Learning device, state prediction device, learning method, and program

Abstract

To provide a learning device, a state prediction device, a learning method, and a program related to learning of a model for predicting a state of an object person with high accuracy or state prediction using the model.SOLUTION: The learning device 1X mainly includes a type determination means 16X and a learning means 17X. The type determination means 16X determines a type of each object person related to an action on the basis of a recommended action which is an amount of an action recommended to each object person, a type of the action, or a combination thereof, and an actual action which is an amount of an actual action of each object person, a type of the action, or a combination thereof. The learning means 17X learns a state prediction model for each type related to behavior based on the life log of each target person classified by type. Here, the state prediction model is a model obtained by performing machine learning on a life log and an index value of a state to be predicted of a person whose life log is measured. The prediction result by the state prediction model is used, for example, for decision making.SELECTED DRAWING: Figure 12

Description

[Technical Field]

[0001] The present disclosure relates to the technical fields of a learning device, a state prediction device, a learning method, and a program for predicting a subject's state. [Background technology]

[0002] There are known devices or systems that predict the state of a subject for the purpose of health promotion, etc. For example, Patent Document 1 discloses a system that predicts the proportions and health state of a subject after a predetermined time has passed based on information on the user's lifestyle, diet, etc., using the user's current proportion information and current health state as a reference. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication WO2019 / 116679 Summary of the Invention [Problem to be solved by the invention]

[0004] If the prediction accuracy of the user's state is low, the user may feel that the state is not changing as predicted, which may result in a decrease in motivation to continue using the system.

[0005] In view of the above-mentioned problems, one of the objectives of the present disclosure is to provide a learning device, a state prediction device, a learning method, and a program for learning a model for predicting the state of a subject with high accuracy or for state prediction using the model. [Means for solving the problem]

[0006] One aspect of the learning device is a type determination means for determining the type of each subject with respect to the behavior based on a recommended behavior, which is the amount of behavior recommended to each subject, the type of behavior, or a combination thereof, and an actual behavior, which is the amount of behavior actually performed by each subject, the type of behavior, or a combination thereof; a learning means for learning a state prediction model for each type of behavior based on the life log of each subject classified by the type, The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. It is a learning device.

[0007] One aspect of the state prediction device is a type determination means for determining a type of the subject regarding the behavior based on a recommended behavior, which is the amount of behavior recommended to the subject, the type of behavior, or a combination thereof, and an actual behavior, which is the amount of actual behavior, the type of behavior, or a combination thereof, of the subject; a model selection means for selecting a state prediction model to be used for predicting the state of the subject from state prediction models trained for each type related to the behavior based on the determined type; a state prediction means for predicting a state of the subject based on the selected state prediction model and a life log of the subject; and The condition prediction model is a model obtained by machine learning of a relationship between a life log and an index value of a condition to be predicted for the person whose life log is measured. It is a state prediction device.

[0008] One aspect of the learning method is: The computer Determine a type of each subject regarding the behavior based on a recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and an actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, performed by each subject; learning a state prediction model for each type of behavior based on the life log of each subject classified by the type; The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. It is a learning method. Note that the "computer" includes any electronic device (or a processor included in an electronic device), and may be configured from multiple electronic devices.

[0009] One aspect of the program is Determine a type of each subject regarding the behavior based on a recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and an actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, performed by each subject; causing a computer to execute a process of learning a state prediction model for each type of behavior based on the life log of each of the subjects classified by the type; The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. It is a program. [Effects of the Invention]

[0010] As an example of an effect of the present disclosure, it is possible to train a model for predicting the state of a subject with high accuracy, or to predict the state using the model. [Brief explanation of the drawings]

[0011] [Figure 1] 1 shows a schematic configuration of a state prediction system. [Figure 2] (A) shows the hardware configuration of the learning device. (B) shows the hardware configuration of the state prediction device. [Figure 3] An overview of the state prediction of a target person using a state prediction model is shown below. [Figure 4]1 is an example of a functional block of a learning device related to learning of a state prediction model. [Figure 5] An overview of learning the state prediction model is shown. [Figure 6] The state prediction models to be trained are listed below. [Figure 7] 1 is an example of a functional block of a state prediction device relating to prediction using a state prediction model. [Figure 8] An overview of state prediction using the state prediction model when A=1 is shown below. [Figure 9] 10 is an example of a flowchart illustrating a learning process of a state prediction model. [Figure 10] 10 is an example of a flowchart illustrating a prediction process using a state prediction model. [Figure 11] 1 shows a schematic configuration of a state prediction system. [Figure 12] FIG. 2 is a block diagram of a learning device. [Figure 13] 10 is an example of a flowchart executed by the learning device. [Figure 14] FIG. 2 is a block diagram of a state prediction device. [Figure 15] 10 is an example of a flowchart executed by the state prediction device. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of a learning device, a state prediction device, a learning method, and a program will be described with reference to the drawings.

[0013] First Embodiment (1) System Configuration FIG. 1 shows a schematic configuration of a state prediction system 100. The state prediction system 100 is a system related to health management of a subject, and performs learning (referring to machine learning, the same applies hereinafter) of a state prediction model that predicts the state of the subject from a life log, which is historical information related to at least one of the subject's lifestyle, behavior, and experiences. The state prediction system 100 also performs state prediction using the state prediction model obtained by learning, and recommends the amount of behavior that the subject should perform using a pre-trained recommendation model.

[0014] Hereinafter, a "subject" refers to a person whose life log is collected and whose behavior is managed by an organization, or an individual user. For example, a subject may be a patient whose health care is managed by a doctor, nurse, or other medical professional. Hereinafter, a subject whose life log is measured to be used for training a state prediction model will be referred to as a "learning subject," and a subject whose state is predicted by the state prediction device 2 will be referred to as a "prediction subject." The "state" to be predicted (i.e., the type of state to be predicted) is represented by a predetermined index, and examples of such indexes include weight, blood glucose level, blood pressure, and any other index that serves as a guide for health management. Items related to the state to be predicted are included in the life log. The "amount of activity" recommended to a subject may be, for example, the calories burned by exercise within a unit period, or may be represented by any other index that represents the amount of activity of the subject.

[0015] The state prediction system 100 mainly comprises a learning device 1, a state prediction device 2, a storage device 3, an input device 4, and an output device 5. Here, the learning device 1 and the storage device 3, and the state prediction device 2 and the storage device 3, perform data communication via a communication network or by direct wireless or wired communication. Similarly, the state prediction device 2 and the input device 4, and the state prediction device 2 and the output device 5 perform data communication via a communication network or by direct wireless or wired communication.

[0016] The learning device 1 performs machine learning of a state prediction model based on the life log stored in the life log memory unit 31 of the storage device 3, and stores the parameters of the state prediction model obtained by machine learning in the state prediction model information memory unit 32 of the storage device 3.

[0017] The state prediction device 2 predicts the state of a person to be predicted whose state is managed and determines a recommended amount of action (also referred to as "recommended amount of action"), and presents information about the state prediction result and the determined recommended amount of action to the person to be predicted. In this embodiment, a target deadline is set, which is the deadline for state management of the person to be predicted. The state prediction device 2 then determines the state and recommended amount of action at the target deadline predicted based on the life log. Hereinafter, the recommended amount of action will indicate the amount of action per unit period (e.g., one day or one week). The target deadline corresponds to the time point at which the state is predicted.

