WO2024024294A1 - Estimation device, estimation method, estimation system, and estimation program - Google Patents

Abstract

Provided is an estimation device that estimates action data and/or outcome data regarding the actions of a subject to make it possible to promote or maintain improvement in the action or outcome of the subject with respect to health, medical care, and lifestyle habit. The estimation device of the present invention comprises: an input unit for inputting cognitive bias data concerning a cognitive bias of the subject; and an output unit that uses an estimation model including a neural network for estimating, on the basis of the cognitive bias data, at least one of action data concerning an action of the subject, attribute data of the subject, position data of the subject, health data of the subject, exercise data of the subject, lifestyle habit data of the subject, and medical care data of the subject, to output at least one of the action data, the attribute data, the position data, the health data, the exercise data, the lifestyle habit data, and the medical care data that have been estimated from the cognitive bias data input by means of the input unit.

Description

Estimation device, estimation method, estimation system, and estimation program

The present invention relates to an estimation device, an estimation method, an estimation system, and an estimation program, and in particular, the present invention is based on data regarding a subject's cognitive biases, cognitive characteristics, and decision-making characteristics (hereinafter referred to as "cognitive biases"). The present invention relates to an estimation device, an estimation method, an estimation system, and an estimation program for estimating data related to behavior.

In recent years, health management devices, health management systems, and health management methods that can appropriately manage a user's health have been provided (see Patent Document 1).

In addition, a learnable model with high prediction accuracy has been proposed that takes into account the consumer's selection behavior, which is subject to cognitive bias, and takes into account the characteristic amounts of the product and the consumer (see Patent Document 2).

JP 2018-49393 Publication Japanese Patent Application Publication No. 2016-115316

It is thought that the subject's cognitive bias affects the subject's behavior and outcomes regarding health, medical care, and lifestyle habits. Furthermore, the behavior of the subjects is thought to have an impact on outcomes related to health, medical care, and lifestyle habits. However, the structure of cognitive bias is complex, as is the structure of target behavior toward health, medical care, and lifestyle habits, so it has been difficult to express it using existing models. In addition, there are large individual differences in cognitive bias and behavior, and existing uniform guidance and interventions are insufficient to encourage and maintain behavior in target subjects, or to improve and maintain outcomes. However, the effectiveness of guidance and intervention by experts is extremely limited. Furthermore, we can model for each individual the target person's behavior and outcomes related to health, medical care, and lifestyle habits that are affected by cognitive biases, and model the outcomes related to health, medical care, and lifestyle habits that are influenced by behavior for each individual person. However, it was not known that it would be a complex model and that a learning algorithm could be constructed.

Furthermore, in order to encourage or maintain behavior related to health, medical care, or lifestyle habits in the target person, or to improve or maintain outcomes related to health, medical care, or lifestyle habits, the target person's cognitive Modeling with bias in mind was unknown. Furthermore, it has not been known to consider and model the behavior of the target person in order to improve or maintain outcomes related to health, medical care, and lifestyle habits for the target person.

The estimation device of the present invention includes an input unit for inputting cognitive bias data regarding a target person's cognitive bias, and based on the cognitive bias data, behavioral data regarding the target person's behavior, attribute data of the target person, and by an estimation model including a neural network for estimating at least one of location data of the subject, health data of the subject, exercise data of the subject, lifestyle data of the subject, and medical data of the subject, outputting at least one of the behavior data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data estimated from the cognitive bias data input by the input unit; and an output section.

Further, the estimation device of the present invention includes an input unit for inputting behavioral data regarding the behavior of the subject, and attribute data of the subject, position data of the subject, health data of the subject based on the behavioral data. , the behavioral data inputted by the input unit by an estimation model including a neural network for estimating at least one of the subject's exercise data, the subject's lifestyle data, and the subject's medical data. an output unit that outputs at least one of the attribute data, the position data, the health data, the exercise data, the lifestyle data, and the medical data estimated from the data.

According to the present invention, by estimating behavior-related data or outcomes based on data related to cognitive bias, it is possible to encourage or maintain a target's behavior regarding health, medical care, and lifestyle habits, and to improve health, medical care, and lifestyle habits. It is possible to improve or maintain a subject's outcomes regarding lifestyle habits.

Furthermore, according to the present invention, by estimating outcome-related data based on behavior-related data, it is possible to improve or maintain a subject's outcomes regarding health, medical care, and lifestyle habits.

FIG. 1 is a block diagram showing an example of the system configuration of an estimation system according to the present embodiment. It is a block diagram showing an example of the system configuration of a machine learning server of this embodiment. FIG. 2 is a diagram schematically showing the configuration of various types of data associated with each other according to the present embodiment. FIG. 2 is a diagram schematically showing the structure of various data of non-target person data associated with target person data of the present embodiment. It is a figure showing the generation process of the estimation model of this embodiment. FIG. 2 is a diagram illustrating multimodal data fusion. FIG. 1 is a block diagram showing an example of the configuration of an estimation device according to the present embodiment. FIG. 3 is a diagram showing a process of transmitting learning data (teacher data) according to the present embodiment. It is a figure showing the output process of estimated data estimated by a learning model (estimation model) of this embodiment. It is a figure showing an example of cognitive bias data according to a target person, and behavioral data estimated from target person data.

(First embodiment)
In the first embodiment, an estimation system for estimating behavioral data or outcomes regarding behavior based on cognitive bias data regarding cognitive biases of a subject will be described.

An estimation system according to an embodiment of the present invention will be explained using the drawings. FIG. 1 is a block diagram showing an example of the system configuration of the estimation system of this embodiment. The estimation system 1 estimates behavior-related data based on cognitive bias-related data.

As shown in FIG. 1, the estimation system 1 includes a machine learning server 4 and a computer (estimation device) 5 that are electrically connected via a network 2 and can communicate with each other.                           

FIG. 2 is a block diagram showing an example of the system configuration of the machine learning server of this embodiment. As shown in FIG. 2, the machine learning server 4 includes a processor (control device) 21, a storage device (for example, ROM, RAM, HDD, etc.) 22, an input section 23, and an output section 25.

The processor (control device) 21 is a control unit such as a CPU, MPU, or GPU, and includes a data acquisition unit 32, a machine learning unit 33, an estimation unit 35, and an identification unit 36. The data acquisition section 32, machine learning section 33, estimation section 35, and identification section 36 are electrically connected by a bus (not shown) and can communicate with each other.

The storage device (storage unit) 22 includes cognitive bias data 38, behavioral data 39, target person data 40, non-target person data 41, a learning model storage unit 46, and an estimated data storage unit 47.

The cognitive bias data 38 is data (including feature data) indicating the psychological tendencies of the subject 6. The cognitive bias data 38 includes at least one of a cognitive bias, a cognitive characteristic, and a decision-making characteristic of the subject 6 (including data on these feature amounts).

Cognitive bias data 38 includes, for example, value functions based on prospect theory, dual process theory, etc., loss aversion, endowment effect, status quo bias, ambiguity aversion, uncertainty avoidance, risk preferences (risk aversion), etc.・Risk love), reference point dependence, diminishing sensitivity, probability weighting function, mental accounting (psychological accounting), ignoring the denominator, time discounting/time preference, social preference (altruism, fairness, reciprocity, non-reciprocity) The target person 6's data (including these feature data) regarding at least one of (fairness avoidance, etc.), framing, and heuristics (anchoring, availability, representativeness) is included.

