Cultivation method and system based on AI large model

Abstract

The invention provides a learning training method and system based on an AI large model, and relates to the technical field of intelligent education. A learner portrait and a knowledge point graph are constructed, a learning path planning module is adopted to plan and optimize a learning path, a target learning path is determined, and a learning material recommendation engine is utilized to perform analysis to obtain customized learning resources. The method comprises the steps of obtaining an AI analysis large model, displaying a target learning path and customized learning resources to a target user for analysis learning, obtaining learning feedback data through a real-time feedback and evaluation module, carrying out dynamic analysis on the feedback data based on a backtracking mechanism adjustment module to obtain a learning optimization strategy, and carrying out training backtracking adjustment. The problem of low learning efficiency caused by insufficient learning resource precision and poor learning feedback and individuation degree is solved, and the effects of improving the learning efficiency and effect and meeting different learning individuation requirements are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent education, and particularly relates to a learning cultivation method and a learning cultivation system based on an AI large model. Background Art

[0002] With the rapid development of information technology, especially the continuous progress of artificial intelligence (AI) technology, the education industry is gradually developing towards the direction of intelligence and personalization. Most traditional education models adopt standardized teaching content and progress, and it is difficult to carry out personalized teaching planning and optimization according to factors such as students' individual differences, learning styles, interests and hobbies. This phenomenon not only leads to low learning efficiency, but also is difficult to meet the needs of different students, thus affecting the learning effect and teaching quality.

[0003] Traditional education methods usually rely on teachers' experience and textbook content, and often adopt a unified teaching plan and progress to impart knowledge. Although this method can ensure the imparting of basic knowledge, it lacks adaptability to students' personalized learning needs. In practical applications, there are significant differences in students' learning speeds, depths of knowledge mastery, learning interests and preferences, etc., and the traditional education model is difficult to provide real-time feedback and adjustment, resulting in problems such as confusion and lagging progress for students during the learning process. In addition, learning resources in the education process, such as textbooks, exercises, videos, etc., are usually unified and cannot be customized and recommended according to students' actual needs. In existing education technologies, although some intelligent education platforms and learning management systems have adopted a data-driven approach for learning progress tracking and evaluation, most of them can only provide limited personalized content recommendations, and the feedback mechanism is lagging, failing to form an effective closed loop. Many existing platforms rely on static course arrangements and textbooks and lack intelligent adjustment functions based on dynamic learning feedback and real-time evaluation. In addition, the current education system is difficult to effectively integrate large-scale and multi-level data analysis and natural language processing technologies to support the intelligent planning of learning paths and the customized recommendation of resources.

[0004] In summary, the existing education system has technical problems such as insufficient applicability and accuracy of learning resources, poor timeliness and personalization of learning feedback, inability to effectively adjust the learning path, resulting in low learning efficiency and unguaranteed teaching quality. Summary of the Invention

[0005] The present application provides a learning cultivation method and a learning cultivation system based on an AI large model, which are used to solve the technical problems existing in the existing education system, such as insufficient applicability and accuracy of learning resources, poor timeliness and personalization of learning feedback, inability to effectively adjust the learning path, resulting in low learning efficiency and unguaranteed teaching quality.

[0006] In view of the above problems, the present application provides a learning method and a learning system based on an AI large model.

[0007] In a first aspect, the present application provides a learning method based on an AI large model, the method comprising:

[0008] Constructing a learner profile and a knowledge point graph through a learning path planning module, a database, and a knowledge base application module; using the learning path planning module to perform learning path planning and optimization on the learner profile and the knowledge point graph to determine a target learning path; using a learning material recommendation engine to analyze the learner profile and the knowledge point graph to obtain customized learning resources; obtaining an AI parsing large model through a large model application service module, and based on a user interface module and the AI parsing large model, presenting the target learning path and the customized learning resources to a target user for parsing and learning, and obtaining learning feedback data through a real-time feedback and evaluation module; dynamically analyzing the learning feedback data based on a backtracking mechanism adjustment module to obtain a learning optimization strategy, and performing learning backtracking adjustment on the target learning path and the customized learning resources through the learning optimization strategy.

