Information processing device, information processing method, and program

Abstract

This information processing device comprises: an acquisition unit that acquires health-related data that is data related to the health of a target user; a health condition prediction unit that inputs each feature amount extracted from the health-related data to a health condition prediction model, which is obtained by machine-learning a relationship between the health-related data and health conditions, to predict the health condition of the target user; an information generation unit that generates explanation information including at least information for prompting the target user to change his or her behavior on the basis of a degree of contribution to the prediction result by each feature amount and a degree of influence on change in behavior of the target user by each feature amount; and an information presentation unit that presents the explanation information to the target user.

Description

Information processing device, information processing method, and recording medium

[0001] The present invention relates to an information processing device, an information processing method, and a recording medium.

[0002] Patent Literature 1 discloses health prediction using machine learning, and Non-Patent Literature 1 discloses LIME (Local Interpretable Model-agnostic Explanations), a technology that presents the basis for prediction results for black-box models.

[0003] Patent No. 6901146

[0004] Marco Tulio Ribeiro, Sameer Singh, Carlos Guestrin, "Why Should I Trust You?": Explaining the Predictions of Any Classifier," Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. [Retrieved August 8, 2023], Internet <URL: https: / / www.kdd.org / kdd2016 / papers / files / rfp0573-ribeiroA.pdf>

[0005] Patent Document 1 and Non-Patent Document 1 simply present health predictions and information that serves as the basis for the predictions, and may not necessarily generate information that will lead to behavioral changes for the target user.

[0006] One aspect of the present invention has been made in consideration of the above-mentioned problems, and one example of a purpose thereof is to generate information that encourages a target user to change their behavior.

[0007] One aspect of an information processing device includes an acquisition unit that acquires health-related data, which is data regarding the health of a target user; a health condition prediction unit that inputs each feature extracted from the health-related data into a health condition prediction model that has undergone machine learning to determine the relationship between the health-related data and the health condition, thereby predicting the health condition of the target user; an information generation unit that generates explanatory information including at least information that encourages the target user to change their behavior based on the contribution of each feature to the prediction result and the influence of each feature on the behavioral change of the target user; and an information presentation unit that presents the explanatory information to the target user.

[0008] In one aspect of the information processing method, at least one processor acquires health-related data that is data regarding the health of a target user, inputs each feature extracted from the health-related data into a health condition prediction model that has learned through machine learning the relationship between the health-related data and the health condition to predict the health condition of the target user, generates explanatory information that includes at least information that encourages the target user to change their behavior based on the contribution of each feature to the prediction result and the influence of each feature on the target user's behavioral change, and presents the explanatory information to the target user.

[0009] One aspect of the storage medium records a program that causes a computer to execute the following processes: acquiring health-related data, which is data regarding the health of a target user; inputting each feature extracted from the health-related data into a health condition prediction model that has learned the relationship between the health-related data and the health condition through machine learning to predict the health condition of the target user; generating explanatory information that includes at least information that encourages the target user to change their behavior based on the contribution of each feature to the prediction result and the influence of each feature on the behavioral change of the target user; and presenting the explanatory information to the target user.

[0010] According to one aspect of the present invention, it is possible to encourage a target user to change their behavior by presenting explanatory information that encourages a behavioral change.

[0011] 1 shows an overall configuration of an information processing system according to the present disclosure. 2 shows a hardware configuration of an information processing device. 3 shows a block diagram of an information processing device according to the present disclosure. 4 shows an example of health-related data. 5 shows an example of training data for a first model. 6 shows an example configuration of a rule DB. 7 shows an example of explanatory information according to the present disclosure. 8 shows a flowchart according to the present disclosure. 9 shows an overall configuration of an information processing system according to the present disclosure. 10 shows an example of training data for a second model. 11 shows an overall configuration of an information processing system according to the present disclosure. 12 shows a schematic diagram of sentence generation using a language model according to the present disclosure. 13 shows an example of explanatory information according to the present disclosure. 14 shows an example of an outline of the operation of a plan generation unit according to the present disclosure. 15 shows an example of explanatory information according to the present disclosure. 16 shows an overall configuration of an information processing system according to the present disclosure. 17 shows an overall configuration of an information processing system according to the present disclosure. 18 shows an overall configuration of an information processing system according to the present disclosure. 19 shows an overall configuration of an information processing system according to the present disclosure. 19 shows a flowchart according to the present disclosure.

[0012] Hereinafter, preferred embodiments of the present disclosure will be described with reference to the drawings.

[0013] First Embodiment (Overall Configuration) In the first embodiment, an information processing system that generates information to encourage behavioral modification will be described. Here, behavioral modification is a term that indicates a change in a person's behavior. For example, in the field of healthcare, behavioral modification may be encouraged in situations such as quitting smoking, making exercise a habit, or improving dietary habits.

[0014] 1 shows the overall configuration of an information processing system 1000 according to the first embodiment. The information processing system 1000 includes a user terminal 20 used by a target user 10, who is a user for whom behavioral modification is to be encouraged, and an information processing device 100.

[0015] The target user 10 inputs health-related data 21, which is data related to the target user's own health, into the user terminal 20. The user terminal 20 transmits the input health-related data 21 to the information processing device 100. The information processing device 100 receives the health-related data 21. The information processing device 100 predicts the health state of the target user 10 using the health-related data 21. The information processing device 100 also generates explanatory information 30 including at least information encouraging the target user 10 to change their behavior, and transmits the explanatory information 30 to the user terminal 20 for display. As will be described in detail below, the information encouraging behavior change is generated based on the contribution of each feature (generated from the health-related data 21) used to predict the health state to the prediction result and the influence of each feature on the behavior change of the target user 10.

[0016] 2 is a block diagram showing the hardware configuration of the information processing device 100. As shown in the figure, the information processing device 100 includes a processor 1, an input / output interface 2, a read-only memory (ROM) 3, a random access memory (RAM) 4, and a storage device 5. Each component is connected to the other via a bus 6, for example.

[0017] The processor 1 is a computer such as a CPU (Central Processing Unit), and executes a program prepared in advance to control the entire information processing device 100. Specifically, the processor 1 may be a CPU, a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof.

[0018] The processor 1 also loads programs stored in the ROM 3, storage device 5, etc. The processor 1 then executes each process coded in the program. The processor 1 also functions as part or all of the information processing device 100. The processor 1 may then execute processes or instructions in flowcharts, which will be described later, based on the program.

[0019] The input / output interface 2 is an interface through which the information processing device 100 transmits and receives data to and from other devices. For example, the information processing device 100 acquires various data related to the target user 10 by communicating with a user terminal 20 via the input / output interface. The information processing device 100 also transmits the generated explanation information 30 to the user terminal 20 via the input / output interface 2.

[0020] The ROM 3 stores various programs executed by the processor 1. The RAM 4 is used as a working memory while the processor 1 is executing various processes.

[0021] The storage device 5 is a non-volatile, non-transitory storage device. For example, the storage device 5 may be a disk-shaped recording medium, a semiconductor memory, or the like. The storage device 5 may be configured to be detachable from the information processing device 100. The storage device 5 stores various programs executed by the processor 1. The storage device 5 may also store machine learning models and learning data used in the health condition prediction process described below.

[0022] 3 is a block diagram showing the functional configuration of the information processing device 100. The information processing device 100 is functionally connected to a user terminal 20 and a rule DB 201. The information processing device 100 functionally includes an acquisition unit 101, a health state prediction unit 102, an information generation unit 103, and an information presentation unit 104.

[0023] The user terminal 20 is a terminal device used by the target user 10, connected to an end of a line or network, and is a device that communicates with other devices. The user terminal 20 is a terminal that acquires health-related data 21 of the target user 10, transmits the health-related data 21 to the information processing device 100, and receives explanatory information 30. For example, the user terminal 20 may be a wearable device such as a smartwatch, or a smartphone.

[0024] The acquisition unit 101 acquires health-related data 21 of the target user 10. The acquisition unit 101 may acquire the health-related data 21 via the user terminal 20, or may acquire data 21 input directly to the information processing device 100 or via a device other than the user terminal 20.