Data regarding the value function, loss aversion, endowment effect, and status quo bias based on the value function is data that indicates the strength or weakness or asymmetry of reactions to gains and losses. Data regarding ambiguity avoidance/uncertainty avoidance is data indicating the strength or weakness of avoidance toward ambiguity. Data regarding risk appetite (risk aversion/risk love) shows the strength or asymmetry of the tendency to avoid risk when in a state of gain (risk aversion) and the tendency to downplay risk when in a state of loss (risk love). This is data showing. The data regarding the reference point dependence is data indicating the relative rate of change from the standard (reference point) set by the subject 6 or the strength or weakness or asymmetry of the reaction to up and down. The data regarding decreasing sensitivity is data indicating the strength of sensitivity with respect to time or frequency, based on the phenomenon that when the same event occurs repeatedly, the subject 6 gradually gets used to the event and the sensitivity decreases.

The data regarding the probability weighting function is data that shows the strength or weakness or asymmetry of the tendency to overestimate when the objective probability is low and underestimate when the objective probability is high. Data related to mental accounting is data that shows the strength or weakness, or asymmetry, of the tendency for decisions made when using money to change depending on the source of the money and what it is used for. Data regarding neglect of the denominator is data that shows the strength or weakness or asymmetry of a tendency to evaluate probability highly when the numerator is large, even when the numerator/denominator probabilities are the same.

Time discount/time preference data is data that shows the discount rate for time, which is how much lower the future value of rewards and gains (delayed rewards) is estimated than the present value (immediate rewards). Subject 6 with a low time discount rate (time preference rate) is able to hold back until future rewards (strong self-control), and subject 6 with a high time discount rate (time preference rate) is able to hold back until future rewards. cannot (strong impulsiveness).

Data on social preferences (altruism, fairness, reciprocity, inequity avoidance, etc.) are important for humans who care not only about their own material rewards but also about the rewards and/or intentions that lead to rewards for reference groups. This is data that shows trends. Data regarding framing changes the impression given depending on what is emphasized, and shows the degree to which it influences decision-making. Data related to heuristics is data that shows an intuitive decision-making process that does not necessarily lead to the correct answer, but can obtain an answer close to the correct answer to a certain level.

Cognitive bias data 38 includes quantitative data, text data, audio data, etc. (including these feature data) obtained by conducting questionnaires, psychological tests, sensitivity tests, games, etc. on the subject 6 regarding cognitive bias ). Further, the cognitive bias data 38 is quantitative data, text data, voice data, etc. (including feature amount data thereof) regarding cognitive bias in SNS, e-mail, blogs, etc.

The behavior data 39 includes data related to at least one of text, images, video, sounds (audio data, etc.), light, smell, and force for urging or maintaining behavior in the subject 6 (characteristic data of these) (including). The behavior data 39 is quantitative data, text data, audio data, etc. (including feature data) regarding behaviors corresponding to each cognitive bias. The behavioral data 39 is quantitative data, text data, audio data, etc. (including these feature data) obtained by administering behavioral questionnaires, psychological tests, sensitivity tests, games, etc. to the subject 6. be. Further, the behavior data 39 is quantitative data, text data, voice data, etc. (including feature amount data of these) related to behavior in SNS, e-mail, blogs, etc. Furthermore, the behavior data 39 includes quantitative data, text data, voice data, etc. (including feature amount data of these) regarding health checkups and behavior at medical institutions. For example, the behavioral data 39 may be quantitative data such as messages that encourage or maintain the behavior of the target person 6 regarding health, medical care, or lifestyle habits, text data, voice data, etc., or may be obtained by natural language processing of these data. It is a feature quantity. The behavior data 39 becomes label data (correct data) of the cognitive bias data 38 in machine learning.

The subject data 40 includes at least one of the subject 6's attribute data, location data, health data, exercise data, lifestyle data, and medical data (including feature amount data thereof). The health data includes at least one of vital data and biomarker data (including their feature data). The subject data 40 becomes label data (correct data) of the cognitive bias data 38 in machine learning.

For example, the subject data 40 includes (1) the subject 6's weight, age, gender, address, diet, drinking, smoking, exercise, stress, medication, socio-economic factors (e.g. income, occupation, etc.), medical history, and family composition data, (2) location information of the subject 6 by GPS etc. (e.g. time, place, surrounding environment, etc.) and map information data, (3) physical activity of the subject 6 (e.g. number of steps, (4) Genetic information, test result information, prescription information, medical treatment of subject 6 at medical institutions This data includes information (for example, surgery, procedure, etc.), medical examination results, and diagnostic information (disease name). Moreover, these data may be statistical values based on a plurality of data. Moreover, these data may be statistical values based on subject data at a plurality of times. Further, these data may be data (for example, food type, amount, calories, etc.) estimated by machine learning based on image data (for example, images of meals, etc.).

Further, as shown in FIG. 2, the subject data 40 includes time-series data 42 and non-time-series data 42 of at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6. Contains time series data 43. The identification unit 36 identifies time series data and non-time series data from at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data, which are the subject data 40.

The time-series data 42 is data that changes over time and can be obtained at multiple times. Time data is associated with the time series data 42. The non-time series data 43 is data that hardly changes with the passage of time (unchangeable data), or data that is substantially impossible to obtain or does not need to be obtained at a plurality of times.

For example, the time series data 42 includes heart rate, electrocardiogram, oxygen saturation, sleep, blood pressure, blood sugar level, and the like. The non-time series data 43 includes genetic information and the like.

The non-target person data 41 includes at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data (including these feature data) of non-target people other than the target person 6. . The health data includes at least one of vital data and biomarker data (including their feature data).

For example, the non-target person data 41 includes (1) weight, age, gender, address, diet, drinking, smoking, exercise, stress, medication, and socio-economic factors (e.g., educational background, income, occupation, position, etc.) of the non-target person; ), past medical history, and family composition data; (2) location information (e.g., time, place, surrounding environment, etc.) of non-target individuals using GPS, and map information; (3) physical activity of non-target individuals. (e.g., number of steps, exercise, muscle strength, muscle mass, etc.), data on heart rate, electrocardiogram, oxygen saturation, blood pressure, blood sugar level, sleep, etc. (4) Genetic information and test result information of non-target persons at medical institutions, This data includes prescription information, medical practice information (for example, surgery, procedure, etc.), medical examination results, and diagnostic information (disease name). Moreover, these data may be statistical values based on data of a plurality of non-target persons. Moreover, these data may be statistical values based on non-target person data at a plurality of times. Further, these data may be data (for example, food type, amount, calories, etc.) estimated by machine learning based on image data (for example, images of meals, etc.).

In addition, non-target data 41 includes quantitative data clarified by previous research for each variable (for example, the odds ratio for women based on men (men are twice as likely to take risks than women). etc.) and the coefficient of the regression equation for age obtained through regression analysis (as your age increases by 1 year, you will take 1% less risk (coefficient = -0.01))). In addition, the non-target person data includes models that parametrically approximate input and output (for example, a value function or a probability weighting function expressed as a mathematical formula) that has been clarified by previous research.

From the non-target person data 41, the cognitive bias data 38, behavioral data 39, and target person data 40 of the target person 6 can be determined or estimated.