[0018] In state prediction, the state prediction device 2 constructs a state prediction model based on parameters stored in the life log storage unit 31 of the storage device 3, and predicts the state of the person to be predicted based on the constructed state prediction model and the person's life log. The state prediction device 2 also constructs a recommendation model based on parameters stored in the recommendation information storage unit 33, and determines recommended behavioral amounts based on the constructed recommendation model and the person's life log. The state prediction device 2 then outputs information related to the predicted state and recommended behavioral amounts of the person to the output device 5. In this case, the state prediction device 2 generates an output signal related to at least one of display and audio, supplies the generated output signal to the output device 5, and displays the information on the output device 5. Thus, the state prediction device 2 presents the person to be predicted with information necessary for condition management, thereby effectively supporting the person to make decisions regarding condition management. The state prediction device 2 may also receive specifications regarding the state to be predicted (which may be any type, such as weight, blood sugar level, or blood pressure) and a target deadline, based on an input signal provided from the input device 4. In this case, at least one of the specified state to be predicted and the target deadline may be used as a parameter of the state prediction model and the recommendation model.

[0019] The input device 4 is an interface that accepts manual input (external input) of information about the person to be predicted. The user who inputs information using the input device 4 may be the person to be predicted himself or herself, or a person who manages or supervises the activities of the person to be predicted. The input device 4 may be, for example, various user input interfaces such as a touch panel, a button, a keyboard, a mouse, or a voice input device. The input device 4 supplies the generated input signal S1 to the state prediction device 2. The output device 5 displays or outputs sound of predetermined information based on the output signal S2 supplied from the state prediction device 2. The output device 5 is, for example, a display, a projector, a speaker, etc.

[0020] The storage device 3 is a memory that stores various information necessary for the processes executed by the learning device 1 and the state prediction device 2. The storage device 3 may be an external storage device such as a hard disk connected to or built into the learning device 1 and the state prediction device 2, or may be a portable storage medium such as a flash memory. The storage device 3 may also be a server device that performs data communication with the learning device 1 and the state prediction device 2. The storage device 3 may also be composed of multiple devices.

[0021] The storage device 3 functionally includes a life log storage unit 31, a state prediction model information storage unit 32, and a recommendation information storage unit 33.

[0022] The life log storage unit 31 stores the life logs of the learning subject and the prediction subject. For example, information on multiple items related to lifestyle, behavior, and experiences is stored in the life log storage unit 31 as the life log of each subject, associated with measurement or input date and time information, the subject's identification information, and the like. Examples of the above items include gender, age, height, menstrual period (female only), weight, body fat percentage, number of steps, exercise item name, time, daily calorie consumption, number of items by meal time period, total calorie intake during meal time periods, meal record registration time, daily sleep time, blood pressure (systolic blood pressure, diastolic blood pressure), body temperature, etc. These items are stored in the life log storage unit 31 as quantified index values. For values ​​of items that change over time, for example, an average value (or other representative value) per unit period (e.g., one week) is calculated, and the calculated average value is used in learning or prediction. The life log includes at least items necessary for determining the recommended behavior amount, items related to the state to be predicted, and items that serve as input for state prediction.

[0023] The life log may be data measured by a sensor that measures each subject, or data input by each subject or their administrator. The above-mentioned sensor may be provided in the state prediction device 2, and the life log of the subject based on the measurement signal output by the sensor may be supplied to the life log storage unit 31 by the state prediction device 2. The sensor may be a wearable device worn by the subject, a camera that captures an image of the subject, a microphone that generates an audio signal of the subject's speech, or a terminal such as a personal computer or smartphone operated by the subject. In this case, the state prediction device 2 may be the above-mentioned terminal such as a personal computer or smartphone. The above-mentioned wearable device, smartphone, etc. may include, for example, a GNSS (Global Navigation Satellite System) receiver, an acceleration sensor, or other sensors that detect biological signals, and the life log may be generated based on the output signals of these sensors.

[0024] The state prediction model information storage unit 32 stores parameters of the state prediction model learned by the learning device 1. The parameters of each state prediction model stored in the state prediction model information storage unit 32 are generated and updated by the learning device 1. The state prediction model is a model that learns the relationship between a life log and a predicted index value of the state to be predicted of the person whose life log is measured. The state prediction model is trained so that, when a life log is input to the state prediction model, a prediction result of the state of the person whose life log is measured (i.e., a predicted index value of the state to be predicted) is output. As will be described later, a plurality of state prediction models according to the type are learned from life logs classified according to the type of the person to be learned. The above types are classified based on whether the person to be learned has behaved in accordance with (i.e., complied with) the recommendation result from the recommendation model. A specific example of a linear model when the state prediction model is a linear model will be described later.

[0025] The state prediction model is not limited to a linear model, and may be a deep learning model based on a neural network or other machine learning model (including a statistical model). When a model based on a neural network such as a neural network is used, the state prediction model information storage unit 32 stores information on various parameters such as the layer structure employed in the model, the neuron structure of each layer, the number and filter size of filters in each layer, and the weight of each element of each filter.

[0026] The recommendation information storage unit 33 stores the parameters of the trained recommendation model. The recommendation model is trained to output a recommended amount of behavior for a person whose life log is measured when an index value of a predetermined item included in the life log is input. A plurality of recommendation models are prepared, for example, according to the number of days (weeks) until a goal deadline. For example, if the condition being managed is weight, the recommendation model outputs a recommended amount of behavior estimated to be necessary for successfully achieving a weight loss goal a predetermined number of weeks (different for each recommendation model) after the start of condition management, which is the start of activities to improve lifestyle habits. The recommendation model may be, for example, a linear model, which is an example of a condition prediction model described below, or a deep learning model based on a neural network or other machine learning model (including a statistical model). Note that the recommended amount of behavior may be determined using a lookup table or the like instead of being calculated using a machine learning model. In this case, for example, the lookup table indicates a correspondence between an index value that can be derived from a life log and a recommended amount of behavior.

[0027] The recommendation information storage unit 33 also stores the recommended behavioral amounts of the person to be predicted output by the recommendation model in association with information on the date and time when the recommendation model was executed. The stored recommended behavioral amounts are used to determine the type of the person to be predicted next time based on whether the person to be predicted has complied with the recommended behavioral amounts.

[0028] The configuration of the state prediction system 100 shown in FIG. 1 is an example, and various modifications may be made to the configuration. For example, at least two of the learning device 1, the state prediction device 2, and the storage device 3 may be implemented by the same device. In another example, the learning device 1 and the state prediction device 2 may each be implemented by multiple devices. In this case, the multiple devices that make up the learning device 1 and the multiple devices that make up the state prediction device 2 exchange information necessary to execute pre-assigned processing between them via direct wired or wireless communication or via communication via a network. In this case, the learning device 1 functions as a learning system, and the state prediction device 2 functions as a state prediction system. In yet another example, the input device 4 and the output device 5 may be integrated into one device. In this case, the input device 4 and the output device 5 may be configured as a tablet terminal that is integrated with or separate from the state prediction device 2.

[0029] (2) Hardware Configuration 2(A) shows the hardware configuration of the learning device 1. The learning device 1 includes, as hardware, a processor 11, a memory 12, and an interface 13. The processor 11, the memory 12, and the interface 13 are connected via a data bus 10.

[0030] The processor 11 functions as a controller (arithmetic unit) that performs overall control of the learning device 1 by executing a program stored in the memory 12. The processor 11 is, for example, a processor such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or a TPU (Tensor Processing Unit). The processor 11 may be composed of multiple processors. The processor 11 is an example of a computer.

[0031] Memory 12 is composed of various types of volatile and non-volatile memory, such as RAM (Random Access Memory), ROM (Read Only Memory), and flash memory. Memory 12 also stores programs for executing the processes performed by learning device 1. Note that some of the information stored in memory 12 may be stored in one or more external storage devices capable of communicating with learning device 1, or may be stored in a storage medium that is detachable from learning device 1.

[0032] Interface 13 is an interface for electrically connecting learning device 1 to other devices. These interfaces may be wireless interfaces such as network adapters for wirelessly transmitting and receiving data to and from other devices, or may be hardware interfaces for connecting to other devices via cables or the like.

