JP7224618B2 - Action recommendation device and action recommendation system - Google Patents

Description

TECHNICAL FIELD The present invention relates to an action recommendation device and an action recommendation system, and more particularly to an action recommendation device and an action recommendation system that recommend actions to be taken by a user using an artificial neural network that has learned statistical data related to interviews of the user. Regarding the system.

In Patent Document 1, the present applicant transmits n-th mission data including a plurality of missions to a plurality of user terminals, and n-th mission data transmitted from each user who has performed the n-th mission data. and the user's mission achievement rate related to the n-th mission data and the user's mission achievement rate related to the n-th mission data. a user who has performed the n+1-th mission data, and transmits to the user terminal the n+1-th mission data in which the load of at least one mission among the plurality of missions is increased when there is an inquiry item to be performed. Considering the user's mission achievement rate related to the n + 1th medical interview data and the n + 1th mission data transmitted from each, and the user's mission achievement rate related to the n + 1th mission data, the n + 1 and means for determining that the one mission contributes to the improvement of the interview item when further improvement is found in the interview item in the interview data of the second time. According to Patent Document 1, it will eventually become possible to know with a very high degree of accuracy what kind of user prefers each mission, thereby giving the user an incentive to carry out missions that are beneficial to him/herself. will work.

Japanese Patent No. 6152313

If the invention disclosed in Patent Document 1 attempts to provide users with detailed, subdivided, and more useful missions, it is conceivable to send a wider variety of interview data to the users. In this case, the number of responses to medical inquiry data sent from each user and the number of mission achievement rate receptions will increase.

Then, it is conceivable that the data processing time for the mission achievement rate and the like in the hypothesis verification device increases, and the cost also increases accordingly. In addition, the method of processing the mission achievement rate adopted in Patent Document 1 uses linear regression analysis, but if so, it can only be analyzed in the case of about 10 items as explanatory variables, and the mission achievement rate etc. If the number of processes is unexpectedly large, it will be almost impossible to process them realistically.

Accordingly, an object of the present invention is to prevent this type of problem from occurring by devising a data processing method.

The action recommendation device of the present invention is
inquiry data transmitting means (for example, communication means 210 shown in FIG. 1) for transmitting inquiry data consisting of a plurality of items to an arbitrary user;
Answer receiving means (for example, communication means 210 shown in FIG. 1) for receiving answers to each item of the medical inquiry data transmitted by the medical inquiry data transmitting means;
Input value generation means (for example, calculation means 230 shown in FIG. 1) that generates input values obtained by changing some item contents of the reply received by the reply reception means;
inputting the input value generated by the input value generating means to an artificial intelligence device having an artificial neural network trained by a plurality of sets of mission data and answers from users who have completed the mission data; , an output value obtaining means (for example, a calculating means 230 shown in FIG. 1) for obtaining an output value corresponding thereto;
mission data transmission means (for example, communication means 210 shown in FIG. 1) for transmitting mission data based on the output value acquired by the output value acquisition means to the arbitrary user;
Prepare.

According to the present invention, data based on answers returned from users are used as input values, and by using a trained artificial neural network, output values that cannot be realized by linear regression are calculated, and mission data are calculated based on the output values. is created, advanced missions can be given to each user based on the task for each user.

The mission data to be transmitted to the arbitrary user can be created based on the degree of difficulty of completing the mission data used when learning the artificial neural network and the answers when the mission data is performed. .

The artificial neural network may be an artificial neural network capable of temporal processing, including a recurrent neural network.

The inquiry data includes the user's usual eating habits, for example, the frequency of eating out, and the frequency of eating boxed lunches sold at convenience stores. In addition, the medical interview data includes some activity levels, including the presence or absence of usual headaches, the presence or absence of usual stomachaches, the susceptibility to catching colds, and the presence or absence of the habit of exercising.

In addition, the action recommendation system of the present invention is
the action recommendation device (for example, the administrator server 200 shown in FIG. 1);
an artificial intelligence device (for example, the AI device 30 shown in FIG. 1) including the artificial neural network;
Prepare.

1 is a schematic configuration diagram of an action recommendation system according to an embodiment of the present invention; FIG. FIG. 2 is a diagram showing an example of medical inquiry data and the like of each user registered in the medical inquiry data DB 400 of FIG. 1; FIG. FIG. 10 is a diagram showing an example of results from a user who has provided mission data; 3 is a flow chart showing an action recommendation operation among the operations of the administrator server 200 shown in FIG. 1; 3 is a flow chart showing an action recommendation operation among the operations of the administrator server 200 shown in FIG. 1; 3 is a diagram showing an example of a comparison table that is created by creating means 260 of FIG. 1 and notified to the user. FIG. 7 is a diagram showing an example of a transition destination screen when a recommended food menu display area 630 in FIG. 6 is selected by clicking or the like; FIG. It is a figure which shows the example of a pop-up screen at the time of selecting the "omelet with plenty of eggs" of Fig.7 (a). 7 is a diagram showing an example of a transition destination screen when the health condition display area 660 of FIG. 6 is selected by clicking or the like; FIG. FIG. 7 is a diagram showing an example of a transition destination screen when the numeric display area 650 of FIG. 6 is selected by clicking or the like; FIG. 7 is a diagram showing an example of a screen that transitions when "Method" on the upper right of FIG. 6 is clicked or otherwise selected; FIG. 3 shows a detailed article supervised by a registered dietitian. FIG. 7 is a diagram showing an example of a screen that transitions when "My List" on the upper right of FIG. 6 is clicked or otherwise selected. Fig. 11(a) is a diagram showing an example of a screen to which a transition is made when a portion of "details" in the My List display area 710 of Fig. 11(a) is selected by clicking or the like.

