Computer system and method for evaluating dementia risk and / or frailty risk

Abstract

To early discover dementia risk and / or frailty risk.SOLUTION: A system stores information on a plurality of check items for calculating a risk score including at least one of a dementia risk score and a frailty risk score. The system calculates a risk score of a first participant on the basis of a predetermined first type risk word that is included in interaction data including a text message posted by the first participant of an interaction group and is associated with each check item of the plurality of check items, determines whether or not an additional question is necessary on the basis of a predetermined second type risk word included in the interaction data or the risk score of the first participant, transmits the additional question to the first participant for updating the risk score of the first participant in response to determination that the additional question is necessary, and updates the risk score of the first participant on the basis of the first participant's response to the additional question.SELECTED DRAWING: Figure 20

Description

[Technical field]

[0001] The present invention relates to assessment of dementia risk and / or frailty risk. [Background technology]

[0002] Due to the extension of life expectancy and the increase in the proportion of elderly people, it is becoming increasingly important to extend the healthy life expectancy of the elderly. In order to maintain the health of the elderly, it is important to detect dementia and frailty risks early and promote preventive actions.

[0003] A related technique of the present disclosure is JP 2016-184403 A (Patent Document 1). Patent Document 1 discloses extracting data based on online communications of a user who accesses web content, identifying health-related problems from the extracted data, and generating content showing suggestions for addressing the health-related problems based on the extracted data. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2016-184403 A Summary of the Invention [Problem to be solved by the invention]

[0005] An effective method for estimating dementia risk and frailty risk is to estimate from the results of a checklist. However, there are a certain number of elderly people who do not respond to the checklist because they are indifferent to dementia or frailty, or because they are afraid that their risk will be revealed. As a result, the response rate of elderly people to the risk checklist is low, and furthermore, the frequency of response is generally once a year. This makes it difficult to frequently track changes in dementia risk and frailty risk.

[0006] In addition, as disclosed in Patent Document 1, it is difficult to estimate the dementia risk or frailty risk with sufficient accuracy and provide appropriate suggestions by making suggestions to address health-related problems based on the user's web access. Therefore, a technology that can detect the dementia risk and / or frailty risk at an early stage is desired. [Means for solving the problem]

[0007] In order to solve the above problem, one aspect of the present invention is a computer system that includes a processor and a storage device that stores a program executed by the processor, wherein the storage device stores information on a plurality of check items for calculating a risk score including at least one of a dementia risk score and a frailty risk score, and the processor calculates the risk score of the first participant in an interaction group based on a predetermined first type risk word associated with each check item of the plurality of check items contained in interaction data including a text message posted by the first participant, determines whether or not additional questions are needed based on a predetermined second type risk word contained in the interaction data or the risk score of the first participant, and in response to the determination that the additional questions are needed, sends an additional question to the first participant for updating the risk score of the first participant, and updates the risk score of the first participant based on the first participant's response to the additional question.

[0008] Further features related to the present invention will become apparent from the description of the present specification and the accompanying drawings. Also, aspects of the present invention may be realized and realized by the elements and combinations of various elements and aspects of the following detailed description and the appended claims.

[0009] It should be understood that the descriptions in this specification are exemplary and illustrative only and are not intended to limit the scope or application of the present invention in any manner. Effect of the Invention

[0010] According to one aspect of the present invention, the risk of dementia and / or the risk of frailty can be detected early. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram showing an example of a functional configuration of a system according to an embodiment of the present specification. [Diagram 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a chat analysis device according to an embodiment of the present specification. [Diagram 3] 1 shows an example of the configuration of training data for an elderly classification estimation model. [Figure 4] 13 shows an example of the configuration of training data for a score improvement behavior estimation model. [Diagram 5] An example of the structure of an exchange group DB is shown below. [Figure 6] 1 shows an example of the configuration of an AC data feature DB. [Figure 7] An example of information contained in the question DB is shown below. [Figure 8] An example of the structure of a direct question list for frailty risk is shown below. [Figure 9] An example of the composition of a frailty risk questionnaire is shown below. [Figure 10] An example of the structure of a direct dementia risk question list is shown below. [Figure 11] 13 shows an example of the configuration of a risk word list related to frailty risk and dementia risk included in the risk word list. [Figure 12] An example of information stored in the response history DB is shown below. [Figure 13] 13 shows an example of the configuration of a frailty risk user response history. [Figure 14] 13 shows an example of the structure of a dementia risk user response history. [Figure 15] 13 shows an example of the configuration of a frailty risk indirect question response history. [Figure 16] 13 shows an example of the structure of a dementia risk indirect question response history. [Figure 17] Here's an example of an indirect question posted in a group chat: [Figure 18]Here is an example of a direct question posted in a group chat. [Figure 19A] An example of a suggested improvement action in chat is shown below. [Figure 19B] 13 shows another example of displaying improvement actions suggested by the chat analysis device. [Figure 19C] 13 shows an example of a response of a participant to an improvement action suggested in a chat room by a chat analysis device. [Figure 20] 13 shows a flowchart of an example of processing performed by a chat analysis device. [Figure 21] 13 shows an example of input data to the score improvement behavior estimation model. [Figure 22] 13 shows an example of the configuration of an improvement score table. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, the embodiments of the present specification will be described with reference to the accompanying drawings. In the accompanying drawings, functionally identical elements may be indicated by the same numbers. Note that the accompanying drawings show specific embodiments and implementation examples according to the principles of the present invention, but these are for understanding the present invention and are by no means used to interpret the present invention in a limiting manner.

[0013] In the present embodiment, the present invention has been described in sufficient detail for those skilled in the art to practice the present invention, but it should be understood that other implementations and forms are possible, and that changes in configuration and structure and substitutions of various elements are possible without departing from the scope and spirit of the technical ideas of the present invention. Therefore, the following description should not be interpreted as being limited thereto.

[0014] Additionally, the embodiments herein may be implemented in software running on a general purpose computer, in specialized hardware, or a combination of software and hardware, as described below.

[0015] When describing each process in the embodiments of this specification with "each processing unit as a program" as the subject (acting entity), the program is executed by a processor (CPU, etc.) to perform the defined process using memory and a communication port (communication control device), so the description may be given with the processor as the subject. <System configuration>

[0016] 1 is a block diagram showing an example of a functional configuration of a system according to an embodiment of the present specification. The system estimates the dementia risk and / or frailty risk of an elderly user through chat analysis, and suggests actions to prevent dementia and / or frailty. This embodiment can promote the prevention of dementia and / or frailty in the elderly.

[0017] Frailty is an intermediate state between a healthy state and a state requiring nursing care, in which vulnerability to health disorders increases due to functional decline, and there is a risk of developing a state requiring nursing care, but the change to a state requiring nursing care can be prevented by appropriate treatment and prevention. The embodiment described below targets both dementia risk and frailty risk, but may target only one of them.

[0018] In a chat system, multiple participants can post (send) messages in real time and read (receive) the posted messages. In one embodiment of the present specification, chats between users are analyzed, and the user's responses (contained in the chats) to messages automatically generated and posted by the system are analyzed to estimate the user's dementia risk and / or frailty risk. Note that the technique described below can also be applied to systems or services other than chats, such as bulletin board systems, in which multiple participants post text messages to each other.

[0019] The system shown in Fig. 1 is a computer system. The computer system includes a chat analysis device 1 and a user terminal 3, which communicate with each other via a network 5. Fig. 1 shows one user terminal 3 as an example, but user terminals of multiple elderly people (users) who participate in chats are connected to the network 5.

[0020] The chat analysis device 1 includes an elderly classification estimation unit 10, a dementia / frailty risk score estimation unit 11, an estimation result integration processing unit 12, a suggested behavior creation unit 13, a suggested behavior interest / effectiveness assessment unit 14, a model training unit 15, an elderly classification estimation model 16, a score improvement behavior estimation model 18, and an interaction data collection unit 19.

