JP7450304B2 - Child support systems, methods and programs - Google Patents

Description

TECHNICAL FIELD The present invention relates to child support systems, methods, and programs.

Traditionally, in compulsory education schools (elementary schools and junior high schools), teachers, school social workers, etc. screen children who may be at risk of poverty, etc. (also called screening), and provide various programs for the screened children. Support is being provided. In such support for children, AI (Artificial Intelligence) is utilized (see Non-Patent Document 1).

Yomiuri Shimbun, "AI learns abuse records, predicts risks to children... Supports child welfare staff in making decisions about temporary protection," [online], November 22, 2021, Internet <https://www. yomiuri.co.jp/national/20211122-OYT1T50124/＞

However, with conventional methods, there is no feedback as to whether or not support for a child has been refused, and even if support for a child is refused and a death due to abuse occurs, it is difficult to improve subsequent support for the child. I couldn't connect it.

Therefore, the present invention aims to provide support to children that reflects the results of support provided to children in the past.

A child support system that is an embodiment of the present invention uses a child status acquisition unit that acquires the child's status, and a model that infers support suitable for the child from the child's status. an inference unit that infers support suitable for the child based on the child's condition; a support result acquisition unit that acquires information regarding the implementation status of the inferred support; and a relearning unit that generates a model in which the execution status of the support is reflected by relearning the support.

Further, the method of the present invention is a method executed by a computer, which includes the steps of acquiring the child's condition and using a model that infers support suitable for the child from the child's condition. the steps of: inferring support suitable for the child from the state; obtaining information regarding the implementation status of the inferred support; and relearning the model using the information regarding the implementation status of the support. and generating a model that reflects the support execution status.

Further, the program of the present invention uses a child state acquisition unit that acquires the child's state, and a model that infers support suitable for the child from the child's state, and uses the obtained child state to support the child. an inference unit that infers support suitable for the support; a support result acquisition unit that acquires information regarding the execution status of the inferred support; and a support result acquisition unit that re-learns the model using information regarding the execution status of the support. This is a program that functions as a relearning unit that generates a model that reflects the execution status.

According to the present invention, support for children can be provided that reflects the results of support provided to children in the past.

1 is a diagram showing an overall configuration according to an embodiment of the present invention. FIG. 2 is a functional block diagram of a child support server according to an embodiment of the present invention. FIG. 2 is a sequence diagram of overall processing according to an embodiment of the present invention. 3 is a flowchart of learning processing according to an embodiment of the present invention. 7 is a flowchart of inference processing according to an embodiment of the present invention. 3 is a flowchart of relearning processing according to an embodiment of the present invention. FIG. 2 is a diagram for explaining an inference model according to an embodiment of the present invention. FIG. 1 is a hardware configuration diagram of a child support server according to an embodiment of the present invention.

Embodiments of the present invention will be described below based on the drawings.

<Explanation of terms>
・“Child” under the Child Welfare Act refers to a person up to the age of 18. The present invention is intended for children aged 0 to 18, including elementary and junior high school students.

<Overall configuration>
FIG. 1 is a diagram showing the overall configuration according to an embodiment of the present invention. The child support system 1 includes a child support server 10, a sensor 20, and a school terminal 30 used by teachers, school social workers, and other school staff working at the school. Each will be explained below.

<<Child support server>>
The child support server 10 is a server that screens children who should receive various types of support. The child support server 10 is composed of one or more computers. The child support server 10 can send and receive data to and from the sensor 20 and the school terminal 30 (terminal operated by the teacher 32 and school social worker 31) via any network.

Specifically, the child support server 10 acquires data indicating the state of the child 21 from the sensor 20 and the school terminal 30, and uses a model to determine whether support is necessary or not and whether support is appropriate for the child 21 based on the state of the child 21. The school terminal 30 infers the support provided and presents the result of the inference to the school terminal 30. In addition, the child support server 10 acquires data indicating the results of support for the child (specifically, information regarding the implementation status of the support) from the sensor 20 and the school terminal 30, and uses the result of the support for the child. By relearning the model, a model that reflects the results of support for the child is generated.

<<Sensor>>
The sensor 20 is a sensor that collects information for identifying the condition of the child 21 and the result of support. For example, the sensor 20 may include a terminal equipped with a GPS (Global Positioning System) function that measures the position of the child 21 (e.g., the child's 21 smartphone), an acceleration sensor that measures the walking speed of the child 21 (e.g., the child's 21 (such as a smartphone), an image sensor that photographs the child 21 (for example, a digital camera installed at various locations, etc.), an eye tracker that measures the viewpoint of the child 21, and the like.