As shown in FIG. 2, the non-target person data 41 includes time series data 44 and non-time series data of at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data. Contains 45. The identification unit 36 identifies time series data and non-time series data from at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data, which are non-target person data 41.

The time-series data 44 is data that changes over time and can be obtained at multiple times. Time data is associated with the time series data 44. The non-time series data 45 is data that hardly changes over time (unchangeable data), or data that is substantially impossible to obtain or does not need to be obtained at a plurality of times.

For example, the time series data 44 includes heart rate, electrocardiogram, oxygen saturation, sleep, blood pressure, blood sugar level, and the like. The non-time series data 45 includes genetic information and the like.

FIG. 3 is a diagram schematically showing the structure of various associated data. As shown in FIG. 3, in addition to cognitive bias data 58 and behavioral data 59, subject data such as subject ID 51, attribute data 52, location data 53, health data 54, exercise data 55, lifestyle data 56, and medical Data 57 and non-target person data 60 (including these feature amount data) are associated.

Note that the non-target person data 60 includes attribute data, location data, health data, exercise data, lifestyle data, and medical data of the non-target person. FIG. 4 is a diagram schematically showing the structure of various data of non-target person data 60 associated with target person data 51 to 57. As shown in FIG. 4, non-target person data 60 includes a non-target person ID 61, attribute data 62, location data 63, health data 64, exercise data 65, lifestyle data 66, and medical data 67 (these feature amount data ) are associated.

The target person data 52 to 57 and the non-target person data 62 to 67 are classified into time series data and non-time series data.

Next, generation of the estimation model will be explained. FIG. 5 is a diagram showing the estimation model generation process of this embodiment.

As shown in FIG. 5, in step S1, the input unit 23 inputs various data (cognitive bias data 38, behavioral data 39, target person data 40, and non-target person data 41). The input unit 23 executes a storage instruction for various data, associates the various data, and stores the data in the storage device 22 (step S2, step S3).

The data acquisition unit 32 acquires various data stored in the storage device 22 by executing the acquisition command (step S4). The machine learning unit 33 generates a learning model (estimated model) using a neural network based on the acquired various data (including these feature amount data) by executing the estimated model generation command (step S5). .

The machine learning unit 33 inputs the cognitive bias data 58 (for example, B-1 to B-5 in FIG. 3) to the input layer, and inputs the behavioral data 59 associated with the cognitive bias data 58 (for example, C- in FIG. 3). A learning model (estimated model) is generated by a neural network that inputs 1 to C-5) as correct values of the output layer.

Further, the machine learning unit 33 inputs the cognitive bias data 58 (for example, B-1 to B-5 in FIG. 3) to the input layer, and the attribute data 52 associated with the cognitive bias data 58 (for example, B-1 to B-5 in FIG. 3). E-11 to E-15), location data 53 (for example, E-21 to E-25 in FIG. 3), health data 54 (for example, E-31 to E-35 in FIG. 3), exercise data 55 (for example, , E-41 to E-45 in FIG. 3), lifestyle data 56 (for example, E-51 to E-55 in FIG. 3), and medical data 57 (for example, E-61 to E-65 in FIG. 3) A learning model (estimated model) is generated by a neural network that inputs at least one of the above as the correct value of the output layer.

The intermediate layer of the neural network is provided with, for example, an Affine layer or a Convolution layer. Downsampling processing or the like may be performed as appropriate. Further, the number of layers, the number of neurons, and the activation function of the intermediate layer are optimally selected so that the estimation result is highly accurate. The configuration of the neural network includes Feed Forward Neural Network (FFNN), Recurrent Neural Network (RNN), Long Short Term Memory (LSTM), attention mechanism, etc. .

The machine learning unit 33 uses parameters (weights) initialized with predetermined values such as random numbers to calculate the value output to the output layer when the cognitive bias data 58 is input to the input layer and the correct answer of the output layer. A loss function representing the deviation from the value (behavior data 59 or/and subject data 52 to 57) is calculated, and the differential value of the loss function is used as the slope to calculate the difference between the value output to the output layer and the correct value of the output layer. A learning model (estimated model) is generated by changing parameters (weights) so that the deviation becomes smaller.

In this way, the machine learning unit 33 creates a learning model (estimated model ) is generated. The learning model (estimated model) includes a neural network including at least one of a convolution layer, a recursive calculation layer, and an attention mechanism.

The learning model (estimated model) includes attribute data 52, location data 53, health data 54, exercise data 55, lifestyle data 56, and It may also include a neural network for estimating cognitive bias data and/or subject data (outcome data) of the subject 6 based on at least one subject data of the medical data 57 (including these feature data). .

In addition, the learning model (estimation model) includes attribute data 62, location data 63, health data 64, exercise data 65, lifestyle data 66, and medical data 67 (these feature amount data) of non-target persons other than the target person 6. It may also include a neural network for estimating cognitive bias data and/or subject data (outcome data) of the subject 6 based on at least one non-target subject data 60 (including the above).

In addition to or separately from the cognitive bias data 58, the learning model (estimation model) includes attribute data 52, location data 53, health data 54, exercise data 55, and lifestyle data 56 of the subject 6. , and medical data 57 (including these feature data), cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and It may also include a neural network for estimating at least one subject data (outcome data) of the medical data.

In this case, the estimated data (especially time series data) becomes the label data (correct data), and the data associated with this data (cognitive bias data of subject 6 and subject data) becomes the input layer of the neural network. A learning model (estimated model) is constructed based on input and learning. For example, by using a learning model (estimation model) learned from Subject 6's past sleep time, other subject data, and cognitive bias data, Subject 6's future sleep time (outcome data) can be estimated. Ru. The learning process is similar to above.

In addition, the learning model (estimation model) includes attribute data 62, location data 63, health data 64, exercise data 65, lifestyle data 66, and medical data 67 (these feature amount data) of non-target persons other than the target person 6. at least one subject data of the subject 6's cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and medical data. It may also include a neural network for estimating (outcome data).

In this case, the estimated data becomes label data (correct data), and the data associated with this data (cognitive bias data of non-target people and data of non-target people) is input to the input layer of the neural network and the learning is performed. A model (estimated model) may be constructed. For example, the cognitive bias data of the target person 6 is estimated by using the cognitive bias data of the non-target person or a learning model (estimated model) learned from the non-target person data. The learning process is similar to above.

In addition, non-target data includes quantitative data clarified by previous studies for each variable (for example, the odds ratio of women based on men (men are twice as likely to take risks than women, etc.). ), the coefficient of the regression equation for age obtained through regression analysis (for each year that your age increases, you will take 1% less risk (coefficient = -0.01)), and the model that parametrically approximates the input and output (for example, (e.g. value function or probability weighting function expressed as a mathematical formula), and based on data quantified by evidence from previous research, cognitive bias data and behavioral data of target person 6 and target person data are calculated from non-target person data. Data (outcome data) may be determined or estimated.

As described above, the learning model (estimation model) includes at least one of a convolution layer, a recursive calculation layer, and an attention mechanism, and connects target person data (or target person data and non-target person data). This is an estimation model that is machine-learned by inputting the acquired data) into the input layer of a neural network. Non-subject data includes statistical values based on data from multiple non-subjects and data on evidence obtained from previous research.