[0021] The chat analysis device 1 further stores a training data DB (database) 21, a question DB 22, an interaction group DB 23, a response history DB 24, an improvement action DB 25, and an interaction data feature DB 27.

[0022] The elderly classification estimation unit 10 estimates the personality type of the elderly user based on the content of the chat using the elderly classification estimation model 16. The dementia / frailty risk score estimation unit 11 estimates the dementia risk score and frailty risk score based on the content of the user's chat. As will be described later, the chat to be analyzed includes messages between users (chats) as well as user responses (messages) to questions from the system.

[0023] The estimation result integration processing unit 12 generates data to be input to the score improvement behavior estimation model 18. The suggested behavior creation unit 13 determines behavior to be suggested to the user to prevent dementia or frailty based on the dementia risk score and the frailty risk score. The suggested behavior interest / effectiveness determination unit 14 determines the interest and effectiveness of the suggestion based on the response to the behavior suggested to the user.

[0024] The model training unit 15 trains an elderly classification estimation model 16 and a score improvement behavior estimation model 18. The elderly classification estimation model 16 estimates a personality type of a user based on the user's chat data. The score improvement behavior estimation model 18 estimates a suggested behavior that may contribute to improving the user's score based on the dementia risk score and the frailty risk score. The interaction data collection unit 19 collects information on interaction groups and chat data for each interaction group from the user terminal 3.

[0025] The user terminal 3 includes a chat input / sending unit 31 and a chat display unit 32. The chat input / sending unit 31 accepts input for chat by a user and transmits it to a server (not shown) that provides a chat service. The chat display unit 32 displays a screen of chat by the user and other users and a screen presenting improvement actions proposed by the chat analysis device 1. The improvement actions can be displayed on a screen of chat with other users, a screen of chat with the chat analysis device 1, or a screen of another application.

[0026] The functional units 10 to 19 in the chat analysis device 1 and the functional units 31 and 32 in the user terminal 3 may be realized by a processor that executes a program, or may be realized by a hardware module implemented in accordance with each functional unit.

[0027] 2 is a diagram showing an example of a hardware configuration of the chat analysis device 1 according to an embodiment of the present specification. In the following, an example of the hardware configuration of the chat analysis device 1 will be described, but the user terminal may also have a similar configuration.

[0028] The chat analysis device 1 includes a CPU (processor) 201 that executes various programs, a memory (main storage device) 202 that stores the various programs, and an auxiliary storage device 203 that stores various data. The CPU 201 can include one or more cores, and the memory 202 is, for example, a DRAM that includes a volatile storage area. The auxiliary storage device 203 is, for example, an HDD (Hard Disk Drive) or a flash memory, and can provide a non-volatile storage area.

[0029] The chat analysis device 1 further includes an output device 204 for presenting information to a user of the device, an input device 205 for inputting instructions, images, etc. from the user, and a communication device 206 for communicating with other devices. These are connected to each other by a bus 207.

[0030] The functional units of the chat analysis device 1 shown in Fig. 1 can be implemented, for example, by the CPU 201 operating according to a program. The CPU 201 reads and executes various programs from the memory 202 as necessary. The memory 202 stores the programs. In Fig. 2, the same names and symbols as those of the functional units shown in Fig. 1 are given to the programs.

[0031] Specifically, the memory 202 stores each program of an elderly classification estimation unit 10, a dementia / frailty risk score estimation unit 11, an estimation result integration processing unit 12, a suggested behavior creation unit 13, a suggested behavior interest / effect determination unit 14, a model training unit 15, an elderly classification estimation model 16, a score improvement behavior estimation model 18, and an interaction data collection unit 19. Each program is loaded into the memory 202 from an auxiliary storage device 203, for example, and executed by the CPU 201. Note that at least a part of the functional units may be configured with a logic circuit.

[0032] The auxiliary storage device 203 stores data referenced or managed by various programs. In Fig. 2, the auxiliary storage device 203 stores a training data DB21, a question DB22, an interaction group DB23, a response history DB24, an improvement action DB25, and an interaction data feature DB27.

[0033] The output device 204 is composed of devices such as a display, a printer, a speaker, etc. The input device 205 is composed of devices such as a keyboard, a mouse, a microphone, etc. The output device 204 presents the results of input from the user and also presents the results of processing by the chat analysis device 1. Instructions from the user are input to the chat analysis device 1 by the input device 205.

[0034] The communication device 206 receives data transmitted from other devices connected via a network including the user terminal 3, and transmits the results of processing by the chat analysis device 1 to the other devices. Note that some devices may be omitted. <Model training>

[0035] 3 shows an example of the configuration of training data 310 for the elderly classification estimation model 16. The training data 310 is included in the training data DB 21. The elderly classification estimation model 16 estimates the personality type of an elderly person from information obtained from a chat in which the elderly person participates.

[0036] 3, the training data 310 indicates explanatory variables input to the elderly classification estimation model 16 and objective variables output to the elderly classification estimation model 16. The explanatory variables are feature amounts input to the model.

[0037] Specifically, the training data 310 includes a user ID column 311 and a period column 312. The explanatory variable column includes a number of posts column 313, a post sentiment analysis column 314, a number of predictive words column 315, an average posting time column 316, and a number of post responses column 317. The objective variable column includes an elderly classification column 318. When the explanatory variables of one record are input to the elderly classification estimation model 16, the elderly classification estimation model 16 outputs an estimated elderly classification.

[0038] Each record of the training data 310 shows information about one elderly person (participant) in a chat in a specific time period in a specific social group (chat room). A social group is a chat group. Multiple participants can post messages in the chat space of the social group at will, and all participants can see the posted messages.

[0039] The user ID column 311 indicates the ID of the target user of the chat analysis device 1. The period column 312 indicates the period of the chat of each record. The number of posts column 313 indicates the number of posts made in the chat by the user indicated in the user ID column 311 during the period indicated in the period column 312. The post emotion analysis column 314 indicates the emotion of the participants obtained from the analysis of the posted content. In the example of FIG. 3, emotions are represented by "+" indicating positive or "-" indicating negative. It should be noted that it is possible to estimate emotions from text using, for example, a machine learning model based on deep learning. Emotion recognition technology is widely known, and details will be omitted.

[0040] The number of predictive words field 315 indicates the number of predictive words appearing in a post. Predictive words are, for example, predictive words for mental illness, and words that are the same as or similar to pre-set words are extracted. The average posting time field 316 indicates the average posting time of messages. The number of post responses field 317 indicates the number of responses to messages posted by others. For example, a "like" reaction corresponds to a response.

[0041] The number of posts, average time spent posting, and number of responses to posts indicate the posting behavior of the user. The results of the post sentiment analysis and the number of predictive words are values ​​based on the content of the user's posts. These pieces of information allow a more appropriate estimation of the user's state. While it is possible to estimate the user's state using only one type of information, using two types of information allows for a more appropriate estimation.

[0042] The elderly classification field 318 indicates the personality type of the user who is elderly.

[0043] The model training unit 15 uses the training data 310 to train the elderly classification estimation model 16. For example, the elderly classification estimation model 16 is configured with a deep neural network, and the model training unit 15 optimizes parameters of the elderly classification estimation model 16 by backpropagation.

[0044] Note that the explanatory variables shown in FIG. 3 are examples, and some of these may be omitted or other explanatory variables may be added. For example, the amount of variation from the average posting time at the time of previous compilation may be added, or may be replaced with the average posting time. Alternatively, the amount of variation in the number of responses to a post may be added, or may be replaced with the number of responses to a post. The elderly classification estimation model 16 may be a rule-based model different from a machine learning model.

[0045] As described above, the input of the elderly classification estimation model 16 can include variables based on the content of the user's posted message, posting frequency, text length, emotion analysis results, chat and reactions to suggested actions, and the amount of variation thereof. Note that at least some of the above examples of the input explanatory variables may be omitted, and other types of variables may be added.

