Ai-based childcare document automation system

Abstract

The present invention relates to an AI-based early childhood education field document automation system comprising a server; a teacher terminal; and a parent terminal, wherein the server performs the following operations: (a) storing curriculum data; (b) generating a hierarchical educational plan using an AI language model based on keywords input from the teacher terminal and referencing the curriculum data; (c) generating childcare documents including observation logs and notice boards by analyzing text or images input from the teacher terminal using an AI language model; (d) evaluating a child's developmental level and generating counseling documents using the AI ​​language model based on the generated childcare documents; and (e) collecting childcare documents and counseling documents generated over a certain period for each child to form a portfolio.

Description

Technology Field

[0001] The present invention relates to an AI-based early childhood education field document automation system, and more specifically, to a system that automatically generates counseling documents and portfolios created in early childhood education fields by utilizing an AI language model. Background Technology

[0003] In the field of early childhood education, the workload of teachers responsible for preparing various administrative documents is continuously increasing. Currently, kindergarten and daycare teachers are manually drafting documents such as annual, monthly, weekly, and daily lesson plans, activity plans, notices, observation logs, counseling documents, event plans, meal announcements, and parent newsletters.

[0004] The aforementioned documents must be systematically prepared in accordance with the curriculum data of the Nuri Curriculum or the Standard Childcare Curriculum, and are characterized by the need to maintain connectivity and consistency among the documents. In particular, they have a hierarchical structure that begins with the annual lesson plan and continues to monthly, weekly, and daily lesson plans, requiring that the content of the higher-level plans be reflected in the lower-level plans.

[0005] Conventional document creation methods involve teachers directly inputting text or copying and modifying existing templates. This approach requires significant time and effort from teachers and presents the problem of difficulty in maintaining consistency between documents.

[0006] Furthermore, conventional notification applications focus primarily on simply delivering text or photos to parents, and thus have limitations in that they fail to provide comprehensive automation functions for early childhood education documents, such as generating lesson plans, writing observation logs, evaluating developmental levels, and creating counseling documents.

[0007] Furthermore, existing systems fail to provide functions such as systematically analyzing a child's developmental level and automatically generating counseling documents based on this analysis, or organizing long-term growth records into a portfolio.

[0008] Therefore, there is a growing need for a system that utilizes AI language models to enable systematic automatic document generation based on curriculum data and efficient document sharing between teacher and parent terminals.

[0009] As prior art, Japanese patent JP2022-180282A deals with an automated childcare document system. The problem to be solved

[0011] This invention aims to resolve the burden of preparing childcare documents for teachers in the field of early childhood education.

[0012] Specifically, in traditional early childhood education institutions, teachers had to manually prepare various childcare documents, such as lesson plans, observation logs, notices, and counseling documents. Since these documents had to be written according to curriculum data and maintain a hierarchical structure ranging from annual to daily plans, they required a significant amount of time and effort.

[0013] In addition, the process of evaluating children's developmental levels based on observation logs and drafting counseling documents was complex, and the system for sharing documents between teachers and parents was inadequate. In particular, it was difficult to systematically utilize information entered by parents for counseling, and organizing a year's worth of counseling logs or portfolios was a significant burden for teachers.

[0014] Accordingly, the present invention aims to improve the work efficiency of teachers and enhance the quality of education by providing an AI-based early childhood education field document automation system that automatically generates the childcare documents using an AI language model and can efficiently share them through teacher terminals and parent terminals. means of solving the problem

[0016] As an embodiment for solving the aforementioned problem, the present invention provides an AI-based early childhood education field document automation system comprising a server, a teacher terminal, and a parent terminal. The server stores curriculum data and can automatically generate a lesson plan with a hierarchical structure using an AI language model based on keywords input from the teacher terminal. Additionally, it can generate childcare documents including observation logs and notice boards by analyzing text or images input from the teacher terminal, generate counseling documents by evaluating the child's developmental level based on the generated childcare documents, and compile a portfolio by collecting documents over a certain period for each child.

[0017] In addition, as an embodiment, the educational plan of the hierarchical structure may include an annual educational plan, a monthly educational plan, a weekly educational plan, and a daily educational plan. The server sequentially generates lower-level educational plans based on upper-level educational plans, and may instruct an AI language model to divide the monthly major themes of the annual educational plan into weekly sub-themes of the monthly educational plan. At this time, when generating the lower level, words having semantic association with the core words used in the upper level can be selected to ensure continuity of the educational content.

[0018] In addition, as an embodiment, when an image is input, the server can use a computer vision model to analyze the child's activities and facial expressions within the image. The analyzed information is converted into text and input into an AI language model, thereby enabling the automatic generation of childcare documents. This supports the creation of systematic documents even if a teacher only takes photos of the child's activities.

[0019] In addition, as an example, the assessment of a child's developmental level may be performed for five developmental areas: physical activity and health, communication, social relationships, artistic experiences, and nature exploration. The server can visualize the assessment results for each developmental area in the form of a radial chart or bar graph and include them in the counseling document, thereby enabling parents to intuitively understand the child's developmental status.

[0020] In addition, as an embodiment, the server can receive video data from a video recording device installed in the classroom and analyze the distance and frequency of interaction between children through an object recognition and object tracking model. By comparing the analyzed interaction data with the disposition data of each child recorded in childcare documents, the degree of positive or negative interaction in the relationships between children can be evaluated, and the results can be reflected in counseling documents. Through this, teachers can objectively analyze relationships between children that are difficult to ascertain visually.

[0021] Additionally, in one embodiment, the portfolio may be organized in at least one of chronological order, classification by developmental area, or grouping by topic. The server may convert the generated portfolio into at least one of a digital file, a printable PDF, or a web-based interactive format and transmit it to a parent's terminal. This allows parents to view their child's growth records in a manner preferred by the parent.