When concatenating target person data and non-target person data, (1) a method of extracting feature amounts after concatenating the raw data of target person data and non-target person data; There are methods such as (3) a method in which feature amounts are individually extracted from raw data of non-target person data and then the feature amounts are connected, and (3) a method in which these are combined. For example, in the case of image data, it is also possible to extract feature amounts in the Convolution layer and then connect them with other data.

In addition, in the case of a neural network that includes a recursive calculation layer (RNN or LSTM) and an attention mechanism, the learning model (estimated model) is based on the cognitive bias of the target person 6 or non-target person output from the intermediate layer or output layer of the neural network. At least one of data, attribute data, location data, health data, exercise data, lifestyle data, medical data, and behavioral data may be input to the input layer of the neural network.

Next, using FIG. 6, a description will be given of multimodal data fusion in which machine learning is performed by inputting concatenated data of target person data and non-target person data to the input layer of a neural network. As shown in FIG. 6, the data input to the input layer of the neural network is data m _concat that is a concatenation of target person data _it and non-target person data z ₁ and z ₂ . As the subject data _it , time-series data output from at least one of Cell State (plays the role of long-term memory) and Hidden State (plays the role of short-term memory) of the neural network is used. As the non-target person data z ₁ and z ₂ , data obtained by dimensionally compressing a plurality of non-target person data Modalities 1 and 2 using Autoencoders 1 and 2, respectively, is used.

In addition to the target person data, cognitive bias data of the target person 6 and non-target person data may be connected and input to the input layer of the neural network.

The generated learning model (estimated model) is stored in the learning model storage unit 46 (step S6).

Next, the computer (estimation device) 5 will be explained using FIG. FIG. 7 is a block diagram showing an example of the configuration of the estimation device of this embodiment. The computer (estimation device) 5 inputs cognitive bias data and outputs behavioral data and/or outcome data acquired by the learning model (estimation model).

As shown in FIG. 7, the computer (estimation device) 5 includes a processor (control device) 210, a storage device (for example, ROM, RAM, HDD, etc.) 220, an input section 230, and an output section 250.

The processor (control device) 210 is a control unit such as a CPU, MPU, or GPU, and includes a data acquisition unit 320, an estimation unit 350, and an identification unit 360. The data acquisition unit 320, the estimation unit 350, and the identification unit 360 are electrically connected by a bus (not shown) and can communicate with each other.

The storage device (storage unit) 220 includes cognitive bias data 380, behavioral data 390, subject data 400, and estimated data storage unit 470.

The input unit 230 inputs cognitive bias data 380 regarding the cognitive bias of the subject 6. The input unit 230 inputs various data (cognitive bias data 380, behavior data 390, subject data 400). The input unit 230 associates various types of data and stores them in the storage device 220 by executing a command to store various types of data. Here, the cognitive bias data 380, behavioral data 390, subject data 400 (attribute data, location data, health data, exercise data, lifestyle data, and medical data) and their feature data are the above cognitive bias data. 38, behavioral data 39, and subject data 40.

The identification unit 360 identifies time series data and non-time series data from at least one of cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and medical data, which are the subject data 400. do.

Next, a process in which the computer (estimation device) 5 transmits learning data (teacher data) to the machine learning server 4 via the network 2 will be described. The computer (estimation device) 5 outputs various data (cognitive bias data 380, etc.) to the machine learning server 4 in order to generate a learning model (estimation model).

FIG. 8 is a diagram showing the process of transmitting learning data (teacher data) in this embodiment. As shown in FIG. 8, in step S10, the input unit 230 inputs various data (cognitive bias data 380, behavior data 390, subject data 400). The input unit 230 associates various types of data and stores them in the storage device 220 by executing a command to store various types of data (step S20, step S30).

The data acquisition unit 320 acquires various data stored in the storage device 220 by executing the acquisition command. The output unit 250 transmits various data to the machine learning server 4 via the network 2 by executing the transmission command (step S40).

The input unit 23 of the machine learning server 4 inputs various data (cognitive bias data 380, behavior data 390, and subject data 400) received from the estimation device 5. The various input data (cognitive bias data 380, behavioral data 390, and subject data 400) are stored in the storage device 22 as cognitive bias data 38, behavioral data 39, and subject data 40, respectively. The time series data and non-time series data are identified by the identification unit 36.

The data acquisition unit 32 of the machine learning server 4 acquires various data stored in the storage device 22, and the machine learning unit 33 of the machine learning server 4 executes the estimation model generation command to obtain the various acquired data. A learning model (estimated model) is generated by a neural network based on the data (including these feature data).

In this way, the computer (estimation device) 5 has a function of outputting learning data (teacher data) to the machine learning server 4. The cognitive bias data 380, the behavior data 390, and the subject data 400 are associated with each other and used as learning data by the machine learning unit 33 of the machine learning server 4. Further, the cognitive bias data 380, the behavior data 390, and the target person data 400 may be used as non-target person data 41 of the target person other than the target person 6.

Next, a function of the computer (estimation device) 5 to output behavior data and other various data estimated using the learning model (estimation model) in the machine learning server 4 will be explained.

Note that in this embodiment, the learning model (estimated model) stored in the learning model storage unit 46 of the machine learning server 4 is used as the learning model (estimated model); By storing the learning model (estimation model) in the storage device 220 of the computer (estimation device) 5 via the network 2, even if the learning model (estimation model) stored in the computer (estimation device) 5 is used. good.

In this embodiment, since the cognitive bias data 380 of the subject 6 and the non-time series data of the subject data are sent from the computer (estimation device) 5 to the machine learning server 4 in advance, the computer (estimation device) 5 The time series data of the subject data of the subject 6 is transmitted to the machine learning server 4. Note that when new cognitive bias data 380 and new non-time series data of the subject data are input from the input unit 230, the computer (estimation device) 5 inputs the new cognitive bias data 380 and the new non-time series data of the subject data. The non-time-series data is sent to the machine learning server 4. The machine learning server 4 that has received the various data updates the learning model (estimated model) using the machine learning unit 33 based on the new data.

FIG. 9 is a diagram showing the output process of estimated data estimated by the learning model (estimation model) of this embodiment. As shown in FIG. 9, in step S10, the input unit 230 inputs various data (subject data 400). The input unit 230 associates various types of data and stores them in the storage device 220 by executing a command to store various types of data (step S20, step S30).

Note that the target person data 400 (attribute data, location data, health data, exercise data, lifestyle data, and medical data) and their feature data are similar to the above-mentioned target person data 40, but the machine learning server This data is different from the learning data (teacher data) by the machine learning unit 33 of No. 4.

The data acquisition unit 320 acquires various data stored in the storage device 220 by executing the acquisition command. The estimation unit 350 transmits various data to the machine learning server 4 via the network 2 by executing the estimation command (step S400).

The input unit 23 of the machine learning server 4 inputs various data (target person data 400) received from the estimation device 5.

The estimation unit 35 of the machine learning server 4 estimates behavioral data using a learning model (estimation model) based on the various acquired data (including these feature data) by executing the estimation command.

In addition to or separately from the cognitive bias data 38, the estimation unit 35 also includes at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6. The behavioral data of the subject 6 may be estimated based on the subject data 40.

In addition to or separately from the cognitive bias data 38, the estimation unit 35 also includes at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6. Based on the subject data 40, at least one of cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6 may be estimated as outcome data.

In addition to or separately from the cognitive bias data 38, the estimation unit 35 also collects attribute data, location data, health data, exercise data, lifestyle data, and medical information of non-target persons other than the target person 6. The behavior data of the target person 6 may be estimated based on at least one non-target person data 41 of the data.