[0025] 4 shows an example of the health-related data 21. The health-related data 21 shown in FIG. 4 includes a user ID (identification) that is identification information of the target user 10, the gender of the target user 10, the date and time when the health-related data 21 was transmitted or generated, and the height, weight, body fat percentage, body age, BMI (Body Mass Index), basal metabolic rate, and visceral fat level of the target user 10. The acquisition unit 101 may also acquire, as the user's health-related data 21, weather, temperature, pedometer measurement information (date, number of steps, number of steps by time period, etc.), blood pressure monitor measurement information (systolic blood pressure, diastolic blood pressure, pulse rate, measurement time, etc.), weight measurement information (skeletal muscle percentage, measurement time, etc.), lifestyle information (mood, physical condition, diet, exercise, sleep, smoking status, drinking status, etc.), attribute information (nickname, gender, date of birth, age, family composition, etc. of the target user 10), and sleep monitor measurement information (measurement date, actual sleep time, time to fall asleep, wake-up time / duration / number of times, number of snores, snoring level, etc.). For example, the user terminal 20 may be a wearable device such as a smart watch, or a smartphone. In this case, the acquisition unit 101 may acquire, as the data 21, measurement information from a pedometer acquired by the user terminal 20. For example, the target user 10 may use the user terminal 20 as an input interface to directly input the user's health-related data 21. Furthermore, the acquisition unit 101 may acquire the user's health-related data 21 from a data provider without using the user terminal 20. For example, if the data provider is a hospital, the acquisition unit 101 may acquire the user's health-related data 21 directly from the hospital without using the user terminal 20.

[0026] The health condition prediction unit 102 predicts the health condition of the target user 10 using a health condition prediction model that has undergone machine learning of the relationship between health-related data and health conditions. The health-related data used in the machine learning of the health condition prediction model may be data collected for any target person. This target person may or may not include the target user 10.

[0027] In this embodiment, an example will be described in which the health condition prediction unit 102 calculates a health condition index 31 as a health condition prediction result. The health condition index 31 is an index that quantifies the user's health condition. The quantification method is not particularly limited, and a numerical value calculated by any method may be used. The health condition prediction model is a model that has been machine-learned using training data 5A for the health condition prediction model, and is stored in the storage device 5. The health condition prediction model may be generated by a known machine learning algorithm (e.g., random forest, support vector machine, naive Bayes, neural network, or the like).

[0028] The training data 5A for the health condition prediction model is data that associates explanatory variables that serve as inputs to the health condition prediction model with dependent variables that serve as outputs from the health condition prediction model. An example of the training data 5A is shown in Figure 5. The training data 5A is data that associates health condition indicators 31, which serve as dependent variables for the health condition prediction model, with health-related data 21, which serve as explanatory variables for the health condition prediction model.

[0029] Information processing device 100 may be equipped with a function for generating training data 5A, i.e., a training data generation unit. The training data generation unit acquires health-related data 21 of subjects whose health condition indicators 31 are known, and generates training data 5A by associating the acquired health-related data 21 with the health condition indicators 31. In addition, the training data generation unit may generate training data 5A after performing preprocessing such as data cleansing, data integration, and data conversion on the health-related data 21.

[0030] Furthermore, information processing device 100 may be equipped with a function for generating a health state prediction model using learning data 5A, i.e., a health state prediction learning unit. The health state prediction learning unit generates the health state prediction model through machine learning using learning data 5A. When health-related data 21 of target user 10 acquired by acquisition unit 101 is input, the health state prediction model generated using learning data 5A outputs a health state index 31 of target user 10 based on the learning results. Health state index 31 may indicate the current health state of target user 10, or may indicate the future health state of target user 10.

[0031] The information generation unit 103 generates explanatory information 30 that includes at least information that encourages the target user 10 to change their behavior based on the contribution of each feature input into the health status prediction model to the prediction result and the influence of each feature on the behavioral change of the target user 10.

[0032] The information generating unit 103 may calculate the above-mentioned contribution degree and influence degree when generating the explanation information 30. Specifically, the information generating unit 103 may use the health condition prediction model and the health-related data 21 acquired by the acquiring unit 101 to calculate the contribution degree of each feature amount to the prediction result of the health condition prediction model.

[0033] For example, the information generating unit 103 can use an XAI (eXplainable AI) algorithm to calculate the contribution of each feature to the prediction result. An example in which the information generating unit 103 uses LIME (Local Interpretable Model-agnostic Explanations), which is an XAI algorithm, will be described below (for details about LIME, see Non-Patent Document 1, for example). The detailed operation of LIME will be described below. The information generating unit 103 may also use another XAI algorithm (for example, SHAP: SHapley Additive exPlanations) instead of LIME.

[0034] The LIME used by the information generating unit 103 will be described below. LIME is a method for estimating the contribution of a feature to a prediction result based on a change in the feature and a change in the model output. The calculation method is expressed by the following equation (1):

[0035]

[0036] Equation (2) is the prediction result for the value obtained by randomly changing the feature quantity used for prediction, and y is the prediction result for the value of the feature quantity used for prediction. ｉ is the feature x ｉ is a parameter (variable) that indicates the degree of contribution of the

[0037] ε ｉ （ｎ） is each feature x ｉ The difference between the features is the difference between the random samples, and indicates the degree of change in the feature when the feature is randomly changed. N is the number of random samples, and d is the number of feature. The information generator 103 calculates a value of a so that the value in the term of Equation (1) is minimized. ｉ Derive and derive a ｉ is the contribution of each feature to the prediction result. In other words, the information generator 103 calculates the prediction result of the health condition prediction model and a ｉ The sum of squares error between the predicted results of the approximation model and the health status prediction model is minimized using a ｉ By solving the optimization problem to find ｉ Calculate.

[0038] Furthermore, the information generating unit 103 may estimate the degree of contribution using a formula including a weighting term for a feature amount that influences behavioral modification, such as the following formula (3).

[0039] λ is the weight (weight of the regularization term) set by the user. ｉ ) is the feature x ｉ is a function that outputs the degree to which Z(x ｉ) may be a rule stored in the rule DB 201 (described later). Formula (3) includes a regularization term expressed using a function indicating the influence of each feature. Therefore, when the contribution degree is estimated using Formula (3), the estimated contribution degree also reflects the influence on behavioral change. Therefore, by using Formula (3), the information generation unit 103 can select feature amounts that have a high contribution degree to the prediction result and a large influence on behavioral change. Then, the information generation unit 103 can use the selected feature amounts to generate explanatory information 30 including information that encourages the target user 10 to change their behavior.

[0040] Alternatively, instead of calculating the contribution degree reflecting the influence degree on the behavioral change, the information generation unit 103 may calculate the contribution degree to the prediction result and the influence degree on the behavioral change, respectively, and select feature quantities to be used for generating the explanatory information 30 based on the calculated contribution degree and influence degree. For example, the information generation unit 103 may select feature quantities with a relatively high contribution degree or a contribution degree equal to or greater than a predetermined threshold, from among the feature quantities whose contribution degrees have been calculated using the above-described formula (1). Then, the information generation unit 103 may select feature quantities from the selected feature quantities using rules (rule DB 201) indicating the influence degree on the behavioral change of the target user 10, and generate explanatory information 30 including the selected feature quantities 32.

[0041] Here, the rules used by the information generation unit 103 to indicate the degree of influence on behavioral change will be described. The rule DB 201 stores data indicating the degree of influence on behavioral change for each item of the health-related data 21. FIG. 6 shows an example of the rule DB 201. The rule DB 201 associates information about the health status of each user, i.e., health-related data, with the user ID of each user of the information processing system 1000. The health-related data are features used to predict each user's health status. Therefore, the rules stored in the rule DB 201 can also be considered as influence-specific information for identifying the degree of influence of each feature on the behavioral change of the target user 10. As shown in the figure, each feature is, for example, weight and body fat percentage. If a feature affects the user's behavioral change, a value of "1" is associated with the feature, and if it does not affect the user's behavioral change, a value of "0" is associated with the feature. For example, for a user with user ID "1," weight is one of the features that affects behavioral change.