100 User Terminal 200 Administrator Server 210 Communication Means 220 Learning Means 230 Calculating Means 240 Registering Means 250 Numerical Means 260 Creating Means 300 Artificial Intelligence (AI) Device 400 Interview Data Database (Interview Data Database)
500 Administrator terminal

Embodiment of the invention

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Description of configuration)
FIG. 1 is a schematic configuration diagram of an action recommendation system according to an embodiment of the present invention. FIG. 1 includes a user terminal 100, an administrator server (behavior recommendation device) 200, an artificial intelligence device (hereinafter referred to as "AI device") 300, and a database such as interview data, which will be described below. 400 (hereinafter referred to as “interview data DB”) and an administrator terminal 500. FIG.

The user terminal 100 is a terminal operated by a user, and mainly receives inquiry data or mission data including a plurality of missions transmitted from the administrator server 200, and responds to the inquiry data of the user or executes the mission. Later responses are sent to the administrator server 200 as appropriate.

The user terminal 100 can be realized by a mobile phone, a smart phone, a personal computer, a video game, an IT device, an AV device, etc., which can perform wireless communication or wired communication with the administrator server 200 .

The administrator server 200 is a server (action recommendation device) managed by the administrator of the action recommendation system of this embodiment. The administrator server 200 includes communication means 210, learning means 220, calculation means 230, registration means 240, digitization means 250, and creation means 260, which will be described below. The functions of each of these means may be realized by, for example, cooperation between the CPU and the memory.

The communication means 210 transmits various data such as medical inquiry data and mission data registered in the medical inquiry data DB 400 to the user terminal 100 . Also, the communication means 210 receives various data such as answers to medical inquiry data transmitted from the user terminal 100 .

In addition, interview data consists of multiple items such as personal data such as gender, age, weight, etc., alcohol consumption, smoking amount, etc. answered by the subject or patient at the time of medical examination or first medical examination at a hospital etc. . In addition, if possible, in addition to these, blood pressure, blood sugar level, whether it is good or bad to wake up, whether you have a headache or abdominal pain, whether you have a chronic disease, how easily you catch a cold, whether you have a habit of moving your body, etc. You may make it ask for various information including. Furthermore, although the user's usual eating habits, for example, the frequency of eating out, and the frequency of eating boxed lunches sold at convenience stores, are included, it is not necessarily limited to medical interviews regarding health conditions.

In addition, it is essential that the timing of sending the interview data is when the user first uses this behavior recommendation system, and then, for example, every month or year from the date of use, or when requested by the user. It is conceivable that any or all of such as when there was In particular, if the notification is made on the same day of every month or on the same day of the same month every year, it is possible to give regular health advice according to the result of the medical examination, which is suitable for the user.

The learning means 220 uses a set of mission data (including medical interview data) and answers of users who have completed the mission (including answers corresponding to the medical interview data) accumulated in the medical interview data DB 400 as learning data. It is used to cause the artificial neural network of the AI device 300 to learn. For example, it is conceivable to input learning data through a programming API.

Here, even if researchers try to recruit monitors in order to collect interview data and their answers, mission data and effects such as achievement rates, there are usually tens of thousands of monitors. Collecting numbers is very difficult.

However, in the case of the action recommendation system of this embodiment, as long as the user can operate the user terminal 100, it is possible to collect mission data and effects such as achievement rates. In terms of rate, the total number of users will be very large.

In particular, as will be described later, desired application software is installed in the user terminal 100, and using the application software, the achievement rate for the mission data can be transmitted by the administrator server 200 without the user explicitly transmitting it. It is preferable to

Calculation means 230 changes some of the answers to the medical question data sent to user terminal 100 to AI device 300 having an artificial neural network that has been trained by learning means 220 as input values. It calculates the output value to be used.

The registration means 240 registers various data of answers to medical inquiry data and mission data transmitted from the user terminal 100 and received by the communication means 210 in the medical inquiry data DB 400 . Using the registered data, the artificial neural network of the AI device 300 can continuously learn.

The quantifying means 250 quantifies various data to be transmitted to the user terminal 100 or the administrator terminal 500 , such as quantifying the achievement rate of mission data by the user transmitted from the user terminal 100 .

The creating means 260 creates an information medium such as screen examples shown in FIGS. The screen created by the creation means 260 is transmitted to the user terminal 100 through the communication means 210 upon request from the user terminal 100 or spontaneously.

The AI device 300 utilizes a mathematical model such as an artificial neural network to specify mission data to be transmitted to the user terminal 100 . As the artificial neural network, it is preferable to use a recurrent neural network that is strong in temporal processing, but it is not limited to this.

In particular, by using an artificial neural network that is strong in temporal processing, it is possible to reduce the number of questions in medical interview data and provide more advanced recommendation information by setting unique tasks for each user.

The medical inquiry data etc. DB 400 stores a set of mission data including medical inquiry data to be transmitted to the user terminal 100 and answers of the user who has completed the mission including answers corresponding to the medical inquiry data transmitted from the user terminal 100. It is registered on a regular basis.

The administrator terminal 500 is a terminal operated by an administrator of the action recommendation system of this embodiment. The administrator terminal 500 can be, for example, a desktop personal computer, although not limited to this. The administrator operates the administrator terminal 500 to register mission data to be transmitted to the user terminal 100 in the pre-learning stage of the artificial neural network of the AI device 300 . Specifically, for example, if there is a fact that it is good to eat an apple as a migraine countermeasure, the event of having a migraine and the mission of eating an apple are linked and registered.

For example, FIG. 1 shows an example in which the AI device 300 and the medical interview data DB 400 are provided outside the administrator server 200. You may make it provide inside.

(Overview of action)
Next, the action of the action recommendation system of this embodiment will be outlined using a specific example. A more detailed description of the action of the action recommendation system of the present embodiment will be given together with the description of the operation described below.

First, for example, it is assumed that interview data such as the following five items were initially transmitted to the user terminal 100 and the answer results were obtained.
A: Do you eat salad at least once a week?
Yes; Eat No; Do not eat B: Do you use stairs more than elevators and escalators?
Yes: Use No: Do not use C: Do you drink less than 4 go a week?
Yes; less than 4 go No; 4 go or more D: Do you feel tired these days?
Yes: No No: Yes E: Have you woken up well?
Yes; Good No; Bad

In the present embodiment, all medical inquiry data can be answered with Yes/No, thereby reducing the burden on the user when answering and the burden of data processing in the administrator server 200 .