[0022] In addition, as an embodiment, the server can collect feedback data on modifications made by teachers to generated documents and use it to further train an AI language model. By improving the satisfaction of teachers' requirements in document generation through this continuous learning, it is possible to generate documents that are more accurate and suitable for the field over time.

[0023] In addition, in one embodiment, the present invention provides an AI-based early childhood education field document automation method performed by a processor of a server. The method may include the steps of storing curriculum data, generating a hierarchical educational plan based on keywords input from a teacher terminal, generating childcare documents by analyzing text or images, evaluating a child's developmental level based on the childcare documents and generating counseling documents, and collecting documents for a certain period for each child to form a portfolio. Effects of the invention

[0025] The AI-based early childhood education field document automation system according to the present invention can provide the following effects.

[0026] Teachers can automatically generate systematic lesson plans ranging from annual to daily with just simple keyword input, which can drastically reduce the time required for document creation.

[0027] By automatically generating observation logs and notices through text or image input, teachers can dedicate more time to direct interaction with children.

[0028] By utilizing AI language models to analyze accumulated childcare documents and objectively evaluate children's developmental levels, systematic and professional counseling documents can be provided.

[0029] By objectively evaluating interactions between children through video analysis, teachers can detect potential relationship problems early that are difficult to identify visually.

[0030] By automatically collecting childcare documents over a certain period and organizing portfolios for each child, the growth process of children can be systematically recorded and managed.

[0031] Through continuous learning that incorporates teacher feedback, you can generate documents that become more accurate and suitable for the field over time. Brief explanation of the drawing

[0033] FIG. 1 is a diagram showing the overall configuration of an AI language model-based early childhood education field document automation system according to one embodiment of the present invention. FIG. 2 is a flowchart illustrating a process for generating a hierarchical educational plan according to an embodiment of the present invention. FIG. 3 is a block diagram showing a hierarchical educational plan based on keywords according to an embodiment of the present invention, referencing curriculum data. FIG. 4 is a block diagram illustrating a method for generating childcare documents by analyzing text or images with an AI language model according to an embodiment of the present invention. FIG. 5 is an illustrative diagram showing an example of a consultation document according to one embodiment of the present invention. FIG. 6 is a diagram illustrating the process of configuring and visualizing a portfolio for each child according to an embodiment of the present invention. Specific details for implementing the invention

[0034] Various embodiments are described with reference to the drawings. In the present invention, various descriptions are provided to facilitate an understanding of the invention. However, it is evident that these embodiments can be practiced without such specific descriptions.

[0035] The term "or" is intended to mean an implicit "or" rather than an exclusive "or." That is, unless otherwise specified or evident from the context, it is intended to mean one of the natural implicit substitutions of "X uses A or B." In other words, it can be interpreted as "X uses A or B" even when X uses A, X uses B, or X uses both A and B. Furthermore, the term "and / or" as used in this invention should be understood to refer to and include all possible combinations of one or more of the enumerated related items.

[0036] Furthermore, the terms “comprising” and / or “comprising” should be understood to mean that such features and / or components are present. However, the terms “comprising” and / or “comprising” should be understood not to exclude the presence or addition of one or more other features, components and / or groups thereof. Additionally, unless otherwise specified or clearly evident from the context to indicate a singular form, the singular in the present invention and claims should generally be interpreted to mean “one or more.”

[0037] And, the term "at least one of A or B" should be interpreted to mean "a case including only A," "a case including only B," and "a case combined with the composition of A and B."

[0039] The present invention will now be described in detail with reference to the drawings. Although omitted for the convenience of explanation, each embodiment of the present invention should generally be interpreted as being performed by a server processor.

[0041] FIG. 1 is a diagram illustrating the overall configuration of an AI language model-based early childhood education field document automation system according to one embodiment of the present invention.

[0042] The AI ​​language model-based early childhood education field document automation system according to the present invention may be configured to include a server (100), a processor (110), a memory (120), a network (130), a teacher terminal (140), and a parent terminal (150).

[0043] The above server (100) is a device that serves as the center of the entire system and can control the execution of document automation functions for early childhood education sites and the transmission and reception of data between users. The server (100) may internally include a processor (110), a memory (120), and a network (130) module, and the processor (110) may be responsible for computation and control functions within the server (100).

[0044] The above processor (110) may correspond to hardware for performing operations such as a CPU, GPU, or APU, and may correspond to various computing devices that perform a similar role.

[0045] The memory (120) is a storage means built into the server (100) and can store various data required for the automation of early childhood education field documents, such as AI language model learning datasets, generated childcare documents, profiles for teachers and parents, various input records, document template information, etc. Additionally, the memory (120) can also be used as a temporary data storage space required by the processor (110) during computation. In this way, the memory (120) can enable rapid data input and output of the system and stable management of large-capacity records. An AI language model refers to a software model that learns natural language (text) data based on artificial intelligence technology and can automatically generate grammatically natural sentences, summaries, answers, and explanatory texts for given input sentences, words, keywords, questions, etc. This model pre-trains a large-scale text corpus (e.g., papers, articles, textbooks, conversation records, etc.) using machine learning or deep learning methods and can output the most suitable linguistic result for given input data (sentences, keywords, queries, etc.).

[0046] The network (130) is a module responsible for data communication between the server (100) and external terminals (140, 150), and can transmit and receive data in real time using wired / wireless communication protocols. Through the network (130), child activity information entered from the teacher terminal (140) can be transmitted to the server (100), and childcare documents generated from the server (100) can be delivered to the parent terminal (150). This enables rapid information exchange between the teacher and the parent, and also enables the realization of a real-time feedback structure.

