Construction method of education prediction large model based on AI students

Abstract

The invention relates to a construction method for an education prediction large model based on AI students, and the method comprises the following steps: 1, constructing a standard model for learners of 0-100 years old, collecting the cognitive behavior data of real learners, and forming a dynamic data set according to the number of days of age as a unit; 2, extracting a 6-18-year-old K12 stage data subset from the dynamic data set, and forming a standard student model by analyzing learning features; step 3, utilizing a large language model to perform inverse derivation on the dynamic data set and the standard student model, and generating a personalized learning path conforming to individual learning rules and cognitive development; and 4, dynamically adjusting a learning path by adopting reinforcement learning and a Monte Carlo tree search algorithm, and performing path prediction in combination with the starting point of the initial state of the student and the terminal point of the model model knowledge base.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence education technology, and in particular to a method for constructing an education prediction model based on AI students. Background Art

[0002] Existing approaches to designing personalized learning paths primarily rely on link prediction based on knowledge point dependencies within knowledge graphs. However, this approach lacks interactivity with learners, making it difficult to achieve highly personalized learning paths. Furthermore, traditional approaches lack the ability to conduct effectiveness experiments and verify the effectiveness of learning paths, resulting in low learning efficiency.

[0003] In existing technologies, there are deficiencies in the design of personalized learning paths, which are manifested in the lack of innovative interactive design, weak student data interaction, and the lack of heuristic and project-based design.

[0004] With existing technologies, personalized learning paths cannot solve the problem of learning anxiety.

[0005] Among existing technologies, the AI ​​teacher route has natural defects: the AI ​​teacher route is not as innovative as the AI ​​student route. The current automatic generation of personalized learning paths is nothing more than simulating real tutors. It only studies the tutors' learning and tutoring methods, and still designs personalized teaching paths from the teacher's perspective and with the teacher as the center. The learning path generated by this AI teacher route is actually not compatible with students' learning. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide a method for constructing an education prediction model based on AI students. It builds a data model of learners aged 0-100, uses the generation ability of the large model to predict and evolve personalized learning paths, and verifies the effectiveness of the path by simulating and predicting the learner's learning process for more than ten years.

[0007] The above-mentioned object of the present invention is achieved through the following technical solutions: A method for constructing an education prediction model based on AI students includes the following steps: Step 1: Build a standard model of learners aged 0-100 years old. By collecting cognitive behavior data of real learners, a dynamic data set is formed according to the age-day unit. Step 2: Extract a K12 data subset of students aged 6-18 from the dynamic data set, and form a standard student model by analyzing learning characteristics; Step 3: Using a large language model to derive the dynamic data set and the standard student model, predict and generate a personalized learning path that conforms to individual learning rules and cognitive development; Step 4: Use reinforcement learning and Monte Carlo tree search algorithm to dynamically adjust the learning path, and make path predictions based on the starting point of the student's initial state and the end point of the role model knowledge base.

[0008] As a further technical solution of the present invention: in step 4, random forest and clustering algorithms are introduced into the Monte Carlo tree search algorithm, and the human growth law is simulated by the L-system algorithm to generate branch paths, and the path weights are adjusted according to real-time feedback.

[0009] As a further technical solution of the present invention: in the step 4, it also includes the construction of a progress bar mechanism, which specifically includes the following steps: first, the real-time learning status is dynamically compared with the preset role model, and then the current learning progress is displayed through a visual interface to be equivalent to the age stage of the role model, and finally the path correction algorithm is triggered based on the comparison results.

[0010] As a further technical solution of the present invention: in step 4, the method for constructing the role model knowledge base includes the following steps: digitizing the growth trajectory of historical celebrities, recording learning behaviors accurately to the minute and second level, establishing a mapping relationship between the celebrity knowledge base and the knowledge graph, and realizing the synchronous simulation of psychological state and knowledge acquisition through interdisciplinary integration.

[0011] As a further technical solution of the present invention: it also includes a method for generating a blank student body, which specifically includes the following steps: first, stripping off the specific knowledge content in the standard model, then retaining the cognitive development laws and learning behavior patterns, and finally supporting the simulation of the initial learning conditions of different individuals through initialization settings.