In addition to or separately from the cognitive bias data 38, the estimation unit 35 also collects attribute data, location data, health data, exercise data, lifestyle data, and medical information of non-target persons other than the target person 6. Based on at least one non-target person data 41 of the data, at least one of the target person 6's cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and medical data is used as outcome data. It may be estimated.

The estimation unit 35 reads the learning model (estimated model) from the learning model storage unit 46, inputs the cognitive bias data 380 into the input layer of the neural network of the learning model (estimated model), and inputs the estimated estimation data (behavior data, At least one of cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and medical data is obtained from the output layer. The estimated data is stored in the estimated data storage section 47.

The estimation data is transmitted by the output unit 25 of the machine learning server 4 to the computer (estimation device) 5 via the network 2. The input unit 230 of the computer (estimation device) 5 receives the estimation data by executing the reception command (step S50). The input unit 230 stores the estimated data in the estimated data storage unit 470.

By executing the estimated data output command, the output unit 250 generates a learning model (estimation model) including a neural network for estimating behavioral data that urges or maintains behavior in the subject 6 based on the cognitive bias data. Accordingly, behavioral data estimated from the cognitive bias data input by the input unit 230 is output (step S60). Furthermore, the output unit 250 outputs other estimated data by executing the estimated data output command (step S60).

The output estimation data is displayed on the display of a communication terminal (tablet, wearable terminal, etc.) electrically connected to the computer (estimation device) 5 or on the display of the computer (estimation device) 5.

FIG. 10 is a diagram illustrating an example of cognitive bias data according to a subject and behavioral data and outcome data estimated from the subject data.

For example, when data on self-control tendency is acquired as cognitive bias data for subject 7, and data on hypertension is acquired as subject data, the input unit 230 inputs cognitive bias data regarding the cognitive bias of subject 7. is input, and the output unit 250 uses an estimation model including a neural network for estimating behavioral data that urges or maintains behavior in the target person 7 based on the cognitive bias data and the target person data. Outputs behavioral data (such as messages emphasizing numerical targets for improving hypertension) and outcome data (estimated blood pressure, etc.) estimated from the input cognitive bias data and subject data. In addition, as behavioral data, the output unit 250 outputs data regarding sounds, lights, odors, and forces that encourage or maintain medication or exercise at a predetermined time to a speaker, a light emitting device, an odor mixture generator, and a driver. It may also be output to a device (for example, a massage device or a motor).

Note that the blood pressure data is acquired from the wearable terminal as time-series data, and new behavioral data and future blood pressure values (outcome data) are estimated according to the acquisition time of the blood pressure data.

Similarly, for subject 8, if data on risk-loving tendencies is acquired as cognitive bias data and high-calorie data (for example, a photographic image of a meal) is acquired as subject data, the output unit 250 Output data (such as audio that emphasizes the benefits of restricting calorie intake) and outcome data (such as estimated body weight). In addition, as behavior data, the output unit 250 outputs data regarding sounds, lights, odors, and forces for suppressing meals and snacks to speakers, light emitting devices, odor mixture generators, and driving devices (e.g., It may also be output to a massage device, motor, etc.

Similarly, when data on risk aversion tendency is acquired as cognitive bias data and high stress data (e.g. comments on SNS) is acquired as subject data regarding subject 9, the output unit 250 outputs behavioral data. (e.g., a message emphasizing the risk of high stress) and outcome data (e.g., an estimated depression scale). In addition, as behavior data, the output unit 250 outputs data related to stress-reducing music, light, smell, and force at a predetermined time to a speaker, a light emitting device, an odor mixture generator, and a drive device (for example, a massage device). or a motor), etc.

Similarly, for the subject 10, if data on impulsive tendency is acquired as cognitive bias data and data on sleep deprivation (for example, sleep time measured by a wearable terminal) is acquired as subject data, the output The unit 250 outputs behavioral data (such as a message proposing an immediately executable method for resolving sleep deprivation) and outcome data (such as estimated sleep time and quality). In addition, as behavioral data, the output unit 250 outputs data regarding sleep-inducing music, light, smell, and force at a predetermined time to a speaker, a light emitting device, an odor mixture generator, and (for example, a massage device or a motor). You can also output it to

In this way, the computer (estimation device) 5 outputs behavior data and/or outcome data estimated by the learning model (estimation model).

As described above, according to the present embodiment, the behavior of the target person regarding health, medical care, and lifestyle habits that are affected by cognitive bias is modeled, and the behavior of the target person regarding health, medical care, and lifestyle habits is encouraged and maintained. It is possible to construct a learning model (estimation model) for

Further, according to the present embodiment, the target person's outcomes regarding health, medical care, and lifestyle habits that are affected by cognitive bias are modeled, and the target person's outcomes regarding health, medical care, and lifestyle habits are improved or maintained. A learning model (estimation model) can be constructed.

Additionally, in addition to or separately from cognitive bias data, estimated data (behavioral data, cognitive bias data, attribute data, location data, health data, exercise data) of the subject based on the subject's own subject data may be used. , lifestyle data, and medical data), highly accurate estimation data that matches the characteristics of the subject can be output.

Furthermore, in order to estimate the target person's estimated data based on non-target person data other than the target person in addition to or separately from the cognitive bias data, the accuracy is improved by taking into account the general characteristics of the person other than the target person. It is possible to output highly estimated data.

In this way, by presenting highly accurate estimation data to subjects, it is expected that nudges that encourage or maintain behavior in subjects, as well as nudges that improve or maintain outcomes, can be effectively generated. Ru.

In order to obtain optimal outcomes regarding health, medical care, and lifestyle habits (e.g., health promotion, disease prevention and treatment), it is necessary to encourage the target person's own behavior in a medically favorable direction.

However, existing guidelines regarding health, medical care, and lifestyle habits are based on statistical values of anonymized population data, so variations in data for individual subjects are statistically treated as errors, and existing guidelines It has been difficult to provide highly accurate guidance and interventions that match the characteristics of the target person. In addition, existing guidance and interventions by experts are based on limited data on subjects due to constraints such as time and place to observe subjects, and are based on guidelines based on group data and in light of experts' experience. Therefore, it was difficult to provide highly accurate guidance and intervention that matched the characteristics of the target person.

According to this embodiment, it is possible to provide highly accurate guidance and intervention that matches the characteristics of the target person based on the target person's data without restrictions on time or place, and provides the optimal guidance and intervention for the target person. It can encourage and maintain behaviors that lead to outcomes.

(Second embodiment)
In the second embodiment, an estimation system for estimating an outcome based on data regarding a subject's behavior will be described. Description of parts similar to those in the first embodiment described above will be omitted.

The generation of the estimation model will be explained. As shown in FIG. 5, in step S1, the input unit 23 inputs various data (behavior data 39, target person data 40, and non-target person data 41). The input unit 23 executes a storage instruction for various data, associates the various data, and stores the data in the storage device 22 (step S2, step S3).

The data acquisition unit 32 acquires various data stored in the storage device 22 by executing the acquisition command (step S4). The machine learning unit 33 generates a learning model (estimated model) using a neural network based on the acquired various data (including these feature amount data) by executing the estimated model generation command (step S5). .

