Student homework AI tutoring method based on parent guidance and authority control

Abstract

The invention discloses a student homework AI tutoring method based on parent guidance and authority control. According to the method, a homework explanation and learning management mechanism with parents as a main control body is constructed based on the natural language understanding and expression ability of a large generative model, and the homework explanation and learning management mechanism comprises terminal binding and parent identity authentication, homework content analysis, explanation strategy scheduling, tutoring / supervision dual-mode permission control and a learning expression precipitation mechanism. Students carry out AI explanation type learning under the authorization of parents, and the system generates multiple rounds of explanations with controllable styles according to the content of questions; and for behaviors such as direct answer asking and the like, permission control and real-time intervention are realized through confirmation of parents. All learning behaviors are recorded by taking parent identities as main control paths, so that the privacy security of juveniles is guaranteed. The system supports multi-terminal cooperative operation, and allows parents to share high-quality explanation content to form a sustainable and optimized cooperative tutoring system.

Description

Technical Field

[0001] This invention relates to the field of educational technology and artificial intelligence integration, and in particular to a student homework tutoring method for home scenarios, based on large-scale model collaborative generation and permission path control. This method uses parents as the core control entity, and through identity authentication, terminal binding, prompt-driven mechanisms, multi-round explanations, and path permission judgment mechanisms, it achieves controllable access to the generative large-scale model, empowering students to use AI for learning while ensuring clear boundaries and process security for learning activities. Background Technology

[0002] In recent years, with the widespread application of generative large-scale model technologies, such as the GPT series, Wenxin Yiyan, and Tongyi Qianwen, in text generation, language understanding, and knowledge representation, artificial intelligence has demonstrated unprecedented empowering potential in the education industry. Especially in student homework tutoring scenarios, generative large-scale models, with their powerful language reconstruction capabilities, multi-angle explanation capabilities, and real-time interactive capabilities, have initially possessed the technical foundation to replace some human explanation tasks, becoming the prototype of "AI teachers."

[0003] However, in practical application, numerous doubts have emerged from various sectors of society, mainly focusing on: 1. Students use large models to "directly obtain answers" instead of thinking independently, which fosters dependence and laziness; 2. Parents are unable to know the specific content of their children's use of AI, and lack the right to know and control. 3. The education system lacks standardized supervision over AI-generated content, resulting in uncontrollable issues with the quality of expression; 4. Minors' learning behavior and identity information on AI platforms are easily over-collected, posing serious privacy risks.

[0004] Against this backdrop, a heated debate has erupted in society regarding whether students should be restricted from using large models. However, this invention proposes a different perspective: large models are not the problem; the problem lies in their unrestrained use. Generative AI, as a powerful and flexible learning tool, should be properly guided and used appropriately, rather than being outright prohibited. We should not throw the baby out with the bathwater, but rather empower students, liberate parents, and regulate boundaries in a systematic way, thereby establishing an "AI-collaborative family learning mechanism" that aligns with the characteristics of our times.

[0005] Traditional homework tutoring software primarily uses question bank retrieval and fixed answer analysis, limiting its educational value and thinking training capabilities. In contrast, large-scale models can generate explanations based on context, possessing "personalized, diversified, and strategic" capabilities, potentially leading to a new learning paradigm: expression-driven learning + strategic explanation + parental boundary control. In this paradigm, students are no longer passive recipients but actively drive the large-scale model through prompts and style templates, obtaining multiple rounds of explanation, multi-faceted analysis, and explanations tailored to their cognitive pace. Parents, acting as guides and defenders, monitor answer acquisition and boundary behavior, constructing a three-way collaborative structure of "AI + Parent + Student."

[0006] Therefore, there is an urgent need for a technical solution that adapts to the capabilities of large-scale models while maintaining a family-controllable structure. This solution should both ensure the development of students' learning abilities using large-scale models and prevent unsupervised abuse, ensuring the interpretability, traceability, and compliance of the learning process. This invention is proposed against this backdrop, aiming to construct a homework tutoring system truly adapted to the characteristics of the generative AI era through a structural approach. Summary of the Invention

[0007] I. Purpose of the Invention

[0008] With the rapid development of generative artificial intelligence (large model) technology, its powerful language generation and expression reconstruction capabilities have shown great potential in educational scenarios, especially in homework explanation, knowledge organization, and personalized learning guidance. At the same time, it has also raised many social concerns, mainly focusing on: Students are overly reliant on AI to directly obtain answers and are avoiding critical thinking training; Parents find it difficult to monitor the boundaries and behaviors of students using AI; The educational process lacks effective mechanisms for recording and collaborating on the expression of information. Learning data privacy is difficult to guarantee, and there is a risk of leakage of minors' information.

[0009] Especially in primary and junior high school, students lack the ability to independently use AI for learning, while parents often lack professional knowledge or explanation skills during tutoring, creating a disconnect between "students not knowing how to learn and parents not knowing how to teach." Against this backdrop, there is an urgent need for an intelligent tutoring mechanism tailored to home learning scenarios, capable of ensuring students can use AI to improve their learning abilities while enabling parents to guide, control, and supervise the learning process.

[0010] The purpose of this invention is to provide an AI-based tutoring method for student homework based on parental guidance and access control, constructing a family learning mechanism with parents as the control center, students as the main learners, and a large-scale model as the intelligent collaborator, thereby solving the following core problems: This allows students to truly learn using large models, rather than simply copying answers; This allows parents to effectively monitor the boundaries and content of the large model's use; Access control mechanisms are used to protect parents' control over the answers and their right to stop questions from going too far. Enables collaborative learning scenarios across multiple terminals, including mobile phones, e-readers, and smart broadcasting devices; Ensure the system does not collect any personal identification information from students, protecting the learning privacy of minors; Support parents in turning high-quality explanations into "parent lesson plans" to achieve family knowledge co-construction.

[0011] To achieve the above objectives, this invention does not involve large model ontology, but focuses only on the invocation structure and boundary mechanism. A systematic methodology with five major steps is designed, covering terminal binding and identity authentication, question collection and context parsing, large model invocation and expression strategy scheduling, pattern recognition and permission path control, and expression accumulation and lesson plan sharing mechanism.

[0012] This method can establish a family AI tutoring mechanism with controllable boundaries, clear roles, rich expression, and auditable path in the era of generative large models. It not only improves students' comprehension and expression abilities but also significantly reduces the tutoring burden on parents, providing a paradigm for future AI education.

[0013] II. Detailed Steps

[0014] Step 1: Terminal Binding and Parental Authentication Mechanism

[0015] This step aims to establish a terminal pairing relationship between students and parents, and to complete system identity registration, login authorization, and usage permission initialization with the parent as the sole controller.

[0016] The system authenticates parents' identities through methods including but not limited to ID card recognition and facial recognition. After successful authentication, the device ID and account information of the smart terminal (such as a mobile phone) held by the parent are used as the parent's master control identifier to complete the binding with the system.