[0012] As a further technical solution of the present invention: in step 3, the generation stage of the personalized learning path introduces a MOBA game mechanism, including: converting knowledge points into combat skill modules, setting a positive feedback mechanism between knowledge mastery and game character capabilities, and enhancing the fun of the path through multi-person collaborative learning tasks.

[0013] As a further technical solution of the present invention: in the step 2, it also includes a computing power resource configuration method, which specifically includes the following steps: calculating the learning data volume at a daily granularity, where the learning data volume includes three parts: textbooks, classrooms, and exercises, setting the token consumption formula: total token amount = number of days × (textbook tokens + classroom tokens + exercise tokens), and calculating the total resource consumption of the K12 stage through the model of 4380 days × 10,000 tokens / day.

[0014] As a further technical solution of the present invention: in the step 3, a dynamic comparison method is also included, which specifically includes the following steps: constructing an L-system growth algorithm to simulate the cognitive development curve, setting a dual-time axis comparison mechanism, the dual time axis is the actual age and the equivalent age of the role model, generating a staged growth report and pushing adaptive learning tasks.

[0015] As a further technical solution of the present invention: in step 1, the arrangement method of the dynamic data set includes the following steps: time series modeling is performed based on 4380 days (12 years × 365 days), and the daily data includes three-dimensional parameters: environmental interaction, physiological indicators, and cognitive load, and typical learning characteristics of age groups are extracted through a sliding window mechanism.

[0016] As a further technical solution of the present invention: it also includes a verification mechanism, which includes the following steps: using a blank student body to design an effectiveness experiment on the learning route, shortening the 18-year learning process simulation to minutes through a time compression algorithm, and setting multidimensional effectiveness indicators for path evaluation. The multidimensional effectiveness indicators include three aspects: knowledge mastery, cognitive development, and psychological matching.

[0017] In summary, the present invention includes at least one of the following beneficial technical effects: 1. This invention discloses a method for constructing a large-scale educational prediction model based on AI student technology. First, a digital simulation is performed based on the learning and cognitive data of learners aged 0-100. A standard model of learners aged 0-100 is then constructed. The generative capabilities of the large-scale model are then leveraged to evolve this basic model into thousands of personalized learning paths. Furthermore, after removing knowledge from the model, a naive learner experiment can be formed. This "blank knowledge learner" is then used to input the initial situation of a real learner. Computers are then used to simulate and evolve a person's learning process over a decade. This method can be used to verify the effectiveness of the designed personalized learning paths.

[0018] 2. This invention uses simulation to construct a data model (AI Student) of learners aged 0-100. By studying learning graphs, this 0-100-year-old learner model develops detailed, personalized learning paths that align with the learning habits of real students. The simulation process is identical to real-life learning, thus improving the accuracy of personalization. By building a data model of learners aged 0-100, the generative power of a large model is leveraged to predict and evolve personalized learning paths. The effectiveness of these paths is verified by simulating learners' learning processes over more than a decade.

[0019] 3. The invention's innovation lies in leveraging educational and biological methods with computer simulation capabilities to design a simulated learner—an AI student—ranging from 0 to 100 years old. This AI student, with no prior knowledge, then learns the curriculum, emulating the learning responses and outcomes of a real student. This computer-aided approach shortens the evolution of human learning.

[0020] 4. The "AI student in a knowledge-free state" of the present invention can serve as a good experimental subject. Named AI student, it can be allowed to simulate a person's learning process starting from the age of 0 using the initial conditions of a real student, and use it to simulate the learning process of an infant from childhood to adulthood. This method can provide a personalized learning path that conforms to the real situation, and then use powerful computing power to simulate the learning process of a real student for more than ten years in just a few minutes. In this way, the effectiveness of the learning path can be fully predicted and iterated, and dynamic adjustments and optimizations can be continuously made to better suit student learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the overall structural diagram of the AI ​​student model of the present invention.

[0022] Figure 2 Flowchart of the present invention. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application; it is obvious that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0024] Example 1: Reference Figure 1 and Figure 2 , a method for constructing an education prediction model based on AI students disclosed in the present invention, comprising the following steps: Step 1: First, preset learners aged 0-100 and build a 36,500-day database. The data of the previous day and the next day are sorted according to the growth patterns of people. Then, the learning data generated by the learners are stored in the corresponding data sets according to the corresponding number of days. Then, by studying the commonalities of the data sets, a high-quality learning data set that simulates real people is formed - AI data people.