[0033] The hardware configuration of the learning device 1 is not limited to the configuration shown in Fig. 2(A). For example, the learning device 1 may further include a display unit such as a display, an input unit such as a keyboard or a mouse, and an audio output unit such as a speaker.

[0034] 2(B) shows an example of the hardware configuration of the state prediction device 2. The state prediction device 2 includes, as hardware, a processor 21, a memory 22, and an interface 23. The processor 21, the memory 22, and the interface 23 are connected via a data bus 20.

[0035] The processor 21 executes a program stored in the memory 22, thereby functioning as a controller (arithmetic unit) that performs overall control of the state prediction device 2. The processor 21 is, for example, a processor such as a CPU, a GPU, a TPU, or a quantum processor. The processor 21 may be composed of multiple processors. The processor 21 is an example of a computer.

[0036] The memory 22 is composed of various types of volatile and non-volatile memory, such as RAM, ROM, and flash memory. The memory 22 also stores programs for executing the processes performed by the state prediction device 2. Some of the information stored in the memory 22 may be stored in an external storage device, such as the storage device 3, that can communicate with the state prediction device 2, or may be stored in a storage medium that is detachable from the state prediction device 2. The memory 22 may also store information stored in the storage device 3 instead.

[0037] The interface 23 is an interface for electrically connecting the state prediction device 2 to other devices. These interfaces may be wireless interfaces such as network adapters for wirelessly transmitting and receiving data to and from other devices, or may be hardware interfaces for connecting to other devices via cables or the like.

[0038] 2B. For example, instead of being connected to the input device 4 and the output device 5 via the interface 23, the state prediction device 2 may have one of these built-in.

[0039] (3) State Prediction Overview An overview of state prediction using the state prediction model will be explained below. Hereinafter, the state prediction model will predict the state based on life logs for at least the most recent "A" weeks (A is an integer equal to or greater than 1). Figure 3 shows an overview of the state prediction of a person to be predicted using a state prediction model. Figure 3 shows an overview of the state prediction from the start of state management of the person to be predicted until the target deadline "B" weeks later (B is an integer of 2 or more). Hereinafter, as an example, state prediction and recommendation of the amount of action will be performed every week from the start of state management until the "B-1" week.

[0040] As shown in FIG. 3 , a life log of the person to be predicted is collected before the start of status management. Based on the collected life log and a status prediction model, the status prediction device 2 predicts the state of the person to be predicted at the target deadline B weeks from now at the time of starting status management. In this case, the status prediction model used at the time of starting status management is a model that outputs a prediction result of the state B weeks from now based on at least the life log for the most recent A weeks before the start of status management. Furthermore, the status prediction device 2 calculates a recommended amount of behavior for the person to be predicted until the target deadline B weeks from now based on the collected life log and the recommendation model. Here, the status prediction device 2 may calculate a recommended amount of behavior for the next week from the time of starting status management that is optimal for the person to be predicted to achieve the goal by the target deadline B weeks from now. The status prediction device 2 presents the state prediction result and the recommended amount of behavior to the person to be predicted or his / her manager.

[0041] In the first week after the start of status management, the status prediction device 2 again predicts the status of the target person at the target deadline "B-1" weeks from now based on the life log for the most recent A weeks and the status prediction model. The status prediction model used in this case is a model that outputs a prediction result of the status "B-1" weeks from now based at least on the life log for the most recent A weeks. As described below, in this case, the status prediction device 2 determines the type of the target person based on whether the target person complied with the recommended amount of behavior calculated at the start of status management, and predicts the status of the target person using a status prediction model corresponding to the determined type. In addition, the status prediction device 2 calculates a recommended amount of behavior for the target person until the target deadline "B-1" weeks from now based on the collected life log and the recommendation model. Here, the status prediction device 2 may calculate a recommended amount of behavior for the next week from the first week after the start of status management that is optimal for the target person to achieve the goal by the target deadline "B-1" weeks from now. The status prediction device 2 presents the predicted status result and the recommended amount of behavior to the target person or their manager.

[0042] The state prediction device 2 then performs the above-described state prediction and action amount recommendation every week. In the "B-1" week from the start of state management, the state prediction device 2 predicts the state of the person to be predicted at the target deadline one week from now based on the life log for the most recent A weeks and the state prediction model. The state prediction model used in this case is a model that outputs a prediction result of the state one week from now based at least on the life log for the most recent A weeks. In this case, the state prediction device 2 determines the type of the person to be predicted based on whether the person to be predicted complied with the recommended action amount output in the "B-2" week, and predicts the state of the person to be predicted using a state prediction model corresponding to the determined type. Furthermore, the state prediction device 2 calculates the recommended action amount for the person to be predicted until the target deadline one week from now based on the collected life log and the recommendation model. The state prediction device 2 presents the state prediction result and the recommended action amount to the person to be predicted or their manager.

[0043] Here, we will provide additional explanation about the issues involved in performing both state prediction and behavioral amount recommendations. State prediction using a state prediction model and behavioral amount recommendations using a recommendation model are performed using different models. Therefore, if state prediction using a state prediction model and behavioral amount recommendations using a recommendation model are performed independently, the state prediction result may still indicate an unimproved state even if the person being predicted behaves as recommended. On the other hand, the state prediction result may also indicate an improved state even if the person does not behave as recommended. In this case, there is a concern that the user's motivation to change their behavior may decrease, leading to the user ceasing use of this system.

[0044] Taking the above into consideration, the state prediction system 100 according to this embodiment learns a state prediction model according to a type classified based on whether or not a recommendation by a recommendation model has been followed, and performs state prediction using the state prediction model. This enables the state prediction system 100 to learn a state prediction model capable of highly accurate state prediction, and to perform state prediction using the state prediction model.

[0045] (4) State prediction model training phase FIG. 4 shows an example of functional blocks of the learning device 1 related to learning a state prediction model. The processor 11 of the learning device 1 functionally includes a life log acquisition unit 15, a subject type determination unit 16, and a type-specific learning unit 17. Note that in FIG. 4, blocks where data is exchanged are connected by solid lines, but the combination of blocks where data is exchanged is not limited to this. The same applies to other functional block diagrams described later. Furthermore, hereinafter, for convenience of explanation, the explanation will be given assuming that state prediction is performed on a weekly basis, but this is not limiting, and state prediction may also be performed on a daily or monthly basis, or on a predetermined number of days basis.

[0046] Here, we will explain the learning of a state prediction model that predicts the state "X" weeks from now (X is an integer that satisfies 1≦X≦B). "X" corresponds to the length of time from the prediction point in time when the state prediction model is used to the target deadline (i.e., the predicted time length). In Figure 3, the state prediction model used at the start of state management corresponds to "X=B", the state prediction model used in the first week corresponds to "X=B-1", and the state prediction model used in the B-1 week corresponds to "X=1".

[0047] The life log acquisition unit 15 acquires, from the life log storage unit 31, the life logs of the learners to be used for learning the state prediction model. In this case, the life log acquisition unit 15 determines, for example, a time point associated with the start of state management (a state management start time in the log) and a time point associated with a target deadline (a target deadline in the log) within a past acquisition period of the life logs of each learner stored in the life log storage unit 31. The life log acquisition unit 15 also determines a time point associated with a prediction time point predicted using the state prediction model (a prediction time point in the log). The life log acquisition unit 15 then acquires life logs necessary for inputting the state prediction model. For example, if the state prediction model is a linear model described below, the life log acquisition unit 15 acquires, for each learner, X weeks' worth of life logs between the prediction time point in the log and the target deadline in the log, and A weeks' worth of life logs immediately preceding the prediction time point in the log. Note that the above-mentioned linear model may be a model that performs state prediction using only A weeks' worth of life logs immediately preceding the prediction time point in the log. In this case, the life log acquisition unit 15 acquires, for each learner, life logs for A weeks immediately preceding the prediction time point on the log. The life log acquisition unit 15 also acquires, from the life log storage unit 31, an index value representing the state to be predicted that is included in the life log at the target deadline on the log, which represents the correct state to be output by the state prediction model. The life log acquisition unit 15 may also extract, in addition to the above-mentioned life logs, life logs necessary for the subject type determination unit 16 to execute the recommendation model from the life log storage unit 31. The life log acquisition unit 15 supplies the acquired life logs to the subject type determination unit 16.