[0047] The teacher terminal (140) is a smart device (e.g., tablet, smartphone, laptop, etc.) used by a childcare teacher and can perform the role of transmitting daily records, activity details, special notes, etc. of children to the server (100). For example, the teacher can use the terminal (140) to input content such as “had a good lunch” or “participated actively in art class” for each child in voice or text. The input data is transmitted to the server (100) via the network (130), processed by the processor (110), and then converted into a daily notice or childcare document.

[0048] Additionally, the teacher terminal (140) can provide an interface to review and modify childcare documents generated by the server (100). Through this, the teacher can improve the quality of the final document by directly entering additional explanations or supplementary comments if necessary.

[0049] The parent terminal (150) is a user device (e.g., smartphone, PC, tablet, etc.) used by a parent who has entrusted their child to a childcare institution, and can receive and verify childcare documents transmitted from the server (100). Through the terminal (150), the parent can check the child's daily activities, meal status, special circumstances, and teacher's opinions in real time, and can leave simple questions or feedback to the teacher if necessary.

[0050] In conclusion, the AI ​​language model-based early childhood education field document automation system of the present invention enables effective information sharing and management between a childcare institution and a home by organically combining each component, such as a server (100), a processor (110), a memory (120), a network (130), a teacher terminal (140), and a parent terminal (150). Through the system, the burden of document work for childcare teachers is drastically reduced, and parents can receive information about their children's lives quickly and accurately.

[0052] FIG. 2 is a flowchart illustrating a process for generating an educational plan with a hierarchical structure according to an embodiment of the present invention. Referring to FIG. 2, the AI-based early childhood education field document automation system of the present invention can automate documents in an early childhood education field through a sequential process from operation (a) to operation (g) from the start stage to the end stage.

[0053] To briefly explain the overall process, the server stores curriculum data and generates a hierarchical lesson plan using an AI language model based on keywords received from the teacher's terminal. Subsequently, it analyzes text or images input from the teacher's terminal to generate childcare documents, and based on these documents, evaluates the child's developmental level to create counseling documents. Finally, the documents generated over a certain period are collected by child to compile a portfolio.

[0054] Now, each action will be explained in detail.

[0055] (a) Operation is the operation of storing curriculum data. Curriculum data refers to data that systematically organizes the goals, content, and methods of early childhood education presented in the Nuri Curriculum or the Standard Childcare Curriculum. By storing curriculum data in memory, the server can secure basic data that an AI language model can refer to in subsequent stages. Curriculum data can be stored in a structured manner by age, developmental area, and topic, and this can be utilized as key reference material when generating educational plans with a hierarchical structure.

[0056] (b) The operation is to generate a lesson plan with a hierarchical structure using an AI language model based on keywords entered from the teacher terminal and referencing the curriculum data. A hierarchical structure refers to a hierarchical structure in which the annual lesson plan is at the top and subdivided into monthly, weekly, and daily lesson plans. When a teacher enters keywords such as "spring," "animal," or "family" through the teacher terminal, the server processor searches for curriculum data related to the keywords and can automatically generate lesson plans for each level by utilizing an AI language model. For example, if the keyword "spring" is entered, the annual lesson plan can set major themes related to spring from March to May, the monthly lesson plan can organize detailed themes for each month, and the weekly and daily lesson plans can generate specific activity content.

[0057] (c) The operation is to generate childcare documents, including an observation log and a notice board, by analyzing text or images input from the teacher terminal using an AI language model. The observation log is a document that records the child's daily activities, developmental status, and special circumstances, while the notice board is a document that conveys the child's daily schedule to parents. If a teacher inputs simple text such as "Young-hee built a tall tower during block play today" or uploads a photo of the child's activity, the AI ​​language model can analyze this to generate a systematic observation log. Additionally, the same content can be reorganized into a format that is easy for parents to understand, and a notice board can also be generated.

[0058] (d) Operation is to evaluate the child's developmental level and generate a counseling document using the AI ​​language model based on the generated childcare document. Evaluation of the developmental level refers to analyzing the degree of growth in each developmental area, such as physical activity, communication, social relationships, artistic experiences, and nature exploration. The server can identify each child's developmental characteristics, strengths, and areas requiring improvement by analyzing the contents of observation logs accumulated over a certain period using the AI ​​language model. Based on this, a counseling document that can be used during parent counseling can be automatically generated, and the counseling document may include the child's overall developmental status, characteristic behavioral patterns, and guidance plans for home.

[0059] (e) Action involves collecting childcare and counseling documents generated over a specific period for each child to form a portfolio. A portfolio refers to a comprehensive collection of documents that systematically records the growth and development process of a child. The server can organize observation logs, notices, counseling documents, activity photos, etc., generated for each child over a year or a specific period in chronological order and classify them by developmental area to form a single integrated portfolio. This allows for a clear overview of the child's growth process and can be provided to parents or used as handover material for teachers in the next grade.

[0060] To explain with an example, when a homeroom teacher of a 3-year-old class inputs the keyword "Spring and Animals and Plants" in early March, the system generates a March lesson plan through action (b). Subsequently, when the teacher inputs the text "Cheolsu observed the forsythia flower and said it was yellow," an observation log is generated through action (c) stating "Cheolsu shows interest in spring flowers, accurately recognizes colors, and can express them in language." As these observation records accumulate, a counseling document is generated through action (d) stating "Cheolsu has excellent observational skills in the nature inquiry area and can express himself using appropriate vocabulary in the communication area," and at the end of the year, Cheolsu's personal portfolio containing all records for the year can be completed through action (e).

[0061] In one embodiment, when a server processor receives the keyword "autumn" from a teacher's terminal, it automatically searches for and extracts all educational elements related to autumn from the curriculum database stored in memory. The processor runs an AI language model to analyze the extracted data and automatically places major themes such as "changes in autumn," "harvest," and "autumn events" into the annual lesson plan for September through November. Subsequently, the processor subdivides the themes by month to generate monthly lesson plans, such as "autumn weather and attire" for September, "autumn fruits and grains" for October, and "fallen leaves and autumn scenery" for November. Furthermore, the processor specifically generates weekly and daily activities, thereby providing a high-quality lesson plan that teachers can use immediately without any modifications.