[0017] Parents can actively add and bind multiple student devices through the main control terminal, including but not limited to e-readers, tablets, mobile phones, computers, and other devices with learning interaction capabilities. Each parent terminal can also bind multiple student devices simultaneously, suitable for family structures where multiple children share a single parent account.

[0018] In addition, auxiliary playback devices (such as smart dolls, headphones, speakers, etc.) also need to be bound and registered by parents to ensure that all content playback behavior is controlled and executed within the permission structure.

[0019] The system does not collect or store any student's personal identification information during this process, including but not limited to fields such as name, age, gender, photo, student ID number, school attended, and grade level. All learning behavior generation and access paths are managed and recorded under the parent's account, achieving privacy protection for student behavior and traceable management of parental responsibility.

[0020] Step Two: Collection of Homework Questions and Context Analysis

[0021] This step aims to achieve structured input and semantic parsing of the student's current learning task, serving as a prerequisite for the large model to generate problem-solving expressions. By collecting the question content, analyzing the knowledge structure, and modeling the associations with the historical learning context, a semantic foundation is laid for the accurate generation and style adaptation of subsequent explanation strategies.

[0022] This invention supports multiple question input paths, including but not limited to: Parents can take photos and upload homework questions from their devices (applicable to parent tutoring mode). Student terminal can take photos or input text (applicable to student self-study mode); Parents or students can directly input the question text or select the question number.

[0023] After receiving the question, the system uses a text-image understanding model and a structure parsing engine to perform semantic structuring on the question content, completing the following core tasks:

[0024] 1. Question structure and semantic element extraction Natural language processing models are used to analyze keywords, question requirements, question types (such as multiple choice, fill-in-the-blank, and problem-solving), and the question setter's intent in the question stem. For image input, OCR text recognition combined with a structural analysis model combining text and images is used to segment and label modules such as the question stem area, answer area, and auxiliary instructions.

[0025] 2. Mapping of knowledge point tags to subject positioning The system maps the parsed semantic elements to a standard knowledge graph, determines the subject area, knowledge nodes, and test abilities (such as reasoning, comprehension, and calculation) involved in the questions, and generates a "knowledge point-question type-ability" triple structure, which serves as the core input factor for subsequent explanation scheduling and difficulty prediction.

[0026] 3. Historical Behavior Chain and Contextual Reference Modeling Without collecting students' personal information, the system can refer to the learning interaction behavior chain recorded under the parent's account, including but not limited to: Historical explanations of similar questions; Parents' or students' style preferences and feedback ratings for the content being presented; The question number (QID), the call prompt, the generated model version, and other information express the path.

[0027] Based on these established expression paths, the system constructs a contextual relationship network between questions and user preferences to support personalized style of explanation content and controllable strategy generation.

[0028] 4. Terminal type and learning scenario compatibility assessment The system identifies the type of device initiating the request (such as a parent's mobile phone, a student's e-reader, or a smart doll) and determines the current mode (tutoring mode / supervisory mode). This information serves as a key parameter for permission scheduling and path generation control, participating in subsequent decisions regarding whether to allow the use of specific teaching strategies or whether parental confirmation is required.

[0029] Ultimately, the output of this step is a structured "question representation graph," which includes elements such as question semantic structure, knowledge point mapping, contextual reference relationships, preference style factors, pattern identifiers, and permission status. This structured information will serve as input conditions to drive the scheduling of the next step's large-scale model explanation and generation strategy.

[0030] Step 3: Large Model Invocation and Explanation of Strategy Scheduling Mechanism

[0031] This step is the core technical aspect of the invention. Relying on a prompt-driven path scheduling mechanism and an expression style generation strategy module, it dynamically calls one or more generative large models to generate personalized problem-solving explanations with diverse styles, adjustable granularity, and context-appropriate content. The system not only supports parents triggering explanation generation during tutoring but also allows students, with parental authorization, to initiate prompt requests and invoke the same strategy scheduling path.

[0032] 1. Multi-model collaborative invocation mechanism This invention supports calling mainstream generative large models, including but not limited to GPT, Wenxin, Xunfei, and Ali Tongyi models, or their customized versions, to construct a multi-model candidate set and scheduling pool, implementing the following mechanisms: Model adaptation scheduling: Dynamically select the appropriate model based on the subject matter, complexity of the question type, and generation goal (explanation / inspiration / analogy, etc.); Task allocation and result fusion: Complex tasks can be addressed by using a "main generation model + auxiliary processing model" chained approach to improve semantic consistency and expression accuracy; Performance evaluation and optimization mechanism: The system continuously records the performance of each model, including generation accuracy, user feedback scores, completion speed, etc., and performs score writing back and subsequent optimization calls based on the three-element path of "strategy-model-result".

[0033] 2. Prompt-driven path scheduling mechanism All explanatory actions must be triggered by prompts. These prompts, serving as the driving entry point for "invocation intent + style preference + structural control," are the core control unit for achieving personalized expression scheduling in this invention. Prompts include, but are not limited to: Please explain in language that children can understand. "Let's put it another way." Say it again. Explaining with diagrams "Focus on the key points" "Use analogy to explain" Tell the story using everyday scenarios. "Teach like a teacher" "Speak like a mother." The system maps prompts to a strategy template pool, with each prompt corresponding to one or more expression strategy paths, controlling the content structure and language style generated by the model.

[0034] 3. Expressive style, strategy, structure, and scheduling During the content generation process, the system will control the following dimensions based on the strategy path matched by the prompts: Granularity of expression: determines whether to expand step by step, whether to explain implicit steps, and whether to omit consensus logic; Language style: You can choose from various expression styles such as everyday language, philosophical, analogical, humorous, parent-child, and teacher-like. Key content: Use strategy templates to control whether to emphasize keywords, common mistakes, frequently misunderstood points, or extended test points; Supplementary dimensions: The model generates question patterns, summaries, and suggestions for variation training, etc., as supplementary expressions.

[0035] The system constructs a complete path diagram of "prompt → strategy template → expression goal → generation of control factors" and records information such as strategy version, style selection, and model call path for each generation behavior, so as to facilitate subsequent behavior chain recording and lesson plan accumulation.

[0036] 4. Nested control of patterns and permissions During the invocation process, the system will determine whether to allow direct invocation of model generation based on the current operating mode (tutoring mode / supervision mode) and the terminal identity (parent / student): When a student requests a prompt like "Give the answer directly," the system will pause generating the prompt and trigger a parental authorization process. If the parents allow the generation of answers, the "give answers" strategy path will be invoked; If parents choose to "explain", then the "explanation and guidance" strategy will be followed. All behavioral paths and permission judgment chains are bound to the parent's identity and recorded in the behavior log.

[0037] Finally, based on the parsed question information, prompt type, and current permission status, the system constructs a complete scheduling request, drives the generative big model to generate multi-style and multi-level explanation and expression results, and distributes them to student terminals or auxiliary devices for playback.