Assume that the inquiry data consisting of [ABCDE] is transmitted to the user α, and the answer from the user α is [11001], where 1 indicates Yes and 0 indicates No. Then, since user α drinks a large amount of alcohol in a week, it is conceivable that this may be the cause of persistent fatigue.

Similarly, if the same medical question data is sent to another user β, and the answer result from the other user β is [00101], the user β has a low intake of vitamins and/or lack of exercise. It is conceivable that this may be the reason why fatigue does not go away.

However, there may be cases where the causal relationship between the above-mentioned answer results of user α and user β is reversed. For example, user β may not have enough time to eat breakfast due to not having a good time waking up, and therefore may be taking less vitamins.

However, taking the case of user β as an example, encouraging user β to take more vitamins and eliminate lack of exercise may lead to improvements in the fact that fatigue does not go away. It can be said that it is useful because it has a nature, and it will be worth trying this.

Therefore, in the present embodiment, in order to make a judgment for recommending what actions should be taken by a certain user, a set of inquiry data transmitted to the user α and the answer result [11001] of the user α, Similarly, a set of the inquiry data transmitted to user β and the answer result [00101] of user β is stored in the inquiry data DB 400 by the registration means 240 for the purpose of learning the artificial neural network of the AI device 300. Register.

Furthermore, as will be described later, mission data to be recommended to users α and β is transmitted to users α and β based on these interview data, and responses from users α and β who have completed the mission data are also registered by the registration means 240. Medical interview data and the like are registered in the DB 400. To give a typical example of these mission data and answers, user α would be given mission data to reduce the amount of alcohol he drinks in a week, and an answer that he would get less tired would be obtained.

When the number of users using the action recommendation system of the present embodiment becomes very large, and therefore a large amount of medical interview data is accumulated in the DB 400 such as medical interview data, it is transferred to the artificial neural network of the AI device 300. let them learn

Here, as for the example of user β, if the artificial neural network of AI device 300 has not been sufficiently learned, data based on interview data and answers to user β is input to AI device 300. However, it is not possible to ascertain whether low vitamin intake and/or lack of exercise are likely to be the cause of persistent fatigue. If the neural network is sufficiently learned, inputting the data based on the interview data and the answers to the user β into the AI device 300 will reveal that the intake of vitamins is low and/or the lack of exercise. , it is possible to ascertain whether or not there is a high possibility that this is the cause of persistent fatigue.

Specifically, by the calculation means 230,
the only variable in item A for vitamin intake is '1' instead of '0'[101];
- only the variable for item B related to lack of exercise is '1' instead of '0'[011];
- the variables for both item A and item B are '1' instead of '0' [111],
- Variables for both item A and item B remain "0" [001],
are set as input values for the AI device 300, and each corresponding output value is obtained.

For example, in DB 400 such as interview data,
・The number of registered answers where [ABC] is [101] and [D] is [1] is 500,
・The number of registered answers where [ABC] is [011] and [D] is [1] is 500,
・When [ABC] is [111], the number of registered answers where [D] is [1] is 1000,
・When [ABC] is [001], the number of registered answers where [D] is [1] is 100,
, if [ABC] corresponding to [1000] with the highest accumulated number is [111], it can be statistically said that fatigue will disappear with a high probability.

In other words, generally speaking, when an answer to medical question data is obtained from a specific user, the calculation means 230 appropriately changes each item (each variable) such as [ABC] in the above example. If these are used as input values for the learned AI device 300, output values corresponding to these can be acquired, and missions based on them can be given to the user.

In this way, it is possible to derive the result that both low vitamin intake and lack of exercise are statistically likely causes of persistent fatigue. If the answer is that it is difficult to get rid of fatigue, we will give user γ a mission to eat salad at least once a week and use the stairs rather than the elevator and escalator. It can be said that if a mission such as "Let's do it" is given, the cause of the user γ's fatigue is statistically eliminated.

By using the learned AI device 300, a certain user is given a mission to take vitamins from vegetables, and another user is given a mission to take vitamins from fruits. It can also result in giving a mission to do so.

After assigning a mission to each user, the administrator server 200 acquires answers from the users who have completed the mission, and accumulates sets of the mission and the answers in the medical interview data DB 400 .

By the way, such a series of processing may increase the number of items of interview data in an attempt to provide a precise and subdivided more useful mission. In that case, it is no longer enough to derive what the preferred mission is from four combinations of about 2×2 as in the above example, and the amount of calculation becomes enormous.

Therefore, in the present embodiment, the learning means 220 also causes the artificial neural network of the AI device 300 to learn a set of missions including medical interview data and their answers accumulated in the medical interview data DB 400. To derive what a mission is preferable for each user, even if it involves a large amount of calculation.

Summarizing the above and generalizing, as an outline of the action of the action recommendation system of the present embodiment, first,
(1) Send detailed and subdivided interview data to multiple (many) users,
(2) Giving a mission that is considered preferable for each user,
(3) The artificial neural network of the AI device 300 is made to learn using a set of missions including interview data and answers when the missions are completed.

Next, after learning the artificial neural network of the AI device 300,
(4) An input value obtained by changing a part (variable) of an answer transmitted from the specific user is input to the AI device 300 having a learned artificial neural network, and a corresponding output value is calculated. death,
(5) selecting a mission based on the calculated output value and giving the specific user a more beneficial advanced mission;
It turns out that.

Furthermore, if a specific user accomplishes an advanced mission assigned to him/herself and then the condition of the specific user improves/improves, that fact is accumulated in the database 400 such as interview data, so the calculated output value accuracy will also improve.

In the series of explanations so far, we have given examples related to health, such as whether fatigue does not go away or whether you wake up well. , it should be noted that it can be used in a wide variety of fields by appropriately selecting the item contents of medical interview data.