[Children's behavioral data]
For example, the child support server 10 may determine whether or not the child 21 is present at the location where support is to be provided, the child's 21's behavior, such as the location of the child's 21 detour, based on the child's 21's position acquired from GPS. I can do it. Furthermore, the child support server 10 can determine the behavior of the child 21, such as whether the child 21 is taking a detour, based on the walking speed of the child 21 acquired from the acceleration sensor. Furthermore, the child support server 10 determines the behavior of the child 21, such as whether or not the child 21 is present at the location where support is to be provided, and where the child 21 takes a detour, based on the image of the child 21 acquired from the image sensor. be able to.

[Children's psychological state data]
For example, the child support server 10 can determine the psychological state of the child 21 based on the child's 21's line of sight acquired from the eye tracker.

<<School terminal>>
The school terminal 30 is a terminal operated by a teacher 32 or a school social worker 31. For example, the school terminal 30 is a personal computer, a tablet, a smartphone, or the like. The school terminal 30 can send and receive data to and from the child support server 10 via the school network. The school terminal 30 can receive the screening service (specifically, presentation of support suitable for the child) provided by the child support server 10 on an application (a web application or a local application) or a website.

Specifically, the school terminal 30 transmits data indicating the state of the child 21 to the child support server 10. Furthermore, the school terminal 30 receives data from the child support server 10 indicating whether or not support is necessary and which support is appropriate for the child, as inferred by the child support server 10 . In addition, the school terminal 30 sends data indicating the status of support to the child (specifically, information on whether or not the execution of support has been refused, the degree of support execution, and the result of support execution) to the child support server. Send to 10.

<Functional block>
As shown in FIG. 2, the child support server 10 includes a learning data acquisition unit 101, a learning unit 102, an inference model storage unit 103, a child status acquisition unit 104, an inference unit 105, and a support result acquisition unit 106. , a relearning unit 107, and the child support server 10 includes a learning data acquisition unit 101, a learning unit 102, a child status acquisition unit 104, an inference unit 105, and a support result by a computer executing a program. It functions as an acquisition unit 106 and a relearning unit 107. Each will be explained below.

The learning data acquisition unit 101 acquires learning data (specifically, input (explanatory variable) is "child's condition", and output (objective variable) is "necessity of support and support suitable for the child"). (learning data).

[Learning data]
For example, the learning data is data that indicates a child's condition and appropriate support for the child, as determined by an experienced school social worker. Further, for example, the learning data is data indicating the state of the child and the support when the support for the child was executed (that is, the execution of the support for the child was not refused).

The learning unit 102 performs machine learning using the learning data acquired by the learning data acquisition unit 101 and generates an inference model. The inference model is a model that, when the child's condition is input, outputs the necessity of support and support suitable for the child. The learning unit 102 causes the inference model storage unit 103 to store the inference model.

The child status acquisition unit 104 acquires the status of the child 21. For example, the child status acquisition unit 104 acquires the child's status from the school terminal 30. Further, for example, the child condition acquisition unit 104 acquires information for identifying the condition of the child 21 collected by the sensor 20 from the sensor 20, and determines the condition of the child.

The inference unit 105 uses a model (that is, an inference model generated by the learning unit 102 and stored in the inference model storage unit 103) that infers the necessity of support and support suitable for the child from the child's condition. The necessity of support and support suitable for the child are inferred from the child's state acquired by the child state acquisition unit 104. Further, the inference unit 105 presents (for example, displays) the inference result on the school terminal 30.

Specifically, the inference unit 105 can screen the child (that is, infer whether support is necessary based on the child's condition). Further, the reasoning unit 105 can present a plurality of supports suitable for the child who should receive support in order of priority.

[Screening analysis]
For example, the child support server 10 uses a decision tree analysis method to present a plurality of questions to the school terminal 30, and based on the checks from the school terminal 30, performs a screening analysis (that is, whether support is necessary or not). and make inferences about appropriate support for the child.

The support result acquisition unit 106 acquires information regarding the execution status of the support inferred by the inference unit 105. Further, the support result acquisition unit 106 can acquire the fact that the support inferred by the inference unit 105 has been performed, and the state of the child after the support inferred by the inference unit 105.