[0046] FIG. 4 shows an example of the configuration of training data 350 for the score-improvement behavior estimation model 18. The training data 350 is stored in the training data DB 21. The score-improvement behavior estimation model 18 estimates behavior that can improve the dementia risk score and / or the frailty risk score based on the elderly classification and the dementia risk score and the frailty risk score. By referring to both the classification and the score, more appropriate estimation is possible. Note that the estimation of the elderly classification and its use may be omitted.

[0047] 4, the training data 350 indicates explanatory variables input to the score-improvement behavior estimation model 18 and objective variables output to the score-improvement behavior estimation model 18. Specifically, the training data 350 includes a user ID column 351.

[0048] The explanatory variable column includes a classification column 352, a dementia risk score at the time of action proposal column 353, a factor A frailty risk score at the time of action proposal column 354A to a factor E frailty risk score at the time of action proposal column 354E, an action classification column 355, an action detail column 356, a number of people in a group column 357, a percentage of the number of people posting in chats column 358, and an average number of chats per day column 359. The objective variable column includes a dementia risk score change column 360 and a factor A frailty risk score change column 361A to a factor E frailty risk score change column 361E.

[0049] Each record of the training data 350 indicates information about one user in a chat of one specific interaction group during a specific period of time. When the explanatory variables of one record are input to the score-improvement behavior estimation model 18, the score-improvement behavior estimation model 18 outputs an estimate of the dementia risk score variance and the frailty risk score variance of each factor.

[0050] The user ID column 351 indicates the user's ID. The classification column 352 indicates the user's personality type. The dementia risk score at time of suggested action column 353 indicates the user's dementia risk score when the score-improvement action estimation model 18 estimates an improvement action. The dementia risk score is a score indicating the degree of risk that the user will progress to dementia. The method of estimating the dementia risk score will be described in detail later, but it is estimated from the number of hits for preset check items.

[0051] The factor A frailty risk score at time of suggested behavior field 354A shows the frailty risk score of the user for factor A when the score improvement behavior estimation model 18 estimates an improvement behavior. The frailty risk score indicates the degree of risk that the user will transition to frailty due to factor A. Fields 354B to 354E show the frailty risk scores for factors B to E, respectively.

[0052] In one embodiment of the present specification, a plurality of factors are defined for frailty risk. As described later, examples of frailty risk factors may include physical strength, nutrition, oral cavity, isolation, and depression. Frailty risk factors may not be defined, some of which may be omitted, and other factors may be added. For example, cognitive factors may be included. Furthermore, multiple factors may be defined for dementia. As described later, in one embodiment of the present specification, the frailty risk for each frailty factor is estimated based on the number of corresponding check items for each frailty factor. This makes it possible to propose more appropriate improvement actions according to the factors.

[0053] The behavior classification field 355 indicates the classification of candidate improvement behaviors proposed by the score improvement behavior estimation model 18. For example, "exercise" indicates making suggestions about walking and exercise, "sleep" means making suggestions about sleep, "food" means making suggestions about food, and "social participation" means making suggestions about interacting with neighbors and joining other social groups (chat groups), etc.

[0054] The behavior details column 356 shows details of the improvement behavior candidates in the classification shown in the behavior classification column 355, which are proposed by the score improvement behavior estimation model 18. The ID of the detailed behavior is shown in FIG. 4. The content of the detailed behavior is registered in the improvement behavior DB 25.

[0055] In the behavioral classification of "walking", there may be behaviors such as "5000 steps / day" and "walking 1000 steps as a habit". These can be proposed as improvement actions for the frailty risk factors of "physical strength" and "depression", for example. Also, an example of an improvement action for dementia risk is "1000 steps / day". Examples of detailed behaviors in the behavioral classification of "social participation" are "participation in a proposed gathering place" and "participation in a cognitive function improvement class". These can be proposed as improvement actions for the frailty risk factors of "social participation" and "dementia risk", for example.

[0056] The number of people in a group field 357 indicates the number of people belonging to an interaction group. The chat poster ratio field 358 indicates the ratio of the number of participants who posted a message within a specified period. Here, a participant who posted even once is counted as a poster. Note that only participants who posted more than a specified number of posts may be counted. The average number of chats / day field 359 indicates the average number of posts per day in an interaction group within a specified period. The unit time for the average number may be a length other than one day. Note that the specified periods for calculating values ​​in different fields may be the same or different.

[0057] The dementia risk score change field 360 indicates the amount of change in the dementia risk score from the dementia risk score at the time of action proposal, calculated from the explanatory variables of the corresponding record. The factor A frailty risk score change field 361A to factor E frailty risk score change field 361E respectively indicate the amount of change in the frailty risk score from the frailty risk scores of factors A to E at the time of action proposal, calculated from the explanatory variables of the corresponding record.

[0058] By inputting information on different proposed actions, in this example, classification and details, into the score improvement action estimation model 18, it is possible to obtain score change values ​​for the different proposed actions. In the operation stage, the chat analysis device 1 inputs a plurality of proposed action candidates together with other data into the score improvement action estimation model 18 and obtains their score change values. By selecting a proposed action that favorably changes the score from the candidate proposed actions and proposing it to the user, it is possible to expect a reduction in the user's risk of dementia and / or frailty. Note that the information on the proposed action may be either a classification or details.

[0059] The model training unit 15 uses the training data 350 to train the score improvement behavior estimation model 18. For example, the score improvement behavior estimation model 18 is configured with a deep neural network, and the model training unit 15 optimizes parameters of the score improvement behavior estimation model 18 by an error backpropagation method. The input of the score improvement behavior estimation model 18 includes information on interaction groups in addition to the user's score. This makes it possible to more accurately estimate an appropriate interaction group for the user.

[0060] 4 are examples, and some of these may be omitted or other explanatory variables may be added. For example, at least some of the classification field 352, the group number field 357, and the chat information fields 358 and 359 may be omitted. The score improvement behavior estimation model 18 may be a rule-based model different from a machine learning model. <Action suggestion>

[0061] The following describes an example of a process of suggesting actions to a user by the chat analysis device 1. The chat analysis device 1 stores an elderly classification estimation model 16 and a score improvement action estimation model 18 trained by a model training unit 15 using training data 310 and 350 in a training data DB 21. The chat analysis device 1 analyzes chats of elderly users including answers to questions from the chat analysis device 1, and suggests actions to the elderly user that will improve their dementia risk score and / or frailty risk score.

[0062] The interaction data collection unit 19 of the chat analysis device 1 collects chat data in each user's interaction group in order to propose an action. The interaction data collection unit 19 analyzes the collected data and updates the interaction group DB 23, the response history DB 24, and the interaction data feature DB 27 based on the analysis results.

[0063] FIG. 5 shows an example of the configuration of the social group DB 23. The social group DB 23 manages information on social groups. A plurality of users belong to a social group and participate in chats of the social group. A user chats with other users in the social group to which the user belongs. A purpose and actions for the purpose are defined for each social group. Participants in the social group post about improvement actions in the chat. Each user belongs to one or more social groups.

[0064] For example, the improvement behavior and purpose of one interaction group is "improving cognitive function by making morning walking a habit." The improvement behavior and purpose of another interaction group is "improving physical strength by fast walking." The improvement behavior and purpose of another interaction group is "preventing depression by getting good quality sleep."

[0065] The interaction data collection unit 19 periodically collects, for example, interaction data between users in interaction groups in which each user participates from multiple user terminals 3, analyzes the collected interaction data, and updates the interaction group DB 23 based on the analysis results. Chat data may be collected from a system that provides a chat service. The collected interaction data includes information on each interaction group itself and chat data for each interaction group.

[0066] 5, the communication group DB 23 includes an improvement action field 231, a purpose field 232, a group member field 233, a chat message submission number ratio field 234, and an average number of chats per day field 235.