[0042] The target user 10 may input features that influence his / her behavioral change through the user terminal 20. For example, the target user 10 may input a value of 1 for each item of the health-related data 21 if the target user 10 determines that the feature will influence behavioral change, or a value of 0 if the target user 10 determines that the feature will not influence behavioral change. The information processing device 100 records the input data as the rule DB 201. The set value may not be a binary value, but may be, for example, a ten-level value that sets the degree of influence on behavioral change. The rule DB 201 may not only be set directly by the target user 10, but may also be set based on lifestyle habits or questionnaire responses from the user, or may be set using a machine learning model.

[0043] The information generation unit 103 selects feature quantities to be included in the explanation information 30 based on the rule DB 201. For example, the information generation unit 103 selects feature quantities whose influence on behavioral change is equal to or greater than a threshold as feature quantities to be included in the explanation information 30. The information generation unit 103 may also use feature quantities whose influence on behavioral change is equal to or greater than a threshold and whose contribution to the prediction result of the health condition prediction model is equal to or greater than a threshold as the explanation information 30. The information generation unit 103 may also further select a certain number of feature quantities or less and generate the explanation information 30. The certain number may be specified by the information generation unit 103 based on the size of the display area or a number specified by the user. For example, the information generation unit 103 may determine the number of feature quantities that can fit within the display area based on the size of the display area, and include the determined number of feature quantities in the explanation information 30.

[0044] Furthermore, the rule DB 201 may be configured based on the contribution of feature values ​​that contribute to the prediction results of the health condition prediction model. Alternatively, an index may be calculated based on the contribution and influence, and the rule DB 201 may be configured based on the index. That is, rules for calculating an index from the contribution and influence may be stored in the rule DB 201. In this case, the information generator 103 may use the rules to calculate an index for each feature value from the contribution and influence of each feature value in predicting the health condition of the target user 10, and determine the feature values ​​to be included in the explanation information 30 based on the index.

[0045] Furthermore, the information generating unit 103 generates explanatory information 30 including feature quantities 32 selected based on the contribution degree calculated by LIME or the like and the influence degree indicated in the rule DB 201. The explanatory information 30 may include at least the feature quantities 32. Furthermore, the form of the feature quantities 32 in the explanatory information 30 is not particularly limited. For example, in the explanatory information 30, the feature quantities 32 may be represented by a character string, or may be represented by a combination of multiple elements such as a character string and a numerical value, or a character string and a graphic (including a graph, etc.).

[0046] The information presenting unit 104 may present the explanatory information 30 to the target user 10 by displaying the explanatory information 30 on the user terminal 20. FIG. 7 shows an example of the explanatory information 30 presented on the user terminal 20 in the first embodiment. The explanatory information 30 may include a health status index 31 in addition to the selected feature 32. The health status index 31 may be a graph showing a time-series change in the health status index predicted by the health status prediction model. The information presenting unit 104 may also display the selected feature 32 as a bar graph representing the contribution of the selected feature 32 to the prediction result of the health status prediction model or the influence on behavioral change. This allows the target user 10 to recognize the contribution to the prediction result and the influence on behavioral change, and encourages behavioral change to maintain or improve their health. Note that the presentation form of the explanatory information 30 is not limited to display, and the explanatory information 30 may be presented in the form of audio, printout, or the like.

[0047] (Information Processing) Fig. 8 is a flowchart of processing by the information processing device 100 of the first embodiment. This processing is realized by the processor 1 shown in Fig. 2 executing a program prepared in advance and operating as each element shown in Fig. 3.

[0048] First, the acquisition unit 101 acquires health-related data 21 of the target user 10 (step S11). Next, the health state prediction unit 102 predicts the health state of the target user 10 from the health-related data 21 acquired in S11 using a health state prediction model that has learned the relationship between the health-related data and health state indicators (step S12). More specifically, the health state prediction unit 102 inputs each feature amount extracted from the health-related data 21 into the health state prediction model. Then, the health state prediction unit 102 predicts the health state of the target user 10 based on the output value output from the health state prediction model.

[0049] Next, the information generator 103 calculates the contribution of each feature used in the prediction to the prediction result based on the health state prediction model used in the prediction in S12 and the health-related data 21 acquired by the acquirer 101 in S11 (step S13). As described above, the contribution can be calculated using, for example, the XAI algorithm.

[0050] Next, the information generation unit 103 selects features to be included in the explanatory information 30 using the contribution calculated in S13 and the rule DB 201 indicating the degree of influence on the behavioral change of the target user 10, and generates the explanatory information 30 including the selected features 32 (step S14). Next, the information presentation unit 104 presents the explanatory information 30 to the user terminal 20 (step S15). This completes the processing.

[0051] As described above, in the first embodiment, the user's health state is predicted, and the contribution of each feature to the health state is calculated. Then, by generating explanatory information 30 to be presented to the user based on the contribution and the influence on behavioral change, it is possible to encourage the target user 10 to change their behavior.

[0052] Second Embodiment Next, a second embodiment of the present disclosure will be described. The second embodiment differs from the first embodiment in the following two respects. First, the acquisition unit 101 acquires behavior-related data 22, which is data related to the user's behavior, in addition to the user's health-related data 21. Second, the information processing device 100 of the second embodiment includes an update unit 105 that updates rules (influence identification information) stored in the rule DB 201 based on the behavior-related data 22, and the information generation unit 103 selects feature quantities using the updated rules. Other than these points, the second embodiment is similar to the first embodiment. Specifically, the overall configuration of the information processing system 1000 and the hardware configuration of the information processing device 100 of the second embodiment are similar to the overall configuration shown in FIG. 1 and the hardware configuration shown in FIG. 2 of the first embodiment, and therefore, description thereof will be omitted.

[0053] (Functional Configuration) Fig. 9 is a block diagram showing the functional configuration of information processing device 100. Information processing device 100 is functionally connected to user terminal 20. As shown in the figure, information processing device 100 of the second embodiment includes an acquisition unit 101, a health condition prediction unit 102, an information generation unit 103, an information presentation unit 104, and an update unit 105. Note that health condition prediction unit 102, information generation unit 103, and information presentation unit 104 are the same as those in the first embodiment, and therefore description thereof will be omitted.

[0054] The user terminal 20 acquires the health-related data 21 and the user behavior-related data 22 of the target user 10, transmits the user health-related data 21 and the user behavior-related data 22 to the information processing device 100, and receives the explanation information 30.

[0055] The acquisition unit 101 acquires the user's health-related data 21, the user's behavior-related data 22, and the explanation information 30. The user's health-related data 21 is the same as in the first embodiment, and therefore a description thereof will be omitted.

[0056] In this embodiment, the explanatory information 30 and health-related data 21 acquired by the acquisition unit 101 are used to update influence identification information for identifying the influence of each feature amount on the behavioral change of the target user 10. The influence identification information may be, for example, the rules (see FIG. 6 ) stored in the rule DB 201 described in the first embodiment, or an influence estimation model for estimating the influence from various data related to the influence on the behavioral change.

[0057] For example, when the acquisition unit 101 acquires explanatory information 30 presented to the target user 10, it acquires behavior-related data 22 related to the behavior of the target user 10 during or after the presentation of the explanatory information 30. The behavior-related data 22 acquired in this manner is data related to the behavior of the target user 10 to whom the explanatory information 30 was presented, in other words, data indicating behavior that may reflect the influence of the presentation of the explanatory information 30. Note that the acquisition unit 101 only needs to acquire at least a portion of the feature amounts included in the explanatory information 30, and does not necessarily need to acquire other information included in the explanatory information 30 (e.g., health status index 31).

[0058] The user's behavior-related data 22 may include information indicating the user ID, gender, date, start date and time, end date and time, behavior category (exercise, meal, etc.), behavior details, calories burned, exercise intensity, etc. The user's behavior-related data 22 may also include emotional response data indicating the user's emotional response after the information presenter 104 presents the explanatory information 30 (which may also be referred to as the emotional response when the explanatory information 30 is presented). For example, if the user terminal 20 is a smartphone, facial image data of the target user 10 obtained from the built-in camera can be analyzed to generate emotional response data indicating the target user 10's emotions or response when the explanatory information 30 is presented. If the user terminal 20 is a smartwatch, various biometric data may be measured when the target user 10 is presented with the explanatory information 30, and the emotional response data may be the result of emotion estimation based on the measurement results. The behavior-related data 22 may also include other items.