[0038] Step 4: Mode Control and Access Control Mechanism (Dual Mode of Coaching / Supervision)

[0039] This step establishes two operating modes—"tutoring mode" and "supervision mode"—by defining the different roles of parents and students in the learning process. The system uses parents as the access control center, constructing an access path management structure of "answer control node + boundary violation recognition + real-time interruption mechanism" to ensure that learning behavior always operates within the boundaries that parents can authorize, supervise, and stop during the semantic generation process enabled by the large model.

[0040] 1. Model 1: Parent-led guidance model (parent-led type)

[0041] This model is suitable for younger students or students with weaker self-motivation. Parents can initiate learning tasks through a mobile app, and the system is controlled by the parents.

[0042] The key features are as follows: Parents can upload photos of the questions or input the question stem via a mobile app; The system analyzes the questions, generates explanations, and pushes them to student terminals (such as e-readers, tablets, etc.). Students can view the explanations on their devices, and parents can monitor the explanation process in real time. The tutoring process is controlled by the parents throughout, and the students can only use the controlled terminals such as e-readers and do not have the right to actively access them. In face-to-face tutoring scenarios, students can also directly view content on their parents' mobile phones without needing a separate device.

[0043] Note: Uploading questions by taking photos on the student's terminal can also be used as an auxiliary method, but it must be run under the parent's authorization system. All explanation requests and model call permissions are still bound to the parent's terminal control logic.

[0044] 2. Mode Two: Parent Supervision Mode (Student Self-Directed)

[0045] This mode is suitable for older students with strong self-learning abilities. Students can study independently on devices such as e-readers, but all key behaviors still require parental authorization or can be supervised by parents.

[0046] The operating mechanism is as follows: Students log in to their accounts through the linked terminal and send a "learning authorization request" to their parents' mobile app; After the parent receives the request, scans the code to confirm or clicks to authorize, the student's terminal enters "learning mode"; Students can use prompts (such as "explain again" or "explain in a different way") to initiate a lecture request, and the system will respond according to their permissions. When a student requests to view the answers directly, the system will pause the process and send a confirmation pop-up to the parent. Parents can choose to "confirm the answer" or "let AI explain the question," and the system will generate an expression based on their selection. The entire learning process is mirrored to the parents' end, allowing them to view the student's questions and actions in real time, and giving them the "right to stop the learning process if it goes too far."

[0047] This model emphasizes active learning by students, with parents not directly intervening in the explanation and generation process, but always maintaining control over the boundaries and key nodes.

[0048] 3. Answer control point design (right to reveal answers)

[0049] To prevent students from using AI as a quiz tool rather than a learning tool, the system establishes a "right to reveal answers" control node: If a student enters a prompt such as "Just tell me the answer" or "What is the answer to this question?", the system will automatically recognize it and stop the generation process. A pop-up request is sent to the parent's terminal, whereby the parent must make a choice between two options: "Provide the answer": The system follows the "answer generation" path and outputs the standard answer; "Explain": The system follows a "strategy explanation" approach, generating step-by-step guided expressions; Parents' choices and the generation path are both recorded as control nodes in the behavior chain.

[0050] This mechanism can flexibly respond to parents' judgments in different scenarios and can also prevent students from using AI to avoid thinking.

[0051] 4. Mechanism for identifying and stopping transgressions (right to stop) The system incorporates a boundary violation detection engine and an instant parental interruption function to prevent students from using AI for non-learning purposes. Boundary violations include, but are not limited to: The question is irrelevant to the topic, such as: "Who is the strongest hero in Honor of Kings?" Upload non-assignment content, such as screenshots, comics, game screenshots, etc.; By using a photo-based question-and-answer format, the system is indirectly made to identify game levels or entertainment content.

[0052] When the system recognizes the above behavior: Automatically intercept requests and display the message "Current behavior does not conform to the scope of learning use"; At the same time, an alarm message is pushed to the parent's terminal; Parents can choose to "force interrupt the current learning process"; The system performs an "interruption" operation and records the chain of actions.

[0053] This mechanism includes the "right to stop" as one of the core rights of parents, ensuring that the boundaries of learning and use are not crossed.

[0054] 5. Permission path management and behavior chain binding mechanism The system constructs a "permission path graph" for all actions such as prompt initiation, model invocation, answer control, and style selection; Each path node is bound to the user role (parent or student), terminal type, authorization status, and behavior result; All behavioral chain data is recorded under the parent's account, forming an auditable and traceable "parent behavioral chain"; All operations performed on the student's end are considered by default to be performed with parental authorization, and the responsibility lies with the parent's account.

[0055] This step, through a mechanism of "role-driven + permission nesting + path blocking," constructs a strong boundary and highly transparent AI learning behavior control system, effectively preventing the risk of abuse. This is one of the key innovations that distinguishes this invention from traditional quiz apps.

[0056] Step 5: Establish a mechanism for sharing lesson plans with parents and reflecting on past experiences.

[0057] This step revolves around the expression and accumulation mechanism of the teaching content and behavioral chain data generated during the learning process. It designs a parent lesson plan system based on parental initiative, centered on diverse expression methods, and incentivized by community sharing. Through structured storage and classification mechanisms, the system encourages the accumulation of high-quality content and supports parental collaboration, while strictly protecting the privacy and control boundaries of students' learning behaviors, thus building a sustainable collaborative knowledge network for family AI tutoring.

[0058] 1. The structured sedimentation mechanism of learning expression Once the AI ​​explanation of a problem is complete, the system automatically encapsulates the generated expression in a structured manner, including but not limited to: Question information: Question image, text, question type, and subject; Explanation content: The explanation process generated by the large model, including multiple rounds of reasoning and alternative versions; Behavioral chain information: prompts used, expression style templates, model version called, feedback scores, etc.; Generate a path summary: Explain the triggering method (parent takes a picture of the question / student asks a question), the call chain, and the parent control node record.

[0059] The system uses the unique identifier of the topic (QID) as the primary key to archive multiple expressions of the topic, providing a foundation for subsequent sharing and evolution.

[0060] All content stored is visible only to the parent by default. The system does not actively upload, analyze, or display any content to protect the privacy of learning activities.

[0061] 2. A lesson plan sharing mechanism initiated by parents The system allows parents to proactively upload their explanations of a specific problem to the "Parent Lesson Plan" module after the lesson is completed. The trigger path is as follows: After the problem is solved, the system prompts parents on the parent's end: "Would you like to save this explanation as a lesson plan?" If the parent selects "Share": The system archives the explanation process of the current question as a lesson plan entry; Automatic naming and categorization, such as: "Chinese Language Arts · Grade 1 · Beijing Area · Parent Lesson Plans"; The lesson plan only records parent-led behaviors and does not collect any student identity data. The content of this lesson plan is only accessible to other parents authorized to access it within the platform.

[0062] The lesson plan content is packaged by the system, and students' terminals have no right to initiate sharing requests or view other people's lesson plan content, ensuring that the responsibility for all actions clearly belongs to the parents.