[0025] Step 2: Extract a dataset for individuals aged 6-18 from the 0-100 AI data set to form a "simulated data set" for the K12 field—AI students. Simultaneously, various information acquisition devices are used to store student learning data generated in the teaching environment in the order of learning for individuals aged 6-18, thus generating a sufficient sample data set for students aged 6-18.

[0026] Step 3: In the 6-18 age group data set, the learning time of real people can be set to 12 years x 365 days = 4380 data subsets. At the same time, the collected data sets of multiple real students are analyzed to form the unique learning characteristics of this age group. The learning characteristics of the 4380 data sets are arranged together in real-life chronological order to form a data model for the learning of standard students aged 6-18 in the K12 field - AI students.

[0027] Step 4: Classify and filter the real-person student data in the subset to identify statistically significant patterns. A large amount of learning data shares common characteristics. Then, leveraging the AI ​​model's ability to draw inferences from one instance, simulate the subset students' potential reactions, confusion points, and eye expressions to each knowledge point. The model then generates learning states identical to those in the subset data. Finally, by incorporating biological research methods into the model to study the laws of biological growth and development, the model can automatically simulate student growth and development, creating a virtual student growth dataset—the AI ​​student.

[0028] Step 5: Designing Living Teaching Materials. For the simulated AI student, the curriculum is meticulously reconstructed, incorporating the design of an "ideal role model student." This involves recreating Newton, aged 0-24, with a 1:1 24-year learning process, detailed down to the minute. This effectively transforms the teaching materials into a living medium, transforming traditional textbooks into a visual learning process for real people. This also allows for interdisciplinary integration, providing psychological guidance and cultivating values ​​for real students. Through the AI ​​Newton design, the two-dimensional textbook can be transformed into a three-dimensional, visual learning process for real people.

[0029] Step 6: Strip the AI ​​student's data set of pure knowledge, retain the basic cognitive and learning laws, and form an AI blank student body with learning ability. It can start growing from 0 years old and grow to 24 years old according to the laws of human growth (which can be extended to 100 years old). Let it learn the course content independently, and then correct it through the "Role Model - AI Celebrity Knowledge Base (AI Newton)", independently run a suitable learning route, and independently summarize a scientific and feasible knowledge point map and a reliable learning path.

[0030] Step 7: The role model is the endpoint of the map, the current student's learning situation is the starting point, and the paths in between are different personalized learning paths. This invention utilizes Monte Carlo tree search algorithms, random forests, and clustering algorithms, mimicking the growth patterns of humans. From the starting point, the algorithm creates a personalized learning path. The starting and end points are fixed, but the process is dynamically adjustable, and each student will have a different personalized learning path. This path leverages the large model's ability to draw inferences from one instance to another, deriving the AI ​​student's data set and adjusting the basic growth patterns into personalized learning paths for each student.

[0031] Step 8: Consumption of computing resources: ① Textbook Reading (approximately 5,000 Tokens) Taking high school textbooks as an example, each page has about 500 words (about 250 tokens for Chinese). If you read for 2 hours a day (about 20 pages of textbooks for nine courses), you will generate about 5,000 tokens. Text containing formula symbols will generate more tokens. Classroom attendance (approximately 1,500-2,500 Tokens / class hour) ② Based on a 45-minute class with an average teacher speaking at 150 words per minute, a single class generates 5,000 words (approximately 2,500 tokens). If multimedia courseware (PPT text + graphic annotations) is used, mixed text and graphic content can increase token generation by 30%-50%.

[0032] ③ Exercise training (approximately 800-1,500 Tokens / subject) Assuming you complete 20 questions per day, you will consume approximately 1,000-1,600 tokens. If step-by-step instructions are included, you may receive an additional 30% tokens.

[0033] 5000+2500+1600=9100 tokens, each student learns about 10,000 tokens of data per day; The estimated data volume of the 4380 subsets is 4380x10000=43,800,000=43.8 million tokens. The generation cost of one million tokens is 10 yuan, so the cost of generating a personalized learning route for each student aged 6-18 is about 400 yuan. Similarly, the generation of this bionic dataset can also be used for the learning routes of learners aged 0-100.