[0059] The update unit 105 updates the influence specifying information based on the behavior-related data 22 acquired by the acquisition unit 101. For example, assume that the influence specifying information is a rule indicating the influence of each feature amount as shown in FIG. 6 . In this case, the update unit 105 identifies a feature amount included in the presented explanation information 30 from among the feature amounts indicated in the rule. Then, the update unit 105 determines whether to update the influence of the identified feature amount based on the behavior-related data 22 corresponding to the identified feature amount.

[0060] For example, suppose the presented explanatory information 30 includes a feature quantity called "weight," and the behavior-related data 22 corresponding to this feature quantity includes "calories burned." In this case, the update unit 105 may update the influence of "weight" to increase it on the condition that the value of "calories burned" in a predetermined period after the presentation of the explanatory information 30 is lower than that in a predetermined period before the presentation of the explanatory information 30. Note that if the influence has already reached its upper limit, the update is omitted. Conversely, if the value of "calories burned" has increased, the update unit 105 may update the influence of "weight" to decrease it.

[0061] As described above, the influence specification information may be an influence estimation model for estimating the influence from various data related to the influence on the behavioral change. In this case, the update unit 105 updates the influence estimation model using the behavior-related data 22 related to the behavior of the target user 10 after the explanation information 30 is presented.

[0062] The influence estimation model is a model generated by machine learning using various information related to the influence. The influence estimation model only needs to be able to be referenced by the information generation unit 103 when estimating the influence, and may be stored in, for example, the storage device 5. The influence estimation model may also be generated using a method such as random forest, support vector machine, naive Bayes, neural network, or neural network.

[0063] FIG. 10 shows an example of training data 5B that can be used to generate an influence estimation model. The training data 5B is data that associates explanatory variables that serve as inputs to the influence estimation model with objective variables that serve as outputs of the influence estimation model. The training data 5B includes, as explanatory variables, information indicating the types of feature quantities described in the explanatory information 30. The training data 5B also includes, as user behavior-related data 22, the start date and time of the behavior, the end date and time of the behavior, the behavior category (exercise, meal, etc.), details of the behavior, calories burned, and exercise intensity. By including data indicating the user's behavior in the explanatory variables, it is possible to generate an influence estimation model that estimates the influence according to the behavior of the target user 10.

[0064] By associating the learning data 5B with the degree of influence, which is the objective variable of the influence estimation model, training data that can be used for machine learning of the influence estimation model can be generated. The degree of influence may be input by the user through the user terminal 20. For example, for each item of the health-related data 21, the user may input the degree of influence in the form of 1 if it is determined that the item has an impact on behavioral change, or 0 if it is determined that the item has no impact on behavioral change. Furthermore, the input numerical value need not be a binary value, and the degree of influence on behavioral change may be set, for example, using a ten-point scale. The acquisition unit 101 can generate training data by associating the input degree of influence with the learning data 5B. Note that the value used to represent the degree of influence is arbitrary. For example, the binary values ​​1 and 0 may represent the presence or absence of an influence, or a numerical value from 1 to 10 may represent the magnitude of the influence.

[0065] The update unit 105 may also estimate the influence of each feature quantity based on the behavior-related data 22. For example, if emotional response data included in the behavior-related data 22 indicates that the target user 10 had a positive emotional response when explanatory information 30 including a certain feature quantity was presented, the update unit 105 may estimate that the influence of the feature quantity is large. On the other hand, if emotional response data included in the behavior-related data 22 indicates that the target user 10 had a negative emotional response when explanatory information 30 including a certain feature quantity was presented, the update unit 105 may estimate that the influence of the feature quantity is small. For example, the update unit 105 may also estimate the influence of the feature quantity included in the explanatory information 30 based on changes in the behavior-related data 22 between a predetermined period before and a predetermined period after the presentation of the explanatory information 30. For example, the update unit 105 may estimate that the influence of the feature quantity is large if the target user 10's daily calorie consumption increases before and after the presentation of explanatory information 30 including the certain feature quantity.

[0066] In this way, by having the update unit 105 estimate the degree of influence, training data can be generated without requiring the user to input the degree of influence. Note that the estimation of the degree of influence may be performed by a processing block different from the update unit 105. For example, a processing block called a training data generation unit may be provided, and this processing block may be made to estimate the degree of influence and generate training data.

[0067] The update unit 105 may update the influence estimation model when a predetermined number of training data or more has been accumulated. Alternatively, the update unit 105 may update the influence estimation model sequentially by online learning (online machine learning) each time the amount of training data increases.

[0068] As described above, by using the user's behavior-related data 22, it is possible to detect, for example, a change in the user's behavior. Then, it is possible to set a high influence on the behavior change of the feature included in the explanatory information 30 that triggered the behavior change. Furthermore, when the user is prompted to input an influence, it is possible to reflect the influence that the user is aware of in the influence identification information (rule DB 201 or influence estimation model).

[0069] The information generation unit 103 uses the rule DB 201 or the influence estimation model to identify or calculate the influence of each feature input to the health condition prediction model and calculate the contribution of each feature. The information generation unit 103 then selects feature(s) based on the influence and contribution, and generates explanatory information 30 including the selected feature(s) 32. For example, the information generation unit 103 selects feature(s) whose influence on behavioral change is equal to or greater than a threshold as feature(s) to be used in the explanatory information 30. The information generation unit 103 may also use feature(s) whose influence on behavioral change is equal to or greater than a threshold and whose contribution is equal to or greater than a threshold as feature(s) to be used in the explanatory information 30. The information generation unit 103 may further select a certain number of feature(s) or less to generate the explanatory information 30. The certain number may be specified by the information generation unit 103 based on the size of the display area or a number specified by the user. For example, the information generation unit 103 may determine the number of feature(s) that can fit within the display area based on the size of the display area, and select the determined number of feature(s) to be included in the explanatory information 30.

[0070] As described above, the information processing device 100 of the second embodiment has features not found in the information processing device 100 of the first embodiment, namely, it acquires the user's behavior-related data 22 and explanatory information 30 and has an update unit 105. This enables the information processing device 100 of the second embodiment to present the user with explanatory information 30 that further encourages behavioral modification.

[0071] Third Embodiment Next, a third embodiment of the present disclosure will be described. The third embodiment differs from the second embodiment in the following two respects. First, the information processing device 100 of the third embodiment includes a sentence generation unit 106 in addition to the components included in the information processing device 100 of the second embodiment. The sentence generation unit 106 generates a sentence (which can also be referred to as text) using selected features 32 and a language model. Second, the information generation unit 103 generates explanatory information 30 including the sentence generated by the sentence generation unit 106. Except for these points, the third embodiment is similar to the second embodiment. Specifically, the overall configuration of the information processing system 1000 and the hardware configuration of the information processing device 100 of the third embodiment are similar to the overall configuration shown in FIG. 1 and the hardware configuration shown in FIG. 2 of the first embodiment, and therefore description thereof will be omitted.

[0072] (Functional Configuration) Fig. 11 is a block diagram showing the functional configuration of the information processing device 100. The information processing device 100 is functionally connected to a user terminal 20. As shown in the figure, the information processing device 100 of the third embodiment includes an acquisition unit 101, a health condition prediction unit 102, an information generation unit 103, an information presentation unit 104, an update unit 105, and a sentence generation unit 106. Note that the acquisition unit 101, the health condition prediction unit 102, and the information generation unit 103 are the same as those in the second embodiment, and therefore description thereof will be omitted.

[0073] The sentence generation unit 106 generates a sentence using the feature 32 and the language model selected by the information generation unit 103. The sentence generation unit 106 transmits the generated sentence to the information generation unit 103.

[0074] (Regarding the Language Model) Here, the language model used by the sentence generation unit 106 in this embodiment will be described. The language model is a model that learns the relationships between words in a sentence and generates related strings related to a target string from the target string. By using a language model that has learned sentences and paragraphs from various contexts, it is possible to generate related strings with appropriate content related to the target string.

[0075] For example, a case where a language model is used in question answering will be described. The language model receives an input question such as "What kind of country is Japan?" as a target string. The language model generates a string such as "Japan is an island country in the Northern Hemisphere..." as an answer to the question.