For example, the support result acquisition unit 106 acquires the support result from the school terminal 30. Further, for example, the support result acquisition unit 106 acquires information for specifying the support result collected by the sensor 20 from the sensor 20, and determines the result of the support.

Note that if the support inferred by the inference unit 105 has not been executed, the support result acquisition unit 106 may present the school terminal 30 with a request to execute the next most suitable support. Furthermore, the support result acquisition unit 106 may notify an administrative agency or the like that the support has been rejected and the support has been changed.

The relearning unit 107 relearns the model (that is, the inference model generated by the learning unit 102 and stored in the inference model storage unit 103) using information regarding the execution status of support for the child, thereby improving the support situation. Generate a model that reflects the Furthermore, the relearning unit 107 can retrain the model using the fact that the support for the child has been performed and the state of the child after the support for the child.

In other words, the relearning unit 107 weights (changes parameters) so as not to infer rejected support, and infers support that was not rejected (and also infers support in which the child's condition has improved after the support). weight (change parameters) to make an inference).

[Relearning data]
For example, the relearning data is data indicating the child's condition and the support when the execution of support for the child is refused. Further, for example, the relearning data is data indicating the state of the child and the support when the support for the child was executed (that is, the execution of the support for the child was not refused).

<<Inference using retrained inference model>>
The inference unit 105 uses the model generated by the relearning unit 107 (that is, the model that reflects the support situation for the child) to provide support suitable for the child based on the child's condition acquired by the child status acquisition unit 104. can be inferred.

Here, we will explain in detail the ``child's condition,'' ``support for the child,'' and ``results of the support.''

<<Children's condition>>
The child's condition is information that is used as a basis for determining whether to provide support to the child 21. For example, a child's condition can be determined by their school attendance status (e.g., information on the number of days they are absent, tardiness, leaving early, etc.), finances (e.g., information on welfare, school assistance, delinquency of various fees, etc.), academic ability (e.g., forgotten items, regular tests, etc.). information on class behavior), health (e.g., information on four measurements, unreceived medical examinations, information on visits to the health room, growth level), community information (e.g., information on how local resources are used, wandering around at night, coming to school in groups, etc.) information regarding events).

School attendance status may be obtained from a database of attendance, tardiness, and early dismissal managed by the school for each child, and periodic academic tests may also be obtained from performance data managed by the school for each child.

The information regarding delinquency of various fees in the economy here indicates the payment status of collections of various expenses necessary for children at schools, administrative agencies, etc., and includes data on the payment status of collections managed by schools, etc. It can be obtained from. Data on welfare and schooling assistance in the economy may be obtained from a database of protection and assistance status managed by administrative agencies for each family or child. In addition, class attitude in terms of academic ability should include, for example, things like how you move your head during class (number and frequency of head movements, how you sit, whether or not your back is straight, how straight your back is, line of sight, how often you raise your hands, what you say, etc.) You may also do so. Among these, the number and frequency of head movements, the way you sit, whether your back is straight, and the condition of your back are measured by taking images with cameras and image sensors installed in classrooms, schoolyards, gymnasiums, etc. The information may be extracted by analysis. As a specific extraction method, it can be extracted from an image by using pattern matching technology or deep learning technology of artificial intelligence. An eye tracker or the like that measures the viewpoint of the child 21 may also be used to extract the line of sight. Regarding the frequency of raising hands, we use cameras and image sensors to determine whether each child has raised their hand or not through image analysis.For example, we can automatically determine whether or not each child has raised their hand for each question asked by the teacher during class. It is also possible to extract it to Regarding the number of utterances and the content of the utterances, the voice of the child who uttered the utterances may be detected through a microphone, and the content of the utterances may be classified using well-known voice analysis technology. For this classification, the voice-analyzed utterances may be applied to pre-categorized utterance content patterns. Furthermore, in addition to the above, the results of a teacher or other measuring staff who visually check and determine the child's condition may be converted into data. Further, the drawings and crafts work may be extracted by capturing an image with a camera or an image sensor and analyzing the image, or the teacher may make a judgment and record it in the database. Furthermore, it is possible to determine the child's performance based on the content of the child's compositions and diaries, to convert the teacher's scoring and judgment results into data, and to convert the text data by reading character strings using OCR technology. The determination may be made by analysis based on well-known natural language analysis technology. It may also include the status of correct answers and incorrect answers to questions in the child's test, the grade (evaluation) for each subject in the report card, and even the contents of the findings recorded in the report card. The findings recorded in the report card may be analyzed and determined based on well-known natural language analysis technology.