[0063] 3. Filing and Classification Logic of Parent Lesson Plans The system supports automatically archiving and tagging parents' lesson plans according to the following dimensions: Subjects (Chinese, Mathematics, English, etc.); Question types (multiple choice, fill-in-the-blank, word problems, etc.); Grade level (selected by parents for explanation, not representing the student's actual grade level); Regional tags (e.g., the location where the parent's account was registered); Expression style (everyday life, humorous, philosophical, metaphorical, etc.); Source methods (uploaded by taking a picture of the question / student asking a question / template trigger, etc.).

[0064] This multi-dimensional tag structure is conducive to forming a clear, rich, and navigable "parent lesson plan database".

[0065] It is worth emphasizing that the "grade" field is only an identifier for the teaching strategy. The system is unaware of and does not collect students' actual grade information, truly reflecting the ability paradigm of "cross-grade understanding" in the era of large models.

[0066] 4. Incentive Mechanism and Shared Rights To encourage parents to actively participate in content creation, the system has set up incentive mechanisms including but not limited to points and badges: Parents can earn platform points for successfully sharing a lesson plan; Earning a certain number of points allows users to redeem usage time, customized features, voice templates, and more. High-quality lesson plans can be recommended by the system to the "Featured Lesson Plans" section, enhancing their display and mutual assistance value; The system follows the principle of "sharing means mutual benefit," and parents who do not actively share cannot view other people's lesson plans.

[0067] This incentive mechanism is based on parental voluntariness, and the system will not disclose any content by default or use it for training, ensuring user data sovereignty.

[0068] 5. Principles of privacy protection for learning activities This invention explicitly states that learning behavior itself is also a form of privacy and needs to be subject to system-level protection mechanisms. Specifically, this includes: The system does not collect, store, or analyze any students' personal identification information; All learning behaviors, expressions, and model interactions are recorded with the parent's role as the main index. All calls made by student terminals are considered to be "executed under parental authorization," and the system defaults to assigning responsibility to the parent. The behavioral chain structure, including the content, question data, and generation path, is not used for training the platform's recommendation engine or as the basis for user profiling.

[0069] This step establishes a mechanism for the accumulation of expression based on voluntary sharing, parental guidance, and strong privacy protection. Combined with platform recommendation and incentive mechanisms, it forms a family AI learning co-construction network centered on parents.

[0070] III. Explanation of Methodological Logic and Structural Closed Loop

[0071] In summary, this invention proposes an AI-based student homework tutoring method based on parental guidance and access control. Focusing on the controllable application requirements of generative large-scale models in homework scenarios, it constructs an intelligent collaborative mechanism with parents as the main controllers, students as the learning subjects, and the model as the expression engine. The entire method progresses through five steps, forming a closed loop: First, the terminal binding and parental authentication mechanism ensures that every model call is conducted with the parent's knowledge and authorization, laying the foundation for a "parent-controlled" structure. Second, the collection and context analysis of homework questions enable a structured understanding of the question content and the construction of a teaching context, supporting the accurate matching of subsequent expression strategies and model invocation; Third, the large model call and explanation strategy scheduling mechanism drives the style strategy module scheduling through prompt path, realizes dynamic generation of explanations for multiple models, multiple styles, and multiple rounds, and builds a differentiated expression system; Fourth, the mode control and permission path management mechanism clarifies the boundary differences between the "parent tutoring mode" and the "student supervision mode" and embeds key permission nodes such as the "right to reveal the answer" and the "right to stop when overstepping boundaries", forming a permission structure of dual mode, dual path and dual control; Fifth, the mechanism for sharing parent lesson plans and the process of refining and refining teaching materials allows parents to voluntarily create "parent lesson plans" from their high-quality explanations, promoting a positive cycle of co-creation and sharing among parents, reuse of teaching materials, and mutual assistance in teaching.

[0072] The five steps are structurally interconnected and logically closed, ensuring that students can actively learn under the collaborative guidance of the large model while also guaranteeing that parents' control and supervision rights are not diminished throughout the process. Through a chain-like structure of identity binding, path authorization, strategy invocation, behavior recording, and expression archiving, this invention realizes a family learning boundary control mechanism empowered by a large model, constructing a new system paradigm for students to use AI learning that is technically feasible, path-controllable, and auditable.

[0073] The focus of this invention is on the structural design and permission control path based on the scenario-based invocation of a large model, rather than the model itself. The method of this invention can be implemented on existing terminal platforms, or it can be deployed and controlled in conjunction with the structure described in "A Student AI Homework Tutoring System Based on Permission Path Control." In this system, the parent terminal, student terminal, and model service platform are coordinated and scheduled through permission paths, which can further support the various execution steps of the method described in this invention.

[0074] IV. Model Call Structure and Protection Boundary Description

[0075] (I) Model Invocation Structure and Policy Scheduling Mechanism The system described in this invention does not rely on a specific large model vendor or model structure. Instead, it is built on a platform structure of "prompt-driven + strategy module scheduling + multi-model collaborative generation," adapting to any large model with natural language generation capabilities, including but not limited to the GPT series, Wenxin Yiyan, Claude, Tongyi Qianwen, Yi, MiniMax, and InternLM. The model call flow structure includes: 1. Prompt-driven layer: Parents or students select preset prompts or input custom expressions through the terminal, and the system matches and calls the strategy accordingly; 2. Strategy Template Scheduling Layer: The system matches explanation strategies (such as analogy, everyday language explanation, highlighting key points, error guidance, reverse reasoning, etc.) based on the type of prompt. 3. Expression Generation Layer: The system calls the target model interface, loads a specific version of the model, and generates structured explanation content; 4. Multi-round feedback generation mechanism: Supports secondary calls triggered by instructions such as "say it again" or "say it another way", ensuring diversity of expression styles and controllability of content.

[0076] (II) Model Deployment Path and Access Control Mechanism To ensure the secure use and access control of large models in home scenarios, this system adopts the following technical mechanisms for model deployment and path management: 1. Path permission control: All calls must be bound to an authenticated parent account, and the system determines their legitimacy based on the call path; 2. Boundary blocking mechanism: Students must access the model through the prompt path or the question-taking path. Directly asking for answers or asking questions that are not related to learning content will trigger the path recognition and blocking mechanism. 3. Model version identification mechanism: All generated content is accompanied by the version number of the model being called and the generation strategy number, which facilitates behavior traceability and content auditing; 4. Edge deployment and remote collaboration mechanism: Some strategy modules can be deployed locally to reduce latency; the core generation capability of large models is deployed in the cloud and uniformly scheduled by the platform. 5. Automatic interception of abnormal behavior: If non-question screenshots, entertainment content input, or behavior that bypasses prompts are detected, the system will block the model call and record the abnormal log. 6. Parental control interface: Parents can stop the student's access process in real time and decide whether to allow the answer to be revealed directly, ensuring the "right to reveal the answer" and the "right to stop".