[0076] The method of training the language model is not particularly limited, but as an example, the language model may be trained to output at least one sentence that includes an input character string.

[0077] As a specific example, the language model is a Generative Pre-Training (GPT) that outputs a sentence including an input string by predicting a string that is likely to follow the input string. Other examples of language models include a Text-to-text Transfer Transformer (T5), a Bidirectional Encoder Representations from Transformers (BERT), a Robustly optimized BERT approach (RoBERTa), and an Efficiently Learning an Encoder that Classifies Token Replacements Accurately (ELECTRA).

[0078] The above is a description of the language model. An outline of a method in which the sentence generation unit 106 generates a sentence using a language model will be described using FIG. 12 . FIG. 12 shows an example of an outline of sentence generation using a language model. First, the sentence generation unit 106 acquires an input sentence (prompt) 40. The prompt 40 is a sentence that commands the generation of a sentence that encourages the user to change their behavior, and includes a word that indicates the feature 32 selected by the information generation unit 103. The prompt 40 may also include either or both of the user's health-related data 21 and the user's behavior-related data 22. For example, in the example of FIG. 12 , the sentence generation unit 106 uses the user's family structure, which is included in the user's health-related data 21, as a sentence to be used as the prompt 40.

[0079] The language model may be stored in any storage device accessible to the information processing device 100, such as the storage device 5. The prompt 40 may be stored in advance in the storage device 5 or may be generated by the sentence generation unit 106. For example, the sentence generation unit 106 can generate the prompt 40 by inputting various information contained in the user's health-related data 21 and words indicating the feature quantities selected by the information generation unit 103 into a pre-generated template. For example, the prompt 40 shown in FIG. 12 can be generated from a template in which the following fields are left blank: "Male in his 50s," "Living in Tokyo," "Wife and daughter," "Hernia," "Weight," and "Body fat percentage." In this case, the sentence generation unit 106 can generate the prompt 40 by inputting the information "Age," "Gender," "Family composition," and "Chronic illness" contained in the user's health-related data 21 and the words "Weight" and "Body fat percentage" indicating the feature quantities selected by the information generation unit 103 into the template. The user may also edit the prompt 40 using the input / output interface 2.

[0080] When the sentence generation unit 106 receives the input of the prompt 40, it uses a language model to generate a comment 33 (sentence) that encourages the user to change their behavior. The sentence generation unit 106 transmits the generated comment 33 to the information generation unit 103.

[0081] The sentence generation unit 106 may change the output of the language model based on conditions specified by the user (such as the size of the display area or the number of characters). For example, the sentence generation unit 106 calculates the number of characters that will fit within the display area based on the size of the display area, and generates a comment 33 that does not exceed that number of characters. For example, the sentence generation unit 106 can use a prompt 40 that specifies a limit on the number of characters to generate a comment 33 that satisfies that limit.

[0082] The information generation unit 103 of the third embodiment differs from the information generation unit 103 of the first and second embodiments in that it generates explanatory information 30 including a comment 33. As shown in FIG. 12 , the comment 33 generated by the sentence generation unit 106 is presented to the target user 10 as part of the explanatory information 30.

[0083] The update unit 105 of the third embodiment differs from the update unit 105 of the first and second embodiments in that it updates the influence identification information using the explanation information 30 including the comment 33. In this way, by using the explanation information 30 including the comment 33, the update unit 105 can, for example, reflect the degree to which the feature value included in the comment 33 affects behavioral change in the rule DB 201. Furthermore, for example, the update unit 105 can also reflect the degree to which the feature value included in the comment 33 affects behavioral change in the influence estimation model.

[0084] FIG. 13 shows an example of explanatory information 30 in the third embodiment. As shown in FIG. 13, explanatory information 30 may include a health condition index 31 in addition to selected feature quantities 32. Furthermore, health condition index 31 may be a graph showing a time-series change in the health condition index. Furthermore, as shown in FIG. 13, information generation unit 103 may display selected feature quantities 32 as a bar graph showing the degree of contribution to the prediction result of the health condition prediction model or the degree of influence on behavioral change. Furthermore, explanatory information 30 shown in FIG. 13 includes a comment 33 generated by sentence generation unit 106.

[0085] As described above, the information processing device 100 of the third embodiment has a feature that is not present in the information processing device 100 of the first and second embodiments, that is, it generates a sentence using the selected feature 32 and a language model. This allows the information processing device 100 of the third embodiment to present the user with explanatory information 30 that further encourages behavioral modification.

[0086] Fourth Embodiment Next, a fourth embodiment of the present disclosure will be described. The fourth embodiment differs from the second embodiment in the following two respects. First, the information processing device 100 of the fourth embodiment includes a plan generation unit 107 in addition to the components included in the information processing device 100 of the second embodiment. The plan generation unit 107 generates an action plan 34 for the target user 10 based on a mathematical optimization calculation method using an objective function that includes the selected feature quantity 32 as an evaluation item. Second, the information generation unit 103 of the fourth embodiment generates explanatory information 30 including the generated action plan 34. Except for these points, the fourth embodiment is similar to the second embodiment. Specifically, the overall configuration of the information processing system 1000 and the hardware configuration of the information processing device 100 of the fourth embodiment are similar to the overall configuration shown in FIG. 1 and the hardware configuration shown in FIG. 2 of the first embodiment, and therefore, description thereof will be omitted.

[0087] (Functional Configuration) Fig. 14 is a block diagram showing the functional configuration of an information processing device 100 according to the fourth embodiment. The information processing device 100 according to the fourth embodiment is functionally connected to a user terminal 20. As shown in the figure, the information processing device 100 according to the fourth embodiment includes an acquisition unit 101, a health state prediction unit 102, an information generation unit 103, an information presentation unit 104, an update unit 105, and a plan generation unit 107. Note that the acquisition unit 101, the health state prediction unit 102, and the information generation unit 103 are the same as those in the second embodiment, and therefore description thereof will be omitted.

[0088] The plan generation unit 107 generates the behavior plan 34 for the target user 10 based on a mathematical optimization calculation method using an objective function that includes the selected feature quantities 32 as evaluation items. In other words, the plan generation unit 107 generates the behavior plan 34 for the target user 10 through mathematical optimization calculation using an objective function that evaluates the behavior of the target user 10 based on the feature quantities included in the explanation information 30.

[0089] A mathematical optimization calculation method is a method for defining a problem as a mathematical formula and finding values ​​of variables that minimize or maximize an objective function while satisfying constraints. For example, dynamic programming, greedy algorithms, and approximation algorithms may be used.

[0090] 15 shows an example of the outline of the operation of the plan generating unit 107. The plan generating unit 107 receives input data 50, creates an objective function 60, and transmits output data 70 to the information generating unit 103. Each of these processes will be described in detail below.

[0091] First, the plan generation unit 107 accepts input data 50. The input data 50 includes an exercise type property 51, a meal property 52, explanatory information 30, and constraints 53. The exercise type property 51 is information about the exercise type of the action plan 34 to be presented to the target user 10. The exercise type indicated in the exercise type property 51 is a candidate for a component of the action plan 34. The exercise type property 51 may include, for example, the name of the exercise type, the exercise intensity of the exercise type, and information indicating the effect on each feature. For example, the exercise type property 51 shown in FIG. 15 indicates that an exercise type called "muscle training A" can be selected as a component of the action plan 34, and also indicates that the value indicating the magnitude of the effect of this exercise on body fat percentage is 7, the value indicating the magnitude of the effect on blood pressure is 5, and the value indicating the magnitude of the effect on muscle mass is 8. In this example, the magnitude of the effect on the feature is expressed on a scale of 1 to 10. The method for evaluating the magnitude of the influence on the feature amount is not particularly limited, and the results of evaluation using any evaluation method may be included in the exercise event property 51.

[0092] The meal property 52 is information about the meals in the action plan 34 to be presented to the target user 10. The meal or ingredient items indicated in the meal property 52 are candidates for components of the action plan 34. The meal property 52 may include, for example, information indicating the item name, the nutrients contained in the item, and the magnitude of the effect on each feature. For example, the meal property 52 shown in FIG. 15 indicates that the item name "vegetables" can be selected as a component of the action plan 34, and that the numerical value indicating the magnitude of the effect of this item on blood pressure is 5. Note that the method for evaluating the magnitude of the effect is not particularly limited, and the results of evaluation using any evaluation method may be included in the meal property 52.