Information on the use of local resources in local information includes, for example, local public facilities, libraries, community centers, transportation facilities, sports equipment, play equipment, teaching materials, vehicles, bicycles, food, fuel, drinks, etc. This includes all kinds of resources that are closely tied to the local community, such as human resources such as support staff, supporters, storytellers, and lecturers. This information on situations in which local resources are used is information about situations in which the use of local resources is required, such as when a disaster occurs that requires food and drink, its extent and content, and how support personnel This includes outdoor training situations that require guidance. Such information may be stored in a database prepared in advance.

Information regarding night wandering, group attendance at school, and events in the area information may also be stored in a database prepared in advance. The child's condition may also include clothing and grooming (shirt tucked into pants, clothing condition, hairstyle, haircut, hair washing, body washing (bathing), etc.). The condition of the child's clothing may be extracted by capturing an image using a camera or image sensor installed in a classroom, schoolyard, gymnasium, etc. and analyzing the image. As a specific extraction method, it can be extracted from an image by using pattern matching technology or deep learning technology of artificial intelligence. The condition of the clothes may be determined by applying a pattern of soiled or damaged clothes to a predetermined pattern. Similarly, personal appearance can be extracted by analyzing images captured by a camera or image sensor. Hairstyles, haircuts, and hairwashing conditions can be extracted by analyzing images captured by cameras and image sensors, but in addition to this, teachers and other measurement staff can actually visually observe and determine the child's condition. The results may be converted into data.

Furthermore, when determining a child's condition based on health, health conditions (caries, disease) may be included. As for the health condition, data from a health checkup measured from a child at a school, health center, hospital, etc. may be used as is. This health condition includes blood pressure measured during a medical checkup, blood test data, presence and details of various diseases (ear, nose, throat, ophthalmology, otorhinolaryngology, dermatology, internal medicine, urinary system, digestive system, respiratory system, etc.), and electrocardiogram. , barium examination, X-ray examination data, etc. may be used. In addition to this, data such as pulse rate, stress level, sleep score, blood sugar level, past medical history data, and information regarding the child's current or past disabilities may be used. When measuring this stress level, for example, a button with an icon indicating the weather (sunny, cloudy, rainy, etc.) that indicates how you feel is displayed on the touch panel of a tablet terminal installed in a classroom, etc., and the child is asked to The state of mind may be detected by inputting the user's feelings through the button.

Regarding dental caries, data from dental examinations may be used as is. In addition to directly measuring measurement data from the body, the health condition may be obtained directly from daily life, such as walking speed, how straight the back muscles are, and the content of conversation. Walking speed and the extent to which the back muscles are stretched can be extracted by analyzing images captured by a camera or image sensor, and the content of the conversation can be determined by detecting the child's voice through a microphone and analyzing what is said using well-known voice analysis technology. Contents may be categorized. Further, the degree of growth may be determined from data such as height, weight, BMI, and chest circumference. Information regarding the status of visits to the health room is recorded in a database, including the date and time of visits of children managed by the health room and details of treatment, and the status of visits to the health room is obtained from the database. Good too. Furthermore, the health condition may be obtained from the results of a physical fitness test of the child, and for example, data such as 50-meter running time, long jump, high jump, and ball throwing distance may be used.

When the child's condition is obtained from the academic ability, the status of the child's homework may also be included. The status of homework may be extracted from whether homework has been submitted or the content of submitted homework. Whether or not homework has been submitted may be obtained from a database managed by the teacher that indicates whether homework has been submitted for each child. The content of the homework may be obtained by recording the scores given by the teacher after evaluating the actual homework content in a database, or if the homework is a work of arts and crafts, etc. Features may be extracted and evaluated by using a pattern matching technique or a deep learning technique of artificial intelligence from an image of the work.

Further, the child's condition may be obtained from communication with the family. Communication with the family is data about the child's condition that the family notifies the school or government agency, such as the information written on the communication form submitted by the family or the SNS (Social Networking Service) sent to the school. ) or the content of the email. If you want to determine a child's condition from the contents of SNS or e-mails, extract the character strings, analyze them using well-known text mining technology, and analyze the contents according to pre-classified patterns. may be categorized. As a categorized pattern, for example, if text information includes characters such as "I'm scared" or "I have strange dreams," it may be included in the category that may lead to mental illness. If the text information includes the words "I don't eat breakfast," it may be included in the category of unstable health.