[0048] The subject type determination unit 16 determines the type of the learner whose life log supplied from the life log acquisition unit 15 is measured. Here, the type of the learner is classified based on whether or not the learner has behaved in compliance with the recommended behavioral amount. Hereinafter, "type 1" refers to a learner who has behaved in compliance with the recommended behavioral amount (i.e., a recommended adherent). Furthermore, "type 2" refers to a learner who has not behaved in compliance with the recommended behavioral amount (i.e., a non-recommended adherent). Furthermore, "type 3" refers to a learner whose compliance with the recommended behavioral amount is unknown (i.e., a person with an unknown tendency).

[0049] A specific example of a type determination method in this case will be described. The subject person type determination unit 16 determines the above-mentioned type based on a comparison result between the recommended amount of activity one week before the prediction time point in the log and the actual amount of activity (also referred to as "executed activity amount") during the period calculated from the life log from the time one week before the prediction time point in the log. In this case, the subject person type determination unit 16 inputs the life log acquired one week before the prediction time point in the log into a recommendation model configured with reference to the recommendation information storage unit 33, and obtains the recommended amount of activity output by the recommendation model. Furthermore, the subject person type determination unit 16 calculates the amount of activity from the life log from the time one week before the prediction time point in the log to the prediction time point in the log. For example, if the life log includes information on calories burned, the subject person type determination unit 16 calculates the amount of activity from the life log based on that information. Alternatively, the subject person type determination unit 16 may calculate the amount of activity from the life log based on any other calculation method. Note that, when the above-mentioned recommended amount of behavior has actually been recommended to the learning subject and the recommended amount of behavior has been stored in the recommendation information storage unit 33 or the like, the subject type determination unit 16 may determine the type using the stored recommended amount of behavior without executing the recommendation model. Similarly, when the above-mentioned amount of execution behavior has already been stored in the recommendation information storage unit 33 or the like, the subject type determination unit 16 may determine the type using the stored amount of execution behavior instead of calculating the amount of execution behavior from the life log.

[0050] The subject type determination unit 16 then determines that a subject whose recommended behavior amount and actual behavior amount are within a predetermined difference is a first type, and that a subject whose recommended behavior amount and actual behavior amount are greater than the predetermined difference is a second type. The predetermined difference is set to a default value stored in advance in the storage device 3 or the memory 12, for example. On the other hand, the subject type determination unit 16 determines that a subject whose life log lacks information for calculating the actual behavior amount is a third type. Note that the recommended behavior amount and actual behavior amount both represent the amount of behavior (such as calories) per common unit period (for example, one day).

[0051] Then, the subject type determination unit 16 supplies the determined type of each learning subject and the life log of each learning subject to the type-specific learning unit 17. Note that when "X=B", that is, when learning a state prediction model to be used at the start of state management, a common state prediction model is used regardless of type, and therefore the subject type determination unit 16 supplies the life log acquired by the life log acquisition unit 15 to the type-specific learning unit 17 without determining the type.

[0052] The type-specific learning unit 17 learns a state prediction model according to the type of learner determined by the learner-type determination unit 16. Specifically, the type-specific learning unit 17 uses the life logs of learners determined to be the first type to learn a state prediction model for the first type. The type-specific learning unit 17 also uses the life logs of learners determined to be the second type to learn a state prediction model for the second type, and uses the life logs of learners determined to be the third type to learn a state prediction model for the third type. Here, if the number of learners determined to be the third type is a predetermined number or less, the type-specific learning unit 17 may learn a state prediction model for the third type using all the life logs of learners of the first and second types. Similarly, if the number of learners determined to be the first type is a predetermined number or less, the type-specific learning unit 17 may learn a state prediction model for the first type using all the life logs of the learners. Similarly, when the number of learning subjects determined to be of the second type is equal to or less than a predetermined number, the type-specific learning unit 17 may learn a state prediction model for the second type using all of the life logs of the learning subjects. Note that when "X=B" is satisfied, that is, when learning a state prediction model to be used at the start of state management, the type-specific learning unit 17 learns a common state prediction model regardless of the type of learning subject, based on the life logs acquired by the life log acquisition unit 15. Then, the type-specific learning unit 17 stores the parameters of the state prediction model obtained by learning in the state prediction model information storage unit 32.

[0053] FIG. 5 shows an overview of learning of a state prediction model by the type-specific learning unit 17. As shown in FIG. 5, the life logs extracted from the life log storage unit 31 are divided according to the type of the learning subject. Here, the life logs are divided into first type life logs, second type life logs, and third type life logs. The type-specific learning unit 17 then learns a state prediction model from the life logs for each type. As a result, a state prediction model for the first type, a state prediction model for the second type, and a state prediction model for the third type are learned. Furthermore, as will be described in detail later with reference to FIG. 6, the type-specific learning unit 17 learns a prediction model for each time point X=1 to B.

[0054] Here, a supplementary explanation will be given about learning when the state prediction model is a linear model.

[0055] When the state prediction model is a linear model, the predicted state (weight, etc.) "y" is expressed as follows using the variable vector "x" of the life log for "A+X" weeks: y=cx+d Here, "c" is a vector with the same number of dimensions as x, and "d" is a coefficient. Each element of vector c and coefficient d correspond to a parameter to be determined by learning.

[0056] When there are "f" indices corresponding to the items used for prediction among the items in a one-week life log (i.e., indices used for state prediction), the vector x has a length of "(f+1)×A+f×X)." The "indices used for state prediction" may be any indices other than the indices representing the state to be predicted. For example, when predicting weight, indices other than weight (calories consumed, sleep time, blood pressure, body temperature, etc.) are used as the "indices used for state prediction." Then, for each learning subject, the type-specific learning unit 17 sets "x" based on the life log for the A weeks immediately preceding the prediction time point in the log and the X weeks from the prediction time point in the log, and sets "y" based on the life log for one week between X-1 weeks and X weeks after the prediction time point in the log. For example, if "y" is weight, the type-specific learning unit 17 may set "y" to the average value of the weight values ​​for one week between X-1 weeks and X weeks after the prediction time point in the log. Then, using pairs of life logs for "A+X" weeks and life logs X weeks later, the number of parameters to be calculated is at least equal to or greater than the number of parameters to be calculated (i.e., the total number of elements of c and coefficient d, i.e., (f+1)×A+f×X+1), and type-specific learning unit 17 calculates vector c and coefficient d using an arbitrary approximate solution. Then, type-specific learning unit 17 stores the parameters of the state prediction model calculated for each type in state prediction model information storage unit 32.

[0057] Fig. 6 shows a list of state prediction models learned by the type-specific learning unit 17. As shown in Fig. 6, the type-specific learning unit 17 learns one state prediction model to be used at the start of state management (i.e., one state prediction model where "X = B"), and learns state prediction models to be used in the first week after the start of state management (i.e., one state prediction model where "X = B-1") for each of the first to third types. Similarly, the type-specific learning unit 17 learns state prediction models to be used in the second week to the "B-1" week after the start of state management, for each of the first to third types.