For example, but not by way of limitation, an advanced instruction menu can be given to a specific athlete. Specifically, if the objective variable related to the AI device 300 is replaced with a running distance or sprint time, etc., a guidance menu is given as mission data, and if the performance obtained when it is practiced is used as an answer, what kind of performance will be obtained? It is possible to grasp whether the guidance menu should be selected. Similarly, by appropriately changing the objective variable, it is possible to specify supplements that the athlete should take.

As a completely different example, if the target variable is replaced with each employee's sales, etc., it is possible to determine which employee should be assigned to which task in order to increase the sales of the company as a whole. can also be used for Generally speaking, if the number of objective variables is relatively small despite the number of explanatory variables being relatively large, this action recommendation system can be suitably used.

(Description of application software)
Next, application software to be installed in the user terminal 100 will be described. Note that this application software does not necessarily have to be installed in the user terminal 100, and the user may appropriately select whether or not to install it.

However, to state the conclusion first, if this application software is installed in the user terminal 100, there is an advantage that the operation burden for the user is reduced, and also for the action recommendation system, it is said that arbitrary answers can be eliminated. There are advantages.

Here, for example, when the user terminal 100 is a smart phone, the smart phone normally has an image capturing means, and thus the user's face image during use of the user terminal 100 or the content of the user's meal is captured. be able to. Also, if the user terminal 100 is a smart phone, the usage time, the call duration, the number of steps taken by the user carrying the user terminal 100, or the movement route (movement distance) can be grasped using the GPS function. You can also

This application software uses the communication means of the smart phone to appropriately transmit such captured face image data, the result of image authentication, or log data of the usage time of the user terminal 100 to the administrator server 200. do.

If the imaged data of the face image transmitted from the user terminal 100 has not undergone image recognition processing, the administrator server 200 performs image processing on it and then performs matching processing or the like to determine whether or not the number of smiles has increased or the unit It is determined whether or not the number of smiles per hour exceeds a predetermined threshold, whether or not the user is eating salad, and the like.

The administrator server 200 also determines whether or not the daily number of steps has increased. As a result of these determinations, if it can be confirmed that the number of steps of the user has increased, in the above example, it is possible to determine that there is a high possibility that the difficulty of getting rid of fatigue is improving. .

Furthermore, if the user terminal 100 is a personal computer, the user can use the personal computer if it is of the type equipped with imaging means, or use a web camera connected to the personal computer if it has one. As in the case where the terminal 100 is a smart phone, the user's face image, etc., as well as the posture, the length of time the user smiles, the length of time the user is likely to talk, etc., may be sent to the administrator server 200 .

Furthermore, regarding the usage time of the user terminal 100, if this application software is equipped with a task manager monitoring function, it is possible to grasp the frequency of use of the user's keyboard or mouse, the number of programs or files used, etc. Therefore, the application software may appropriately transmit information indicating the usage time of the user terminal 100, such as captured face image data, to the administrator server 200. FIG.

In this way, by transmitting the information obtained by using the application software from the user terminal 100 to the administrator server 200 almost in real time or every several minutes to several hours, fatigue can be quickly relieved by completing the mission. It is possible to judge whether or not there is an improvement trend in the health condition such as difficulty in falling out.

If an attempt is made to obtain an answer from a user who has completed the mission data without using application software, it is conceivable that the user will have to spend more time operating the user terminal 100. , is suitable from the viewpoint of reducing the burden on the user.

In addition, as an effect obtained by using application software, the administrator server 200 can objectively judge the state based on image recognition, etc. It becomes possible to determine whether or not there is an improvement trend in the health condition. Furthermore, it is practically impossible for the user to recognize the number of smiles per unit time and transmit the number to the administrator server 300 without using application software. It is sufficient if the evaluation is sent from the user.

However, even if no application software is installed on the user terminal 100, the administrator server 200 can easily get rid of fatigue by transmitting inquiry data to the user terminal 100 and obtaining answers from the user. It is possible to judge whether or not there is an improvement trend.

Regardless of whether the application software is installed or not, if the AI device 300 continues to learn the artificial neural network by sequentially updating the answer results to the medical interview data accumulated in the administrator server 200, the user can An additional effect is also obtained in that the mission data to be given to is also improved.

Furthermore, for example, it may be possible to find a causal relationship between currently unknown missions and their effects. For example, in order to increase the number of times a user smiles per unit time, assuming that it was not known what to do, as a result of a large number of users performing some mission, If the number of 's' increases, the mission may be statistically responsible for increasing the number of smiles.

If so, as a secondary effect of using the action recommendation system of this embodiment, it is possible to investigate this causal relationship in the future, and furthermore, as a result, it may contribute to preventive medicine. Also, the range of future missions to be given to other users will be expanded.

That is, as an example of this, for example, when trying to increase the number of smiles, it is assumed that there is some kind of mission that has been granted so far, but there are some circumstances that make it difficult to carry out the mission. On the other hand, there are cases where it is easy to carry out missions that are found to increase the number of smiles by using this action recommendation system.

In this way, the user can select missions in order of effectiveness, or select missions that have a high probability of obtaining a certain effect and have a high mission achievement rate in the past. If it is possible to do so, the range of future mission choices will also expand.

(Description of data registered in medical interview data DB 400)
2 and 3 are diagrams showing examples of medical interview data and the like of each user registered in the medical interview data DB 400 of FIG. 1. FIG.

As shown in FIG. 2 as an example, items of medical interview data include "gender", "age", "weight", "blood sugar level", "blood pressure", right and wrong of "falling asleep", right and wrong of "awakening", " The items of "drinking amount" and "smoking amount" are shown. Note that these items are examples, and the presence or absence of chronic diseases, including those with headaches, may be provided.

In order to simplify input/selection of medical inquiry data by the user, five options, for example, can be prepared in the answer column of the medical inquiry data. 2 and 3 show examples in which numerical values corresponding to this are registered. Specifically, for example, for gender, "1" corresponds to male and "2" to female, and for weight, "1" is too thin, "2" is slightly thin, and "3" is. We usually have options such as "4" being overweight and "5" being overweight.