Further, the child's condition may be determined based on friendships. Deterioration of friendships is considered to be one of the important items that reflects children's isolation and loneliness. For example, by conducting a multiple regression analysis with friendship relationships as the dependent variable, it is possible to quantify in the form of coefficients which factor among multiple related factors influences the deterioration of friendship relationships and to what extent. I can do it. Then, the child's condition may be determined using numerical data on friendships calculated by using a large weighting coefficient for factors that have a large influence on the deterioration of friendships. For example, as a result of multiple regression analysis using friendship relationships as the dependent variable, it was found that information about the home environment such as language and appearance, as well as school information such as homework status, influenced friendships. If it is found that there is a significant correlation with the deterioration of friendships, information about the home environment such as ``language'' and ``appearance'' and information about the school such as ``homework status'' have a significant impact on the deterioration of friendships. The child's condition may be determined using numerical data on friendships calculated by using large weighting coefficients for influencing factors. This friendship relationship may be acquired through a connection network between friends, and for example, an application or software that can acquire human relationships through SNS may be used. In addition to this, friendship relationships may be determined based on the distance between children during free time such as lunch breaks. It can be determined that the closer the distance between the children, the deeper the friendship, and the farther the distance between the children, the shallower the friendship. This distance between children may be obtained by attaching a wearable terminal to the child and acquiring location information through GPS, etc., or it may be determined through image analysis using cameras installed in classrooms, schoolyards, etc. You can. By acquiring these data, it is possible to determine not only friendship relationships but also the number of friends.

Further, the child's condition may be obtained from whether or not the child is enrolled in a support class and the frequency of enrollment. In such a case, the presence or absence of enrollment in the support class and the period of enrollment may be recorded in a database, and the information may be obtained from the database.

Further, the child's condition may be acquired from the way the child eats and the amount of food. The amount of food may be determined based on whether the school lunch served at school or the like is completely eaten, partially left, or almost all left. This information can be obtained, for example, by a homeroom teacher observing each child and manually recording the situation in a database, or by capturing images of the child's lunch container using a camera installed in the classroom, etc. The degree of remaining may be automatically determined using image analysis technology. Furthermore, the manner of eating a meal may be determined based on eating speed, frequency of conversation during eating, manners and etiquette during eating, etc. These may also be recorded in the database manually by the homeroom teacher, or may be automatically determined through image analysis using a camera installed in a classroom or the like.

The child's condition may also be obtained from the child's previous attribute information, such as the history of changing schools, previous history as a victim or perpetrator of bullying, past history of incidents such as shoplifting, parent information, etc. It may also be based on recorded information such as a history of abuse from a child.

Further, the child's condition may be acquired from information regarding the home environment. Information about the family environment includes, for example, the child's family structure, whether the child's mother is pregnant, whether the child's guardian has a disability or illness, whether the child lives with the grandparents, and whether the care recipient is at home. whether or not the applicant has been affected by a disaster, whether the parents are divorced or not, income (household income), whether or not the applicant is on public assistance, whether the applicant has siblings and their age structure, and whether the applicant is a biological child of the parents or one of the parents. You may also include information such as whether or not the information is accepted. These data may be obtained from a database kept by an administrative agency or the like, or may be obtained from a database into which information obtained through actual interviews with children and guardians is input.

Further, the child's condition may be acquired from information regarding the child's usual behavior. This child's behavior may include, for example, being timid, looking at the facial expressions of parents or other adults, easily touching or interfering with others, showing abnormal attachment to food, or showing abnormal anxiety when undressing. It includes various behaviors based on the mind and body, such as being extremely restless, violent, and emotionally unstable. Such behavior may be extracted by capturing an image with a camera or an image sensor and analyzing the image. As a specific extraction method, it can be extracted from an image by using pattern matching technology or deep learning technology of artificial intelligence. The voice may be obtained by analyzing the voice recorded by a microphone using voice recognition technology.

<<Support for children>>
Support for children is the content of support for children 21. For example, support for children includes support from specialized organizations such as child guidance centers, support using local projects and resources (e.g. children's cafeteria, learning support, home education support, etc.), and support using resources within the school. Support (e.g., a team of teachers such as homeroom teachers, school nurses, special needs coordinators, school counselors, and school social workers).

The support provided by specialized institutions such as counseling centers includes, for example, in addition to support provided by child guidance centers, support provided by medical institutions, support provided during the health and welfare period, and the like.