[0058] In this way, the life log acquisition unit 15, the subject type determination unit 16, and the type-specific learning unit 17 change X between 1 week and B weeks, and sequentially perform processing to learn the state prediction models to be used from the start of state management through the "B-1" week. As a result, "3 × B-2" state prediction models are learned. Then, the type-specific learning unit 17 stores the parameters of these learned state prediction models in the state prediction model information storage unit 32.

[0059] Note that when the input is "X=B," predicting B weeks into the future using "B+1" weeks' worth of input is equivalent to predicting what the indicator value of the state to be predicted will be for the week from B-1 week to week B counting from the start of state management, assuming that the type of lifestyle one week before the start of state management is B+1 weeks into the future. The life log for period A contains indicator values ​​for the state to be predicted X weeks into the future, but the life log for period X does not contain indicator values ​​for the state to be predicted. The indicator value for the state to be predicted X weeks into the future is the target of prediction.

[0060] Here, each component of the life log acquisition unit 15, the subject type determination unit 16, and the type-specific learning unit 17 can be realized, for example, by the processor 11 executing a program. Alternatively, each component may be realized by recording the necessary program in an arbitrary non-volatile storage medium and installing it as needed. Note that at least a portion of each component may not be realized by software programs, but may be realized by any combination of hardware, firmware, and software. Also, at least a portion of each component may be realized using a user-programmable integrated circuit, such as an FPGA (Field-Programmable Gate Array) or a microcontroller. Also, at least a portion of each component may be configured by an ASSP (Application Specific Standard Produce), an ASIC (Application Specific Integrated Circuit), or a quantum processor (quantum computer control chip). In this way, each component may be realized by various hardware. The same applies to other embodiments described below. Furthermore, each component may be realized by the cooperation of multiple computers, for example, using cloud computing technology.

[0061] (5) Prediction phase using state prediction models 7 is an example of a functional block of the state prediction device 2 related to prediction using a state prediction model. The processor 21 of the state prediction device 2 functionally includes a life log acquisition unit 25, a subject type determination unit 26, a state prediction unit 27, a recommendation unit 28, and a UI control unit 29. Here, a case where a state is predicted X weeks from now will be described. In FIG. 3, the state prediction model used at the start of state management corresponds to "X=B", the state prediction model used in the first week corresponds to "X=B-1", and the state prediction model used in the B-1 week corresponds to "X=1".

[0062] The life log acquisition unit 25 acquires the life log of the prediction target person from the life log storage unit 31. In this case, the life log acquisition unit 15 acquires, for example, the life logs for the most recent A weeks required for input to the state prediction model from the life log storage unit 31. Then, the life log acquisition unit 15 supplies the life logs for the most recent A weeks required for input to the state prediction model to the target person type determination unit 16. In addition, the life log acquisition unit 25 supplies the life logs required for the target person type determination unit 16 to execute the recommendation model to the recommendation unit 28.

[0063] The subject type determination unit 26 determines the type of the predicted subject whose life log is measured and supplied from the life log acquisition unit 25. In this case, the subject type determination unit 26 refers to the recommendation information storage unit 33 and obtains the recommended behavior amounts recommended to the predicted subject in the previous week. The subject type determination unit 26 also calculates the amount of execution activity based on the life log from the previous week to the present, and determines the type of the predicted subject based on the comparison result between the calculated amount of execution activity and the recommended amount of execution activity. The subject type determination unit 26 determines a predicted subject whose recommended behavior amount and the execution activity amount are within a predetermined difference as a first type, and a predicted subject whose recommended behavior amount and the execution activity amount are greater than the predetermined difference as a second type. On the other hand, the subject type determination unit 26 determines a predicted subject whose life log lacks information for calculating the execution activity amount as a third type. The subject type determination unit 26 supplies the determined type of the predicted subject and the life log to the state prediction unit 27.

[0064] In addition, when "X=B", that is, when the prediction time is the time when status management begins, a common status prediction model is used regardless of type, so the subject type determination unit 26 supplies the life log to the status prediction unit 27 without determining the type.

[0065] The state prediction unit 27 predicts the state of the person to be predicted X weeks from now. In this case, the state prediction unit 27 extracts parameters of a state prediction model corresponding to the type determined by the person type determination unit 26 from the state prediction model information storage unit 32. Then, the state prediction unit 27 acquires a prediction result of the state of the person to be predicted X weeks from the state prediction model based on the state prediction model configured using the extracted parameters and the life log. This makes it possible to obtain an accurate prediction result of the state of the person to be predicted X weeks from now based on the state prediction model learned specifically for the type of the person to be predicted. Then, the state prediction unit 27 supplies the prediction result of the state of the person to be predicted X weeks from now to the UI control unit 29. Note that when "X=B," i.e., when the prediction time is the time when state management starts, the state prediction unit 27 predicts the state of the person to be predicted based on the state prediction model that is not dependent on the type determination result and the life log.

[0066] The recommendation unit 28 determines a recommended amount of action to be newly recommended to the person to be predicted. In this case, the recommendation unit 28 extracts parameters of a recommendation model with a target deadline of X weeks from the recommendation information storage unit 33, and acquires a recommended amount of action from the recommendation model based on the recommendation model configured using the extracted parameters and the life log. Then, the recommendation unit 28 supplies the acquired recommended amount of action to the UI control unit 29. The recommendation unit 28 stores the acquired recommended amount of action in the recommendation information storage unit 33 in association with identification information and date and time information of the person to be predicted.

[0067] The UI control unit 29 controls the user interface. For example, the UI control unit 29 displays and / or outputs by voice the prediction result of the state after X weeks generated by the state prediction unit 27 and the recommended amount of action determined by the recommendation unit 28 on the output device 5. The UI control unit 29 may also receive information specifying the setting of model parameters and a target deadline from the input device 4.

[0068] Here, a supplementary explanation will be given of the state prediction using the state prediction model by the state predictor 27 when the state prediction model is a linear model.

[0069] FIG. 8 shows an overview of state prediction using the state prediction model when "A=1." When "A=1," the state prediction unit 27 uses the life logs of the person to be predicted that were actually collected in the week immediately preceding the prediction time as input to the state prediction model. When the state prediction model is the linear model described above, as described above, in addition to the life logs for the most recent A (here, 1) weeks, life logs for X weeks from the prediction time to the target deadline are required. In this case, the state prediction unit 27 generates pseudo life logs (pseudo life logs) for X weeks from the prediction time to the target deadline based on the life logs (collected life logs) for A weeks (here, 1 week) collected before the prediction time. In this case, for example, the state prediction unit 27 generates pseudo life logs that are copies of the collected life logs. That is, the state prediction unit 27 generates the pseudo life logs by assuming that the person to be predicted will continue living the same lifestyle as in the week immediately preceding the prediction time for X weeks until the target deadline. If A is 2 or more, the state prediction unit 27 may perform statistical processing such as averaging on the life logs for the most recent A weeks to generate a pseudo life log.

[0070] Then, the state prediction unit 27 predicts the state of the person to be predicted X weeks from now using the life logs for "X+1" weeks, which are a combination of the collected life logs and the pseudo life logs, and the state prediction model. Specifically, the state prediction unit 27 generates a vector x having a length of "(f+1)×1+f×X" from the life logs for "X+1" weeks, and calculates y, which represents the state of the person to be predicted X weeks from now, based on "y=cx+d" using the vector c and coefficient d obtained by learning. Note that the state prediction model may be a model that predicts the state using only the collected life logs (i.e., the life logs for A weeks immediately before the prediction point in the log), instead of predicting the state from the life logs for "X+A" weeks, which are a combination of the collected life logs and the pseudo life logs.