The machine learning unit 33 inputs the behavioral data 59 (for example, C-1 to C-5 in FIG. 3) to the input layer, and inputs the subject data (for example, C-1 to C-5 in FIG. 3) associated with the behavioral data 59 to the input layer. C-5) to the input layer, attribute data 52 (for example, E-11 to E-15 in FIG. 3) associated with behavior data 59, and position data 53 (for example, E-21 to E in FIG. -25), health data 54 (for example, E-31 to E-35 in Figure 3), exercise data 55 (for example, E-41 to E-45 in Figure 3), lifestyle data 56 (for example, E-51 to E-55) and medical data 57 (for example, E-61 to E-65 in FIG. 3) are inputted as correct values of the output layer. generate.

The machine learning unit 33 uses parameters (weights) initialized with predetermined values such as random numbers to determine the value output to the output layer when the behavioral data 59 is input to the input layer and the correct value of the output layer. (Target person data 52 to 57) A loss function representing the deviation from the output layer is calculated, and the differential value of the loss function is used as the slope, so that the deviation between the value output to the output layer and the correct value of the output layer becomes small. A learning model (estimated model) is generated by changing parameters (weights).

In this way, the machine learning unit 33 generates a learning model (estimated model) using a neural network that inputs the behavior data 59 into the input layer and inputs the subject data 52 to 57 as correct values in the output layer. The learning model (estimated model) includes a neural network including at least one of a convolution layer, a recursive calculation layer, and an attention mechanism.

The learning model (estimated model) includes attribute data 52, location data 53, health data 54, exercise data 55, lifestyle data 56, and medical data of the subject 6 in addition to or separately from the behavioral data 59. The neural network may include a neural network for estimating the subject data (outcome data) of the subject 6 based on at least one subject data of 57 (including these feature data).

In addition, the learning model (estimation model) includes attribute data 62, location data 63, health data 64, exercise data 65, lifestyle data 66, and medical data 67 (these feature amount data) of non-target persons other than the target person 6. The neural network may include a neural network for estimating target person data (outcome data) of the target person 6 based on at least one non-target person data 60 (including the above).

The learning model (estimated model) also includes attribute data 52, location data 53, health data 54, exercise data 55, lifestyle data 56, and Based on at least one subject data of the medical data 57 (including these feature data), behavioral data, attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6 are determined. It may also include a neural network for estimating at least one subject data (outcome data).

In this case, the estimated data (especially time series data) becomes the label data (correct data), and the data associated with this data (behavior data of subject 6 and subject data) is input to the input layer of the neural network. A learning model (estimation model) that has been trained using the following methods is constructed. For example, by using a learning model (estimation model) learned from Subject 6's past sleep time, other subject data, and cognitive bias data, Subject 6's future sleep time (outcome data) can be estimated. Ru. The learning process is similar to above.

In addition, the learning model (estimation model) includes attribute data 62, location data 63, health data 64, exercise data 65, lifestyle data 66, and medical data 67 (these feature amount data) of non-target persons other than the target person 6. At least one subject data (outcome data) of the subject 6's attribute data, location data, health data, exercise data, lifestyle data, and medical data is based on at least one non-target subject data 60 of may include a neural network for estimating.

In this case, the estimated data becomes label data (correct data), and the data associated with this data (behavior data of non-target people and data of non-target people) is input to the input layer of the neural network to learn the learning model. (estimation model) may be constructed. For example, the outcome data of the target person 6 is estimated by using a learning model (estimation model) learned from the non-target person's behavior data and the non-target person data. The learning process is similar to above.

In addition, non-subject data is a model that parametrically approximates the quantitative data and inputs and outputs clarified by previous research for each variable, and is based on the data quantified by the evidence of previous research. Behavioral data and subject data (outcome data) of the subject 6 may be determined or estimated from the subject data.

As described above, the learning model (estimation model) includes at least one of a convolution layer, a recursive calculation layer, and an attention mechanism, and connects target person data (or target person data and non-target person data). This is an estimation model that is machine-learned by inputting the acquired data) into the input layer of a neural network. Non-subject data includes statistical values based on data from multiple non-subjects and data on evidence obtained from previous research.

Next, the computer (estimation device) 5 will be explained using FIG. The computer (estimation device) 5 inputs behavioral data and outputs outcome data acquired by a learning model (estimation model).

Next, a process in which the computer (estimation device) 5 transmits learning data (teacher data) to the machine learning server 4 via the network 2 will be described. The computer (estimation device) 5 outputs various data (cognitive bias data 380, etc.) to the machine learning server 4 in order to generate a learning model (estimation model).

As shown in FIG. 8, in step S10, the input unit 230 inputs various data (behavior data 390 and subject data 400). The input unit 230 associates various types of data and stores them in the storage device 220 by executing a command to store various types of data (step S20, step S30).

The data acquisition unit 320 acquires various data stored in the storage device 220 by executing the acquisition command. The output unit 250 transmits various data to the machine learning server 4 via the network 2 by executing the transmission command (step S40).

The input unit 23 of the machine learning server 4 inputs various data (behavior data 390 and subject data 400) received from the estimation device 5. The input various data (behavior data 390 and subject data 400) are stored in the storage device 22 as behavior data 39 and subject data 40, respectively. The time series data and non-time series data are identified by the identification unit 36.

The data acquisition unit 32 of the machine learning server 4 acquires various data stored in the storage device 22, and the machine learning unit 33 of the machine learning server 4 executes the estimation model generation command to obtain the various acquired data. A learning model (estimated model) is generated by a neural network based on the data (including these feature data).

In this way, the computer (estimation device) 5 has a function of outputting learning data (teacher data) to the machine learning server 4. The behavior data 390 and the subject data 400 (correct data) are associated with each other and used as learning data by the machine learning unit 33 of the machine learning server 4. Further, the behavior data 390 and the target person data 400 may be used as non-target person data 41 of the target person other than the target person 6.

Next, the function of the computer (estimation device) 5 to output the outcome and other various data estimated using the learning model (estimation model) in the machine learning server 4 will be explained.

In this embodiment, since the behavioral data 390 of the target person 6 and the non-time series data of the target person data are sent from the computer (estimation device) 5 to the machine learning server 4 in advance, the computer (estimation device) 5 The time series data of the subject data of the person 6 is transmitted to the machine learning server 4. Note that when new behavior data 390 and new non-time series data of the subject data are input from the input unit 230, the computer (estimation device) 5 inputs the new behavior data 390 and new non-time series data of the subject data. Send the time series data to the machine learning server 4. The machine learning server 4 that has received the various data updates the learning model (estimated model) using the machine learning unit 33 based on the new data.

FIG. 9 is a diagram showing the output process of estimated data estimated by the learning model (estimation model) of this embodiment. As shown in FIG. 9, in step S10, the input unit 230 inputs various data (subject data 400). The input unit 230 associates various types of data and stores them in the storage device 220 by executing a command to store various types of data (step S20, step S30).

The data acquisition unit 320 acquires various data stored in the storage device 220 by executing the acquisition command. The estimation unit 350 transmits various data to the machine learning server 4 via the network 2 by executing the estimation command (step S400).

The input unit 23 of the machine learning server 4 inputs various data (target person data 400) received from the estimation device 5.

By executing the estimation command, the estimation unit 35 of the machine learning server 4 calculates subject data (outcome data) using a learning model (estimation model) based on the various acquired data (including these feature data). Estimate.