In addition, it is set so that the numerical value is higher for options that are generally considered to be good (in the above example, "normal" is 5 points, "too thin" is 0 points, "slightly thin" is 3 points, etc.). ), for example, quantifying medical interview data with a perfect score of 100 is also effective because it leads to the numerical target of the user.

The arrows shown in FIG. 3 are created based on changes in the data example shown in FIG. Here, it is shown that the quality of "awakening" improved for users with a high mission "achievement rate".

For convenience of explanation of the data registered in the medical interview data DB 400, it is shown here as if the data example shown in FIG. If it is before completion of learning of the neural network, it may be preferable not to overwrite because it is desired to increase learning data.

(Explanation on learning data)
As described above, in the action recommendation system of the present embodiment, before the learning of the artificial neural network of the AI device 300 is completed, in order to accumulate the learning data of the artificial neural network, the interview data etc. DB 400 A set of mission data containing data and its answers is accumulated.

The first mission to be given to user α for that purpose is
(1) Eat a salad at least once a week,
(2) use stairs rather than elevators and escalators;
(3) Up to 5 cigarettes per day;
Suppose there were three.

Then, it is assumed that the user α has successfully performed each mission for the first time, and then answered about the state of health. In this case, a set of the content of each mission for the first time and the answers thereto is stored in the medical interview data DB 400 .

Next, for example, a second mission with the following contents can be given to the user α. Here, an example is shown in which the second mission is a partial change of the first mission, but it may be completely different from the first mission.
(1) Eat salad at least twice a week,
(2) use stairs rather than elevators and escalators;
(3) Up to 5 cigarettes per day;

Assume that the user α is actually given the second mission described above, and then the user α also completes the second mission, and then answers about the state of health. In this case, a set of the content of each mission for the second time and the answers thereto is stored in the medical interview data DB 400 .

Next, for example, the following third mission can be given to the user α. Here, an example is shown in which the third mission is a partial modification of the first and second missions, but it may be completely different from the first and second missions.
(1) Eat salad at least twice a week,
(2) use stairs rather than elevators and escalators;
(3) limit the number of cigarettes to 3 per day;

Actually, assume that the user α is given the third mission described above, and then the user α also completes the third mission, and then answers about the health condition. In this case, a set of the content of each mission for the third time and the answers thereto is stored in the medical interview data DB 400 .

In this way, the content of the mission is changed as appropriate, and when a response is received about the health condition after that, a set of the content of each mission and the response to this is accumulated in the medical interview data etc. DB 400. . This is performed not only for user α but also for other users.

Then, in the medical interview data DB 400, sets of mission data including a large amount of medical interview data and their answers will eventually be accumulated for hundreds or thousands of users. Using these as learning data, the artificial neural network of the AI device 300 is made to learn.

Then, for example, when medical question data is transmitted to another user terminal 100 and an answer corresponding to the medical question data is returned from the user terminal 100, the calculation means 230 sends the learned AI device 300 Then, some of the answers to the interview data are changed as input values, the corresponding output values are calculated, and based on the output values, a preferable mission for obtaining some specific effect for the user terminal 100 It is possible to select data and transmit it by means of communication 210 .

In other words, the action recommendation system can statistically grasp what kind of effect will be obtained by carrying out what kind of mission as a highly accurate result that is not affected by individual differences. This brings about an effect that the mission data can be given to the user.

Secondarily, for example, it is possible to give a more effective mission to the user, so that the user has a sense of confidence that if the given mission is carried out, it will be effective with a high probability. As a result, an incentive to carry out the mission will work, and as a result, the action recommendation system will be able to operate in a good cycle such that the accuracy of the answers from the user fed back to the administrator server 200 will increase. .

Next, periodically or irregularly, for example, every week, the learning means 220 uses a set of the mission data including the medical interview data accumulated in the DB 400 such as the medical interview data and the answers to the mission data as learning data. Train an artificial neural network.

Note that the number of missions does not have to be three, and may be about five. However, if there are too many, the mission achievement rate may decrease, so it is preferable to set it to about 3 to 5.

(Description of operation)
FIGS. 4 and 5 are flow charts showing an action recommendation operation among the operations of the administrator server 200 shown in FIG. FIG. 4 shows the procedure before completing the initial setting among the operations related to action recommendation. Note that each process shown in FIGS. 4 and 5 is performed for each user.

<Processing before completion of initial settings>
First, the user operates the user terminal 100 to download the application software for the action recommendation system as described above and install it on the user terminal 100 . After that, when the user activates the application software, inputs communication information such as name, gender, age, e-mail address, etc., and taps a send button, the administrator server 200 is requested to send medical interview data. is done.

In the administrator server 200, when the communication means 210 receives the transmission request of the medical interview data from the user terminal 100 (step S11), the registration means 240 stores the user's communication information and the like transmitted together with the transmission request in the medical inquiry data DB 400. , and the medical interview data prepared in advance is transmitted to the user terminal 100 that requested the transmission (step S12).

After that, the user terminal 100 receives the medical interview data, and if the user inputs/selects his own medical interview data to the user terminal 100 and taps the send button, the user terminal 100 sends the administrator server 200 , the medical interview data of the user is transmitted. Of course, the inquiry data may be transmitted to the user terminal 100 by a push type instead of such a pull type.

In the administrator server 200, when the communication means 210 receives the medical inquiry data (step S13), the registration means 240 registers the medical inquiry data in association with the communication information of the user already registered in the medical inquiry data DB 400. (Step S14).

By the above processing, processing for initial setting such as member registration in the action recommendation system of the present embodiment is completed. Note that the initial setting shown in FIG. 4 may be performed once for each user.

<Procedures after completion of initial settings>
FIG. 5 shows the procedure after the initial setting up to step S14 in FIG. 4 is completed. First, the administrator server 200 selects mission data from the mission data registered in the medical interview data DB 400 based on the response to the received medical interview data (step S21).

Here, when the artificial neural network of the AI device 400 has not yet undergone sufficient learning processing, such as when the operation of the action recommendation system of this embodiment has just started, responses to the received interview data , the mission data registered by the administrator using the administrator terminal 500 will be transmitted. In the above example, if the user has a migraine headache, the mission data of eating an apple corresponding to the event of having a migraine headache is to be transmitted.