[0067] The interaction group DB 23 further includes a column showing the number of people in each elderly classification in the interaction group. In Fig. 5, a column 236 showing the number of people in classification (A) and a column 237 showing the number of people in classification (B) are shown as examples. The interaction data collection unit 19 can determine the number of people in each classification from the estimation result of the elderly classification estimation model 16 based on the past interaction data.

[0068] The improvement action column 231 indicates the action that the participants of the communication group talk about in the chat. The improvement action and purpose are defined in advance for each communication group. For example, if the improvement action is "walking", the participants post messages in the chat room (communication group) reporting the results of their own walking, useful information related to walking, responses to other people's posts, etc. The purpose column 232 indicates the purpose of the communication group. The number of people in group column 233 indicates the number of people belonging to the group.

[0069] The chat poster ratio field 234 indicates the ratio of the number of participants who posted messages within a specified period. Here, a participant who posted at least once is counted as a poster. It is also possible to count only participants who posted more than a specified number of posts. The average number of chats / day field 235 indicates the average number of posts per day in the interaction group within a specified period. The average unit of time may be different from one day.

[0070] The group number column 233, the chat poster percentage column 234, and the average number of chats per day column 235 respectively store the same type of data as the group number column 357, the chat poster percentage column 358, and the average number of chats per day column 359 in the training data 350 of the score improvement behavior estimation model 18.

[0071] The data in the improvement action column 231 and the objective column 232 are fixed for each interaction group. The data in the other columns 233 to 237 are updated as needed according to the analysis results of the interaction data collection unit 19.

[0072] 6 shows an example of the configuration of the interaction data feature DB 27. The interaction data feature DB 27 stores features to be input to the elderly classification estimation model 16 and the score estimation model 17. The interaction data collection unit 19 periodically collects interaction data between users in interaction groups in which each user participates from a plurality of user terminals 3, analyzes the collected interaction data to extract predetermined features, and stores information on the analysis results in the interaction data feature DB 27. The interaction data collection unit 19 stores data indicating the contents of the chat in the interaction data feature DB 27 or another database not shown.

[0073] The interaction data feature DB 27 includes items similar to the explanatory variables of the training data 310 of the elderly classification estimation model and the score estimation model shown in Fig. 3. Each record indicates information about posts of one user in one interaction group during a specific period.

[0074] Specifically, the interaction data feature DB 27 includes a user ID column 271 and a period column 272. It further includes a number of posts column 273, a post sentiment analysis column 274, a number of predictive words column 275, an average posting time column 276, a number of post responses column 277, and an interaction group column 278. These columns store the same types of data as the columns with the same names in the training data 310 for the elderly classification estimation model and the score estimation model. The interaction group column 278 indicates the interaction group of the chat.

[0075] In the example of FIG. 6, a user with ID1 belongs to social group A and social group B, and information about posts by the user in each social group is periodically collected and stored. A user with ID2 belongs to social group A, and information about posts by the user in the group is periodically collected and stored. Old records may be deleted in a timely manner.

[0076] The information used to estimate the dementia risk score and the frailty risk score for each factor will be described below. The chat analysis device 1 holds information on check items for calculating the dementia risk score and information on check items for calculating the frailty risk score for each factor. The chat analysis device 1 analyzes chat messages of the target elderly person, judges the check items of dementia risk and frailty risk for each factor, and determines each risk score. In one example, the number of applicable check items represents the risk score.

[0077] The chat analysis device 1 analyzes chats (posted messages) of participants, independent of the questions asked by the device 1 for estimating dementia risk and frailty risk, and determines whether the target elderly person's message indicates that the check item corresponds to the check item, and determines whether additional questions are necessary to make a more accurate judgment on the check item. The analysis of the participant's chat can increase the chances of making a judgment on the check item. The additional questions can be questions expressed in indirect or direct terms, which allows for a more accurate judgment. The sending of additional questions may be triggered by the risk score exceeding a predetermined value.

[0078] The chat analysis device 1 also posts messages of indirect questions to the chat between participants to determine whether or not a check item is met. A determination is made regarding the check item based on the target elderly person's response to the indirect question. The chat analysis device 1, for example, analyzes the response message of the target elderly person to the indirect question, and determines whether or not the message corresponds to the check item, or determines whether or not an additional question is required. Also, depending on whether or not there is a response to the indirect question, it can determine whether or not a direct additional question is required. The chat analysis device 1 may post messages other than questions for estimating the risk of dementia and frailty in the chat room (interaction group).

[0079] By using indirect questions, it is possible to increase the possibility of obtaining answers to the check items from the elderly. In addition, direct follow-up questions allow for more accurate judgment. In addition to the analysis of direct questions, it is also possible to analyze only the messages of chat participants independent of the risk estimation questions or the responses to indirect questions.

[0080] 7 shows an example of information included in the question DB 22. The question DB 22 stores a list of direct questions about frailty risk 41, a list of indirect questions about frailty risk 42, a list of direct questions about dementia risk 43, a list of indirect questions about dementia risk 44, and a risk word list 45. The information in the question DB 22 is set in advance.

[0081] 8 shows an example of the configuration of the frailty risk direct question list 41. The frailty risk direct question list 41 includes an ID column 411, a direct question column 412, a points column 413, and a frailty risk factor column 414. Each record shows information on a question sentence in direct language about a check item for estimating frailty risk.

[0082] The ID column 411 indicates the ID of the direct question regarding frailty risk. The direct question column 412 indicates a directly expressed question sentence. The points column 413 indicates points for the answer to the direct question. The answer is expected to be either YES (affirmative) or NO (negative). The addition of points means that the risk of frailty increases. The direct question is a question that obtains an answer indicating the point of frailty risk, that is, a question that directly asks YES or NO. Note that the points do not have to be two-valued, but may be multi-valued.

[0083] The frailty risk factor column 414 shows frailty risk factors estimated from the answers to each frailty direct question. Here, the frailty risk factors of physical strength, nutrition, oral health, isolation, and depression are defined as examples. The types and number of factors may vary depending on the design.

[0084] FIG. 9 shows an example of the configuration of the frailty risk indirect question list 42. The frailty risk indirect question list 42 shows one or more indirect questions for each direct question (check item) shown in the frailty risk direct question list 41. The frailty risk indirect question list 42 includes an ID column 421, an indirect question column 422, and a direct question ID column 423. Each record shows information on a question sentence in an indirect manner regarding a check item for estimating frailty risk. The indirect question is intended to indirectly obtain a YES / NO answer instead of directly asking a YES or NO to indicate a point as in a direct question.

[0085] The ID column 421 indicates the ID of the frailty risk indirect question. The indirect question column 422 indicates the question sentence expressed in an indirect manner. The direct question ID column 423 indicates the ID of one or more direct questions (check items) to which the indirect question corresponds. As shown in FIG. 9, one indirect question may correspond to multiple direct questions. An indirect question is a question intended to obtain a message from the elderly person corresponding to a YES or NO answer to the corresponding direct question, and is a roundabout expression that reduces discomfort that may be caused by the question. Indirect questions make it easier to obtain a response from the elderly person.

[0086] 10 shows an example of the configuration of the dementia risk direct question list 43. The dementia risk direct question list 43 includes an ID column 411, a direct question column 412, and a point column 413. Each record shows information on a question sentence in direct language about a check item for estimating dementia risk.

[0087] The ID column 431 indicates the ID of a direct dementia risk question. The direct question column 432 indicates a directly expressed question sentence. The points column 433 indicates points for an answer to the direct question. The answer is expected to be either YES (affirmative) or NO (negative). The addition of points means that the dementia risk increases. Note that multiple factors may be defined for dementia risk, such as frailty risk. Points do not have to be binary, but may be multivalued.