[0034] Step 9: Dynamically compare the teaching patterns and learning content summarized from the subsets using a progress bar. This provides real-time corrections to learning direction and informs students of their current progress. Simultaneously, a biological "L-system algorithm" generates a real-time student growth model. Research based on frontline teaching data has shown that this method can effectively alleviate students' learning anxiety. The progress bar also provides information on which year and learning stage students are currently in during Newton's time, and how they can achieve the same level of knowledge and ability by the age of 18.

[0035] Step 10: Curriculum Reconstruction: Within the data subset, knowledge can be restructured and integrated into MOBA game elements to enhance learning addictiveness. Simultaneously, bionic AI students, stripped of knowledge, can utilize MOBA game play to develop personalized learning paths.

[0036] Step 11: Thousands of personalized learning paths for students aged 6-18 will be generated. By comparing them against standard role models, this provides an academic assessment method for measuring different students' learning levels. This method compares students' knowledge mastery, abandoning traditional score-based evaluation methods and focusing on individual student growth. At the same time, students nationwide can use AI students as a benchmark to align their abilities. With each new piece of content learned, students' abilities will be visually demonstrated, visualizing their growth and learning outcomes. This gives learning meaning and makes the learning process as engaging as a game.

[0037] Real personalized teaching should not be a rote learning method, but a new personalized innovative learning path designed with project-based and heuristic teaching. A real-life learning data model of Newton aged 0-18 is designed. This basic learning model simulates Newton's knowledge base and growth process one-to-one. Then, after setting the initial conditions of the real student, a virtual AI student starts learning the teaching materials from the age of 0. The "AI student" uses the computer's simulation evolution capability to simulate the 18-year learning process. Then, as the real student grows, the AI ​​student is used for guidance. At the same time, in order to avoid accuracy issues in the generation, the standard Newton learning model is used for comparison and correction to ensure the reliability of the personalized learning path designed by the present invention. Based on the basic learning model and the knowledgeless AI student, an algorithm can be used to generate hundreds of millions of different personalized learning paths. If a student studies physics, he or she can choose the Newton model. If he or she likes to study Chinese classics, a growth model of a Chinese classics master can be designed. At the same time, the present invention introduces a "progress bar mechanism" to display the learning of real students and the interaction of AI students in the form of a progress bar. After studying the course, the AI ​​students compare with the standard model and give a progress bar for the learning task, informing the real students how much knowledge reserves they have at this age and time period, which is equivalent to that of Newton. This technical route is equivalent to using the future self to guide the present self in learning and development.

[0038] For example, if we make a model of an AI doctor, it will tell you how much knowledge reserves you have when you are in high school, which is equivalent to that of the doctor, and then guide you towards a doctorate. This is project-based teaching and heuristic teaching, which helps students become the person they want to be, so that they can acquire the knowledge base of the person they want to be by following the role model.

[0039] The innovation of this invention is to design a role model learning model - AI virtual Newton, which simulates Newton's learning and growth process and simulates his data one-to-one. Then, when students are anxious, they can compare the AI ​​student with the role model and tell the real students that your current level is equivalent to the 16-year-old state of Newton in history. Don't worry, you are already great. As long as you keep studying, you can also have Newton's level of knowledge.

[0040] Therefore, the innovation of the personalized learning route of the present invention lies in designing the learning path with students as the center, simulating the learning model of students aged 0-24, guiding students with the student data model, and generating the student's learning path. It is equivalent to a 24-year-old student returning to the age of 0, and then using 24 years of memory to guide the 0-year-old self on how to grow and learn. This can solve the problem of matching the personalized path with the learner, and is also conducive to the learner's growth and learning according to his future self.