[0093] The explanatory information 30 includes feature quantities 32 selected by the information generation unit 103. The explanatory information 30 shown in FIG. 15 includes body fat percentage, skeletal muscle percentage, and blood pressure, which are selected as items that have a high impact on behavioral change. Constraint conditions 53 are conditions that must be satisfied by the behavior plan 34. The constraint conditions 53 in FIG. 15 include the constraint conditions "calories consumed are X (kcal) or more" and "eat vegetables once a day." The constraint conditions 53 can be set and changed by the target user 10.

[0094] Next, the plan generation unit 107 generates an objective function 60. The objective function 60 is a function to be maximized or minimized in the mathematical optimization calculation method. The evaluation items included in the objective function 60 use selected feature values ​​32 included in the explanatory information 30. The objective function 60 shown in FIG. 15 evaluates the action plan of the target user 10 and includes three evaluation items: "body fat percentage deviation," "skeletal muscle percentage deviation," and "blood pressure deviation." Weight values ​​α, β, and γ are respectively set for each evaluation item. Each evaluation item corresponds to the feature value 32 selected by the information generation unit 103. The weight values ​​can be calculated by inverse reinforcement learning using learning data that associates an action plan given an evaluation result with input data (such as exercise event properties) corresponding to the action plan. The weight values ​​may be set based on the rule DB 201 or may be set by the target user 10.

[0095] Finally, the plan generation unit 107 outputs output data 70. The output data 70 is an action plan 34 calculated by the plan generation unit 107 using the input data 50 and the objective function 60, which maximizes the evaluation of the action plan represented by the objective function 60. The action plan 34 indicates the type of exercise to be performed on each day of the week and the meals to be eaten. The example of the action plan 34 shown in FIG. 15 recommends running and eating vegetables on Mondays. This method is not limited to the example shown in FIG. 15 , and can generate an action plan 34 in any format. The plan generation unit 107 transmits the action plan 34 to the information generation unit 103. For example, the input data 50, the objective function 60, and the output data 70 may be stored in the storage device 5.

[0096] The information generating unit 103 of the fourth embodiment differs from the information generating unit 103 of the first and second embodiments in that it generates explanation information 30 including an action plan 34 .

[0097] FIG. 16 shows an example of explanatory information 30 in the fourth embodiment. As shown in FIG. 16, explanatory information 30 may include a health state index 31 in addition to selected feature quantities 32. Furthermore, health state index 31 may be a graph showing a time-series change in the health state index. Furthermore, as shown in FIG. 16, information generator 103 may display selected feature quantities 32 as a bar graph showing the degree of contribution to the prediction result of the health state prediction model or the degree of influence on behavioral change. Furthermore, explanatory information 30 shown in FIG. 16 includes an action plan 34 generated by plan generator 107.

[0098] As described above, the information processing device 100 of the fourth embodiment has a feature not found in the information processing devices 100 of the first and second embodiments, in that the plan generation unit 107 generates the behavior plan 34 for the target user 10 based on a mathematical optimization calculation method using an objective function that includes the selected feature quantity 32 as an evaluation item. This allows the information processing device 100 of the fourth embodiment to present the user with explanatory information 30 that further encourages behavioral change.

[0099] Fifth Embodiment Next, a fifth embodiment of the present disclosure will be described. The fifth embodiment differs from the second embodiment in the following three points. First, the acquisition unit 101 of the fifth embodiment acquires attribute information of the target user 10 as part of the user's health-related data 21, in addition to the user's behavior-related data 22 and explanatory information 30. Second, the information processing device 100 of the fifth embodiment includes a similar user identification unit 108 in addition to the components included in the information processing device 100 of the second embodiment. The similar user identification unit 108 selects similar users who are similar to the target user 10 based on the attribute information acquired by the acquisition unit 101. Third, the information generation unit 103 of the fifth embodiment generates explanatory information 30 including similar user information 35, which is information related to similar users. Other than these points, the fifth embodiment is similar to the second embodiment. Specifically, the overall configuration of the information processing system 1000 and the hardware configuration of the information processing device 100 of the fifth embodiment are similar to the overall configuration shown in FIG. 1 and the hardware configuration shown in FIG. 2 of the first embodiment, and therefore will not be described again.

[0100] (Functional Configuration) FIG. 17 is a block diagram showing the functional configuration of an information processing device 100 according to the fifth embodiment. The information processing device 100 according to the fifth embodiment is functionally connected to a user terminal 20 and a user information DB 202. The user information DB 202 stores information about each user who uses the information processing system 1000. For example, the user information DB 202 stores attribute information about each user, a health condition index calculated for each user by the health condition prediction unit 102, health-related data 21 for each user, feature quantities 32 selected for each user, and behavior-related data 22 for each user. As shown in the figure, the information processing device 100 according to the fifth embodiment includes an acquisition unit 101, a health condition prediction unit 102, an information generation unit 103, an information presentation unit 104, an update unit 105, and a similar user identification unit 108. The health condition prediction unit 102 and the information generation unit 103 are the same as those in the second embodiment, and therefore will not be described here.

[0101] The acquisition unit 101 of the fifth embodiment acquires user health-related data 21 including attribute information of the target user 10, user behavior-related data 22, and explanatory information 30. The attribute information is information that characterizes the target user 10 and indicates at least one of the properties and characteristics of the target user 10. For example, the age, gender, place of residence, family structure, occupation, income, etc. of the target user 10 can also be used as attribute information. In addition, at least a portion of the health condition index calculated by the health state prediction unit 102, the health-related data 21, the selected feature amount 32, and the behavior-related data 22 can also be used as attribute information.

[0102] The similar user identification unit 108 identifies similar users who are similar to the target user 10 from among the users registered in the user information DB 202, based on the attribute information acquired by the acquisition unit 101. Then, the similar user identification unit 108 generates similar user information 35 indicating the identified similar users, and transmits it to the information generation unit 103.

[0103] The method by which the similar user identification unit 108 selects similar users is not particularly limited. For example, the similar user identification unit 108 may calculate the Euclidean distance between the attribute information of each user as the similarity between the users, and identify similar users based on the calculated similarity. For example, the similar user identification unit 108 plots the attribute information of each user (age, height, weight, blood pressure, cholesterol, blood sugar level, etc.) on respective coordinate axes, calculates the Euclidean distance between the attribute information of each user, and determines that the closer the distance, the more similar the attribute information of the users is. For example, the similar user identification unit 108 may identify users whose similarity to the target user 10 is equal to or greater than a predetermined threshold as similar users.

[0104] The similarity used by the similar user identification unit 108 is not limited to Euclidean distance, and may be, for example, cosine similarity. The similar user identification unit 108 may also identify similar users who are similar to the target user 10 by classifying the users shown in the user information DB 202 into users who are similar to the target user 10 and users who are dissimilar to the target user 10 using a classification method such as the K-nearest neighbor method.

[0105] The information generating unit 103 of the fifth embodiment differs from the information generating unit 103 of the first and second embodiments in that it generates explanation information 30 including similar user information 35 .

[0106] The update unit 105 of the fifth embodiment differs from the update units 105 of the first and second embodiments in that it updates the influence identification information using similar user information 35. By using the explanation information 30 including the similar user information 35 in this manner, the update unit 105 can, for example, reflect the degree to which the feature amounts included in the similar user information 35 (the feature amounts included in the explanation information 30 presented to the similar user) influence the behavioral change of the target user 10 in the rule DB 201. Furthermore, for example, the update unit 105 can also reflect the degree to which the feature amounts included in the explanation information 30 presented to the similar user influence the behavioral change of the target user 10 in the influence estimation model.