[0088] A dementia risk indirect question list 44 is defined for the dementia risk direct question list 43. The relationship between the dementia risk indirect question list 44 and the dementia risk direct question list 43 is similar to the relationship between the frailty risk indirect question list 42 and the frailty risk direct question list 41, and the above explanation applies. Here, the dementia risk indirect question list 44 is omitted from the illustration.

[0089] 11 shows an example of the configuration of a risk word list related to frailty risk and dementia risk, included in the risk word list 45. The risk word list 45 shows a combination of single or multiple words that triggers the addition of points to a frailty risk score or dementia risk, or the posting of a direct question in chat by the chat analysis device 1, in a chat between users or in a user's response to an indirect question.

[0090] 11, the risk word list 45 includes a direct question ID column 451, a point calculation risk word column 452, and an additional question trigger risk word column 453. The direct question ID column 451 indicates the IDs of the direct questions on frailty risk and the direct questions on dementia risk. The direct question ID column 451 indicates some or all of the IDs indicated in the ID column 411 of the direct question list for frailty risk 41 and the ID column 431 of the direct question list for dementia risk 43.

[0091] The point calculation risk word column 452 indicates risk words that indicate whether or not a check item in a direct question applies in chat between participants (which may include the chat analysis device 1) or answers corresponding to indirect questions. Risk words are composed of one or more words. In addition to indicating specific words, risk words can be defined using word attributes, wild cards, etc. Depending on the direct question, either YES or NO increases the risk score, and an answer that increases points is an answer that applies to the check item.

[0092] The additional question trigger risk word field 453 indicates risk words that trigger additional questions in chats between participants or answers to indirect questions. A risk word is composed of one or more words. In addition, risk words can be defined by using word attributes, wild cards, etc., in addition to indicating specific words.

[0093] Based on the detection of the trigger risk word, the chat analysis device 1 posts a corresponding indirect question or direct question in the chat room. For example, the chat analysis device 1 may post an indirect question to the user when the number of times that indirect questions of the direct question (check item) have been posted is less than a specified number, and may post a direct question when the specified number has already been reached. The specified number is 1 or a number greater than this. In response to the detection of the additional question risk word, a direct question may always be posted.

[0094] At least one of the point calculation risk word and the follow-up question trigger risk word may be defined for each of the chat between participants independent of the indirect question and the answer to the indirect question. The follow-up question trigger risk word may be defined for each of the indirect question and the direct question.

[0095] 12 shows an example of information stored in the response history DB 24. The response history DB 24 shows whether or not a user has answered a check item associated with a direct question, and a history of responses of elderly users to indirect questions and direct questions. The response history DB 24 includes a frailty risk user response history 51, a dementia risk user response history 52, a frailty risk indirect question response history 53, and a dementia risk indirect question response history 54.

[0096] 13 shows an example of the configuration of the frailty risk user answer history 51. The frailty risk user answer history 51 holds and manages answer results corresponding to direct questions (check items) on frailty risk of each user. The answers here are responses indicating YES / NO corresponding to the direct questions, and can be obtained from a message to the direct question, a message to an indirect question, or a message independent of these.

[0097] The frailty risk user answer history 51 includes a user ID column 511 , a frailty risk direct question ID column 512 , a number of indirect questions posted column 513 , a number of indirect question responses column 514 , and an answer column 515 .

[0098] The user ID column 511 indicates the ID of the user, and the frailty risk direct question ID column 512 indicates the ID of the frailty risk direct question. The indirect question posting count column 513 indicates the number of times an indirect question corresponding to the frailty risk direct question has been posted in chat by the chat analysis device 1. The indirect question response count column 514 indicates the number of times an elderly user has responded to an indirect question. The answer column 515 indicates the YES / NO answer result by the user corresponding to the frailty risk direct question. The answer is obtained from the response message to the indirect question or the risk word in another message, or from the answer to the direct question.

[0099] Here, it is assumed that the total number of posts and replies in all chat rooms (interaction groups) that the user participates in is counted. The number of posts and replies may be counted for each chat room.

[0100] Fig. 14 shows an example of the configuration of the dementia risk user answer history 52. ​​The dementia risk user answer history 52 holds and manages answer results corresponding to direct dementia risk questions (check items) for each user. The answers here indicate YES / NO answer results corresponding to the direct questions. The answers are obtained from the response message to the indirect question or the risk words in other messages, or from the answers to the direct questions.

[0101] The dementia risk user answer history 52 includes a user ID column 521, a dementia risk direct question ID column 522, a number of indirect question posts column 523, a number of indirect question responses column 524, and an answer column 525. The user ID column 521 indicates the ID of the user, and the dementia risk direct question ID column 522 indicates the ID of the dementia risk direct question.

[0102] The number of times an indirect question has been posted column 523 indicates the number of times an indirect question corresponding to the direct question about dementia risk has been posted by the chat analysis device 1. The number of times an indirect question has been answered column 524 indicates the number of times an elderly user has answered an indirect question. The answer column 525 indicates the YES / NO answer result given by the user to the direct question about dementia risk. Here, it is assumed that the total number of posts and responses in all chat rooms in which the user participates is counted. The number of times may be counted for each chat room.

[0103] The chat analysis device 1 collects data on one or more chat rooms in which the target elderly user participates, analyzes the messages of the target elderly user, and updates the frailty risk user answer history 51 and dementia risk user answer history 52 at any time. Regarding the presence or absence of a response to an indirect question, for example, a message posted within a predetermined time from the posting of an indirect question or a "Like" post to the question may be determined as a response to the question, and further, the content of the message may be analyzed to determine the relevance of the message to the question as a response to the question. In addition, the information of each user may be reset, for example, at predetermined intervals.

[0104] 15 shows an example of the configuration of the frailty risk indirect question response history 53. The frailty risk indirect question response history 53 stores information on the response history of multiple users to indirect questions about frailty risk. In the example shown in FIG. 15, the frailty risk indirect question response history 53 includes a frailty risk indirect question ID column 531 and a response ratio column 532.

[0105] The frailty risk indirect question ID column 531 indicates the ID of the frailty risk indirect question, and the response rate column 532 indicates the rate of responses from users to the indirect question. The response rate can be calculated from the number of indirect questions posted and the number of responses to all users. This information is shown in the frailty risk user response history 51. The frailty risk indirect question response history 53 provides information on indirect questions that are likely to elicit responses from users.

[0106] 16 shows an example of the configuration of the dementia risk indirect question response history 54. The dementia risk indirect question response history 54 stores information on the response history of multiple users to indirect questions about dementia risk. In the example shown in FIG. 16, the dementia risk indirect question response history 54 includes a dementia risk indirect question ID column 541 and a response ratio column 542.

[0107] The dementia risk indirect question ID column 541 indicates the ID of the frailty risk indirect question, and the response rate column 542 indicates the rate at which users responded to the indirect question posts. The response rate can be calculated from the number of indirect questions posted and the number of responses for all users. This information is shown in the dementia risk user response history 52. ​​The dementia risk indirect question response history 54 provides information on indirect questions that are likely to elicit responses from users.

[0108] Here, an example of posting of an indirect question and a direct question by the chat analysis device 1 will be described. FIG. 17 shows an example of posting of an indirect question in a group chat. FIG. 17 shows an image of a chat application on a user terminal 3 of a certain user. In FIG. 17, a dashed ellipse is added for explanatory purposes.

[0109] 17, the chat analysis device 1 posts an indirect question 611 in the chat room. This indirect question 611 may be associated with, for example, the frailty risk direct questions D1 and D2 or the dementia risk direct question 16. The post by the chat analysis device 1 may be sent to all participants in the chat room or only to specific participants.

[0110] Response messages 612 and 613 from two users are obtained in response to the indirect question 611. Therefore, the number of responses to the indirect question is counted for these two users. The chat analysis device 1 analyzes the response messages 612 and 613 based on the risk word list 45, and determines whether or not there is a point calculation risk word for dementia risk or frailty risk of each factor, or whether or not there is an additional question trigger word.