[0062] In addition, as an embodiment, the server processor can simultaneously process observation records of 30 children input from multiple teacher terminals during the day. The processor analyzes the text and images input for each child using a parallel processing method to generate observation logs and notices for each child in real time. For example, if Teacher A inputs "Minji exchanged conversations while playing with dolls with a friend" and Teacher B inputs "Hyunwoo completed a number puzzle," the processor can utilize an AI language model for each input to simultaneously generate an observation log regarding Minji's social and language development, and an observation log regarding Hyunwoo's cognitive development and problem-solving ability. Furthermore, the processor can automatically classify and store the generated documents in each child's database and immediately send the notice to the corresponding child's parent terminal.

[0063] In addition, as an example, the server processor can periodically analyze childcare documents accumulated over three months to automatically detect developmental patterns or peculiarities of a specific child and provide notifications to the teacher. If the processor analyzes a child's observation log using an AI language model and finds that patterns such as "preferring solitary play," "having little interaction with peers," or "refusing to participate in group activities" appear repeatedly, the processor can synthesize this information to generate and send a notification to the teacher's terminal stating, "Attention is needed regarding the social development of child OO. Based on observations over the past three months, it has been analyzed that interaction with peers is significantly low." At the same time, the processor can automatically generate a lesson plan including customized intervention strategies and activity suggestions for the child, thereby supporting the teacher in immediately initiating appropriate educational interventions.

[0064] In addition, as an embodiment, when the server receives keywords related to the season from a teacher terminal, the AI ​​language model can set a major theme for an annual educational plan representing that season. When the server generates a monthly educational plan based on this major theme, it can instruct the AI ​​language model to divide the major theme into multiple weekly sub-themes. At this time, the AI ​​language model can ensure continuity of educational content by selecting one or more words semantically related to a core word in a higher layer and placing them in each week. For example, sub-themes related to weather, natural phenomena, changes in flora and fauna, and lifestyles that represent the characteristics of the season may be included.

[0065] In addition, as an example, if the server sets a major theme related to nature or the environment, the AI ​​language model can analyze the semantic association between superordinate and subordinate concepts when dividing it by week. The AI ​​language model can extract various subordinate concepts associated with the core words included in the major theme and arrange them appropriately for the developmental stages of the curriculum. The sub-themes for each week can consist of specific activities that include the core concepts of the superordinate major theme while also being connected to children's daily experiences, thereby maintaining consistency and systematicity in education.

[0066] In addition, as an example, when a server generates an annual educational plan based on a major theme including life topics or social concepts and then subdivides it into monthly educational plans, the AI ​​language model can consider various aspects of the relevant theme. The AI ​​language model can classify the superordinate concept into several sub-categories and generate specific sub-topics for each category that take into account the child's developmental level and interests. Furthermore, the system can be structured to enable systematic learning by arranging the sub-topics to expand from simple concepts to complex concepts and from concrete experiences to abstract understanding according to weekly progress.

[0068] The AI ​​language model of the present invention is a natural language processing model specifically designed and trained for document automation in the field of early childhood education. The AI ​​language model is based on a large-scale language model, but can be implemented through additional training and optimization specialized for the early childhood education domain.

[0069] The learning process of an AI language model can consist of two stages: pre-training and fine-tuning. In the pre-training stage, general language comprehension skills are learned using a large-scale text corpus. This corpus may include educational literature, child development theories, curriculum documents, and research papers related to childcare, and can consist of billions of tokens. In the fine-tuning stage, the model is specialized using actual data from documents used in early childhood education settings, such as lesson plans, observation logs, notices, and counseling documents. In this process, detailed contents of the Nuri Curriculum and the Standard Childcare Curriculum, age-specific developmental indicators, and examples of educational activities can be utilized as training data.

[0070] The model structure may be based on a Transformer architecture. The Transformer consists of an encoder and a decoder and can effectively process contextual information through a multi-head attention mechanism. The AI ​​language model of the present invention may have billions of parameters, and each layer may consist of an attention sublayer and a feedforward neural network sublayer. The attention mechanism selectively focuses on important information in input keywords or sentences, enabling the generation of educationally meaningful content.

[0071] The preprocessing stage is the step of converting input data into a form that the model can process. Text input from the teacher terminal first undergoes the tokenization process. Tokenization is the process of breaking down a sentence into its smallest meaningful units; in the case of Korean, morphological analysis or subword tokenization methods can be used. For example, the input "Young-hee played with blocks" can be broken down into tokens such as "Young-hee," "block," "play," and "did."

[0072] Tokenized data is converted into a vector form through an embedding process. Each token is represented as a point in a high-dimensional vector space, and semantically similar words can be placed in close proximity. Positional encoding is added to preserve the order information of each token within a sentence. Additionally, special tokens are added to indicate the beginning and end of a sentence, document type, and the like.

[0073] During the model's inference process, appropriate output is generated based on the input information. For example, if the keyword "spring" is input, the model can generate an educational plan by internally comprehensively considering spring-related educational activities, seasonal characteristics, and connections to child development. In the generation process, decoding strategies such as Beam search or top-k sampling can be used to select the most appropriate text.

[0074] The post-processing stage is the step of refining the output generated by the model into a final document form. The generated text first undergoes grammar checking to correct awkward expressions or errors. Subsequently, formatting is performed to suit the document type. In the case of lesson plans, they are structured in a table format, while observation logs can be organized into items such as date, observation content, and evaluation.

[0075] In particular, the AI ​​language model of the present invention may include a consistency verification module. The module verifies whether the generated document conforms to curriculum standards, is appropriate for the developmental stage, and does not contradict previously generated documents. For example, if an activity intended for 3-year-olds exceeds the developmental level of that age, it can detect and correct it.