[0077] (III) Boundary Protection and Platform Anti-Surrounding Mechanism To prevent others from circumventing the protection scope of this invention by changing the model structure, bypassing the prompt path, or altering the device form, this system proposes the following structural protection strategies for boundary control: 1. Structural logic protection: This invention does not protect the large model itself, but rather the system structure of "prompt path-driven strategy module + model call scheduling + nested access control". 2. Model call path blocking mechanism: All non-policy-driven model calls are considered illegal paths and will be recorded and blocked; 3. Behavior chain recording mechanism: All model call behaviors form an auditable call chain, including information such as QID, strategy path, prompt, called model, generated version, and terminal identifier; 4. Anti-fake strategy path mechanism: The system requires that the path must include a complete prompt binding field and a behavior chain authorization field to prevent others from directly calling the model through empty commands or bypassing the path; 5. End-to-end anti-bypass deployment structure: The system is deployed in a closed collaborative structure consisting of "parent main terminal + student secondary terminal + model server terminal + edge broadcast terminal" to prevent the system functions from being copied by simply calling the large model.

[0078] (iv) Declaration of the use of generative large models in this invention To avoid misunderstanding of the technical boundaries of this invention during the review process, the following statement is hereby made: The "generative large model" described in this invention includes, but is not limited to, the GPT series, the Claude series, Wenxin Yiyan, Tongyi Qianwen, GLM, and other general-purpose language models with natural language generation capabilities.

[0079] This invention does not involve the research, development, training, fine-tuning or parameter design of the above-mentioned model ontology, nor does it claim any protection rights to the model algorithm architecture itself.

[0080] The core of the technological innovation of this invention lies in: Utilize existing or third-party large model capabilities to build a structured invocation mechanism suitable for homework scenarios; Design a model collaborative invocation structure with prompt-driven, strategy-scheduled, access control, and expression generation paths; In a multi-terminal collaborative environment, a parent-led and boundary-controlled model usage mode is achieved, ensuring the auditability and anti-circumvention of the learning process.

[0081] Therefore, the focus of protection claimed by this invention is the system structure design outside the model, including control mechanisms such as prompt path generation, call permission determination, behavior chain construction, and boundary blocking strategy.

[0082] Based on the principle of "calling large models rather than building our own models", this invention aims to provide a model empowerment mechanism that is adapted to family education scenarios and has controllable boundaries, in the context of the widespread application of generative large models, so as to ensure that children can carry out AI-assisted learning in a safe, controlled and effective framework.

[0083] V. Terminology Definitions and Identification Boundaries

[0084] To clarify the meaning of the core terms involved in this invention, avoid semantic ambiguity during implementation, and clearly define the technical boundaries and scope of protection, several key terms in this invention are defined and explained below:

[0085] 1. "Prompt Path" This term refers to natural language prompts, expression guidance templates, or explanation style instructions input, selected, or invoked by parents or students in the interactive terminal. These prompts trigger the system's preset explanation strategy module and drive the large model to generate personalized expression content. The prompt path, as one of the only entry points for model invocation, constitutes a necessary prerequisite for the system's legitimate expression generation.

[0086] Prompt paths include, but are not limited to, the following types: Say it again. "To put it another way" Explain using a metaphor. Speak slowly. Explain it in a way that a fourth grader can understand. The system automatically records the triggering command, calling strategy, and expression version of all explanatory content called through the prompt path, and writes them into the behavior chain.

[0087] 2. "Question-answering behavior chain" This term refers to the user's operational trajectory and system response sequence recorded during a complete AI-assisted homework tutoring session. This behavioral chain covers the entire process from question acquisition, prompt triggering, model generation, parental confirmation, student viewing, and result feedback, and records the following information fields in a structured manner: QID (Question Unique Identifier) Prompt Types and Content Model version number called Strategy template ID used Output summary Parental confirmation markers (e.g., whether revealing the answers is allowed). Student feedback behavior (such as liking, requesting a re-examination) The answer behavior chain, as an important basic structure for behavior tracing, path monitoring, permission determination, and data accumulation in the system, is one of the core mechanisms that distinguishes this invention from traditional operating systems.

[0088] 3. "Right to reveal the answer" This term refers to the situation where, when a student directly requests to "view the answer" or initiates a clear request such as "give the answer directly" on the terminal, the system does not immediately generate the answer content, but instead suspends the model calling process and sends a permission confirmation request to the bound parent control terminal.

[0089] The system only responds when a parent selects "Allow Answer Generation"; otherwise, it switches to an explanation-based approach, such as "Let me teach you how to solve this problem step by step." This mechanism is a key control point in this invention to prevent students from abusing AI.

[0090] 4. "Right to Halt" This term refers to the parent's real-time monitoring rights during the student's use of the system, and their ability to monitor the student's behavior beyond the learning boundaries (e.g., when the student initiates actions that exceed the learning boundaries). Asking questions about non-academic topics (such as entertainment or games); Upload irrelevant images (such as screenshots or emojis); (e.g., repeatedly bypassing the prompts and directly attempting to call the answer interface) Parents can terminate the execution of a particular behavior with a single click and add the behavior block record to the behavior chain. The right to stop ensures parents' ability to intervene at the boundaries of the entire learning process.

[0091] 5. "Coaching Model" and "Supervision Model"

[0092] Tutoring Mode: Parents initiate question input, photo capture, or content selection on the main control terminal. The system generates explanation content and synchronizes it to the student's terminal. This mode is suitable for younger students or user scenarios with high dependency, where students do not need an independent terminal, and parents complete all operation control.

[0093] Supervision Mode: Students log in on their terminals and initiate learning requests. The system then sends authorization confirmation information to the parents. After parental confirmation, students can access the explanation function, but the system retains parental confirmation rights at two key points: first, the generation of "directly requesting answers" must be confirmed by the parent; second, parents can stop students from asking "out-of-bounds questions."

[0094] The underlying data structure and path logic are consistent for both modes. Parents are the responsible parties for operations, and all behavioral chain data is attached to the parent's account. The system will not transfer the authority subject due to different usage modes.

[0095] 6. "Model call path validity" The system pre-defined prompt trigger paths, problem-solving paths, and other legal model call flows. All model expression generation must be driven by a clearly defined path. The system will perform path validity checks on all call requests to determine if the following conditions are met: Is a legitimate parent account linked? Are there any prompts or instructions? Is it within the permitted time period? Is it a valid request within the current action chain context?

[0096] Requests that do not activate the model through a legitimate path will be automatically blocked by the system, preventing responses and generating abnormal behavior records.

[0097] 7. "Privacy of Learning Behavior" This term refers to all data generated by students during their learning process using the AI ​​system, including all behaviors, requests, content, questions, and feedback, which are considered part of the privacy of minors' learning activities. System default: We will not collect students' personal identification information such as names, ages, genders, photos, and schools. It is not mandatory to link the student's actual grade level; All learning data is linked to the parent's account for access control and behavior recording; The system must not perform personalized modeling or profiling analysis of students based on learning data.