[0041] Previous personalized learning routes could not prove the effectiveness of learning, and there was no experiment to prove that the personalized routes generated by the large model were specific and effective. The data prediction and generation of AI personalized routes provided by this invention are based on the learning data of 6-18 years old generated in the teaching environment for simulation, in which each learning detail generated can be verified with the process Previous personalized learning routes were based on AI teacher routes, which studied how students should design them from the teacher's perspective. This is a relatively backward AI teacher concept. These routes are not necessarily the learning routes required by students. The learning effect of routes generated from the teacher's perspective is very poor. The AI ​​student route of this invention can be well designed from the student's perspective and based on the daily learning data of students, perfectly fitting the student's learning environment. The previous AI teacher route is equivalent to giving sunlight to saplings and giving fertilizer to saplings when they need it. However, the AI ​​student route is equivalent to studying what saplings need every day during their growth process, studying the daily growth patterns of saplings, and at the same time using large models to simulate the process of saplings (students) growing into big trees (role model). Biology has a detailed research process for plant growth, but pedagogy lacks a detailed process for tracking and studying student growth. Trees change statically, and students change dynamically. However, we can use interdisciplinary integration to study the student growth route in the field of education. At the same time, through information technology, we can predict students' future detailed learning routes to guide current students in their studies and successfully complete the informatization of education. Previous AI tutoring approaches were simply recommendation algorithms or problem-solving tools. These were based on simple data generation capabilities based on large models, but large models cannot handle complex design and construction tasks—for example, generating the tens of tons of blueprints for an aircraft carrier, which this LLM model clearly cannot accomplish. The AI ​​student designed by this invention is like designing an aircraft carrier from scratch. Then, the complete blueprints and design details of the aircraft carrier are fed into the large model, allowing the large model to complete the complex design work. This can transform a single aircraft carrier design into one capable of generating a variety of functions (the large model's ability to draw inferences from one example). The personalized learning path generated by this invention leverages the large model's ability to draw inferences from one example to another, evolving thousands of personalized learning path models from a detailed standard learning model. This requires a significant amount of computing power, approximately 40 million computing power tokens per student.

[0042] Innovation in the education evaluation mechanism: In the past, students were evaluated solely based on scores. This method is a single indicator and cannot directly reflect the students' actual learning level. This method provides a standard learning model as a benchmark. It uses a progress bar to measure the learning and mastery levels of different students, and provides the gap with the standard model, so that students can have a detailed learning plan down to the minute. This changes the previous extensive education model to a cost-effective, personalized and precise learning model.

[0043] The best teacher is the student at age 18. This invention can generate a future student to guide current students in their learning. Computer technology can shorten the time it takes for humans to evolve, verifying the effectiveness of learning paths for students aged 0-18. Previously, it took 18 years to determine if a learning path was effective. However, using initial student information and conditional initialization using an AI student, the AI ​​student is made to mimic the real student from age 0 to 18, generating a learning and growth path that suits the student. Furthermore, because the generated path is detailed, this invention can introduce a self-checking mechanism. This allows for the rapid iteration of the optimal solution for personalized learning paths, which previously took 18 years to verify, in just minutes, through computer technology. This completes the 18-year validation process in just a few minutes.

[0044] In step six of the present invention, the generation of the AI ​​student data set includes the following steps: Step a1: Record the student's current status (homework images, class videos, eye state, and other data during the teaching process); Step a2: Call the LLM model to analyze learning data indicators such as pictures, videos, and eye contact, and provide an analysis report; Step a3: Call the LLM model to generate prompt words, and then call the LLM large model to analyze the test paper; Step A4: Call the LLM model to understand and generate the context of 4380 data sets, read the materials, and understand the content of the materials: LLM analyzes the test paper and provides a logical chain for the question. Hint: The standard solution to this question is: If you are a teacher, please help me analyze the possible confusions students may have (list ten directions, and eliminate those that do not conform to students' learning patterns). If you are a student, please help me find learning resources that can solve the above problems and design a method for learning this knowledge point. Step a5: Generate a detailed study plan of 4380 days of learning methods and learning resources based on the knowledge points; Step a6: Generate data for the learning process of 4380 data sets in sequence; Step a7: The knowledge points and learning data in the 4,380 subsets are divided into locations similar to those on a map. The generated "model student" is the end point, and the initial diagnosis of the learning situation is the starting point. The collected learning path data of real-life top students is the optimal solution of the knowledge map. Then, based on the optimal solution, influencing factors are introduced to generate a suboptimal solution with influencing factors.