Support that utilizes local businesses and resources includes meal support in the form of boxed lunches and support through food banks.

Support for children also includes support that utilizes educational resources, such as providing a place for learners to study using local learning facilities such as libraries, introducing textbooks and supplementary materials, and providing support for social studies. This also includes providing a place for tours.

Support for children also includes financial support, including various options for financial support, introductions to support organizations, housing support, and employment support for parents.

In addition, when support for children is provided by medical institutions, in addition to regular referrals to clinics, clinics, hospitals, etc., support for children also includes referrals to medical institutions related to mental care and dental health checkups.

Support for children also includes support for the use of public institutions such as libraries, community centers, sports fields, stadiums, grounds, play facilities, parks, halls, etc., as well as reservations and introductions.

Support for children also includes support for inviting lecturers and instructors, including inviting people who are active in various industries to be lecturers, and inviting experts and people with various valuable experiences to be lecturers. It will be done.

Support for children also includes support for providing equipment, such as providing crutches and wheelchairs for injured children and children with disabilities in their legs and hips, and canes and guide dogs for children with visual impairments. You may also provide support. Additionally, support may be provided to children with other respiratory or internal diseases by providing them with a variety of equipment to support their normal lives.

Further, support for children includes support for providing educational tools, and may include support for providing musical instruments, tools necessary for arts and crafts, and tools necessary for physical education time.

Support for children also includes support for ensuring safety, including the placement of instructors and lifeguards on the route to school, and security within the school and school areas.

Further, the support for children includes support for daily life, and if the child's family needs help with housework, support such as dispatching a housework support worker may be provided.

Further, support for children includes support for withdrawn children, and supporters who support withdrawn children may provide continuous visiting support.

Support for children also includes support for children who have fallen into delinquency or crime, and also includes support provided by probation officers at rehabilitation support centers.

Support for children also includes support for drug abuse, including support through drug reabuse prevention programs based on cognitive behavioral therapy.

<<Results of support>>
The support result is the result of attempting to provide the child 21 with the support inferred by the child support server 10. For example, the result of the assistance is that an attempt to provide assistance to child 21 was refused. For example, the result of the support is that the support was provided to the child 21, and the child's condition after the support was provided (for example, the number of absences, tardiness, leaving early, and forgetting things decreased. , academic ability improved).

The results of this support are not limited to these, but include the results of all types of support shown in the above-mentioned support for children. The result of this support may include not only the fact that the support has been performed, but also the degree to which the support has been performed. The degree of implementation of support is not only when all support is implemented without omission, but also when only half of the support itself is implemented, or when it is hardly implemented at all. good. Furthermore, the state of the child for whom support has been implemented may be indicated by rank, in addition to whether the child's academic ability has improved or not. In other words, the result of this support is not limited to the case where it is simply expressed as a binary value of whether it is effective or not, but the degree of the result of the support may be expressed as a rank of three or more levels. .

Here, rejection will be explained in detail. Refusal includes "refusal by the child," "refusal by the parent," "refusal by the school social worker," "refusal decision due to failure to complete within the designated period," and "local organization such as a children's cafeteria refusing the child's residence." This includes "refusal due to distance from".

FIG. 3 is a sequence diagram of the entire process according to an embodiment of the present invention.

In step 1 (S1), the child support server 10 generates a model that infers the necessity of support and support suitable for the child based on the child's condition. The model can output multiple "supports appropriate for the child" in order of priority, and select or deselect the multiple "supports appropriate for the child" using the "results of support" (without relearning). ).

In step 2 (S2), the sensor 20 and the school terminal 30 transmit the child's status to the child support server 10.

In step 3 (S3), the child support server 10 uses the inference model generated in S1 to infer support suitable for the child from the child's condition acquired in S2.

In step 4 (S4), the child support server 10 presents to the school terminal 30 the necessity of support and the support suitable for the child inferred in S3.

In step 5 (S5), the sensor 20 and the school terminal 30 transmit the result of the support in S4 to the child support server 10.

In step 6 (S6), the child support server 10 uses the result of the support in S5 (specifically, information regarding the support situation) to relearn the inference model generated in S1, thereby determining the support execution status. Generate a model that reflects the Thereafter, using the model retrained in S6, the need for support and support suitable for the child are inferred from the child's condition.

<Processing method>
Hereinafter, learning processing will be described with reference to FIG. 4, inference processing will be described with reference to FIG. 5, and relearning processing will be described with reference to FIG. 6.