[0009] In a second aspect, the present application provides a learning system based on an AI large model, the system comprising:

[0010] A knowledge graph construction unit for constructing a learner profile and a knowledge point graph through a learning path planning module, a database, and a knowledge base application module; a path planning unit for performing learning path planning and optimization on the learner profile and the knowledge point graph using the learning path planning module to determine a target learning path; a learning customization unit for using a learning material recommendation engine to analyze the learner profile and the knowledge point graph to obtain customized learning resources; an analysis and learning unit for obtaining an AI parsing large model through a large model application service module, and based on a user interface module and the AI parsing large model, presenting the target learning path and the customized learning resources to a target user for parsing and learning, and obtaining learning feedback data through a real-time feedback and evaluation module; a dynamic analysis unit for dynamically analyzing the learning feedback data based on a backtracking mechanism adjustment module to obtain a learning optimization strategy, and performing learning backtracking adjustment on the target learning path and the customized learning resources through the learning optimization strategy.

[0011] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0012] A learning method based on an AI large model provided by the present application constructs a learner portrait and a knowledge point map through a learning path planning module, a database, and a knowledge base application module; uses the learning path planning module to plan and optimize the learning path for the learner portrait and the knowledge point map to determine the target learning path; uses a learning material recommendation engine to analyze the learner portrait and the knowledge point map to obtain customized learning resources; obtains an AI parsing large model through a large model application service module, and based on a user interface module and the AI parsing large model, presents the target learning path and the customized learning resources to the target user for parsing and learning, and obtains learning feedback data through a real-time feedback and evaluation module; dynamically analyzes the learning feedback data based on a backtracking mechanism adjustment module to obtain a learning optimization strategy, and performs backtracking adjustment of learning for the target learning path and the customized learning resources through the learning optimization strategy, solving the technical problems existing in the existing education system, such as insufficient applicability and accuracy of learning resources, poor timeliness and personalization degree of learning feedback, inability to effectively adjust the learning path, and thus resulting in low learning efficiency and inability to guarantee teaching quality, achieving the technical effects of personalized learning path planning, customized learning resource recommendation, and dynamic optimization based on real-time feedback, thereby improving learning efficiency and effect and meeting the personalized needs of different learners. Description of the Drawings

[0013] Figure 1 FIG. is a schematic flow chart of a learning method based on an AI large model provided by the present application.

[0014] Figure 2 FIG. is a schematic structural diagram of a learning system based on an AI large model provided by the present application.

[0015] Description of the reference numerals: knowledge graph construction unit 11, path planning unit 12, learning customization unit 13, parsing and learning unit 14, dynamic analysis unit 15. Detailed Embodiments

[0016] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0017] Embodiment 1, as Figure 1 shown, the present application provides a learning method based on an AI large model, and the method includes:

[0018] Construct a learner portrait and a knowledge point map through a learning path planning module, a database, and a knowledge base application module.

[0019] The learning path planning module is used to plan and optimize the learning path for the learner profile and the knowledge point graph, and determine the target learning path.

[0020] Specifically, the learning path planning module is an intelligent module based on data analysis. Its main function is to construct a learner profile for students according to their learning data, and determine their learning progress, knowledge mastery, and learning preferences. By collecting data such as the basic information, learning history, performance records, and learning preferences of learners through the learning path planning module, these data form the basis of the learner profile. The learner profile is a comprehensive description of the personal characteristics of learners, covering dimensions such as learners' learning ability, knowledge mastery, and learning style. For example, the learning speed of learners, the depth of mastery of a certain knowledge point, and the preferred learning methods, such as visual learning or auditory learning, can all be reflected in the profile. The learner profile can not only help the system accurately grasp the learning needs of learners, but also provide personalized data support for subsequent learning path planning.