[0107] FIG. 18 shows an example of explanatory information 30 according to the fifth embodiment. As shown in FIG. 18 , the explanatory information 30 may include a health status index 31 in addition to the selected feature quantities 32. The health status index 31 may be a graph showing the time-series changes in the health status index. As shown in FIG. 18 , the information generator 103 may display the selected feature quantities 32 as a bar graph showing the contribution of the selected feature quantities 32 to the prediction results of the health status prediction model or the influence on behavioral change. The explanatory information 30 shown in FIG. 18 also includes similar user information 35. The similar user information 35 may include comments from similar users, graphs showing time-series changes in the health status index of similar users, and the like. In addition, the similar user information 35 may include, for example, the names (not necessarily real names) and attribute information of similar users, feature quantities that had a significant influence on the behavioral change of similar users, and the like, as long as the privacy of the similar users is not violated.

[0108] As described above, the information processing device 100 of the fifth embodiment has a feature not found in the information processing device 100 of the first and second embodiments, in that it includes the similar user identification unit 108 and generates explanatory information 30 including similar user information 35 related to the similar users identified by the similar user identification unit 108. This allows the information processing device 100 of the fifth embodiment to present explanatory information 30 to the user that further encourages behavioral modification.

[0109] Sixth Embodiment Next, a sixth embodiment of the present disclosure will be described. The sixth embodiment differs from the second embodiment in the following two respects. First, the information processing device 100 of the sixth embodiment includes a detection unit 109 in addition to the components included in the information processing device 100 of the second embodiment. The detection unit 109 detects a change in the behavior (behavioral modification) of the target user 10 based on the user behavior-related data 22 acquired by the acquisition unit 101. Second, the update unit 105 of the sixth embodiment generates a second model using training data 5B including the detection result of the detection unit 109. Other than these points, the sixth embodiment is similar to the second embodiment. Specifically, the overall configuration of the information processing system 1000 and the hardware configuration of the information processing device 100 of the sixth embodiment are similar to the overall configuration shown in FIG. 1 and the hardware configuration shown in FIG. 2 of the first embodiment, and therefore, description thereof will be omitted.

[0110] (Functional Configuration) Fig. 19 is a block diagram showing the functional configuration of an information processing device 100 according to the sixth embodiment. The information processing device 100 according to the sixth embodiment is functionally connected to a user terminal 20. As shown in the figure, the information processing device 100 according to the sixth embodiment includes an acquisition unit 101, a health condition prediction unit 102, an information generation unit 103, an information presentation unit 104, an update unit 105, and a detection unit 109. Note that the acquisition unit 101, the health condition prediction unit 102, the information generation unit 103, and the information presentation unit 104 are the same as those in the second embodiment, and therefore description thereof will be omitted.

[0111] The detection unit 109 detects a change in the behavior (behavioral modification) of the target user 10 based on the user behavior-related data 22 acquired by the acquisition unit 101, and transmits the detection result to the update unit 105. First, the detection unit 109 receives the user behavior-related data 22 acquired by the acquisition unit 101. The user behavior-related data 22 is the same as that described in the second embodiment. Next, the detection unit 109 detects whether the user behavior has changed based on the user behavior-related data 22.

[0112] The method by which the detection unit 109 detects a behavioral change is not particularly limited. For example, the detection unit 109 may detect a behavioral change using acceleration data obtained from the user terminal 20 as the behavior-related data 22. When the target user 10 carrying the user terminal 20 exercises, the acceleration data measured by the user terminal 20 changes in accordance with the movement of the target user 10. Therefore, the detection unit 109 can determine whether the target user 10 is exercising by analyzing the change in the acceleration data. For example, the detection unit 109 may determine whether the target user 10 is exercising by calculating the amplitude of the acceleration. The detection unit 109 calculates the amplitude of the acceleration by calculating the square root of the value obtained by squaring the acceleration in each axial direction. The detection unit 109 determines that the target user 10 has exercised when the amplitude of the acceleration exceeds a predetermined threshold.

[0113] The detection unit 109 acquires the exercise record of the target user 10 using the above-described method. Then, by analyzing the exercise record for a certain period of time, the detection unit 109 determines whether exercising has become a lifestyle habit for the target user 10 based on a predetermined rule. If the detection unit 109 determines that the target user 10 has acquired an exercise lifestyle habit, the detection unit 109 detects a behavioral change in the target user 10. The predetermined rule may be, for example, a change in the frequency or duration of exercise over a predetermined period of time. In other words, the detection unit 109 may detect that a behavioral change has occurred when the frequency or duration of exercise over a predetermined period of time has changed. Note that the predetermined rule may include a change amount or rate of change in the frequency or duration of exercise. In this case, the detection unit 109 detects that a behavioral change has occurred when the frequency or duration of exercise exceeds a set change amount or rate of change. The detection unit 109 may detect both positive behavioral changes (e.g., acquiring an exercise habit) and negative behavioral changes (e.g., a decrease in exercise frequency or exercise time) of the target user 10, or may detect only one of them.

[0114] When the detection unit 109 detects a behavioral change, the detection result includes the category of the behavior in which the behavioral change was detected, the date and time when the behavioral change was detected, and the explanatory information 30 that was presented to the target user 10 during the period between the time before the behavioral change was detected and the time after the behavioral change was detected. The detection result is not limited to these, and may include other items.

[0115] Finally, the detection unit 109 transmits the detection result to the update unit 105. The detection result of the detection unit 109 may be stored in any storage device such as the storage device 5.

[0116] The update unit 105 of the sixth embodiment updates the influence identification information by regarding a feature included in the explanatory information 30 (explanatory information presented before the behavioral change was detected) that is thought to have triggered the behavioral change detected by the detection unit 109 as a feature with high influence. In this way, the update unit 105 can reflect in the influence identification information (the rule DB 201 or the influence estimation model) that the feature has a high influence for a feature that is likely to have actually triggered the behavioral change of the target user 10.

[0117] As described above, the information processing device 100 of the sixth embodiment has a feature not found in the information processing device 100 of the second embodiment, in that the detection unit 109 detects a behavioral change of the target user 10 and the update unit 105 updates the influence level identification information based on the detection result. This allows the information processing device 100 of the sixth embodiment to present the user with explanatory information 30 that further encourages behavioral change.

[0118] 20 is a block diagram showing the functional configuration of an information processing device 100 according to a seventh embodiment. The information processing device 100 according to the seventh embodiment includes an acquisition unit 101, a health state prediction unit 102, an information generation unit 103, and an information presentation unit 104.

[0119] The acquisition unit 101 acquires health-related data 21, which is data related to the health of the target user 10. The health state prediction unit 102 inputs each feature amount extracted from the health-related data 21 into a health state prediction model that has undergone machine learning to determine the relationship between the health-related data 21 and the health state, thereby predicting the health state of the target user 10. The information generation unit 103 generates explanatory information 30 that includes at least information encouraging the target user 10 to change their behavior, based on the contribution of each feature amount to the prediction result and the influence of each feature amount on the behavioral change of the target user 10. The information presentation unit 104 presents the explanatory information 30 to the target user 10.

[0120] FIG. 21 is a flowchart of a process (information processing method) performed by the information processing device 100 according to the seventh embodiment. The acquisition unit 101 acquires health-related data 21 of the target user 10 (step S71). Next, the health state prediction unit 102 inputs each feature extracted from the health-related data 21 of the target user 10 into a health state prediction model that has machine-learned the relationship between the health-related data 21 and the health state of the target user 10, thereby predicting the health state of the target user 10 (step S72). Next, the information generation unit 103 generates explanatory information 30 including at least information encouraging the target user 10 to change their behavior based on the contribution of each feature used in the prediction in S72 to the prediction result and the influence of each feature on the behavioral change of the target user 10 (step S73). Then, the information presentation unit 104 presents the explanatory information 30 to the target user 10 (step S74).

[0121] According to the information processing device 100 of the seventh embodiment, it is possible to encourage the target user 10 to change his / her behavior by generating information that encourages a change in behavior.

[0122] Although the present invention has been described above with reference to exemplary embodiments, the present invention is not limited to the above-described exemplary embodiments. Various modifications within the scope of the present invention that would be understood by those skilled in the art are possible in the configuration and details of the present invention. The present disclosure may include embodiments in which the details described herein are appropriately combined or substituted as necessary. For example, details described using a specific embodiment may also be applied to other embodiments to the extent that no inconsistency arises. For example, although multiple operations are described in sequence in flowchart form, the order of description does not limit the order in which the operations are performed. Therefore, when implementing each embodiment, the order of the multiple operations can be changed as long as it does not interfere with the content. Furthermore, the processing of each step shown in each flowchart can be executed by a single processor or can be shared and executed by multiple processors. In other words, the information processing method according to each of the above-described exemplary embodiments may be executed by a single processor or by multiple processors.