On the other hand, when the AI device 400 has sufficiently learned the artificial neural network, the calculation means 230 outputs to the trained AI device 300, among the results transmitted from the user who has completed the mission data, The corresponding output value is calculated by using the input value with some changes.

Since this point is important, it will be explained once again using another example. for example,
・Statistical data A that eating an apple at least once a week alleviates migraine headaches,
Statistical data B that migraine headaches are relieved even if you do light exercise 2 to 3 times a week,
・Statistical data C that migraines almost never occur if you eat an apple once a week or more and do light exercise several times a week,
existed.

Moreover,
Regarding statistical data A, it is a fact that migraines are relieved in 30% of people,
Regarding statistic B, the fact is that 40% of people experience migraine relief,
Regarding statistic C, the fact is that 95% of people experience migraine relief,
and

Then, from the user α who has completed the initial settings,
・I don't eat much apples (variable 0),
・There is a lack of exercise (variable 0),
・ Migraine headaches,
Suppose there is an answer.

Here, if a person selects a mission, typically, based on statistical data C, for user α, eat an apple at least once a week and do light exercise several times a week. It is expected that the mission will be selected. However, in this embodiment,
(1) A variable 0 indicating that apples are rarely eaten is set to a variable 1 and input to the AI device 300 having a trained artificial neural network,
(2) Input the variable 0, which indicates a lack of exercise, to the AI device 300 having a learned artificial neural network, with the variable 1 as the variable 1;
(3) Both of these variables 0 are set to variable 1 and input to the AI device 300 having a trained artificial neural network.
That is executed by the calculation means 230 .

Then, since the AI device 300 is not human, even if the effect itself of the statistical data A to C cannot be recognized, according to the calculation means 230,
In the case of (1), an output value corresponding to the effect on 30% of people is obtained,
In the case of (2), an output value corresponding to the effect on 40% of people is obtained,
In the case of (3), an output value corresponding to the effect on 95% of people is obtained.

That is, the above statistical data A to C are analysis results obtained by regression analysis, respectively, and each output value is an analysis result obtained by performing deep learning on the learned AI device 300. Theoretically they are equivalent.

Therefore, based on these output values, the administrator server 200 can grasp that (2) is preferable to (1) and (3) is preferable to (2). Then, the mission of eating an apple corresponding to (3) once a week or more and doing light exercise several times a week is selected.

The mission data is transmitted to the user terminal 100 by the communication means 210 (step S22).

Assume that the user then performs a mission given by the administrator server 200 . Typically, when the above-described application software is not installed in the user terminal 100, the user operates the user terminal 100 to provide a corresponding answer when receiving new inquiry data. It can be entered/selected and sent to the administrator server 200 .

On the other hand, typically, when the above-described application software is installed in the user terminal 100 and the application software is activated, the user's step count or movement using the GPS function is performed on the user terminal 100 as appropriate. The distance and the like can be acquired and transmitted to the administrator server 200 .

When the answer from the user is received, the administrator server 200 registers a set of the answer and the mission data related to the answer in the medical interview data DB 400 (step S23). After that, it is determined whether or not the accumulated number of sets of mission data and their answers in the medical interview data DB 400 is equal to or greater than a predetermined threshold (step S24). Note that the threshold for mission selection can be set by the administrator. The next mission data is transmitted to the user terminal 100 by the communication means 210 .

As a result of the determination, if the set of mission data and its answers in the medical interview data DB 400 is not equal to or greater than a predetermined threshold value, the process proceeds to step S21 and the next mission data is selected. Note that the n-th mission data and the (n+1)-th secret letter data may have the same content.

On the other hand, as a result of the determination in step S24, if the accumulated number of sets of mission data and their answers in the medical interview data DB 400 is equal to or greater than a predetermined threshold, the learning means 220 determines whether the mission data accumulated in the medical interview data DB 400 A learning process is performed on the AI device 300 using the set of data and its answers (step S25).

By the above processing, using the AI device 300 having a trained artificial neural network, it is possible to assign a desired mission to the user. Artificial neural networks can also be trained continuously.

Subsequently, the creating means 260 performs various processes for displaying information on the displays of the user terminal 100 and the administrator terminal 500 so that the user and administrator can browse the information described later, and creates display data.

6 to 11 are diagrams showing examples of screens created by the creating means 260 shown in FIG. FIG. 6 shows an example of the TOP screen displayed after the user Taro logs in through the web page in order to receive the service provided by this action recommendation system. As for information necessary for login, a login ID and a password may be given to the user at the time of initial setting as described above.

Here, as an example, a comment display area 610 showing comments on Taro, an article display area 620 showing feature articles for all users, and a meal menu (mission data) recommended for Taro are displayed. A recommended food menu display area 630, a topics display area 640 showing various topics, a numerical value display area 650 showing the achievement rate of Taro's mission data, etc. and a health condition display area 660 in which values and ranks are shown.

Contents displayed in the comment display area 610 , the recommended food menu display area 630 , the numerical value display area 650 and the health condition display area 660 are mainly created by the digitizing means 250 . The contents displayed in the article display area 620 and the topics display area 640 are mainly input through the administrator terminal 500 .

FIG. 7A shows an example of a transition destination screen when the recommended food menu display area 630 in FIG. 6 is selected by clicking or the like. Here, the types of meals such as "Western-style" and "Japanese-style" are shown, and "Western-style" is selected by Taro. It shows the state where the menu is displayed. Each menu also displays a numerical value indicating how the pseudo annual income displayed in the numerical display area 650 and the health condition display area 660 changes when the menu is eaten.

These menus themselves may be unique to Taro or may be common to all users. However, since the menu to be presented is mission data for Taro, it is not necessarily presented to users other than Taro. Specifically, these menus will be presented to Taro and users of similar characteristics to Taro.