[0076] In addition, the AI ​​language model of the present invention may include a personal information protection module. When processing data containing a child's name, photo, personal characteristics, etc., anonymization or pseudonymization processing is performed to ensure that personal information is protected. In the generated document, personally identifiable information may also be masked or generalized as needed.

[0077] The AI ​​language model of the present invention can possess continuous learning capabilities. When teachers review and modify generated documents and this feedback is collected, it can be used to further train the model. Through this, over time, it becomes possible to generate documents that are more accurate and suitable for the field.

[0078] Model lightweighting techniques can be applied for performance optimization. Through methods such as knowledge distillation, quantization, and pruning, model size can be reduced while maintaining performance. This can reduce the computational burden on the server and improve response speed.

[0080] FIG. 3 is a block diagram showing a hierarchical educational plan based on keywords according to an embodiment of the present invention, referencing curriculum data.

[0081] Referring to FIG. 3, the process of generating an educational plan of the present invention begins with a processor receiving keywords from a teacher terminal. The keywords are words or phrases representing educational topics or activity content, such as "spring," "family," "animals," "color play," etc.

[0082] When keywords are received, the server's processor generates a lesson plan through three core processes: curriculum hierarchy mapping, sequential generation of plans by hierarchy, and the lesson plan structure.

[0083] First, curriculum hierarchy mapping is the process of connecting input keywords with the content systems of the Standard Early Childhood Education Curriculum or the Nuri Curriculum stored in the curriculum database. For example, if the keyword "spring" is input, the system can map it to curriculum elements such as "Observing Seasonal Changes" in the Nature Exploration domain or "Expressing Spring Landscapes" in the Arts Experience domain.

[0084] Second, the sequential generation of hierarchical plans is a process of creating educational plans sequentially from the upper level to the lower level. An annual educational plan is generated first, followed by a monthly plan based on it, and then weekly and daily plans based on these. This sequential generation method can ensure consistency and continuity of educational content.

[0085] Third, the educational planning structure refers to the way in which generated educational plans are systematically organized. Educational plans at each level are structured to concretize and subdivide the content of the higher-level plan.

[0086] Looking at the bottom part of Figure 3, the annual education plan forms the top layer. The annual education plan includes the overall educational direction for the year and major themes for each month. For example, major themes such as "Spring and Animals and Plants" in March, "Me and My Family" in April, and "Our Neighborhood" in May may be set.

[0087] Below the annual education plan are monthly education plans, such as the January education plan, February education plan, and March education plan. Each monthly education plan subdivides the main theme of the corresponding month into four weekly sub-themes. For example, the March main theme "Spring and Animals and Plants" could be structured as follows: Week 1 "Spring Has Arrived," Week 2 "Observing Spring Flowers," Week 3 "Planting Seeds," and Week 4 "Spring Animal Friends."

[0088] Each monthly lesson plan is further subdivided into weekly lesson plans for weeks 1, 2, 3, and 4, and, in some cases, up to week 5. The weekly lesson plans organize daily activities based on the sub-theme of the corresponding week.

[0089] At the lowest level are the daily lesson plans for Monday, Tuesday, Wednesday, Thursday, and Friday. Each daily lesson plan includes specific activity details, materials, activity methods, and evaluation plans for the corresponding day. For example, during the "Observing Spring Flowers" week, activities such as "Finding Spring Flowers While Walking" could be planned for Monday, "Drawing Spring Flowers" for Tuesday, and "Learning the Names of Spring Flowers" for Wednesday.

[0090] The advantage of this hierarchical structure is that it ensures the systematicity and continuity of education. The goals and content of higher-level plans are naturally connected to lower-level plans, enabling children to have step-by-step and systematic learning experiences. Furthermore, AI language models can understand this hierarchical structure and automatically generate consistent lesson plans by considering the relationships between each level.

[0091] The system of the present invention supports teachers in systematically generating an entire year's worth of educational plans with just simple keyword input through this hierarchical structure, which can significantly reduce the workload of teachers.

[0093] FIG. 4 is a block diagram illustrating a method for generating childcare documents by analyzing text or images with an AI language model according to an embodiment of the present invention.

[0094] Referring to FIG. 4, the childcare document generation system of the present invention has a structure that automatically generates childcare documents, such as observation logs and notice boards, by processing text or images received from a teacher terminal (140).

[0095] First, the input method through the teacher terminal (140) is broadly divided into two types. The first is text input, where content entered by the teacher directly typing or voice is converted into text and transmitted to the server. For example, text in the form of a simple memo, such as "while the child is doing a block stacking activity," can be entered. The second is image input, where a photo taken by the teacher of the child's activity is transmitted to the server. For example, a "photo showing the child playing with blocks" can be entered.

[0096] When text is entered, the server's AI language model analyzes it to generate childcare documentation. The AI ​​language model analyzes the simple input text to understand the context, assigns educational meaning, and expands it into professional childcare documentation. For example, in response to the input "While the toddler was engaged in a block-stacking activity," the AI ​​language model can generate an observation log like the following:

[0097] Today, OO creatively built a tower with the toddlers using blocks of various shapes. During the stacking process, they demonstrated improvements in fine motor skills and spatial perception. In particular, they showed patience and problem-solving skills while rebuilding the collapsed blocks.

[0098] The above AI language model can also convert the same content into a notice for parents. The notice is written in a more friendly and easier-to-understand style than an observation log and can be generated as follows:

[0099] OO really enjoyed playing with blocks today. I was so proud to see them concentrating on building a tall tower. I could see them growing as they tried again even when the blocks collapsed.