[0098] This privacy protection structure is one of the important boundary statements of this invention, which distinguishes it from the centralized collection of student data and individual behavior recognition of traditional systems.

[0099] 8. "Parent's Lesson Plan" This term refers to the action taken by a parent who, after completing a full lesson plan using this system, actively chooses to archive and share the lesson plan. This lesson plan is a structured unit of explanation generated, named, archived, and shared by the parent, and has the following characteristics: Active data collection: The system will not automatically collect, store, or disclose any teaching activities. All "lesson plans" are only generated after parents explicitly select "share to lesson plans". Structured Recording: The system automatically records and organizes data such as prompts, explanation strategies, model responses, and student feedback in a lecture process to form a complete explanation path; Category Tags: The system automatically categorizes lesson plans into the lesson plan library based on tags selected by parents, such as "Chinese Language Arts, Grade 1, Beijing Area, Parent Lesson Plans"; Only visible to parents: Only the parent's device has publishing permissions; the student's device cannot initiate a sharing request. Mutual assistance and incentives are possible: Parents' lesson plans can be viewed by other parents, and the system provides incentive feedback mechanisms such as points and badges to encourage positive sharing.

[0100] "Parent lesson plans" are not only a form of knowledge expression, but also a result of parent-led AI tutoring mechanisms. They are a unique family learning structure asset unit of this invention.

[0101] 9. "Style Strategy" This term refers to a system that uses prompts or context recognition to invoke control templates with different explanation styles and expressions to adapt to students' comprehension preferences and question types. Style strategies include, but are not limited to: Humorous style: "Explain this problem in a funny way"; A conversational style: "Explaining using examples we can see in everyday life"; Philosophical style: "Is there any logic to this question?"; Metaphorical style: "Could you give an example?" Exam-oriented: "How would you answer this in an exam?" Step-by-step explanation type: "Can we do it step by step?" Encouraging feedback: "Say it again and I'll get it."

[0102] Upon receiving a prompt or strategy guidance instruction, the system invokes a preset "explanation style template" and schedules the corresponding large model expression path to generate content in the appropriate style. This strategy mechanism allows users to switch styles multiple times during a single question explanation, which is a key feature distinguishing this invention from fixed-answer software.

[0103] VI. Path Control and Abnormal Detour Blocking Mechanism

[0104] To prevent others from circumventing the protection path of this invention by changing the model structure, bypassing the strategy process, or altering the terminal form, this system has constructed a path control and abnormal bypass blocking mechanism. This mechanism aims to ensure that the complete structure of "prompt-driven + strategy scheduling + permission node control" cannot be disassembled, replaced, or bypassed. The specific mechanism is as follows:

[0105] 1. Call path blocking mechanism All calls to large models must enter through the system's embedded policy path, meaning they must be driven by a complete request flow consisting of bound prompts, style templates, and authorization status. Any large model call initiated directly without going through the policy path (such as skipping prompts or entering questions without authorization) will be considered an illegal act by the system, and will be rejected, the call chain recorded, and a blocking policy triggered.

[0106] 2. Behavior chain generation and verification mechanism The system generates a structured behavior chain record for each model call, including fields such as unique question identifier (QID), prompt type, style strategy label, called model version, parent / student terminal identifier, and generation timestamp. This behavior chain serves as an auditable path for tracing, correcting deviations, and identifying anomalies, ensuring that the entire generation process is verifiable and monitorable.

[0107] 3. Mechanism for identifying missing prompts and pseudo-strategies The system performs real-time verification of the completeness and validity of prompts in the request path. Any request that does not contain a prompt field or uses a forged prompt (such as an empty string or an invalid template ID) will be immediately marked as a "fake policy path" and the system will interrupt the model call and block the access prompt.

[0108] 4. Unauthorized command identification and halting mechanism During the learning process on the student's end, the system continuously performs semantic recognition and risk assessment on the input content. If it detects behaviors that are irrelevant to the assignment, attempt to bypass the strategy to obtain answers, upload non-assignment images, or attempt to call unauthorized expression paths, the system will immediately freeze the request and notify the parent to perform a "allow or not" confirmation operation. If the student violates the rules repeatedly, the parent can "stop" the session with one click to ensure that the system boundaries are not breached.

[0109] 5. Terminal form factor cannot bypass the mechanism This system employs a deployment structure of "master terminal binding + secondary terminal authorization + behavior mirroring" to ensure that the model calling capabilities of any type of terminal, such as mobile phones, e-readers, tablets, and smart dolls, must rely on the parent's authorization path to complete, thus avoiding the circumvention of system permissions and path control logic by changing the hardware form.

[0110] In summary, this invention effectively constructs an indivisible, irreplaceable, and unbypassable call protection framework through multiple structural designs, including strategy path uniqueness verification, behavior chain recording and verification mechanisms, illegal request instant interception mechanisms, and permission node confirmation mechanisms. This ensures the integrity, security, and unavoidability of the system's operational logic and the inviolability of patent boundaries.

[0111] VII. Beneficial Effects

[0112] The "AI-based tutoring method for student homework based on parental guidance and access control" proposed in this invention is a typical product of the era of large-scale models, aiming to solve two major core educational problems that are prevalent in society today: 1. The paradox of our times: students' inability to properly use AI-assisted learning; 2. The structural contradiction that parents bear a heavy burden and feel powerless in helping with homework.

[0113] While generative AI technologies (such as large language models) have been widely applied across various industries, the primary and secondary education sector, which should be prioritized for AI adoption, is effectively banned due to societal concerns about misuse, overstepping boundaries, and cheating. Although students possess the technical skills to acquire AI capabilities, the lack of proper usage methods, boundary guidance, and accountability mechanisms prevents AI-driven education from being truly implemented.

[0114] Meanwhile, homework tutoring has evolved into a "second job" for parents. Especially in dual-income families, parents often face a large number of homework problems that need to be explained after a tiring day, and many of these problems are beyond their ability to explain. This rigid demand has not been systematically addressed for many years.

[0115] Therefore, this invention innovatively constructs a homework tutoring system mechanism with parents as the main controllers, students as learners, and a large model as the collaborative expression entity, which has the following significant beneficial effects and innovative features:

[0116] 1. For the first time, a boundary-controllable structure for "AI-empowered students" has been achieved. This invention does not simply provide problem-solving tools, but establishes a complete expression generation path of "prompt-driven + strategy scheduling + model invocation + permission nesting". While ensuring that students can learn through AI, it ensures that the large model will not be abused for non-learning scenarios such as directly copying answers or asking questions for entertainment.

[0117] 2. Establish a tutoring mechanism based on "parental authorization + access control". This invention ensures that all AI-generated behaviors occur with parental authorization through parental QR code login, permission confirmation, and process monitoring. The system empowers parents with substantial control over the boundaries of AI use through a dual mechanism of "right to reveal answers" and "right to stop overstepping limits," truly achieving "visible, controllable, and interruptible by parents."