[0071] Each of the components, including the life log acquisition unit 25, the subject type determination unit 26, the state prediction unit 27, the recommendation unit 28, and the UI control unit 29, can be realized, for example, by the processor 21 executing a program. Alternatively, each component may be realized by recording the necessary program in an arbitrary non-volatile storage medium and installing it as needed. Note that at least a portion of each of these components may not necessarily be realized by software programs, but may be realized by any combination of hardware, firmware, and software. Also, at least a portion of each of these components may be realized using a user-programmable integrated circuit, such as an FPGA or a microcontroller. Also, at least a portion of each of these components may be configured by an ASSP, an ASIC, or a quantum processor (quantum computer control chip). In this way, each component may be realized by various hardware. The same applies to other embodiments described below. Furthermore, each of these components may be realized by the cooperation of multiple computers, for example, using cloud computing technology.

[0072] (6) Processing Flow FIG. 9 is an example of a flowchart illustrating the learning process of the state prediction model executed by the learning device 1.

[0073] First, the learning device 1 extracts the life log of each learner from the life log storage unit 31 (step S11). The extracted life log includes, for example, items necessary for determining recommended behavior amounts, items related to the state to be predicted, and items used as input for state prediction. The learning device 1 then determines the type of each learner (step S12). In this case, as described above, the learning device 1 acquires the recommended behavior amount at the time immediately preceding the prediction time point in the log and the actual behavior amount from the immediately preceding time point to the prediction time point, and determines the type of each learner based on the difference between the recommended behavior amount and the actual behavior amount. The learning device 1 then learns a state prediction model according to the type (step S13). In this case, the learning device 1 divides the life log into groups according to the type of each learner determined in step S12, and uses the life log of each group to learn a state prediction model of the corresponding type. The learning device 1 then stores the parameters of the state prediction model obtained by learning in the state prediction model information storage unit 32.

[0074] Next, the learning device 1 determines whether learning of the state prediction model has been completed (step S14). For example, when it is determined that learning of all state prediction models that may be used in each week from week 0 to week "B-1" shown in FIG. 6 (i.e., each state prediction model where "X=1, 2, ..., B-1, B") has been completed, the learning device 1 determines that learning has been completed. When it is determined that learning has been completed (step S14; Yes), the learning device 1 ends the processing of the flowchart. On the other hand, when it is determined that learning has not been completed (step S14; No), the learning device 1 returns the processing to step S11. For example, the learning device 1 sequentially executes learning of the state prediction model for each X of "X=1, 2, ..., B-1, B" in steps S11 to S13, and learns all state prediction models by executing steps S11 to S13 B times. When learning a state prediction model for the case where "X=B", the learning device 1 does not execute step S12, but instead executes learning of the state prediction model using the life log acquired in step S11 in step S13.

[0075] Fig. 10 is an example of a flowchart showing a prediction process using a state prediction model executed by the state prediction device 2. The state prediction device 2 executes the process of the flowchart shown in Fig. 10, for example, when it is time to predict the state of the person to be predicted and recommend the amount of activity. The timing to predict the state of the person to be predicted and recommend the amount of activity may be specified by a user input via the input device 4, or may be at a preset date and time.

[0076] First, the state prediction device 2 extracts the life log of the person to be predicted from the life log storage unit 31 (step S21). Next, the state prediction device 2 determines whether or not an action amount has been recommended to the person to be predicted (step S22). If an action amount has been recommended to the person to be predicted (step S22; Yes), the state prediction device 2 determines the type of the person to be predicted and selects a state prediction model corresponding to the determined type (step S23). In this case, for example, as described above, the state prediction device 2 acquires the recommended action amount recommended immediately before the prediction time and the executed action amount, which is the actual action amount from the recommendation to the prediction time, and determines the type of the person to be predicted based on the difference between the recommended action amount and the executed action amount. Then, the state prediction device 2 extracts parameters of the state prediction model corresponding to the determined type from the state prediction model information storage unit 32 and configures a state prediction model corresponding to the type of the person to be predicted.

[0077] On the other hand, if there is no record of recommending an activity amount to the person to be predicted (step S22; No), the state prediction device 2 proceeds to step S24. In this case, the state prediction device 2 extracts parameters of a state prediction model corresponding to "X=B" from the state prediction model information storage unit 32, and configures a state prediction model that is independent of the type of person to be predicted. In practice, the state prediction device 2 may assume multiple "Xs," prepare third-type state prediction models according to X, and use the third-type state prediction models depending on X.

[0078] Next, the state prediction device 2 predicts the state of the person to be predicted based on the life log extracted in step S21 and the state prediction model (step S24). Also, in step S23, the state prediction device 2 determines recommended behavior amounts based on the life log and the recommendation model. Then, the state prediction device 2 outputs the prediction result of the state of the person to be predicted and the recommended behavior amounts via the output device 5 (step S25).

[0079] (7) Variations Next, preferred modifications of the above-described embodiment will be described. The following modifications may be applied in combination. (Variation 1) The types of subjects are not limited to being divided into three types.

[0080] For example, the learning device 1 may determine the type based on the result of comparing the recommended behavior amounts and actual behavior amounts for the most recent period, as well as the result of comparing the recommended behavior amounts and actual behavior amounts for a period prior to that period.

[0081] For example, if X is "B-2" or less, the learning device 1 may determine the type by further considering the results of comparing the recommended amount of behavior with the actual amount of behavior over the past week, as well as the results of comparing the recommended amount of behavior with the actual amount of behavior over the past two weeks to one week. In this case, there are 9 (=3×3) possible combinations of the comparison results. Similarly, if X is "B-3" or less, the type may be determined by further considering the results of comparing the recommended amount of behavior with the actual amount of behavior over the past three weeks to two weeks. In these cases, the learning device 1 learns a state prediction model for each type based on the life log divided by type, and stores the parameters of the learned state prediction model for each type in the state prediction model information storage unit 32. The state prediction device 2 also determines the type of the person to be predicted using the same method as the learning device 1, and selects a state prediction model to use for state prediction based on the determination results.

[0082] Furthermore, the learning device 1 and the state prediction device 2 may classify the types of subjects in more detail based on attributes such as the gender and age of the subjects.

[0083] For example, the learning device 1 determines the type based on the tendency of whether or not to adhere to the recommended behavioral amounts and the gender of the subject. In this case, Type 1 to Type 3 are provided for men and Type 1 to Type 3 are provided for women, resulting in a total of six types. The learning device 1 learns a state prediction model for each type based on the life logs divided into types, and stores parameters of the learned state prediction model for each type in the state prediction model information storage unit 32. Similarly to the learning device 1, the state prediction device 2 determines the type of the person to be predicted based on the tendency of whether or not to adhere to the recommended behavioral amounts and the gender of the subject, and selects a state prediction model to be used for state prediction based on the determination result. This allows the state prediction device 2 to classify the person to be predicted in more detail and perform state prediction with high accuracy using a state prediction model suited to the person to be predicted.

[0084] (Variation 2) In the above description, the amount of activity is recommended, but the target of the recommendation is not limited to the amount of activity, and may be a type of activity, or a combination of an amount of activity and a type of activity. In this case, the recommendation model is a model that is trained to output at least one of the amount of activity or type of activity recommended for the person whose life log is measured when an index value of a predetermined item included in the life log is input. The recommended amount of activity, type of activity, or a combination of these is also called a "recommended action."

[0085] Furthermore, even when determining a type of a subject based on whether the subject complies with a recommended behavior, the state prediction device 2 identifies the actual amount of behavior, the type of behavior, or a combination thereof, instead of calculating the amount of execution behavior. Hereinafter, the subject's actual amount of behavior, the type of behavior, or a combination thereof will also be referred to as "executed behavior." The state prediction device 2 then determines the type of the subject based on the results of comparing the subject's execution behavior with the recommended behavior. Here, when the type of behavior is used, it is preferable that the type of execution behavior and the type of recommended behavior at least match as a necessary or sufficient condition for a person to comply with the recommendation.