[0100] When an image is input, a computer vision model on the server first analyzes it. The computer vision model recognizes and analyzes the child's activities, facial expressions, teaching aids used, and the surrounding environment within the image. For example, in a photo of block play, it can identify the child's posture, the shape and color of the blocks, and the characteristics of the constructed structures.

[0101] The analysis results of the computer vision model are converted into text and passed to an AI language model, which then generates childcare documents based on this. An example of an image-based notice is as follows:

[0102] OO completed a wonderful tower with colorful blocks! As you can see in the photo, the child is happily participating in the play. I think the child will enjoy it even more if you play with blocks together at home.

[0103] In addition, honorific documents can be generated through image analysis. Honorific documents are used in more formal situations and can be written as follows:

[0104] Child OO actively participated in block play activities that help develop fine motor skills and enhance creativity. As you can see in the activity photos, the child demonstrated concentration and experienced a sense of accomplishment in completing the structure.

[0105] Looking at the childcare document generation results shown at the bottom of Figure 4, the AI ​​language model can generate documents of various forms based on input information. The generated documents may include evaluations by child developmental area, specific activity details, educational significance, and home linkage plans.

[0106] The AI ​​language model of the present invention converts simple observations into educationally meaningful documents by referring to developmental indicators and educational goals stored in a curriculum database. In addition, it can automatically adjust the appropriate writing style and content level according to the purpose of the document and the reader.

[0107] This automated childcare document generation system enables teachers to create professional and systematic childcare documents with just simple notes or photos, significantly reducing the time spent on document creation and allowing them to dedicate more time to direct interaction with children.

[0109] FIG. 5 is an illustrative diagram showing an example of a consultation document according to one embodiment of the present invention.

[0110] The counseling document of the present invention is a child development assessment and counseling material automatically generated by an AI language model that comprehensively analyzes accumulated childcare documents. As a core document utilized during counseling between teachers and parents, the counseling document may include content that systematically organizes and analyzes the child's overall developmental status.

[0111] Looking at the process of generating counseling documents, the server processor inputs childcare documents, such as observation logs, notices, and activity records accumulated over a certain period, into an AI language model. The AI ​​language model comprehensively analyzes these documents to identify the child's developmental characteristics and, based on this, can generate a structured counseling document.

[0112] In one example, for the case of Kim Min-su, a 4-year-old child, if records such as "enjoys playing with blocks," "shows interest in number concepts," and "plays cooperatively with friends" appear repeatedly in the observation log accumulated over 3 months, the AI ​​language model can analyze this and generate a comprehensive evaluation such as "Min-su is developing spatial perception and mathematical thinking skills, and his social skills are also developing in a balanced way through cooperative play with peers."

[0113] In one embodiment, for a child with slow language development, an AI language model can identify patterns in an observation log such as "expressing oneself only with words," "having little interest in picture books," and "primarily using non-verbal expressions," and present specific evaluations and guidance plans in a counseling document, such as, "OO's language development is currently somewhat slower than that of peers, but their non-verbal communication skills are excellent. Increasing language stimulation through activities like reading picture books and language play would be helpful."

[0114] In one example, a counseling document for an emotionally sensitive child may include content such as, "OO has excellent ability to recognize and empathize with the emotions of others. However, as the child sometimes struggles to regulate their own emotions, we are providing guidance on emotional regulation methods in stages. Please also accept the child's emotions at home and teach them appropriate ways to express them."

[0115] In one embodiment, for a gifted child, an AI language model can generate an evaluation such as, "OO is showing superior ability compared to peers in the field of natural inquiry, based on observation records such as 'deep interest in science experiments,' 'recognition of complex patterns,' and 'questions at a higher level than peers.' Since OO is highly curious and has excellent inquiry skills, they will be able to further develop their talents through advanced science activities or project-based learning."

[0116] In one example, an AI language model may analyze physical activity observation records in a counseling document for a child with excellent physical development to include content such as, "OO has outstanding athletic ability, particularly excellent balance and agility. Providing opportunities to participate in various sports activities will further develop the child's talents. Currently, the kindergarten is increasing physical activity time to guide the child in releasing their energy in a positive way."

[0117] In one embodiment, for a child with outstanding creativity, an AI language model can synthesize observation records from art activities, music activities, dramatic play, etc., to generate an evaluation such as, "OO has excellent artistic sense, and is particularly outstanding in color sense and rhythm. They demonstrate creativity through free expression activities and frequently present original ideas. Please support the child's creativity by providing various art materials at home."

[0118] In one example, a counseling document for a child experiencing difficulties in peer relationships may present the current situation and directions for improvement, such as, "OO tends to prefer playing alone, but recently, interactions with friends have been gradually increasing. We are providing opportunities to practice social skills through small group activities and supporting the child in building positive peer experiences."

[0119] These counseling documents serve as objective and systematic resources during parent consultations and can play a crucial role in suggesting customized educational directions tailored to the individual characteristics of the child. The AI ​​language model of this invention comprehensively analyzes data extracted from various childcare documents to identify developmental patterns or characteristics that teachers might otherwise overlook, and incorporates them into the counseling documents.

[0121] FIG. 6 is a diagram illustrating the process of configuring and visualizing a portfolio for each child according to an embodiment of the present invention.

[0122] The portfolio of the present invention is an integrated document that comprehensively displays the growth and development process of a child by systematically collecting and organizing childcare documents, counseling documents, activity photos, artwork, etc., generated over a certain period for each child. The portfolio is not merely a collection of documents, but a meaningful record containing the child's growth story analyzed and reconstructed by an AI language model.

[0123] The portfolio creation process begins with a server processor collecting child-specific documents stored in a database. The processor can search and extract all relevant documents for a set period, for example, one year or one semester. The collected documents may include daily observation logs, weekly notices, monthly evaluation reports, quarterly counseling documents, activity photos, digital images of children's artwork, etc.