[0111] For example, in the response message 612, words such as the date "3 / 13", "walking", and the number of steps "12259" may be included in the risk words for the indirect question. Also, in the response message 613, words such as "a week in a row" and "can't go outside" may be included in the risk words for the indirect question.

[0112] For example, when a point calculation risk word is detected a predetermined number of times within a predetermined period, the chat analysis device 1 actually judges that the check item is applicable or not applicable. The predetermined number of times is an arbitrary set value of 1 or more.

[0113] The chat analysis device 1 may determine whether or not to post an additional question based on the detection of additional question risk words and non-response to indirect questions. For example, when an additional question risk word is detected a predetermined number of times within a predetermined period, or non-response to an indirect question reaches a predetermined number or more, the chat analysis device 1 may post a corresponding direct question. The predetermined number is an arbitrary value equal to or greater than 1. Alternatively, the determination may be made based on the total number of detections of additional question risk words and non-response.

[0114] The chat analysis device 1 may acquire the relationship between location information and date and time information from the user terminal 3, and determine the accuracy of the risk words in the user's response message, and depending on the radius result, may determine whether or not points should be added for dementia risk and frailty risk, or whether or not additional questions should be asked.

[0115] For example, if the response message includes a date, the accuracy of the date may trigger an addition of a dementia risk score or an additional question. In this way, the chat analysis device 1 may determine the accuracy of a response message to an indirect question based on data other than chat that can be acquired from the user terminal 3, and may determine whether to add points for dementia risk or frailty risk or ask an additional question based on the determination result.

[0116] Fig. 18 shows an example of a direct question posted in a group chat. Fig. 18 shows an image of a chat app on a user terminal 3 of a certain user. In Fig. 18, a dashed ellipse is added for explanation. In the example of chat shown in Fig. 18, the chat analysis device 1 posts direct questions 625 and 626 in the chat room according to the analysis results of user posted messages 621, 622, and 623.

[0117] For example, the user posted message 621 includes the phrase "no appetite," and the user posted message 623 includes the phrase "skipping lunch." The chat analysis device 1 posts direct-phrase follow-up questions 625 and 626 in response to these phrases. The follow-up questions 625 and 626 are direct questions about frailty risk, respectively.

[0118] In the example shown in Fig. 18, a direct question is posted to all participants in the group. In another example, the direct question may be sent only to a target user. When the number of participants who have not received an answer to the direct question to be posted is less than a predetermined number, the chat analysis device 1 may post a direct question individually to the participants who have not yet answered.

[0119] For example, the chat analysis device 1 may post a direct question to all users in the group when a risk word of a specific direct question is detected by more than a predetermined number of participants within a predetermined period and the number of participants who have not answered the direct question is more than a predetermined number. In other cases, a direct question may be sent to each user who detected a risk word. The above two predetermined numbers may be set individually and may be 0 or a number greater than 0. This explanation may also be applied to sending an indirect question triggered by a risk word.

[0120] 19A shows an example of a suggestion of an improvement action in a chat. In the example of FIG. 19A, the chat analysis device 1 posts a message 705 indicating an improvement action in a group chat. The improvement action estimated to be effective for preventing dementia and / or frailty for many participants can be shown to all participants in the exchange group.

[0121] FIG. 19B shows another display example of improvement actions suggested by the chat analysis device 1. In the example of FIG. 19B, the chat analysis device 1 presents improvement actions on a suggested action presentation screen dedicated to one participant. In this example, together with the improvement action, the current risk score of the participant and an estimated improvement score 713 based on the improvement action are displayed. With the display of this example, improvement actions can be presented only to users who need them. Participants can return a "Like" response 711 to the suggested action, and also input their actual number of steps 712.

[0122] FIG. 19C shows an example of a response of a participant to an improvement action proposed in a chat room by the chat analysis device 1. An improvement action suggestion message 705, a frailty risk score for a specific factor, and an estimated improvement score 723 are displayed. In response to this message, a "Like" 721 and a related message 722 are returned. The chat analysis device 1 may correct the score variation value of the learning data based on the responses 721 and 722 from the participants to the proposed improvement action. Similarly, the chat analysis device 1 may correct the score variation value of the learning data based on a "Like" response 711 and a change in the actual number of steps 712 (change in actual behavior in response to the action suggestion) shown in FIG. 19B.

[0123] An example of the processing of the chat analysis device 1 will be described below. FIG. 20 shows a flowchart of an example of the processing of the chat analysis device 1. In step S11, the interaction data collection unit 19 collects user interaction data (chat data) from each user terminal 3. The interaction data collection unit 19 updates the interaction group DB 23 and the interaction data feature amount DB 27 based on the collected interaction data. Data for estimating dementia risk and frailty risk, which will be described later, is also included in the collected data. The interaction data is collected during the execution of the steps described below.

[0124] The process for one elderly user in one social group is described below. The process described below can be performed for each user in each social group.

[0125] In step S12, the elderly classification estimation unit 10 estimates the type of the target elderly person based on the data stored in the interaction data feature DB 27, using the elderly classification estimation model 16. The elderly classification estimation unit 10 inputs the latest data of the target elderly person in the target interaction group to the elderly classification estimation model 16 to determine the type of the target elderly person. The input data are the explanatory variables shown in FIG. 3.

[0126] In step S13, the dementia / frailty risk score estimation unit 11 estimates the dementia risk score of the target elderly person and the frailty risk score for each factor (S13). The frailty risk score for each factor of each user is managed in the frailty risk user answer history 51, and the dementia risk score is managed in the dementia risk user answer history 52. ​​The dementia / frailty risk score estimation unit 11 updates the frailty risk user answer history 51 and the dementia risk user answer history 52 based on an analysis of the collected interaction data.

[0127] For example, the answer column 515 of the frailty risk user answer history 51 indicates information on answers corresponding to the direct questions (check items) of each user. The frailty risk direct question list 41 indicates the points of each direct question, the relationship between the answer and the factors, in columns 413 and 414. From these, the dementia / frailty risk score estimation unit 11 can obtain the frailty risk score of each factor of the user. The dementia risk score can also be obtained from the dementia risk user answer history 52 and the dementia risk direct question list 43.

[0128] The dementia / frailty risk score estimation unit 11 analyzes chats between users without sending questions. When a point calculation risk word for any direct question is detected a predetermined number of times within a predetermined period in a message from a target user, the answer column 515 or 525 of the corresponding record in the frailty risk user answer history 51 or dementia risk user answer history 52 is updated. The risk words are registered in the point calculation risk word column 452 of the risk word list 45. The predetermined number is an arbitrary setting value of 1 or more. The number of times each risk word is detected is managed for each user.

[0129] The dementia / frailty risk score estimation unit 11 may, for example, periodically post indirect questions for estimating the dementia risk score or the frailty risk score of each factor in the group chat. The indirect questions to be posted are selected from the frailty risk indirect question list 42 or the dementia risk indirect question list 44.

[0130] For example, it is possible to post indirect questions for check items selected in sequence in a round robin manner. Alternatively, check items whose indirect questions exceed a threshold may be selected based on the participants' unanswered rate, for example. Indirect questions for each check item may be selected based on a response ratio, for example. The response ratio is managed in the frailty risk indirect question response history 53 and the dementia risk indirect question response history 54. The dementia / frailty risk score estimation unit 11 may select, for example, an indirect question with the highest response rate from indirect questions that have not been sent within a predetermined period. It is to be noted that the periodic posting of indirect questions may be omitted.

[0131] The dementia / frailty risk score estimation unit 11 analyzes the responses of participants to indirect questions. When a point calculation risk word corresponding to each direct question is detected a predetermined number of times within a predetermined period, the answer column 515 or 525 of the corresponding record in the frailty risk user answer history 51 or dementia risk user answer history 52 is updated. The risk words are registered in the point calculation risk word column 452 of the risk word list 45, and in this example, are common to responses to indirect questions and user-to-user chats. The number of detections is counted by a common counter.

