits the calculation result of the importance π _k,c of each lifestyle habit by work style directly as display data to the user terminals UT1 to UTn. However, the present invention is not limited to this. For example, the factor behavior inference information output processing unit 17 may compare the calculated importance π _k,c of each lifestyle habit with a preset threshold value, identify a lifestyle habit whose importance π _k,c is equal to or greater than the threshold value, generate display data representing the identified lifestyle habit, and transmit the display data to the user terminals UT1 to UTn. In addition, the factor behavior inference information output processing unit 17 may add the importance π _k,c to the identified lifestyle habit.

(Action and Effects)
As described above, in one embodiment, the data acquisition processing unit 11 first acquires data representing the user's health condition and a plurality of lifestyle habits for each of the work styles (for example, when working from home and when working at the office), the data extraction processing unit 13 extracts sets of data corresponding in time from the above data, the variable configuration processing unit 14 obtains the feature values of each data constituting this set of data, and the representative value calculation processing unit 15 calculates the representative value of the feature values related to each lifestyle for each work style based on each obtained feature value. Then, the model learning processing unit 16 applies the feature values of each of the above data and their representative values to the model structure obtained by the model configuration processing unit 12 to learn parameters representing the importance of each lifestyle habit that affects changes in the user's health condition. Then, based on the learned parameters Θ corresponding to each lifestyle obtained by the above learning, the factor behavior estimation information output processing unit 17 generates display data representing the importance of each lifestyle for each work style and transmits it to the user terminals UT1 to UTn.

Therefore, the importance of each lifestyle habit that affects health changes is estimated for each working style for each fixed period, and the estimated results are presented to the user. This allows the user to identify the lifestyle habits they should pay attention to in order to maintain their health for their own working style, that is, the behaviors that are the cause of health changes, for each fixed period.

[Other embodiments]
In the embodiment, the processing functions of the factorial behavior inference device SV are provided in a server computer on the Web or cloud, but the processing functions of the factorial behavior inference device SV may be provided in a server computer located on a local area network such as a workplace or community, or in a personal computer shared by multiple users. The functions of the factorial behavior inference device SV may be distributed among multiple server computers and personal computers used by users, etc.

Other aspects of the invention may be modified in various ways without departing from the spirit and scope of the invention, including the type of health condition, the type of lifestyle to be estimated, the type of working style, the method for calculating the features of each of the data, the structure of the learning model, and the configuration of the display data presented to the user.

Although the embodiments of the present invention have been described in detail above, the above description is merely an example of the present invention in every respect. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. In other words, when implementing the present invention, specific configurations according to the embodiments may be appropriately adopted.

In short, this invention is not limited to the above-described embodiment as it is, and in the implementation stage, the components can be modified and embodied without departing from the gist of the invention. Furthermore, various inventions can be formed by appropriately combining multiple components disclosed in the above-described embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, components from different embodiments may be appropriately combined.

SV...factorial behavior inference device UT1 to UTn...user terminal NW...network 1...control unit 2...program storage unit 3...data storage unit 4...communication I/F unit 5...bus 11...data acquisition processing unit 12...model configuration processing unit 13...data extraction processing unit 14...variable configuration processing unit 15...representative value calculation processing unit 16...model learning processing unit 17...factorial behavior inference information output processing unit 31...health condition storage unit 32...lifestyle storage unit 33...work pattern storage unit 34...learning model storage unit 35...parameter storage unit

Claims

A first processing unit that acquires data representing a health condition, a plurality of lifestyle habits, and a working style of a target user at a predetermined fixed period;
a second processing unit that extracts feature amounts from each of the data representing the health condition, the plurality of lifestyle habits, and the working pattern, and outputs the extracted feature amounts as a first variable;
a third processing unit that calculates a representative value of characteristic amounts for each of the plurality of lifestyle habits from the characteristic amounts and outputs the calculated representative value of characteristic amounts as a second variable;
a fourth processing unit that derives a parameter that represents a degree of influence of the plurality of lifestyle habits on a change in the health state by the work style by applying the first variable and the second variable to a model that represents a relationship between the health state by the work style and the plurality of lifestyle habits;
a fifth processing unit that generates and outputs information indicating behaviors that are factors that cause changes in the health state for each of the work styles based on the parameters.
A sixth processing unit extracts a set of the data that correspond in time from the data representing the health condition, the plurality of lifestyle habits, and the working style,
2. The factor behavior inference device according to claim 1, wherein the second processing unit extracts the feature amount from each of the data constituting the extracted data set, and outputs the extracted feature amount as the first variable.
The factor behavior inference device according to claim 1, wherein the fourth processing unit uses a linear regression model as a model representing the relationship between the health state by the work type and the plurality of lifestyle habits, and calculates the parameters using the least squares method.
The factor behavior inference device according to claim 1, wherein the fifth processing unit calculates, for each of the work styles, a ratio of the absolute value of the parameter of each of the lifestyle habits to the sum of the absolute values of the parameters of the plurality of lifestyle habits as information representing the importance of each of the lifestyle habits, and outputs the calculated information representing the importance of each of the lifestyle habits as information representing behavior that is a factor in the change in the health state.
The factor behavior inference device according to claim 1, wherein the fifth processing unit calculates, for each of the work styles, a ratio of the absolute value of the parameter of each of the lifestyle habits to the sum of the absolute values of the parameters of the plurality of lifestyle habits as information representing the importance of each of the lifestyle habits, and outputs the lifestyle habits whose calculated importance is equal to or exceeds a preset threshold as information representing behavior that is a factor in the change in the health state.
A factor behavior inference method executed by an information processing device including a processor and a storage medium,
acquiring data representing the health condition, a plurality of lifestyle habits, and working patterns of the target user at predetermined intervals;
extracting feature values from each of the data representing the health condition, the plurality of lifestyle habits, and the working pattern, and outputting the extracted feature values as a first variable;
calculating a representative value of characteristic amounts for each of the plurality of lifestyle habits based on the characteristic amounts, and outputting the calculated representative value of characteristic amounts as a second variable;
a step of deriving a parameter representing a degree of influence of the plurality of lifestyle habits on a change in the health state by the working style by applying the first variable and the second variable to a model representing a relationship between the health state by the working style and the plurality of lifestyle habits;
generating and outputting information indicating behavior that is a factor of a change in the health state for each work style based on the parameters.
A program that causes a processor provided in the factor behavior inference device to execute at least one of the processes performed by each processing unit provided in the factor behavior inference device described in any one of claims 1 to 5.