The figures attached to the menu are the basis for the above-mentioned pseudo annual income. This numerical value is calculated by the administrator server 200 by referring to the AI device 300 or the interview data database 400 based on the mission achievement rate, that is, the degree of difficulty, and the effect of the mission achievement. 500 can be calculated.

In general, the higher the difficulty and the higher the effect, the higher the numerical value. When utilizing such a rule, even the most effective mission does not always have the highest value in the presented menu if its achievement rate is low.

FIG. 7(b) shows an example of a pop-up screen when "omelet with plenty of eggs" in FIG. 7(a) is selected. The content of this pop-up screen example is the same as the content of FIG. 10(a) described later. However, they may have different contents.

FIG. 8 shows an example of a transition destination screen when the health condition display area 660 in FIG. 6 is selected by clicking or the like. Here, "¥7,400,000" is displayed in the upper left of the screen as a figure corresponding to the pseudo annual income corresponding to Taro's current health condition. In the middle left of the screen and the bottom left of the screen, as the grounds for the pseudo annual income, whether it was obtained by improving diet or by improving lifestyle habits is added.

On the right side of the screen, the evidence based on the amount of usage of the keyboard, mouse, etc. (Key, Mouse, respectively), the number of files opened, the number of programs used, etc. (File, Program items), and the grounds based on work hours, number of smiles, number of steps, distance traveled, etc. (items of Time, Smile, Walk, and GPS, respectively) are also shown. However, the display contents shown in FIG. 8 are examples, and information other than these may be shown as the basis, or information different from these may be displayed.

Here, it can be seen that the number of steps decreased while the usage of the keyboard increased. If the user has a desk job and is busy with work and cannot secure sufficient exercise time, such a transition in performance will occur. At this time, it is also possible to add a function of asking the user terminal 100 from the action recommendation system, "The number of steps has decreased, but are you busy?"

Taking the number of steps as an example, by measuring the number of steps on a daily basis, the average number of steps, the maximum number of steps, and the minimum number of steps per unit period such as every day, week, and month are acquired, and the average number of steps is used as a reference. , the activity level is determined as high or low, and a high day is "¥1,000", and a low day is "¥-1,000".

In addition, if the number of steps of the user on holidays, typically on weekends, is also acquired, the number of steps on holidays is generally considered to be less than the number of steps on weekdays. Another method would be to separately manage the number of steps and the number of steps on weekdays.

Also, if the number of steps exceeds the maximum on a day, or if the number of steps increases by, for example, 20% or more relative to the average number of steps, it can be estimated that there has been a sudden behavior described later. In this case, for example, "¥+2,000" can be used. On the other hand, when the number of steps is less than the minimum number of steps on a day, or when the number of steps is decreased by, for example, 20% or more from the average number of steps, it can be estimated that there was a visit to the hospital, which will be described later. In this case, for example, it can be "¥-2,000".

It should be noted that although forcing the user to perform manual input will impose a burden on the user, a user carrying a physical pedometer separately without using the pedometer function of the smartphone can use the pedometer through the user terminal 100 . It is not excluded to manually input the number of steps measured in and send it to this action recommendation system.

If the user terminal 100 is a video game, an IT device, an AV device, or the like, the number and frequency of use of these devices may be used to measure the effectiveness of mission execution, and the information shown in FIG. 8 may be substituted for these. Specifically, the IT device by the user, such as turning on the power of the IT device (and the time until it is turned off), logging in via the network (and the time until logging out). A numerical value corresponding to the pseudo annual income can be displayed based on the number of actions for the like.

FIG. 9 shows an example of a transition destination screen when the numeric display area 650 in FIG. 6 is selected by clicking or the like. It should be noted that even when the part labeled "see chart" in FIG. 8 is selected by clicking or the like, the screen example shown in FIG. 9 is displayed.

Here, it is an example of a screen showing the use transition of the keyboard, mouse, etc. for one week, etc., shown on the right side of the screen in FIG. Here, usage transitions and the like are shown in line graphs or stacked bar graphs according to the characteristics of the grounds, but the types of graphs are not limited to these, and the user can customize and select the display form as appropriate. can be Furthermore, instead of using a graph, the answers to the missions are represented by dots such as black and white, and the missions that are highly related to the answers and the missions that are not related to the answers are color-coded. It is also a method to show

FIG. 10(a) shows an example of a screen that transitions when "Method" on the upper right of FIG. 6 is clicked or otherwise selected. FIG. 10(b) shows an example of a screen that transitions when "details" in the article in FIG. 10(a) is clicked or otherwise selected.

The screen example in FIG. 10(a) presents Taro with the mission "Eat 3 eggs a day", and if this mission is accomplished, the number corresponding to the above-mentioned pseudo annual income is displayed. shows how much it increases. In addition, the expected effects and reasons for eating eggs, recommended articles for Taro, egg-related recipe information, shop information, and the like are displayed.

FIG. 10(b) shows a detailed article supervised by a registered dietitian. This example shows what the protein contained in eggs means to the body, the good points of omelettes and steamed rice bowls as menus, and simple cooking methods.

FIG. 11(a) shows an example of a screen that transitions when "My List" on the upper right of FIG. 6 is clicked or otherwise selected. FIG. 11(a) shows a mylist display area 710 for sending a message when a mission is completed, and a result display area for displaying the response results based on the frequency of each mission sent through the mylist display area 710. 720, a breakdown display area 730 showing the breakdown of the entire missions that have become a habit, and a mission display area 740 showing a list of missions assigned to Taro.

The mission display area 740 displays date information of each of the multiple missions presented to Taro so far, in addition to the mission "Eat 3 eggs a day" described with reference to FIG. , information indicating the frequency of mission execution by asterisks, and information attached to the date information on which mission execution became a habit is also displayed.