[0132] The dementia / frailty risk score estimation unit 11 further updates the frailty risk user answer history 51 or the dementia risk user answer history 52. ​​Specifically, in the record of the participant's ID, the indirect question posting count column 513 or 523 and the indirect question response count column 514 or 524 may be updated, and the answer column 515 or 525 may be updated. Additionally, the frailty risk indirect question response history 53 and the dementia risk indirect question response history 54 are updated.

[0133] In step S14, the dementia / frailty risk score estimation unit 11 determines whether an additional question is necessary, and if it is determined that an additional question is necessary (S14: Y), it transmits the additional question in step S15. In one example, the dementia / frailty risk score estimation unit 11 determines whether an additional question is necessary based on the risk score estimate and the detected risk word.

[0134] For example, when the frailty risk score estimate value or dementia risk score estimate value of any factor of the target user exceeds a predetermined value, the dementia / frailty risk score estimation unit 11 determines to send an additional question corresponding to the risk score estimate value. The additional question is an indirect question or a direct question. Note that only one of the risk score estimate value condition and the risk word condition may be determined.

[0135] The dementia / frailty risk score estimation unit 11 may post the additional question in the group chat. In other words, it is referred to by all participants. The dementia / frailty risk score estimation unit 11 may send the additional question to the target user individually. The dementia / frailty risk score estimation unit 11 may select the destination of the additional question according to a predetermined condition. For example, when sending the same additional question to multiple participants, the additional question is posted in the group chat.

[0136] The dementia / frailty risk score estimation unit 11 selects, as a target for additional questions, direct questions (check items) about the target factor frailty risk or dementia risk that have not yet been answered by the target user. Furthermore, the dementia / frailty risk score estimation unit 11 determines whether to send an indirect question or a direct question to obtain an answer to the direct question. For example, an indirect question may be sent to the relevant user when the number of times that indirect questions for the direct question (check item) have been posted is less than a specified number, and a direct question may be sent when the specified number has already been reached. The specified number is 1 or a number greater than that. A direct question may always be sent.

[0137] When sending an indirect question, the dementia / frailty risk score estimation unit 11 may select an indirect question to send based on a response ratio. The response ratio is managed in the frailty risk indirect question response history 53 and the dementia risk indirect question response history 54. The dementia / frailty risk score estimation unit 11 may select, for example, an indirect question with the highest response rate from indirect questions that have not been sent within a predetermined period. The response ratio may not be referred to and the selection may be made in a round robin manner.

[0138] In this example, the dementia / frailty risk score estimation unit 11 determines whether an additional question is necessary based on the detection of an additional question trigger risk word. The additional question trigger risk word is registered in a risk word list 45. The dementia / frailty risk score estimation unit 11 analyzes messages from the target user and detects risk words. When the target user detects a risk word associated with an unanswered direct question (check item) a predetermined number of times within a predetermined period, the dementia / frailty risk score estimation unit 11 sends an additional question about the direct question. The predetermined number of times is an arbitrary set value of 1 or more.

[0139] The additional question may be an indirect question or a direct question. For example, the dementia / frailty risk score estimation unit 11 may post an indirect question to the target user when the number of times that the indirect question of the direct question (check item) has been posted is less than a specified number, and may post a direct question when the specified number has already been reached. The specified number is 1 or a number greater than 1. In response to the detection of an additional question risk word, a direct question may always be posted. When sending an indirect question, the selection of the indirect question and the selection of the destination of the additional question may be the same as described above.

[0140] The dementia / frailty risk score estimation unit 11 updates the response history DB 24 according to the response results from each chat participant to the additional question.

[0141] In step S16, the dementia / frailty risk score estimation unit 11 updates the dementia risk score of the target user and the estimated values ​​of the frailty risk score for each factor by referring to the frailty risk user answer history 51 and the dementia risk user answer history 52.

[0142] In step S17, the chat analysis device 1 estimates a fluctuation estimate value of the dementia risk and the frailty risk score for each factor for each improvement action based on the elderly person type, the dementia risk score, and the frailty risk score for each factor.

[0143] The estimation result integration processing unit 12 generates data to be input to the score-improvement behavior estimation model 18. Fig. 21 shows an example of input data 400 to the score-improvement behavior estimation model 18. Items of the input data set shown in Fig. 21 are the same as the items of the explanatory variables of the training data 350 of the score-improvement behavior estimation model shown in Fig. 4.

[0144] Specifically, the input data 400 is composed of a classification field 402, a dementia risk score at the time of suggested action field 403, frailty risk score fields at the time of suggested action for each factor 404A-404E, an action classification field 405, an action detail field 406, a number of people in a group field 407, a chat poster ratio field 408, and an average number of chats / day field 409. The data periods for calculating values ​​in different fields may be the same or different.

[0145] Each record indicates one input data set. Multiple input data sets are sequentially input to the score improvement behavior estimation model 18, and the dementia risk score variation amount and the frailty risk score variation amount for each factor are output. All records of the input data shown in FIG. 21 indicate combinations of information on the target elderly person in the target interaction group and different suggested behaviors.

[0146] The estimation result integration processing unit 12 stores the type of the target elderly person in all cells of the classification column 402, acquires the dementia risk score and the frailty risk score at the time of behavior proposal for each factor, and stores them in all cells of the score at the time of behavior proposal columns 403 to 404 E. The estimation result integration processing unit 12 stores information on different actions that are set in advance in the cells of the action classification column 405 and the action detail column 406.

[0147] Furthermore, the estimation result integration processing unit 12 acquires the values ​​of the number of people in the group, the ratio of the number of people posting in chat messages, and the average number of chat messages per day for the target interaction group from the interaction group DB 23. The estimation result integration processing unit 12 stores the acquired values ​​in all cells of the group number of people field 407, the ratio of the number of people posting in chat messages field 408, and the average number of chat messages per day field 409.

[0148] Furthermore, the suggested behavior creation unit 13 sequentially inputs the input data sets (records) of the input data 400 into the score improvement behavior estimation model 18, and obtains estimated values ​​of the variation in the dementia risk score for each input data set and estimated values ​​of the variation in the frailty risk score for each factor.

[0149] In one embodiment of the present specification, the proposed behavior creation unit 13 may correct the dementia risk score and the frailty risk score for each factor based on the user's reaction to a predetermined number of past (e.g., the previous) proposed behaviors. This makes it possible to re-propose more appropriate behavior candidates. Note that the user's reaction to the previous proposed behavior does not need to be referenced.

[0150] The proposed behavior interest / effect judgment unit 14 searches for the user's reaction to the previous proposal in the chat data collected by the exchange data collection unit 19. The reaction can be obtained from the posting of "Like" to the proposed message or the contents of the message posted after the proposal. Note that the method of text analysis is widely known, so the details are omitted.

[0151] For example, when a "Like" is posted or a related message is posted, the proposed behavior interest / effectiveness determination unit 14 may determine that there is interest or an effect, and improve the dementia risk score of the same behavior as the previous time and the frailty risk score for each factor by a predetermined value. When there is no response, or when a message denying the proposed behavior is posted, the proposed behavior interest / effectiveness determination unit 14 may determine that there is no interest or effect, and worsen the dementia risk score of the same behavior as the previous time and the frailty risk score for each factor by a predetermined value. The correction value may be changed according to the number of responses or the number of related message posts. When information on the user's actual behavior is available, the presence or absence of behavior according to the proposal may be referenced.

[0152] Next, in step S18, the chat analysis device 1 executes the processes of steps S12 to S17 for each of the other participants in the target communication group. As a result, an estimated value of the variation in the dementia risk score and an estimated value of the variation in the frailty risk score for each factor for each of the other participants in the target communication group are obtained.