[0086] In this modified example, the state prediction device 2 also accurately determines the type of the subject based on whether or not the subject is a recommendation adherent, and can predict the state of the subject with high accuracy using a state prediction model corresponding to the determined type.

[0087] (8) Application Examples Based on the state prediction result of the person to be predicted, the state prediction device 2 may recommend a predetermined action to the person to be predicted or may automatically perform a predetermined action such as ordering online on behalf of the person to be predicted.

[0088] In this case, the state prediction device 2 may determine health products (including diet foods, low-calorie foods, and supplements) to be recommended to the person to be predicted based on the state prediction result, and may present the determined health products to the person to be predicted or place an online order using e-commerce. Instead of health products, the state prediction device 2 may also present recommended facilities (including nearby sports gyms) to the person to be predicted or recommended stretches and exercises to the person to be predicted. Furthermore, the state prediction device 2 may add a process to automatically make a tentative reservation for a weekend during a time slot when the facility is open, based on information on availability of recommended facilities (nearby gyms or facilities for specific sports) and information on recommended sports. The state prediction device 2 may also add a process to cancel the tentative reservation if the person to be predicted does not officially complete the reservation procedure within a specified time. Instead of outputting this recommended information using the output device 5, the state prediction device 2 may link with systems of facilities that sell health products or transportation systems, and display recommended products on displays such as digital signage. In this case, the state prediction device 2 may identify the position of the walking target person by performing personal authentication of the pedestrian using an image generated by a camera installed in these systems, and display information on a display near the identified position of the target person. Furthermore, the state prediction device 2 may provide an incentive such as points that can be used by the target person when the target person performs the recommended behavior.

[0089] Second Embodiment 11 shows a schematic configuration of a state prediction system 100A. The state prediction system 100A according to the second embodiment is a server-client model system, in which a state prediction device 2A functioning as a server device performs the processing of the learning device 1 and the state prediction device 2 according to the first embodiment. Hereinafter, the same components as those in the first embodiment will be appropriately designated by the same reference numerals, and their description will be omitted.

[0090] 11, the state prediction system 100A mainly includes a state prediction device 2A that functions as a server, a storage device 3 that stores the same data as in the first embodiment, and a terminal device 8 that functions as a client. The state prediction device 2A and the terminal device 8 communicate data via a network 7.

[0091] The terminal device 8 is a terminal having an input function, a display function, and a communication function, and functions as the input device 4 and the output device 5 shown in Fig. 1. The terminal device 8 may be, for example, a personal computer, a tablet terminal, a PDA (Personal Digital Assistant), etc. The terminal device 8 is provided with a sensor for measuring the user's life log, and transmits a measurement signal output by the sensor or an input signal based on a user input to the state prediction device 2A.

[0092] The state prediction device 2A has the hardware configuration shown in FIG. 2(A) and the functional block configurations shown in FIGS. 4 and 7. After executing a learning process for the state prediction model, the state prediction device 2A predicts a state using the state prediction model and recommends an action amount using a recommendation model. In this case, the state prediction device 2A (specifically, the UI control unit 29 in FIG. 7) transmits an output signal related to the state prediction result and the recommended action amount to the terminal device 8 via the network 7, based on a request from the terminal device 8. In this case, the terminal device 8 functions as the output device 5 in the first embodiment.

[0093] As described above, the state prediction system 100A according to the second embodiment can perform learning of a state prediction model, state prediction using the learned state prediction model, and the like, and can suitably present the results of the state prediction to the user of the terminal device 8. Note that in the second embodiment, a device other than the state prediction device 2A may execute the learning process of the recommendation model.

[0094] <Third embodiment> 12 is a block diagram of a learning device 1X. The learning device 1X mainly includes a type determination unit 16X and a learning unit 17X. Note that the learning device 1X may be configured by a plurality of devices.

[0095] The type determination means 16X determines the behavioral type of each subject based on the recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and the actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, of each subject. The type determination means 16X can be, for example, the subject type determination unit 16 of the learning device 1 in the first embodiment or the subject type determination unit 16 of the state prediction device 2A in the second embodiment.

[0096] The learning means 17X learns a state prediction model for each behavioral type based on the life log of each subject classified by type. Here, the state prediction model is a model obtained by machine learning of the relationship between the life log and the predicted index value of the state to be predicted of the person whose life log is measured. The learning means 17X can be, for example, the type-specific learning unit 17 of the learning device 1 in the first embodiment or the type-specific learning unit 17 of the state prediction device 2A in the second embodiment.

[0097] 13 is an example of a flowchart executed by the learning device 1X. The type determination means 16X determines the behavioral type of each subject based on the amount of behavior recommended for each subject, the type of behavior, or a combination thereof, which is a recommended behavior, and the amount of actual behavior, the type of behavior, or a combination thereof, which is an actual behavior of each subject (step S31). Then, the learning means 17X learns a state prediction model for each behavioral type based on the life log of each subject classified by type (step S32). Here, the state prediction model is a model obtained by machine learning the relationship between the life log and a predicted index value of the state to be predicted of the person whose life log is measured.

[0098] According to the third embodiment, the learning device 1X can learn a state prediction model for accurately predicting the state of a subject from a life log.

[0099] <Fourth embodiment> 14 is a block diagram of a state prediction device 2X. The state prediction device 2X mainly includes a type determination means 26X, a model selection means 27Xa, and a state prediction means 27Xb. Note that the state prediction device 2Y may be configured by a plurality of devices.

[0100] The type determination means 26X determines the type of the subject regarding behavior based on recommended behavior, which is the amount of behavior recommended to the subject, the type of behavior, or a combination thereof, and actual behavior, which is the amount of actual behavior, the type of behavior, or a combination thereof, of the subject. The type determination means 26X can be, for example, the subject type determination unit 26 of the state prediction device 2 in the first embodiment or the subject type determination unit 26 of the state prediction device 2A in the second embodiment.

[0101] The model selection means 27Xa selects a state prediction model to be used for predicting the state of the subject from state prediction models learned for each behavioral type based on the determined type. Here, the state prediction model is a model obtained by machine learning of the relationship between the life log and the predicted index value of the state to be predicted of the person whose life log is measured.

[0102] The state prediction means 27Xb predicts the state of the subject based on the selected state prediction model and the subject's life log. The model selection means 27Xa and the state prediction means 27Xb can be, for example, the state prediction unit 27 of the state prediction device 2 in the first embodiment or the state prediction unit 27 of the state prediction device 2A in the second embodiment.

[0103] FIG. 15 is an example of a flowchart executed by the state prediction device 2X. The type determination means 26X determines the subject's behavioral type based on the amount of behavior recommended for the subject, the type of behavior, or a combination thereof (recommended behaviors) and the subject's actual amount of behavior, the type of behavior, or a combination thereof (actual behaviors) (step S41). Then, the model selection means 27Xa selects a state prediction model to be used for predicting the subject's state from state prediction models learned for each behavioral type based on the determined type (step S42). Here, the state prediction model is a model obtained by machine learning the relationship between the life log and the predicted index value of the state to be predicted for the person whose life log is measured. The state prediction means 27Xb predicts the subject's state based on the selected state prediction model and the subject's life log (step S43).

[0104] The state prediction device 2X according to the fourth embodiment can appropriately select a state prediction model according to the type based on whether or not the subject has taken the recommended action, thereby accurately predicting the subject's state.

[0105] In the above-described embodiments, the program can be stored using various types of non-transitory computer-readable media and supplied to a computer processor or the like. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic storage media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical storage media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program may also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can supply the program to a computer via a wired communication path such as an electric wire or optical fiber, or via a wireless communication path.