In addition to or separately from the behavioral data 39, the estimating unit 35 also collects at least one of the target person's attribute data, location data, health data, exercise data, lifestyle data, and medical data of the target person 6. The subject data (outcome data) of the subject 6 may be estimated based on the data 40.

In addition to or separately from the behavioral data 39, the estimating unit 35 also collects at least one of the target person's attribute data, location data, health data, exercise data, lifestyle data, and medical data of the target person 6. Based on the data 40, at least one of attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6 may be estimated as outcome data.

Further, in addition to or separately from the behavioral data 39, the estimation unit 35 may collect attribute data, location data, health data, exercise data, lifestyle data, and medical data of non-target persons other than the target person 6. The outcome data of the subject 6 may be estimated based on at least one non-target subject data 41.

Further, in addition to or separately from the behavioral data 39, the estimation unit 35 may collect attribute data, location data, health data, exercise data, lifestyle data, and medical data of non-target persons other than the target person 6. Based on at least one subject data 41, at least one of cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, and medical data of the subject 6 is estimated as outcome data. Good too.

The estimation unit 35 reads the learning model (estimated model) from the learning model storage unit 46, inputs the behavior data 390 to the input layer of the neural network of the learning model (estimated model), and inputs the estimated estimation data (attribute data, position data, health data, exercise data, lifestyle data, and medical data) from the output layer. The estimated data is stored in the estimated data storage section 47.

The estimation data is transmitted by the output unit 25 of the machine learning server 4 to the computer (estimation device) 5 via the network 2. The input unit 230 of the computer (estimation device) 5 receives the estimation data by executing the reception command (step S50). The input unit 230 stores the estimated data in the estimated data storage unit 470.

By executing the estimated data output command, the output unit 250 outputs information input from the input unit 230 using a learning model (estimation model) including a neural network for estimating outcome data for the subject 6 based on behavioral data. Outcome data estimated from the behavior data is output (step S60). Furthermore, the output unit 250 outputs other estimated data by executing the estimated data output command (step S60).

As described above, according to the present embodiment, the outcomes for health, medical care, and lifestyle habits that are affected by the behavior of the target person are modeled, and the outcomes for health, medical care, and lifestyle habits are modeled to improve or maintain the outcomes for health, medical care, and lifestyle habits. A learning model (estimation model) can be constructed.

In addition to or apart from behavioral data, estimated data (behavioral data, cognitive bias data, attribute data, location data, health data, exercise data, lifestyle data, (at least one of habit data and medical data), highly accurate estimation data that matches the characteristics of the subject can be output.

Furthermore, since the estimated data of the target person is estimated based on non-target person data other than the target person in addition to or separately from the behavioral data, highly accurate estimation takes into account the general characteristics of the other person. Data can be output.

In this way, by presenting highly accurate estimation data to subjects, it is expected to effectively nudge subjects to improve or maintain outcomes.

Although the embodiments according to the present invention have been described above, the present invention is not limited to these, and can be modified and modified within the scope of the claims.

The computer (estimation device) 5 may be a laptop computer, a desktop computer, a tablet terminal, a smartphone, or a wearable terminal.

Further, the computer (estimation device) 5 is electrically connected to other communication terminals (laptop computer, desktop computer, tablet terminal, smartphone, wearable terminal, etc.), and the input unit 230 is connected to other communication terminals (laptop computer, desktop computer, tablet terminal, smartphone, wearable terminal, etc.). Various data may be input from the terminal, and the output unit 250 may output various data to other communication terminals.

Furthermore, the machine learning server 4 can also serve as the estimation device of the present invention. In this case, the functions of the input unit 23, output unit 25, data acquisition unit 32, estimation unit 35, identification unit 36, and estimated data storage unit 47 of the machine learning server 4 are the functions of the input unit 230 of the computer (estimation device) 5, They correspond to the functions of the output section 250, the data acquisition section 320, the estimation section 350, the identification section 360, and the estimated data storage section 470, respectively.

Further, in this embodiment, the machine learning server 4 generates the learning model (estimated model), but the computer (estimation device) 5 may generate the learning model (estimated model).

Furthermore, the behavioral data may include data regarding incentives for encouraging the target person to take action. For example, the behavioral data may be data regarding coupons, points, money, and special rights that can be used at a predetermined location.

The present embodiment also includes a step of inputting cognitive bias data regarding the target person's cognitive bias, and based on the cognitive bias data, behavioral data regarding the target person's behavior, attribute data of the target person, by an estimation model including a neural network for estimating at least one of location data of the subject, health data of the subject, exercise data of the subject, lifestyle data of the subject, and medical data of the subject, outputting at least one of the behavior data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data estimated from the cognitive bias data input by the input unit; The method includes the steps of:

The present embodiment also includes a step of inputting behavioral data regarding the behavior of the subject, and based on the behavioral data, attribute data of the subject, location data of the subject, health data of the subject, Estimated from the behavioral data input by the input unit by an estimation model including a neural network for estimating at least one of the subject's exercise data, the subject's lifestyle data, and the subject's medical data. and outputting at least one of the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data.

The present embodiment also includes cognitive bias data regarding the target person's cognitive bias, behavioral data regarding the target person's behavior labeled with the cognitive bias data, attribute data of the target person, and location data of the target person. , Machine learning for causing a neural network to perform machine learning using at least one of the subject's health data, the subject's exercise data, the subject's lifestyle data, and the subject's medical data as training data. an input unit for inputting cognitive bias data regarding the cognitive bias of the subject; and the behavioral data and the attributes estimated from the cognitive bias data input by the input unit by an estimation model including the neural network. and an output unit that outputs at least one of data, the location data, the health data, the exercise data, the lifestyle data, and the medical data.

The present embodiment also includes behavioral data regarding the behavior of the subject, attribute data of the subject labeled with the behavioral data, location data of the subject, health data of the subject, and a machine learning unit for causing a neural network to perform machine learning using at least one of exercise data, lifestyle data of the subject, and medical data of the subject as training data; the attribute data, the position data, the health data, the exercise data, the lifestyle data estimated from the behavioral data input by the input unit, using an input unit to input, and an estimation model including the neural network; and an output unit that outputs at least one of the medical data.

The present embodiment also includes an estimation program executed by a computer in which a processor estimates the behavior of a target person, wherein the computer labels cognitive bias data regarding the cognitive bias of the target person and the cognitive bias data. behavioral data regarding the behavior of the subject, attribute data of the subject, location data of the subject, health data of the subject, exercise data of the subject, lifestyle data of the subject, and an acquisition function for acquiring at least one of medical data of a person; the cognitive bias data; the behavior data; the attribute data; the location data; the health data; the exercise data; the lifestyle data; The estimation program includes a machine learning function for causing a neural network to perform machine learning using at least one of the data as training data.

The present embodiment also includes an estimation program executed by a computer in which a processor estimates the behavior of a target person, wherein the computer generates behavior data regarding the behavior of the target person, and the behavior data labeled with the behavior data. Acquire at least one of attribute data of the subject, location data of the subject, health data of the subject, exercise data of the subject, lifestyle data of the subject, and medical data of the subject. The acquisition function, the behavior data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and at least one of the medical data are used as teacher data to cause a neural network to perform machine learning. It includes a machine learning function for, and an estimation program that realizes.