[0123] Some or all of the above-described embodiments can be described as follows: However, some or all of the above-described embodiments are not limited to the following.

[0124] (Supplementary Note 1) An information processing device comprising: an acquisition unit that acquires health-related data, which is data regarding the health of a target user; a health condition prediction unit that inputs each feature extracted from the health-related data into a health condition prediction model that has learned the relationship between the health-related data and the health condition through machine learning to predict the health condition of the target user; an information generation unit that generates explanatory information including at least information that encourages the target user to change their behavior based on the contribution of each feature to the prediction result and the influence of each feature on the behavioral change of the target user; and an information presentation unit that presents the explanatory information to the target user.

[0125] (Supplementary Note 2) The information processing device described in Supplementary Note 1 is characterized in that the information generation unit identifies the degree of influence using influence identification information for identifying the degree of influence of each of the feature quantities on the behavioral change of the target user, and the acquisition unit further acquires behavior-related data which is data regarding the behavior of the target user to which the explanatory information has been presented, and further includes an update unit which updates the influence identification information based on the behavior-related data.

[0126] (Supplementary Note 3) The information processing device according to Supplementary Note 1 or 2, wherein the information generation unit calculates the contribution by performing a mathematical optimization calculation on an optimization problem including a regularization term expressed using a variable indicating the influence of each of the feature quantities.

[0127] (Appendix 4) An information processing device described in any one of Appendices 1 to 3, further comprising a sentence generation unit that generates sentences using the features contained in the explanatory information and a language model, and the information presentation unit presents the explanatory information including the sentences generated by the sentence generation unit.

[0128] (Appendix 5) An information processing device described in any one of Appendices 1 to 4, further comprising a plan generation unit that generates an action plan for the target user by mathematical optimization calculation using an objective function that evaluates the behavior of the target user based on the features included in the explanatory information, and the information presentation unit presents the explanatory information including the action plan generated by the plan generation unit.

[0129] (Supplementary Note 6) The information processing device described in any one of Supplementary Notes 1 to 5 is characterized in that the acquisition unit further acquires attribute information of the target user and further includes a similar user identification unit that identifies similar users who are similar to the target user based on the attribute information, and the information presentation unit presents the explanatory information including similar user information related to the similar users identified by the similar user identification unit.

[0130] (Supplementary Note 7) The information processing device according to Supplementary Note 2, wherein the behavior-related data includes data indicating an emotional reaction of the target user after the explanatory information is presented.

[0131] (Appendix 8) An information processing device as described in Appendix 2 or 7, further comprising a detection unit that detects behavioral changes of the target user based on the behavior-related data, and the update unit updates the impact identification information based on the detection results by the detection unit.

[0132] (Supplementary Note 9) An information processing method in which at least one processor acquires health-related data that is data regarding the health of a target user, inputs each feature extracted from the health-related data into a health condition prediction model that has learned through machine learning the relationship between the health-related data and the health condition to predict the health condition of the target user, generates explanatory information that includes at least information that encourages the target user to change their behavior based on the contribution of each feature to the prediction result and the influence of each feature on the behavioral change of the target user, and presents the explanatory information to the target user.

[0133] (Supplementary Note 10) A recording medium storing a program that causes a computer to execute the following steps: acquiring health-related data related to the health of a target user; inputting each feature extracted from the health-related data into a health condition prediction model that has learned the relationship between the health-related data and the health condition through machine learning to predict the health condition of the target user; generating explanatory information including at least information encouraging the target user to change their behavior based on the contribution of each feature to the prediction result and the influence of each feature on the target user's behavior change; and presenting the explanatory information to the target user. Note that the recording medium may be a computer-readable, non-transitory tangible recording medium. The above Supplements 9 and 10 can be expanded into the forms of Supplements 2 to 8, similar to Supplementary Note 1.

[0134] REFERENCE SIGNS LIST 1 Processor 2 Input / output interface 3 ROM 4 RAM 5 Storage device 5A First model learning data 5B Second model learning data 6 Bus 10 Target user 20 User terminal 21 Health-related data 22 Behavior-related data 30 Explanation information 31 Health condition index 32 Selected feature amount 33 Comment 34 Behavior plan 35 Similar user information 40 Input sentence 50 Input data 51 Exercise type property 52 Meal property 53 Constraint condition 60 Objective function 70 Output data 100 Information processing device 101 Acquisition unit 102 Health condition prediction unit 103 Information generation unit 104 Information presentation unit 105 Update unit 106 Sentence generation unit 107 Plan generation unit 108 Similar user identification unit 109 Detection unit 201 Rule DB 202 User information DB 1000 Information Processing Systems

Claims

1. An acquisition unit that acquires health-related data, which is data about the health of the target user, A health status prediction unit inputs each feature extracted from the aforementioned health-related data into a health status prediction model that uses machine learning to analyze the relationship between health data and health status, and predicts the health status of the target user. An information generation unit generates explanatory information that includes at least information that encourages behavioral change in the target user, based on the contribution of each of the features to the prediction results and the influence of each of the features on behavioral change in the target user. An information display unit that presents the aforementioned explanatory information to the target user, An information processing device equipped with the following features.

2. The information generation unit identifies the degree of influence of each feature quantity using influence identification information to identify the degree of influence of the target user's behavioral change, The acquisition unit further acquires behavior-related data, which is data relating to the actions of the target user who has been presented with the explanatory information. The system further includes an update unit that updates the impact identification information based on the aforementioned behavior-related data. The information processing apparatus according to feature 1.

3. The information generation unit calculates the contribution by performing mathematical optimization calculations on an optimization problem that includes a regularization term expressed using a variable indicating the degree of influence of each feature. The information processing apparatus according to feature 1.

4. The system further comprises a sentence generation unit that generates sentences using the features and language model included in the explanatory information. The information presentation unit presents the explanatory information, including the sentences generated by the sentence generation unit. The information processing apparatus according to feature 1.

5. The system further comprises a planning generation unit that generates an action plan for the target user by performing a mathematical optimization calculation using an objective function that evaluates the target user's actions based on the features included in the explanatory information, The information presentation unit presents the explanatory information, including the action plan generated by the plan generation unit. The information processing apparatus according to feature 1.

6. The acquisition unit further acquires the attribute information of the target user, The system further includes a similar user identification unit that identifies similar users to the target user based on the attribute information mentioned above. The information presentation unit presents the explanatory information, which includes similar user information related to the similar user identified by the similar user identification unit. The information processing apparatus according to any one of claims 1 to 5.

7. The aforementioned behavioral data includes data showing the emotional response of the target user after the explanatory information has been presented. The information processing apparatus according to feature 2.

8. The system further includes a detection unit that detects changes in the behavior of the target user based on the aforementioned behavior-related data. The update unit updates the impact identification information based on the detection result from the detection unit. The information processing apparatus according to claim 2 or 7.

9. At least one processor, We obtain health-related data, which is data about the health of the target users. Each feature extracted from the aforementioned health-related data is input into a health status prediction model that uses machine learning to analyze the relationship between health-related data and health status, thereby predicting the health status of the target user. Based on the contribution of each of the aforementioned features to the prediction results and the influence of each of the aforementioned features on the behavioral change of the target user, explanatory information is generated that includes at least information that encourages behavioral change in the target user. An information processing method for presenting the aforementioned explanatory information to the target user.

10. The process of obtaining health-related data, which is data about the health of the target user, The process involves inputting each feature extracted from the aforementioned health-related data into a health status prediction model that uses machine learning to analyze the relationship between health-related data and health status, in order to predict the health status of the target user. A process for generating explanatory information that includes at least information that encourages behavioral change in the target user, based on the contribution of each of the features to the prediction results and the influence of each of the features on behavioral change in the target user; A process of presenting the aforementioned explanatory information to the target user, A program that causes a computer to execute something.