[0124] Once document collection is complete, the AI ​​language model classifies and organizes the collected documents based on various criteria, such as chronological order, developmental domain, and topic. During this process, duplicate or similar content is integrated, and key developmental indicators and meaningful episodes are extracted. For example, by extracting all records related to "block play" from a year's worth of observation logs, the developmental process of a child's constructive skills can be organized in a chronological order.

[0125] The components of the portfolio can be broadly categorized into developmental tracking, photographic materials, artwork collections, and a comprehensive evaluation. The developmental tracking section tracks the child's developmental process across five domains: physical activity and health, communication, social relationships, artistic experiences, and nature exploration. For each domain, developmental levels by period are visualized using graphs or charts to allow for a clear overview of growth trends.

[0126] The photo section provides photos capturing children's various activities, organized by month and theme. An AI language model automatically generates descriptions for each photo, allowing the content to be structured as a meaningful record of growth rather than a simple array of images. For example, descriptions illustrating developmental stages can be added, such as "March: Attempting to cut with scissors for the first time," "June: Skillfully cutting shapes with scissors," and "September: Precisely cutting out complex patterns."

[0127] The Artwork Collection section digitizes and stores artwork, craft results, and observation logs created by children over the course of a year. For each piece, the production date, materials used, the child's description of the work, and the teacher's evaluation can be recorded. An AI language model analyzes the evolution of the artwork to generate text that explains the development of the child's expressiveness and creativity.

[0128] The comprehensive evaluation section is the final part of the portfolio, summarizing overall growth and development over the past year. The AI ​​language model comprehensively analyzes all data to generate a comprehensive evaluation report that includes the child's key growth points, special talents or interests, and future developmental directions.

[0129] Various infographic elements are utilized in the portfolio visualization process. Radial charts showing growth by developmental area, line graphs representing changes over time, heatmaps indicating activity frequency, and percentile graphs showing developmental levels compared to peers can be automatically generated. These visual elements help parents intuitively understand their child's developmental progress.

[0130] The system of the present invention supports outputting portfolios in various forms. It can be transmitted to a parent's terminal as a digital file or converted into a printable PDF file and provided. Additionally, it can be produced as a web-based interactive portfolio, allowing parents to explore detailed content online by period and subject area.

[0131] A portfolio is not merely a collection of records, but contains a child's growth narrative analyzed and reconstructed by an AI language model. For example, growth stories such as, "Minsu preferred playing alone in March, but began showing interest in playing with friends starting in June, and by December, he demonstrated leadership in leading peer play," can be automatically generated through data analysis.

[0132] This AI-based portfolio creation system can effectively support children's development by drastically reducing teachers' document organization tasks while providing more systematic and meaningful growth records.

[0134] Embodiments that can be further extended to the present invention are disclosed.

[0135] The following describes in more detail, as an additional embodiment of the present invention, a method for analyzing relationships between children using image data and evaluating the health or potential risks to the formation of social relationships among children.

[0136] Specifically, the present invention can acquire time-series video data of children's movement and behavior patterns within a classroom using a plurality of video recording devices installed within a childcare institution, including CCTV. For example, a CCTV camera installed on the ceiling or wall of a classroom can film the entire indoor area to continuously record the children's activities and transmit this in real time to a processor (110) of a server (100).

[0137] The processor (110) can individually identify the location and movement of each child by running an artificial intelligence-based object detection and object tracking model based on the input image data. At this time, the location of the child can be expressed as specific coordinate values ​​(e.g., x, y coordinates), and the distance between the children can be calculated using the Euclidean distance between the coordinate values ​​or a similar distance measurement technique.

[0138] The processor (110) of the present invention can continuously store the distance data calculated in this way and separately record cases where two or more children are located close to a predefined threshold distance or continue to interact within a specific time interval. For example, if the set distance standard is set to 1 meter, a situation in which two children are active together within that distance for 10 seconds or more continuously can be determined as close interaction.

[0139] The processor (110) can quantitatively calculate the frequency and cumulative time of the close interaction data collected over a certain period (e.g., one day, one week, or one month, etc.) to derive the mutual closeness ratio for each child. For instance, it can quantify what percentage of the total activity time two specific children spent together or interacted during the day.

[0140] Meanwhile, the processor (110) of the present invention can compare and analyze the personal tendencies and behavioral characteristics of the children with respect to the closely interacting children by utilizing existing childcare document data, such as observation logs, counseling documents, and AI-based developmental evaluation data, stored in the memory (120) of the server (100). The personal tendency data can be specified, for example, as sociability, introversion, extraversion, emotional sensitivity, aggression, and receptiveness, and can be automatically extracted through AI natural language analysis of past observation records and counseling data.

[0141] In this way, the processor (110) can objectively analyze the similarities and differences between dispositional characteristics of closely interacting children through machine learning-based algorithms (e.g., Euclidean distance, cosine similarity, etc.). In particular, if a combination of specific indicators among dispositional characteristics, such as aggression, dominant disposition, passivity, and emotional sensitivity, shows a difference in the level of risk, the processor (110) can evaluate, based on this, whether there is a possibility that the relationship between the two children will develop into a healthy social relationship in the long term, or whether there is a high possibility that one child will become the target of bullying or conflict by the other child, using a quantitative probability value or a grade (e.g., high, medium, low).

[0142] For example, if one child has a relatively high aggression tendency and another child who interacts closely with them has high passivity or emotional sensitivity, the processor (110) of the present invention may determine that there is a high likelihood that this relationship will develop into a bullying situation in the long term and may automatically generate a counseling document recommending specific intervention measures along with a warning notification to the teacher terminal (140).

[0143] Conversely, if two children who maintain close interactions have similarly high levels of sociality or receptiveness and similarly low levels of aggression or emotional sensitivity, they are assessed as having a high potential to develop a healthy and stable social relationship, and educational programs or activities that support the development of a positive relationship can be recommended to the teacher.