[0118] 3. Propose a collaborative mechanism where "AI is the teacher and parents are the coordinators". Parents no longer need to explain the questions themselves; they simply need to take a picture of the question and upload it, or authorize the student to study independently. The system then uses a large model to generate explanations suitable for the child's understanding. Parents shift from "teaching content" to "managing learning paths," greatly reducing their teaching burden and improving family learning efficiency.

[0119] 4. Dual-mode adaptation: Coaching + Supervision running in parallel The system is compatible with two family education scenarios: a "tutoring mode" where parents lead the explanation of questions by taking photos, and a "supervisory mode" where students operate independently and parents have access to monitoring functions. Regardless of whether parents are familiar with the knowledge content, they can use the system to complete high-quality tutoring.

[0120] 5. Protect the privacy and behavioral independence of minors This invention does not collect students' names, ages, grades, or other identity information. All learning behaviors are recorded under the parents' account names, avoiding the creation of profiles of minors, effectively protecting students' privacy, and strengthening the parental responsibility mechanism.

[0121] 6. Build a shareable and collaborative "parent lesson plan system" Parents can choose to submit a specific lesson plan to the platform's lesson plan library, creating a structured and reusable version of their family's explanations. This mechanism not only promotes mutual assistance and sharing among parents but also builds a non-textbook-style learning resource system that is relevant to the family context.

[0122] 7. The path is auditable, the call can be blocked, and the anti-bypass mechanism is complete. The system employs mechanisms such as behavior chain structure, path hash verification, and prompt-driven binding to prevent others from directly calling the model by bypassing the strategy process, forming a technical "blocking anti-bypass call structure".

[0123] In summary, this invention is not a "learning machine" or "question bank-type answering tool" in the traditional sense, but rather a structural mechanism invention for how generative AI can be correctly, controllably, and sustainably implemented in basic education. It opens up a new paradigm path for the integration of AI and family education in the future that is regulated, scalable, and shareable. Attached Figure Description

[0124] Figure 1 : A schematic diagram of the overall structure of the system of the present invention; Figure 2 Method path flowchart; Figure 3 Terminal binding and parental identity authentication flowchart; Figure 4 Homework question collection and context analysis structure diagram; Figure 5 Diagram illustrating the large model invocation and explanation strategy scheduling mechanism; Figure 6 : Expression behavior chain structure diagram; Figure 7 : Path control and abnormal detour blocking mechanism diagram; Figure 8 Flowchart for Parent Lesson Plan Generation and Sharing; Figure 9 Comparison of control structures between coaching and supervision models. Detailed Implementation

[0125] To make the objectives, technical solutions, and beneficial effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. This invention is not limited to the specific embodiments described below; any equivalent substitutions or improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0126] Example 1: Recording the Homework Explanation Process and Behavioral Chain in a Parent-Led Tutoring Model

[0127] In this embodiment, parents use a mobile app to photograph math homework questions and select a "structured explanation" prompt template (template name: Math_Structure_G3). After the system analyzes the questions, it matches the prompt with a "modular derivation + everyday metaphor" strategy, calls the large model service, and generates multiple rounds of explanation content.

[0128] Students receive instruction via e-reader terminals. Throughout the process, the system records the following action chain fields: QID: Unique question ID (e.g., MATH2025-BJ-G3-0145) Parent ID and terminal identifier (e.g., ParentWXID-27381) Strategy template path: Prompt → Strategy module → Style component (e.g., Math_Structure_G3 → Decompose → Life_Analogy) Model call information: model version, response time, output digest fingerprint Authorization confirmation node (e.g., request to show answers before answering questions is rejected) Timestamps and Operation Sequences

[0129] All information is linked to the parent's account for future review and sharing.

[0130] Example 2: Control Point Triggering and Path Authorization under Student Self-Supervision Mode

[0131] A fourth-grade student logged into the e-reader and requested an explanation of a Chinese reading comprehension question. The student entered the prompt "understand this passage from a different perspective." The system matched the "perspective-shifting explanation strategy" (template name: CH_Read_VPShift_G4) and prepared to generate the expression content.

[0132] At this point, the system detects a student's request to view the standard answers, immediately halts the process, and simultaneously pushes a "Confirm Answer" pop-up to the linked parent's mobile phone. The parent selects "Explanation instead of direct answers," and the system continues executing the model call, with the following path structure: StudentTerm → PromptEntry → StrategySelect → PermissionCheck →ModelCall If a student attempts to exceed the boundaries (such as entering entertainment topics), the system path will be interrupted, and a "behavior interception log" will be generated, including: Crossing the line (entertainment keyword) Interception policy version Should parental terminals intervene? This mechanism ensures that the path is compliant, controllable, and traceable.

[0133] Example 3: Parent Lesson Plan Generation and Templated Naming Structure

[0134] After a parent completes an explanation of a "proportion word problem" and expresses satisfaction with the generated content, they click the "Share to Parent Lesson Plan" button. The system automatically archives and names the lesson plan. Example title: Math, Grade 4, Word Problems, Parent Lesson Plans, Beijing Area Fields included: Question ID, Explanation Strategy Path, Generated Version, Behavior Chain Summary, Student Terminal Device Identifier (anonymized). Strategy template name: MATH_App_G4_ExplainBySteps The lesson plan can then be recommended to other parents' devices by the platform, with access limited to users who also choose to share.

[0135] The system does not save the original audio or images, but only retains the structured expression chain summary to ensure privacy compliance; the parent account receives 10 points to participate in the "mutual assistance incentive" system.

[0136] Example 4: Schematic diagram of auxiliary device access and permission closure structure

[0137] In first-grade Chinese language learning, when students click the "Audio Playback" button, the system redirects the content to a linked smart doll device for audio playback. The playback path includes: E-reader terminal triggers → Permission check → Broadcast content signature verification → Doll device decryption and broadcast The system only allows forwarding to the doll broadcasting module when the explanatory content in the e-book is marked as "playable" (i.e., the strategy output is accompanied by the AllowVoice flag).

[0138] The doll device does not have the ability to call models; it can only play authorized content. The broadcast path is closed within the system to prevent others from using ordinary speakers to replace the execution of system logic.

[0139] Example 5: Multi-terminal binding and student privacy protection practice mechanism

[0140] In one family, one parent linked three student devices (two e-readers and one tablet). Each device was successfully linked after the parent's identity was verified via facial recognition. The parent set policy preferences in a mobile app (such as a preference for life-related explanations and a slow speaking style), and these settings were synchronized to all student devices.

[0141] Throughout the entire usage process, the system does not collect the following information: Student's name, age, grade, gender, and school No learning profiles or behavioral scoring indicators are formed. The system independently selects its explanation strategy based on the question type and prompt strategy, rather than relying on student profiles.

[0142] All explanation request behavior chains are recorded under the parent's control path to ensure that the responsible party for the behavior is clear, data management is compliant, and boundary control is clear.