FIG. 11(b) shows an example of a screen that transitions when the "details" section in the My List display area 710 of FIG. 11(a) is selected by clicking or the like. Here, there are guidelines for determining the number of stars that indicate the frequency of mission execution, the input method, and how the pseudo annual income changes depending on the number of stars determined. There is an explanation that

If screen examples as shown in FIGS. 6 to 11 can be given to the user, the following effects can be expected.
・Since it is possible to provide each user with a unique mission that is suitable for each user, each user can be expected to achieve excellent results by completing the mission assigned to him/herself.
・By providing missions to a large number of users and obtaining the results of each mission as feedback, it is possible to increase the accuracy of effect measurement and to assign suitable unique missions according to the characteristics of users.
・By presenting the user with changes in familiar behavior (keyboard usage frequency, etc.) after completing the mission, let the user understand the effect of completing the mission and continue the mission to the user. incentive can be given.
- By devising a user interface (presentation of pseudo annual income, etc.), it is possible to keep the user motivated to accomplish the mission.

As described above, in the present embodiment, the behavior recommendation system has been described mainly by taking the improvement of the user's health condition as an example. sleep time, meal amount, trip to the hospital, sudden behavior, etc. can be used to measure the effect of mission accomplishment, so for example, the user may be able to select items to be displayed on the right side of FIG. .

Incidentally, just to be sure, if the user is sick, moving to the hospital is a positive action to try to cure it, so it can be said to be a desirable action. In addition, in order for this action recommendation system to appropriately judge the user's health condition, it is desirable to grasp the history of illness, hospital visits, and the presence or absence of chronic diseases. is also one method.

In this way, for example, when the user wishes to view the history of disease onset at any timing, it is possible to respond. The presence or absence of movement to a hospital can be predicted based on the GPS function and map application software provided in a smartphone or the like.

In addition, sudden actions include, for example, actions such as going on a trip or watching a sports game or watching a movie that the user does not usually do. Such behavior is considered to have a positive effect on the user's mind and body as a result, and can be said to be a behavior with a high degree of activity. For such behavior, the numerical value corresponding to the pseudo-annual income should be increased. The presence or absence of sudden behavior may also be determined by input from the user or using a smartphone or the like, as in the case of hospital visits.

In addition, it is preferable to use application software installed in a smartphone to determine the presence or absence of a sudden action, because it is possible to obtain where and what the user is doing without troubling the user. However, for example, when both playing tennis and watching tennis are selected as items to be displayed on the right side of FIG. is playing tennis or watching tennis.

Therefore, in this embodiment, when the GPS function and the map application software identify that the user is on the tennis court, the above identification is performed in consideration of the step count information obtained from the pedometer. there is Specifically, when playing tennis, the possibility of having a smartphone or the like in one's pocket is not zero, but this possibility is not considered to be very high. If the counter does not count the number of steps, it should be determined that the user is playing tennis, and if not, it should be determined that the user is watching tennis. Also, if it is difficult to determine any of these, the user may be selectively inquired about.

Similarly, for example, when the user moves only within a predetermined range for a certain period of time or longer, such as when the user is browsing exhibits such as paintings in an art museum, the GPS function and the map application software enable the user to It is also possible to determine the level of activity by specifying the current location of the user.

Furthermore, in recent years, some department stores, for example, have a futsal field on the roof in order to make effective use of the roof of the building. Normally, it would be difficult to determine whether a user is visiting a department store for shopping or futsal. In addition to detecting the presence or absence of a slight movement of the smartphone or the like in a short period of time such as 1 to 2 minutes, the above determination is performed.

Specifically, for example, if the GPS function of the smartphone, etc., does not confirm the movement of the smartphone, etc. for 10 minutes, it is highly likely that the user is playing futsal with the smartphone, etc. in the bag. Since it can be assumed, it is determined that futsal is being played in this case.

On the other hand, if the movement of the smartphone or the like is confirmed by the GPS function of the smartphone or the like every 1 to 2 minutes, for example, the user puts the smartphone or the like in the bag and visits several tenant shops, etc. Since it can be assumed that there is a high possibility that the user is looking at the product while moving, it is determined that the user is shopping in this case.

However, although it is optional, there are cases where it cannot be immediately determined which one is applicable. In such a case, the user may be inquired as in the case of the tennis example. It does not matter how the inquiry is made. If the judgment cannot be made, the inquiry may be made immediately or after a certain period of time has passed.

Furthermore, since it is highly likely that users who behave in the same pattern on a regular basis are likely to play futsal on a regular basis, the inquiry to the user can be made by saying, "Good ○ ○" when requesting the transmission of medical interview data. ○Are you playing futsal on the roof of the department store?"

If the user answers that "I play futsal for one hour from 15:00 every Saturday on the roof of XX department store", then the GPS function and the map tell that the user went to XX department store. If it is confirmed by the application software, it is possible to grasp whether or not the user has completed the mission data and the type without performing unnecessary determination processing in this time period.

Claims (5)

inquiry data transmission means for transmitting inquiry data consisting of a plurality of items for which a causal relationship is presumed to an arbitrary user;
an answer receiving means for receiving an answer to each item of the medical inquiry data transmitted by the medical inquiry data transmitting means;
When one answer received by the answer receiving means includes a numerical value indicating a negative cause and a result , the input value is all patterns obtained by changing the item that is the numerical value indicating the negative cause. an input value generating means for generating as
Each input value generated by the input value generating means is converted into a numerical value indicating a positive result by a plurality of sets of mission data and responses to each item of medical interview data sent to the user who performed the mission data. output value acquisition means for inputting to an artificial intelligence device having an artificial neural network in which the number of registrations including
mission data transmission means for transmitting mission data corresponding to an input value relating to the maximum value among the output values acquired by the output value acquisition means to the user who has given the one answer;
action recommendation device. The mission data transmitted to the arbitrary user includes the degree of difficulty indicating the achievement rate of mission data used in learning the artificial neural network and interview data transmitted to the user who performed the mission data. 2. The action recommendation device according to claim 1, wherein the action recommendation device is created based on the answers to each item. 2. The action recommendation device according to claim 1, wherein said artificial neural network is an artificial neural network capable of temporal processing including a recurrent neural network. 2. The action recommendation device according to claim 1, wherein said inquiry data includes the user's usual lifestyle habits and health condition. a behavior recommendation device according to claim 1;
an AI device including the artificial neural network;
An action recommendation system comprising:

Images