[0045] Based on the initial conditions of a real-life student, this invention constructs a virtual student, aged 0-18—an "AI student." This student is used to learn, simulating the growth and learning of a human from age 0 to 18. This virtual AI student can be used to learn most of the difficult textbook knowledge, and can also be used for cross-disciplinary learning. As the real-life student matures, the real-life student can leverage the imitation ability of the brain to replicate the AI ​​student's process one-to-one. This AI student's knowledge base can provide assistance, allowing the real-life student to grow and learn in the same way as the AI ​​student, and any discrepancies can be dynamically adjusted. The knowledge accumulated in the current academic community has surpassed what human learning can handle, and traditional textbooks no longer meet current educational requirements. There is a need to create a virtual AI student that can inherit knowledge and learn knowledge beyond the human lifespan, thereby better serving teaching and the development of human knowledge. Student learning should be like open-source programming: there's no need to reinvent the wheel or start from scratch over and over again. Instead, students should focus on innovation, leveraging the human brain's creative capabilities to the fullest.

[0046] AI student + textbook = AI living textbook. It's a virtual human with 18 years of real-life learning content. It possesses 18 years of knowledge learned according to human development patterns, and it's visualized, allowing real-life students to imitate and learn. In this embodiment, the AI ​​living textbook is explained in detail: the AI ​​virtual student has the same knowledge and cognitive structure as a human, with visualized knowledge modules that students can access. Students can access the knowledge points they need, tailored to the cognitive needs of the human brain. The human brain's learning and inheritance are age-limited and will encounter bottlenecks. After reaching 100 years old, further development becomes difficult. However, the growth of an AI brain can be continued. The biggest academic challenge is the lifespan of the human brain. AI students can infinitely construct knowledge flows, surpassing the upper limit of the human brain's learning capabilities. This design enables AI students to become the primary carrier of future knowledge. While humans can only learn for 100 years, AI students can continue to learn and inherit, iterating to become the most powerful learning brain.

[0047] A human brain might cease learning after a hundred years, and then a new brain would take over. However, AI brains are different. They can continuously build on the research of their predecessors, and as more data accumulates, academic capabilities emerge. AI living textbooks are like a knowledge transmission tower. They follow the learning patterns of humans and simulate real people, but they can live indefinitely, constantly learning. Once they have absorbed all human knowledge, real people can access this visualized knowledge on demand. Like an external hard drive in the human brain, the human brain can directly access and utilize this knowledge structure. This allows for direct access to any knowledge needed, allowing the knowledge of a small group to be shared by the majority. This ensures equal access to knowledge and reduces the learning burden for future students. Therefore, the innovation of AI living textbooks lies in mimicking the learning patterns of real human brains, following the progression of learning from 0 to 100 years.

[0048] Evaluation system for AI students: AI students can bring about a new evaluation system. Because AI students virtualize student data, they can introduce a progress bar to indicate the degree of completion of knowledge points. Furthermore, different students' virtual AI students can be compared and evaluated for their completion levels. Using a perfect AI student as a model for measurement, different students' completion levels can be evaluated and ranked based on completion, rather than traditional score rankings.

[0049] AI students complete student data collection and interaction, use AI students to communicate and interact with real students, and then use AI students to communicate with the virtual data world. AI students become a bridge for communication between the data world and the real world.

[0050] Common data structure for AI data people: AI students can be expanded to cover ages 0-100. For example, the 18-24-year-old AI student architecture can be used to absorb knowledge points from specific industries and fields in production and life, forming an 18-24-year-old AI data person. This person is identical to a real person and can absorb the computer programming knowledge of 18-24-year-olds to help programmers accumulate private domain data in the programming field. It can also absorb the robotics knowledge of 18-24-year-olds to help robotics companies build their own knowledge bases. It can also absorb the aircraft carrier manufacturing knowledge of 18-24-year-olds to form a knowledge base in the aircraft carrier manufacturing field. AI students aged 0-18, or even the complete AI student aged 0-100—the AI ​​data person—can become a universal industry data structure, collecting data from each industry in the form of a bionic real person. Simultaneously, the content of this data architecture can be learned through the capabilities of large models. After training, it becomes a proprietary predictive large model for the industry. It has been found that predictive large models are the best solution for large models to be applied in specialized fields.

[0051] The future AI internet will connect and communicate through different AI data humans. Each isolated industry will have its own unique AI data human (e.g., a database simulating real people with accumulated industry data spanning 0-100 years). Each human will communicate using a common protocol, and different AI data humans can be called upon to complete tasks in the AI ​​era. In the future AI universe, different AI data humans will be independent individuals with virtual IDs. Each AI data human will be a reflection of a real-world professional identity. They will be more powerful than real people and will have achieved a highly digitized representation of the real world, leading to the ultimate AGI.