FIG. 4 is a flowchart of learning processing according to an embodiment of the present invention.

In step 11 (S11), the learning data acquisition unit 101 of the child support server 10 determines that the learning data (specifically, the input (explanatory variable) is "child's state" and the output (objective variable) is " Obtain learning data on the need for support and the support appropriate for the child.

In step 12 (S12), the learning unit 102 of the child support server 10 performs machine learning using the learning data acquired in S11 to generate an inference model.

In step 13 (S13), the learning unit 102 of the child support server 10 stores the inference model generated in S12 in the inference model storage unit 103.

FIG. 5 is a flowchart of inference processing according to an embodiment of the present invention.

In step 21 (S21), the child status acquisition unit 104 of the child support server 10 acquires the status of the child 21.

In step 22 (S22), the inference unit 105 of the child support server 10 uses the inference model (that is, the inference model generated in S12 of FIG. 4 and stored in the inference model storage unit 103 in S13) to Based on the child's condition obtained from the above, it is inferred whether support is necessary and what support is appropriate for the child.

In step 23 (S23), the inference unit 105 of the child support server 10 presents the result of the inference in S22 (the necessity of support inferred in S22 and the support suitable for the child) to the school terminal 30.

FIG. 6 is a flowchart of relearning processing according to an embodiment of the present invention.

In step 31 (S31), the support result acquisition unit 106 of the child support server 10 acquires the support result (that is, the support result in S23 in FIG. 5).

In step 32 (S32), the relearning unit 107 of the child support server 10 performs relearning using the result of the support obtained in S31, and generates an inference model in which the result of the support obtained in S31 is reflected.

In step 33 (S33), the relearning unit 107 of the child support server 10 stores the inference model generated in S32 in the inference model storage unit 103.

When performing this relearning, a learning method such as transfer learning or fine tuning may be applied, for example. Transfer learning is a method of performing machine learning using learning data on a trained model that has been previously trained using data different from the learning data to be learned from now on. In this transfer learning, learning results from prior learning are transferred to machine learning, which improves learning efficiency compared to starting learning from a completely non-learning state.

Although transfer learning and fine tuning have different strict definitions, they have in common that machine learning is performed on a trained model that has been trained in advance. Therefore, in this embodiment, transfer learning includes a broad concept including fine tuning.

In transfer learning, the parameters of the middle layer are readjusted using a learning data set different from that used in the prior learning, although the parameters of the middle layer are based on the parameters of the middle layer that have been adjusted in advance through prior learning. This allows the results of child support to be reflected in the learning dataset through relearning, making it possible to perform more optimal inferences.

Further, FIG. 7 shows an example in which two or more types of children's states are input to the inference model and child support is output as a search solution. There may be not only one type of child status described above, but also two or more types, and an inference model is constructed by learning a dataset with output data (child support) for these. As a result, when two or more children's states are input, child support suitable for them is output as a search solution.

For example, if a child's condition A is that their clothes are dirty, and the child's condition B is that their head moves more than average during class, mental health The highest priority is given to dispatching a related counselor, and the next priority is dispatching a supporter for housework.

Further, as input data, various external environments other than the child's state may be learned as search solutions (child support). In other words, the input data is one or more types of child's condition and information on the external environment, and the output data is support for the child.

The external environment may include all kinds of circumstances, such as economic conditions, social conditions, technological evolution, and political conditions. In addition, external environment information may include weather and climate information such as temperature, humidity, rainfall, weather (sunny, cloudy, thunderstorm), cloud cover, etc., as well as information on the prevalence of infectious diseases, events such as sports, etc. The information may also be included.

Depending on these external environments, the content of optimal support to be provided to the child may change. For example, even in cases where support such as provision of public facilities such as libraries to children would normally be considered optimal, in cases of epidemics of infectious diseases or weather conditions such as typhoons, other support may be necessary. In some cases, the best option may be to provide a study program that can be received at home via email. By learning such external environment information as learning data, it is possible to propose optimal support suitable for the external environment.

When making such inferences that include the external environment, first, the state of the child is obtained, and information about the external environment at the time the inference is made is also obtained.

Next, using the inference model shown in FIG. 7, support suitable for the child is inferred from the acquired information on the child's condition and external environment.