[0143] Example 6: Intercepting Mechanism for Abnormal Calls Triggered by Missing Policy Paths

[0144] In actual use, a student attempted to skip the prompts and directly request the system to generate answers to questions on the e-reader. Because this behavior was not authorized by the parent, not bound to a policy template, and lacked a valid prompt, the system identified the request path as a "non-policy path call," thus triggering the path blocking mechanism.

[0145] The specific processing procedure is as follows: Path integrity verification: The system found that the current request was missing the PromptID, the Policy TemplateID, and the Parent TraceID, and determined that the path was invalid; Model call interception: The system interrupts the call request and does not pass input content to any large model to prevent illegal generation; Abnormal Behavior Chain Record: The system generates a "Path Abnormal Interception Record", which includes fields such as terminal identifier, trigger time, abnormality type, and path summary, and writes it to the behavior chain log under the parent account; Platform alerts and user feedback: The platform automatically pushes a notification to the parent's app that "an illegal request from the student's end was detected and successfully blocked," and suggests reviewing whether the current policy settings are complete; Blocking tag archiving and auditing support: This behavior record is tagged with "path blocking" and enters the path optimization and auditing module to provide data basis for subsequent model permission scheduling.

[0146] This embodiment demonstrates the system's ability to accurately identify, immediately block, and trace unauthorized paths, strengthens the structural boundaries of policy-driven model calls, ensures that students complete their use within the framework set by parents, and prevents bypassing and abuse.

[0147] Example 7: Dynamic Scheduling and Control Mechanism in Multi-Model Collaborative Scenarios

[0148] A parent enabled the "student self-study" mode for their fifth-grade student and authorized them to use different styles of explanation templates in the system. When answering a Chinese reading comprehension question, the student used two different prompts: "life-like metaphor explanation" and "philosophical inductive summary," and requested that the system generate different styles of content to deepen their understanding.

[0149] The system execution flow is as follows: Prompt Recognition and Strategy Matching: The system analyzes the two prompts input by the student and matches them to "Life-like Metaphor Template V2.1" and "Philosophical Expression Template V3.0" respectively; Multi-model scheduling mechanism activated: Based on the strategy template settings, the system schedules Model-A (popular style) and Model-B (abstract style) respectively to generate explanation content; Behavior chain recording and structure encapsulation: Each request generates an independent behavior chain node, which is structured and includes fields such as QID, prompt type, strategy template ID, called model ID, generated summary, and model output score, and is uniformly included in the parent control chain; Parent-visual tracking interface: Parents can view the prompts, model names, output styles and behavior scores used by students in real time through the app, which helps them understand students' learning preferences and learning outcomes. Strategy Difference Analysis and Feedback Optimization: The system automatically scores based on the style differences between two calls and prompts parents to recommend a certain expression style as a future teaching preference setting.

[0150] This embodiment demonstrates the system's intelligent scheduling and path management capabilities in a multi-model environment. It not only supports students in switching styles according to their needs, but also achieves personalized optimization of learning paths through behavior chain recording and scoring mechanisms, and ensures that parents have real-time supervision and control over the entire process.

Claims

1. A student homework AI tutoring method based on parental guidance and access control, characterized in that, The method includes the following steps: (1) Terminal binding and parental identity authentication mechanism: The system receives parents' identity information and terminal device binding requests through the platform, and completes the authorization binding between parents' mobile terminals and students' learning terminals. The system uses methods including but not limited to ID card verification, facial recognition, and device fingerprint authentication to confirm the identity of parents and establishes a behavior chain structure with parents as the main controllers. The student terminals include, but are not limited to, devices such as e-readers, mobile phones, and tablets. (2) Collection of homework questions and context analysis: The system receives images or text of homework questions uploaded by parents or students. It performs structured analysis of the question information, identifies the question type, subject, keywords and difficult content, and generates multiple rounds of explanation context based on the expression goals set by parents. (3) Large model invocation and explanation strategy scheduling mechanism: After receiving structured question information, at least one generative large model is invoked to generate personalized explanation content based on the prompt-driven strategy module. The prompts include style tags, guidance methods, explanation objectives, etc. The system supports multi-model collaborative generation and provides style scheduling and expression version management mechanisms. (4) Mode control and permission path management: The system supports dual-path operation of "parent tutoring mode" and "student self-management mode". Parents have the right to reveal the answer and the right to stop the process. When a student requests to view the answer directly, the system will automatically pop up a confirmation interface for the parent. Parents can authorize "provide the answer" or "continue the explanation" through the control panel. At the same time, the system has a mechanism for recognizing and interrupting behavior that exceeds the scope of the rules. (5) Mechanism for sharing lesson plans with parents and reflecting on their learning: After completing the explanation of the questions, parents can choose to actively share the explanation process to the "Parent Lesson Plan" module. The system will structure and archive the explanation content and display it according to tags such as subject, grade, and region. The system will record the sharing behavior based on the parent's account and give incentive feedback such as points. The student terminal is unable to initiate the lesson plan sharing operation.

2. The method as described in claim 1, wherein the terminal binding mechanism supports binding one parent account to multiple student terminals and multiple parent collaboration terminals, and all terminal behavior records are affiliated with the main parent account for auditing and path tracing.

3. The method as described in claim 1, wherein the methods for collecting homework questions include, but are not limited to, uploading photos, handwriting recognition, text input, speech transcription, etc., and the system performs image and text recognition and knowledge point matching processing on the uploaded questions.

4. The method as described in claim 1, wherein the prompt-driven strategy module calls different expression style templates according to different prompt types, and the style types include, but are not limited to, colloquial, humorous, philosophical, structured, metaphorical explanations, etc.

5. The method as described in claim 1, wherein in the access control mechanism, the parent has the final authorization right to request "view the answer", and the system triggers a parent confirmation pop-up when the student initiates the request and records the authorization behavior to the behavior chain.

6. The method as described in claim 1, wherein the system has an illegal path identification and bypass blocking mechanism, and any request that does not call the model through the policy prompt will be identified as an illegal path and interrupted, blocked and logged.

7. The method as described in claim 1, wherein the system generates a behavior chain record for all calling behaviors, including fields such as QID, calling prompt, generated model identifier, style tag, device terminal number, and timestamp, for auditing and tracing.

8. The method as described in claim 1, wherein the student's learning behavior does not collect any personal identification information in the system, including but not limited to name, age, school, and actual grade, and all learning behavior data is only recorded under the parent's identity node.

9. The method as described in claim 1, wherein the "Parent Lesson Plan" module is a content sharing area that is visible and operable only by the parent's terminal, the system automatically generates a content title and tag for each lesson plan, and restricts only the sharing user to browse the lesson plan content of others.

10. The method as described in claim 1, wherein the broadcast content on the student terminal is limited to the explanation content generated with parental authorization, and auxiliary devices such as smart dolls, speakers and other playback devices can only call the content marked as "broadcastable" to perform voice output, and cannot generate explanation content independently.

Images