According to the present invention, by estimating behavioral data and/or outcome data regarding a subject's behavior, it is useful as an estimation device, etc. that promotes and maintains improvements in the subject's behavior and outcomes regarding health, medical care, and lifestyle habits. It is.

1...Estimation system 2...Network 4...Machine learning server 5...Estimation device 22...Storage device 23...Input unit 25...Output unit 32...Data acquisition unit 33...Machine learning unit 35...Estimation unit 36...Identification unit 38...Cognitive bias Data 39...behavior data 40...target person data 41...non-target person data 41...target person data 42...time series data 43...non-time series data 44...time series data 45...non-time series data 46...learning model storage unit 47 ... Estimated data storage section 58 ... Cognitive bias data 59 ... Behavioral data 60 ... Non-target person data 220 ... Storage device 230 ... Input section 250 ... Output section 320 ... Data acquisition section 350 ... Estimation section 360 ... Identification section 380 ... Cognitive bias data 390...Behavior data 400...Target data 470...Estimated data storage unit

Claims (17)

1. an input unit for inputting cognitive bias data regarding the subject's cognitive bias;
   Based on the cognitive bias data, behavioral data regarding the subject's behavior, attribute data of the subject, location data of the subject, health data of the subject, exercise data of the subject, life of the subject the behavioral data and the attribute data estimated from the cognitive bias data input by the input unit by an estimation model including a neural network for estimating at least one of habit data and medical data of the subject; an output unit that outputs at least one of the location data, the health data, the exercise data, the lifestyle data, and the medical data;
   An estimation device comprising:
2. an input unit for inputting behavioral data regarding the target person's behavior;
   Based on the behavioral data, the subject's attribute data, the subject's location data, the subject's health data, the subject's exercise data, the subject's lifestyle data, and the subject's medical data. The attribute data, the position data, the health data, the exercise data, and the lifestyle habits estimated from the behavioral data input by the input unit using an estimation model including a neural network for estimating at least one of the following: an output unit that outputs at least one of data and the medical data;
   An estimation device comprising:
3. The estimation device according to claim 1, wherein the cognitive bias data is data indicating a psychological tendency of the subject.
4. The estimation device according to claim 1 or 2, wherein the behavior data includes data regarding at least one of text, image, video, sound, light, smell, and force regarding the subject's behavior. .
5. The estimation device according to claim 1 or 2, wherein the behavior data includes data regarding an incentive for encouraging the target person to take action.
6. The estimation model is for estimating the behavioral data of the subject based on at least one subject data of the subject's attribute data, location data, health data, exercise data, lifestyle data, and medical data. The estimation device according to claim 1, comprising a neural network.
7. The estimation model estimates the behavior of the target person based on at least one non-target person data of attribute data, location data, health data, exercise data, lifestyle data, and medical data of the non-target person other than the target person. 7. The estimation device according to claim 1, further comprising a neural network for estimating data.
8. The estimation model calculates the cognitive bias data and the behavior of the subject based on at least one subject data of the subject's attribute data, location data, health data, exercise data, lifestyle data, and medical data. Claim 1 or claim 1, further comprising a neural network for estimating at least one of data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data. The estimation device according to item 2.
9. The estimation model is based on at least one non-target person data of non-target people other than the target person, such as attribute data, location data, health data, exercise data, lifestyle data, and medical data. comprising a neural network for estimating at least one subject data of cognitive bias data, the behavior data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data. The estimating device according to claim 8, characterized in that:
10. The estimation model includes the neural network including at least one of a convolution layer, a recursive calculation layer, and an attention mechanism, and connects the target person data and the non-target person data and inputs the connected data to the input layer of the neural network. 10. The estimation device according to claim 9, wherein the estimation model is machine learned by performing machine learning.
11. Identifying time series data and non-time series data from at least one of the cognitive bias data, the behavior data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data. Equipped with an identification section to
    The estimation model includes at least one of a convolution layer, a recursive calculation layer, and an attention mechanism, and performs machine learning by connecting the time series data and the non-time series data and inputting the resultant data to the input layer of the neural network. The estimating device according to claim 1 or 2, wherein the estimating model is an estimating model.
12. inputting cognitive bias data regarding the subject's cognitive bias;
    Based on the cognitive bias data, behavioral data regarding the subject's behavior, attribute data of the subject, location data of the subject, health data of the subject, exercise data of the subject, life of the subject the behavioral data and the attribute data estimated from the cognitive bias data input by the input unit by an estimation model including a neural network for estimating at least one of habit data and medical data of the subject; outputting at least one of the location data, the health data, the exercise data, the lifestyle data, and the medical data;
    An estimation method comprising:
13. entering behavioral data regarding the subject's behavior;
    Based on the behavioral data, the subject's attribute data, the subject's location data, the subject's health data, the subject's exercise data, the subject's lifestyle data, and the subject's medical data. The attribute data, the position data, the health data, the exercise data, and the lifestyle habits estimated from the behavior data input by the input unit using an estimation model including a neural network for estimating at least one of the following: outputting at least one of data and the medical data;
    An estimation device comprising:
14. Cognitive bias data regarding the subject's cognitive bias,
    Behavioral data regarding the behavior of the subject labeled in the cognitive bias data, attribute data of the subject, location data of the subject, health data of the subject, exercise data of the subject, lifestyle data and at least one of the medical data of the subject,
    A machine learning section that uses the neural network for machine learning as training data,
    an input unit for inputting cognitive bias data regarding the subject's cognitive bias;
    The behavior data, the attribute data, the position data, the health data, the exercise data, the lifestyle data, which are estimated from the cognitive bias data input by the input unit, by the estimation model including the neural network; an output unit that outputs at least one of the medical data;
    An estimation system comprising:
15. Behavioral data regarding the behavior of the target person,
    Attribute data of the subject labeled in the behavioral data, location data of the subject, health data of the subject, exercise data of the subject, lifestyle data of the subject, and medical care of the subject. at least one of the data,
    A machine learning section that uses the neural network for machine learning as training data,
    an input unit for inputting behavioral data regarding the behavior of the subject;
    At least one of the attribute data, the position data, the health data, the exercise data, the lifestyle data, and the medical data estimated from the behavior data input by the input unit by the estimation model including the neural network. an output section that outputs one;
    An estimation system comprising:
16. An estimation program executed on a computer in which a processor estimates the behavior of a subject,
    The computer includes:
    Cognitive bias data regarding the subject's cognitive bias,
    Behavioral data regarding the behavior of the subject labeled in the cognitive bias data, attribute data of the subject, location data of the subject, health data of the subject, exercise data of the subject, lifestyle data and at least one of the medical data of the subject,
    an acquisition function to acquire;
    the cognitive bias data;
    At least one of the behavioral data, the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data,
    As training data, a machine learning function for making the neural network perform machine learning,
    An estimation program characterized by realizing the following.
17. An estimation program executed on a computer in which a processor estimates the behavior of a subject,
    The computer includes:
    Behavioral data regarding the behavior of the target person,
    Attribute data of the subject labeled in the behavioral data, location data of the subject, health data of the subject, exercise data of the subject, lifestyle data of the subject, and medical care of the subject. at least one of the data,
    an acquisition function to acquire;
    The behavioral data;
    At least one of the attribute data, the location data, the health data, the exercise data, the lifestyle data, and the medical data,
    As training data, a machine learning function for making the neural network perform machine learning,
    An estimation program characterized by realizing the following.