<Hardware configuration>
FIG. 8 is a hardware configuration diagram of the child support server 10 according to an embodiment of the present invention. The child support server 10 includes a CPU (Central Processing Unit) 1001, a ROM (Read Only Memory) 1002, and a RAM (Random Access Memory) 1003. The CPU 1001, ROM 1002, and RAM 1003 form a so-called computer. Further, the child support server 10 can include an auxiliary storage device 1004, a display device 1005, an operating device 1006, an I/F (Interface) device 1007, and a drive device 1008. Note that each piece of hardware of the child support server 10 is interconnected via a bus B.

The CPU 1001 is a calculation device that executes various programs installed in the auxiliary storage device 1004. Each process described in this specification is performed by the CPU 1001 executing the program.

ROM 1002 is a nonvolatile memory. The ROM 1002 functions as a main storage device that stores various programs, data, etc. necessary for the CPU 1001 to execute various programs installed in the auxiliary storage device 1004. Specifically, the ROM 1002 functions as a main storage device that stores boot programs such as BIOS (Basic Input/Output System) and EFI (Extensible Firmware Interface).

The RAM 1003 is a volatile memory such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory). The RAM 1003 functions as a main storage device that provides a work area in which various programs installed in the auxiliary storage device 1004 are expanded when they are executed by the CPU 1001.

The auxiliary storage device 1004 is an auxiliary storage device that stores various programs and information used when the various programs are executed.

The display device 1005 is a display device that displays the internal state of the child support server 10 and the like.

The operating device 1006 is an input device through which the operator of the child support server 10 inputs various instructions to the child support server 10.

The I/F device 1007 is a communication device that connects to a network and communicates with other devices.

A drive device 1008 is a device for setting a storage medium 1009. The storage medium 1009 here includes a medium for recording information optically, electrically, or magnetically, such as a CD-ROM, a flexible disk, or a magneto-optical disk. Furthermore, the storage medium 1009 may include a semiconductor memory that electrically records information, such as a ROM or a flash memory.

Note that the various programs to be installed in the auxiliary storage device 1004 are installed by, for example, setting the distributed storage medium 1009 in the drive device 1008 and reading out the various programs recorded in the storage medium 1009 by the drive device 1008. be done. Alternatively, various programs to be installed in the auxiliary storage device 1004 may be installed by being downloaded from the network via the I/F device 1007.

Although the embodiments have been described above, changes can be made in form and details without departing from the spirit and scope of the claims.

1 Child support system 10 Child support server 20 Sensor 21 Child 30 School terminal 31 School social worker 32 Teacher 101 Learning data acquisition unit 102 Learning unit 103 Inference model storage unit 104 Child status acquisition unit 105 Inference unit 106 Support result acquisition unit 107 Re Learning section 1001 CPU
1002 ROM
1003 RAM
1004 Auxiliary storage device 1005 Display device 1006 Operating device 1007 I/F device 1008 Drive device 1009 Storage medium

Claims (6)

a child status acquisition unit that acquires the child's status;
an inference unit that infers support suitable for the child from the child's condition acquired by the child status acquisition unit using a model that infers support suitable for the child from the child's condition;
a support result acquisition unit that acquires information regarding the execution status of the inferred support;
A child support system, comprising: a relearning unit that generates a model in which the execution status of the support is reflected by relearning the model using information regarding the execution status of the support. The child support system according to claim 1, wherein the relearning unit further uses the execution status of the support and the state of the child after the support to relearn the model. The child support system according to claim 1 or 2, wherein information regarding the child's condition and the execution status of the support is acquired by a sensor. The child state acquisition unit further acquires information on the external environment at the time of inference,
The inference unit uses a model that infers support suitable for the child from the child's condition and external environment, and infers support suitable for the child from the child's condition and external environment information acquired by the child condition acquisition unit. The child support system according to claim 1 or 2, which infers. A method performed by a computer, the method comprising:
obtaining the child's status;
Inferring support suitable for the child from the obtained child condition using a model for inferring support suitable for the child from the child's condition;
obtaining information regarding the execution status of the inferred support;
A method including the step of relearning the model using information regarding the execution status of the support to generate a model that reflects the execution status of the support. computer,
a child status acquisition unit that acquires the child status;
an inference unit that infers support suitable for the child from the obtained child condition using a model that infers support suitable for the child from the child's condition;
a support result acquisition unit that acquires information regarding the execution status of the inferred support;
A program for functioning as a relearning unit that generates a model in which the execution status of the support is reflected by relearning the model using information regarding the execution status of the support.

Images