[0153] The suggested action creation unit 13 stores the estimated variation values ​​of the dementia risk scores of all participants in the target interaction group and the estimated variation values ​​of the frailty risk scores for each factor in an improvement score table.

[0154] FIG. 22 shows an example of the configuration of the improvement score table 450. In the example of FIG. 22, the user with ID1 is the target user, and the users with ID2 and ID3 are other participants in the target social group. The improvement score table 450 stores the dementia risk score and the estimated fluctuation value of the frailty risk score for each factor due to each improvement action of each user. "Social group A" or "Social group B" in the improvement action means participation in social group A or social group B (chat group). Other improvement actions indicate improvement actions other than participation in social groups, such as taking a walk or participating in local activities.

[0155] In the next step S19 and thereafter, the suggested behavior creation unit 13 determines the improvement behavior to be proposed and the range of users to which it is proposed. In step S19, the suggested behavior creation unit 13 determines the improvement behavior to be proposed to the target user based on the dementia risk score of the target user and the fluctuation value of the frailty risk score for each factor. For example, when the risk score fluctuation value of the improvement behavior exceeds each threshold, it is determined that the improvement behavior is to be presented to the target user. Alternatively, it may be determined based on the sum of the risk score fluctuation values ​​or a value input to a predetermined function. There may be one or more improvement actions presented. For example, an improvement action estimated to be the best from the risk score fluctuation values ​​may be selected. Note that only some of the score fluctuation values ​​may be referenced.

[0156] Next, the suggested action creation unit 13 estimates the effect range of the improvement action selected for the target user based on the statistical value of the score fluctuation estimate value of the improvement action. For example, the suggested action creation unit 13 calculates the average value and variance value of the risk score fluctuation value within the interaction group. The average value is an index indicating the improvement effect of the entire interaction group, and the variance value is an index indicating the variation of the improvement effect within the interaction group.

[0157] The average value may be a simple average value (with a common weighting coefficient of 1) or a weighted average value, and the median value may be used instead of the average value. The standard deviation or the maximum value of the difference between the fluctuation values ​​may be used instead of the variance value.

[0158] The proposed action creation unit 13 compares the average value of the risk score variation value with each threshold, and further compares the variance value of the risk score variation value with each threshold. If the average value of the risk score variation value exceeds the threshold and the variance value is less than the threshold (S19: YES), it is estimated that there is an effect on the entire interaction group. If not (S19: NO), it is determined that the effect on the entire interaction group is small.

[0159] If the improvement action is effective for the entire interaction group (S19: YES), in step S20, the suggested action creation unit 13 presents the improvement action in the group chat (S20). If the improvement action is not effective for the entire interaction group (S19: NO), in step S21, the suggested action creation unit 13 presents the improvement action only to the target elderly person. After that, the flow returns to step S13, and the score is re-estimated, the score variation value is re-estimated, and the improvement action is re-suggested.

[0160] In one embodiment of the present specification, the suggested action creation unit 13 selects a sentence to be presented to the user from the improvement action DB 25. For example, the suggested action creation unit 13 selects a message sentence from the improvement action DB 25 to propose the improvement action that has been decided to be proposed. The method of creating the message to be presented is known as a so-called chatbot technology, and the details will be omitted here.

[0161] In the above example, an improvement action indicating specific content is input to the score improvement action estimation model 18. The input improvement action may be a template indicating the content, or may be a message actually presented to the user. When a template is input, a message to be presented may be selected from the improvement action DB 25 based on the template and actual chat content.

[0162] The input to the score improvement action estimation model 18 may be only the type of improvement action. The suggested action creation unit 13 selects the text of a message from the improvement action DB 25 based on the type of improvement action decided to be proposed and the content of the message in the chat. The method of creating the message to be presented is known as the so-called chatbot technology. The suggested action creation unit 13 may process the message stored in the improvement action DB 25 according to the chat content.

[0163] In the above example, an improvement action is proposed to the target elderly person. The chat analysis device 1 may present the risk estimation value to the target elderly person or transmit it to a predetermined system without proposing an improvement action. The predetermined system may be, for example, a system of a medical institution that treats the target elderly person or a system of a family member.

[0164] The present invention is not limited to the above-described embodiment, but includes various modified examples. For example, the above-described embodiment has been described in detail to clearly explain the present invention, and is not necessarily limited to those including all of the configurations described. In addition, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. In addition, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

[0165] In addition, the above-mentioned configurations, functions, processing units, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. In addition, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the program, table, file, etc. that realizes each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card or SD card.

[0166] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0167] 1. Chat analysis device 3. User terminal 5. Network 10 Elderly Classification and Estimation Department 11 Dementia and frailty risk score estimation section 12 Estimation result integration processing unit 13 Proposal Action Creation Department 14 Proposal Action Interest and Effect Assessment Section 15 Model Training Department 16 Elderly classification estimation model 18 Score improvement behavior estimation model 19 AC Data Collection Department 21 Training Data DB 23 Exchange Group DB 24 Response history DB 25 Improvement action DB 27 Exchange Data Feature DB 201 CPU 202 Memory 203 Auxiliary storage device 206 Communication Devices

Claims

1. A processor; A storage device for storing a program executed by the processor; Including, The storage device includes: storing information on a plurality of check items for calculating a risk score including at least one of a dementia risk score and a frailty risk score; The processor, Calculating a risk score for the first participant in the exchange group based on a predetermined first-type risk word associated with each check item of the plurality of check items included in exchange data including a text message posted by the first participant in the exchange group; Determine whether or not an additional question is required based on a predetermined second-type risk word included in the interaction data or the risk score of the first participant; In response to determining that a follow-up challenge is required, sending a follow-up challenge to the first participant to update the risk score of the first participant; A computer system that updates a risk score for the first participant based on the first participant's response to the follow-up question.

2. 2. The computer system of claim 1, the storage device stores information on direct questions and indirect questions corresponding to each of the plurality of check items; The processor selects a direct question or an indirect question from the information of direct questions and indirect questions for the follow-up question.

3. 3. The computer system of claim 2, The processor, A computer system that posts, in the social networking group, an indirect question selected from the information on direct questions and indirect questions before determining whether or not an additional question is required.

4. 4. The computer system of claim 3, The storage device includes: Information on one or more code word questions corresponding to each of the plurality of check items; and storing information on a response rate for each of the one or more codeword questions; The processor selects one of the euphemism questions to send from the one or more euphemism questions based on the response ratio.

5. 2. The computer system of claim 1, The risk score includes a dementia risk score and a frailty risk score of different factors; The information on the plurality of check items includes information on one or more check items for each of the dementia risk score and the frailty risk score of different factors, The processor calculates the risk score of the first participant for each risk score, the dementia risk score and the frailty risk score for different factors, determines whether or not additional questions are necessary, sends the additional questions, and updates the risk scores.

6. 2. The computer system of claim 1, The processor proposes a risk improvement action to the first participant determined based on the risk score of the first participant.

7. 7. The computer system of claim 6, The processor, Based on the risk score, a change in the risk score due to each candidate improvement action that improves the risk score is estimated; A computer system that determines the risk improvement action to be proposed to the first participant based on the change in the risk score.

8. A method for assessing dementia risk and / or frailty risk, comprising: The system stores information on a plurality of check items for calculating a risk score including at least one of a dementia risk score and a frailty risk score, The method further comprises the steps of: Calculating a risk score for the first participant in the exchange group based on a predetermined first-type risk word associated with each check item of the plurality of check items included in exchange data including a text message posted by the first participant in the exchange group; Determine whether or not an additional question is required based on a predetermined second-type risk word included in the interaction data or the risk score of the first participant; In response to determining that a follow-up challenge is required, sending a follow-up challenge to the first participant to update the risk score of the first participant; updating a risk score for the first participant based on the first participant's response to the follow-up question.

Images