[0144] Such an embodiment of the present invention is clearly differentiated from existing relationship analysis methods based on simple observation in that it utilizes automated distance and behavior analysis based on real-time video data and comprehensive and objective comparative analysis of personal tendency data. Furthermore, it can demonstrate an advanced effect in that it helps childcare teachers detect and intervene in potential social problems that are not easily identifiable visually at an early stage through the automated interaction pattern analysis provided by the AI ​​language model and object recognition tracking model of the present invention.

[0145] The above embodiments of the present invention may be specifically reflected in the creation of childcare documents, counseling documents, and portfolios in the following ways.

[0146] First, the processor (110) of the server (100) can add content that objectively describes the social relationships between the children when creating a childcare document, based on the results of the closeness data and tendency comparison analysis between the children obtained through video. For example, if a healthy peer relationship is expected based on the closeness and tendency analysis results between specific children, the processor (110) can automatically write content such as, “OO often engages in activities with △△, and their interactions are stable and positive. We plan to guide them to further strengthen their friendship through joint play activities in the future.” in the childcare document.

[0147] In addition, if the closeness analysis results classify a relationship as negative or potentially dangerous, this can be highlighted as a separate item in the observation log and automatically generated in the form of, “Although OO and △△ have been together very frequently recently, there is a possibility that △△ may experience stress due to OO’s active nature and △△’s passive nature, so careful observation and intervention are required,” thereby supporting the teacher to respond more quickly and appropriately.

[0148] The processor (110) of the server (100) can also utilize video-based closeness analysis results to provide information about a child's peer relationships in a way that is easier to understand, in order to provide more specific and practical information to parents when creating a notice. For example, the notice can be generated to include clear feedback and recommendations, such as, “Today, OO spent most of their activity time together with △△ and maintains a pleasant and harmonious relationship. I hope you will encourage conversation about △△ at home to support the formation of positive peer relationships.”

[0149] Finally, the processor (110) of the present invention can reflect the results of the video-based closeness analysis in an integrated document that records the long-term social development of a child. That is, the integrated document may additionally include graphs and charts that can visually represent changes in major peer relationships and tendency analysis for each period. For example, if a specific child maintains a positive relationship with a specific friend and maintains a high level of closeness for a certain period, this can be recorded in the integrated document along with a closeness graph visualized as a “social development indicator.” Conversely, for relationships where conflict is continuously observed, this can be specified along with a visual warning indicator to be used as a basis for long-term counseling and intervention.

[0150] Such an embodiment can more effectively achieve the objective of the present invention by substantially adding objective data-based social relationship analysis information to existing documents such as childcare documents, counseling documents, and portfolios, thereby enabling more accurate management of children's social skills and active intervention in relationship formation.

[0152] For the purposes of this invention, the present invention is not limited to the above embodiments, and various modifications and changes are possible by those skilled in the art within the technical scope of the invention.

[0153] For example, each component, procedure, function, etc. described in this specification may be optionally combined, modified, added, or omitted within the technical scope of the present invention, and the present invention may be equally applied to such various modified embodiments.

[0154] In addition, the embodiments and drawings described in this specification are merely examples to aid in understanding the invention, and the scope of the invention should be defined by the appended claims, and all variations falling within the equivalents of the claims should also be interpreted as being included within the scope of the invention.

Claims

Claim 1 An AI-based early childhood education field document automation system comprising: a server; a teacher terminal; and a parent terminal, wherein the server comprises: (a) the Nuri Curriculum or the Standard Childcare Curriculum By structuring curriculum data by developmental domain and age (b) an operation of saving; based on keywords entered from the teacher terminal, and referencing the curriculum data, an operation of generating an educational plan with a hierarchical structure including an annual educational plan, a monthly educational plan, a weekly educational plan, and a daily educational plan using an AI language model, wherein the server sequentially generates lower-level educational plans based on upper-level educational plans, and when generating the monthly educational plan, divides the monthly major themes of the annual educational plan into weekly sub-themes, when generating the weekly educational plan, organizes activity content for each day based on the sub-themes of the monthly educational plan, and when generating the daily educational plan, organizes specific activities for the corresponding day based on the activity content of the weekly educational plan, and when generating each lower level, selects words having semantic association with the core words used in the upper level to ensure continuity of educational content; (c) an operation of generating childcare documents including an observation log and a notice board by analyzing text or images entered from the teacher terminal using an AI language model, wherein if an image is entered, the child's activities and facial expressions within the image are analyzed using a computer vision model, and after converting the analyzed information into text, the AI The operation of automatically generating childcare documents by inputting them into a language model; (d) classifying the childcare documents accumulated over a certain period into five developmental areas of physical movement and health, communication, social relationships, artistic experience, and nature exploration, analyzing the classified data with the AI ​​language model to evaluate the child's developmental level, and including the evaluation results for each developmental area in the form of a radial chart or bar graph. A system characterized by: (e) receiving video data from a video recording device installed in the classroom and analyzing the distance and frequency of interaction between children through an object recognition and object tracking model, evaluating the degree of positive or negative interaction in the relationship between children by comparing the analyzed interaction data with the tendency data of each child recorded in the childcare document, and reflecting the evaluation result in the counseling document; (f) collecting childcare documents and counseling documents generated over a certain period for each child, extracting developmental indicators from the collected documents and organizing them in chronological order, configuring a portfolio in at least one of classification by developmental area and grouping by topic, converting the configured portfolio into at least one of a digital file, a printable PDF, or a web-based interactive format and transmitting it to a parent terminal; and (g) collecting the content modified by the teacher on the generated documents as feedback data, and performing an action to improve the accuracy of document generation by further training the AI ​​language model using the collected feedback data. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 delete Claim 6 delete Claim 7 delete Claim 8 delete

Images