[0052] The implementation principle of this invention is as follows: The invention discloses a method for constructing an AI-based educational prediction model. First, a digital simulation is performed based on the learning and cognitive data of learners aged 0-100. Then, a standard model of learners aged 0-100 is constructed. Then, using the generative capabilities of large models, this basic model is evolved into thousands of personalized learning paths. At the same time, after removing knowledge from the model, a knowledge-free learner experiment can be formed. This "blank knowledge learner" is used to input the initial situation of a real learner. Then, a computer is used to simulate and predict a person's learning process over a period of more than ten years. This method can verify the effectiveness of personalized learning paths.

[0053] The embodiments of this specific implementation method are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing an AI-based student education prediction model, characterized in that: The following steps are involved: Step 1: Build a standard model of learners aged 0-100 years old. By collecting cognitive behavior data of real learners, a dynamic data set is formed according to the age-day unit. Step 2: Extract a K12 data subset of students aged 6-18 from the dynamic data set, and form a standard student model by analyzing learning characteristics; Step 3: Using a large language model to derive the dynamic data set and the standard student model, predict and generate a personalized learning path that conforms to individual learning rules and cognitive development; Step 4: Use reinforcement learning and Monte Carlo tree search algorithm to dynamically adjust the learning path, and make path predictions based on the starting point of the student's initial state and the end point of the role model knowledge base.

2. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 4, random forest and clustering algorithms are introduced into the Monte Carlo tree search algorithm, and the human growth law is simulated by the L-system algorithm to generate branch paths, and the path weights are adjusted according to real-time feedback.

3. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 4, the construction of a progress bar mechanism is also included, which specifically includes the following steps: first, dynamically compare the real-time learning status with the preset role model, then display the current learning progress equivalent to the age stage of the role model through a visual interface, and finally trigger the path correction algorithm based on the comparison results.

4. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 4, the method for constructing the role model knowledge base includes the following steps: digitizing the growth trajectory of historical celebrities, recording learning behaviors accurately to the minute and second level, establishing a mapping relationship between the celebrity knowledge base and the knowledge graph, and realizing synchronous simulation of psychological state and knowledge acquisition through interdisciplinary integration.

5. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: It also includes a method for generating a blank student body, which specifically includes the following steps: first, stripping away the specific knowledge content in the standard model, then retaining the cognitive development laws and learning behavior patterns, and finally supporting the simulation of the initial learning conditions of different individuals through initialization settings.

6. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 3, the generation stage of the personalized learning path introduces MOBA game mechanisms, including: converting knowledge points into combat skill modules, setting a positive feedback mechanism between knowledge mastery and game character capabilities, and enhancing the fun of the path through multi-person collaborative learning tasks.

7. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 2, a computing resource configuration method is also included, which specifically includes the following steps: calculating the learning data volume at a daily granularity, where the learning data volume includes three parts: textbooks, classroom, and exercises; setting the token consumption formula: total token amount = number of days × (textbook token + classroom token + exercise token); and calculating the total resource consumption of the K12 stage through a model of 4380 days × 10,000 tokens / day.

8. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 3, a dynamic comparison method is also included, which specifically includes the following steps: constructing an L-system growth algorithm to simulate the cognitive development curve, setting a dual-time axis comparison mechanism, wherein the dual time axis is the actual age and the equivalent age of the role model, generating a stage-by-stage growth report and pushing adaptive learning tasks.

9. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: In step 1, the method for arranging the dynamic data set includes the following steps: performing time series modeling based on 4380 days (12 years × 365 days), wherein the daily data includes three-dimensional parameters of environmental interaction, physiological indicators, and cognitive load, and extracting typical learning characteristics of age groups through a sliding window mechanism.

10. The method for constructing an AI student-based education prediction model according to claim 1, characterized in that: It also includes a verification mechanism, which includes the following steps: using a blank student body to design an effectiveness experiment on the learning route, shortening the 18-year learning process simulation to minutes through a time compression algorithm, and setting multidimensional effectiveness indicators for path evaluation. The multidimensional effectiveness indicators include knowledge mastery, cognitive development, and psychological matching.

Images