[0106] In addition, part or all of the above-described embodiments (including variations, the same applies below) may also be described as, but are not limited to, the following supplementary notes. Furthermore, not only the devices, methods, and storage media described in the supplementary notes, but also various hardware, software, various recording means (including storage media) for recording software, or systems may be made to depend on part or all of the configurations described in the supplementary notes, as long as they do not deviate from the above-described embodiments.

[0107] [Appendix 1] a type determination means for determining the type of each subject with respect to the behavior based on a recommended behavior, which is the amount of behavior recommended to each subject, the type of behavior, or a combination thereof, and an actual behavior, which is the amount of behavior actually performed by each subject, the type of behavior, or a combination thereof; and a learning means for learning a state prediction model for each type related to the amount of activity based on the life log of each of the subjects classified by the type, The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. Learning device. [Appendix 2] The learning device described in Appendix 1, wherein the type determination means determines whether each subject has complied with the recommended behavior based on the recommended behavior and the actual behavior, and determines the type based on the determination result of whether each subject has complied with the recommended behavior. [Appendix 3] The learning device described in Appendix 2, wherein the type determination means further determines at least one of the gender or age of each subject, and determines the type based on the determination result of whether each subject has complied with the recommended behavior and the determination result of at least one of the gender or age. [Appendix 4] The learning device according to claim 1, wherein the learning means learns a plurality of state prediction models having different lengths from a prediction time point at which the state is predicted using the state prediction model to a predicted time point at which the state is predicted. [Appendix 5] a type determination means for determining a type of the subject regarding the behavior based on a recommended behavior, which is the amount of behavior recommended to the subject, the type of behavior, or a combination thereof, and an actual behavior, which is the amount of actual behavior, the type of behavior, or a combination thereof, of the subject; a model selection means for selecting a state prediction model to be used for predicting the state of the subject from state prediction models trained for each type related to the behavior based on the determined type; a state prediction means for predicting a state of the subject based on the selected state prediction model and a life log of the subject; and The condition prediction model is a model obtained by machine learning of a relationship between a life log and a predicted index value of a condition to be predicted of a person whose life log is measured. A state prediction device having the following. [Appendix 6] 6. The state prediction device according to claim 5, further comprising a recommendation unit that calculates the recommended action to be newly recommended to the subject based on a life log of the subject. [Appendix 7] 7. The state prediction device according to claim 6, further comprising an output control unit configured to output the state prediction result and the recommended action calculated by the recommendation unit through an output device. [Appendix 8] The method further includes a life log acquisition means for acquiring the life log of the subject for a predetermined period immediately before the prediction time point, The state prediction device described in Appendix 5, wherein the state prediction means generates a life log from the prediction time point to the predicted time point at which the state is predicted based on a life log for the specified period, and predicts the state of the subject based on the life log for the specified period, the life log from the prediction time point to the predicted time point, and the state prediction model. [Appendix 9] The computer Determine a type of each subject regarding the behavior based on a recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and an actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, performed by each subject; learning a state prediction model for each type of behavior based on the life log of each subject classified by the type; The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. How to learn. [Appendix 10] Determine a type of each subject regarding the behavior based on a recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and an actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, performed by each subject; causing a computer to execute a process of learning a state prediction model for each type of behavior based on the life log of each of the subjects classified by the type; The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. program. [Appendix 11] The computer Determine the type of the subject with respect to the behavior based on the recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended to the subject, and the actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, of the subject; selecting a state prediction model to be used for predicting a state of the subject from state prediction models trained for each type related to the behavior based on the determined type; predicting a state of the subject based on the selected state prediction model and the life log of the subject; The condition prediction model is a model obtained by machine learning of a relationship between a life log and a predicted index value of a condition to be predicted of a person whose life log is measured. State prediction methods. [Appendix 12] Determine the type of the subject with respect to the behavior based on the recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended to the subject, and the actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, of the subject; selecting a state prediction model to be used for predicting a state of the subject from state prediction models trained for each type related to the behavior based on the determined type; causing a computer to execute a process of predicting a state of the subject based on the selected state prediction model and a life log of the subject; The condition prediction model is a model obtained by machine learning of a relationship between a life log and a predicted index value of a condition to be predicted of a person whose life log is measured. program. [Appendix 13] A storage medium storing the program described in Appendix 10 or 12.

[0108] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above embodiments. Various modifications within the scope of the present invention that would be understood by those skilled in the art can be made to the configuration and details of the present invention. In other words, the present invention naturally includes various modifications and alterations that would be possible for those skilled in the art based on the entire disclosure, including the claims, and the technical ideas. Furthermore, the disclosures of the above-cited patent documents and other documents are incorporated herein by reference. [Explanation of symbols]

[0109] 1. 1X learning device 2, 2A, 2X condition predictor 3 Storage device 4 Input Devices 5 Output Devices 8 Terminal Equipment 100, 100A Condition Prediction System

Claims

1. a type determination means for determining the type of each subject with respect to the behavior based on a recommended behavior, which is the amount of behavior recommended to each subject, the type of behavior, or a combination thereof, and an actual behavior, which is the amount of behavior actually performed by each subject, the type of behavior, or a combination thereof; a learning means for learning a state prediction model for each type of behavior based on the life log of each subject classified by the type, The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. Learning device.

2. The learning device according to claim 1, wherein the type determination means determines whether each subject has complied with the recommended behavior based on the recommended behavior and the actual behavior, and determines the type based on the determination result of whether each subject has complied with the recommended behavior.

3. The learning device described in claim 2, wherein the type determination means further determines at least one of the gender or age of each subject, and determines the type based on the determination result of whether each subject has complied with the recommended behavior and the determination result of at least one of the gender or age.

4. 2. The learning device according to claim 1, wherein the learning means learns a plurality of state prediction models having different lengths from a prediction time point at which the state is predicted using the state prediction model to a predicted time point at which the state is predicted.

5. a type determination means for determining a type of the subject regarding the behavior based on a recommended behavior, which is the amount of behavior recommended to the subject, the type of behavior, or a combination thereof, and an actual behavior, which is the amount of actual behavior, the type of behavior, or a combination thereof, of the subject; a model selection means for selecting a state prediction model to be used for predicting the state of the subject from state prediction models trained for each type related to the behavior based on the determined type; a state prediction means for predicting a state of the subject based on the selected state prediction model and a life log of the subject; and The condition prediction model is a model obtained by machine learning of a relationship between a life log and a predicted index value of a condition to be predicted of a person whose life log is measured. A state prediction device having the following.

6. The state prediction device according to claim 5 , further comprising a recommendation unit that calculates the recommended action to be newly recommended to the subject based on a life log of the subject.

7. The state prediction device according to claim 6 , further comprising output control means for outputting the state prediction result and the recommended action calculated by the recommendation means by an output device.

8. The method further includes a life log acquisition means for acquiring the life log of the subject for a predetermined period immediately before the prediction time point, The state prediction device of claim 5, wherein the state prediction means generates a life log from the prediction time point to the predicted time point at which the state is predicted based on a life log for the specified period, and predicts the subject's state based on the life log for the specified period, the life log from the prediction time point to the predicted time point, and the state prediction model.

9. The computer Determine the type of each subject in terms of the amount of behavior based on the recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and the actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, of each subject; learning a state prediction model for each type of behavior based on the life log of each subject classified by the type; The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. How to learn.

10. Determine a type of each subject regarding the behavior based on a recommended behavior, which is the amount of behavior, the type of behavior, or a combination thereof, recommended for each subject, and an actual behavior, which is the amount of behavior, the type of behavior, or a combination thereof, performed by each subject; causing a computer to execute a process of learning a state prediction model for each type of behavior based on the life log of each of the subjects classified by the type; The condition prediction model is a model obtained by machine learning of the relationship between a life log and a predicted index value of a condition to be predicted for the person whose life log is measured. program.

Images