[0021] At the same time, the knowledge point graph is constructed with the help of the database and knowledge base application module. The database and knowledge base application module establish the knowledge point graph by storing and organizing course content, learning resources, and their relationships. The knowledge point graph is a graphical representation that stores and visualizes each knowledge point and its relationships through graph database technology, showing the relationships between different knowledge points, such as the order of learning and the relevance of knowledge points. Each knowledge point is connected to other knowledge points through different relationships, forming a hierarchical knowledge structure. For example, the knowledge point of "algebra" can be associated with the knowledge point of "equation", and the equation can be further refined into sub-knowledge points such as "linear equation" and "quadratic equation". Through the construction of the graph, the system can understand the logical relationship before and after knowledge points, thus providing a scientific basis for learning path planning. The role of the graph database here is to store and efficiently query a large number of knowledge points and their relationships to ensure that the connection relationships between knowledge points can be clearly presented.

[0022] Next, based on the collected learning data, the learning path planning module analyzes the learning history of learners through natural language processing (NLP) technology, and extracts the key characteristics of the learning situation. The key characteristic set of the learning situation includes learning speed, knowledge point mastery, and learning style. Learning speed refers to the speed at which learners absorb and master new knowledge within a specific time period; knowledge point mastery measures the depth of understanding and application ability of learners for the knowledge they have learned; learning style refers to the preferred learning methods of learners, such as learning through videos, texts, interactions, etc. By analyzing these key characteristics, the system can analyze the personalized needs of learners, and thus plan the most suitable path for their learning progress and methods according to the different characteristics of different students.

[0023] During this process, the learning path planning module also combines the key learning situation feature set in the learner profile with the knowledge point graph for the planning and optimization of the learning path. The system plans a learning path suitable for the learner based on the learner's current knowledge level, combining the relationships between different knowledge points in the knowledge point graph, the sequential learning order, and the interdependence of required knowledge points. In this way, the learning path can not only meet the personalized needs of students but also ensure the coherence and systematicness of learning. For example, for learners with relatively poor basic knowledge, it may be recommended to start from basic knowledge points and gradually progress to more complex knowledge points; for learners with a certain foundation, based on their learning speed and mastery level, it will be recommended to skip some mastered basic knowledge and directly enter advanced learning content.

[0024] Next, through in-depth analysis of the learner's personalized needs, the learning path is optimized through a series of algorithms and models to determine the target learning path. This is not only an adjustment of the learning progress but also an adaptation of the learning content. For example, the system will dynamically adjust the learning path to ensure that students can smoothly transition to the next related knowledge point after mastering a knowledge point. This mode can effectively support subsequent learning path recommendations and resource customization, ensuring that the entire learning process conforms to the characteristics and needs of the learner. At this time, the collaborative work between the learning path planning module and the knowledge base application module constitutes the basic framework of the entire system, ensuring that learners can master the required knowledge to the greatest extent in the shortest time, providing the most accurate learning support for each learner, and greatly improving learning efficiency and effectiveness.

[0025] In short, through the collaborative work of the above modules, the entire learning path planning process will ensure that learners can smoothly and efficiently master and understand knowledge in a personalized learning path.

[0026] Analyze the learner profile and the knowledge point graph using the learning material recommendation engine to obtain customized learning resources.

[0027] Optionally, the learning material recommendation engine is designed to analyze the learner's personal learning profile and knowledge point map to provide personalized learning resources. This engine uses a variety of data analysis, machine learning, and natural language processing techniques to intelligently recommend the most suitable textbooks, video tutorials, exercise sets, or other teaching aids based on factors such as the learner's learning history, preferences, learning efficiency, and knowledge point mastery. The learning material recommendation engine first receives the detailed data of the learner profile and knowledge point map provided by the learning path planning module. Then, it uses advanced data analysis and natural language processing techniques to deeply analyze the profile and map, that is, to identify the learner's weaknesses and interests through algorithms, match appropriate learning content and difficulty, and thus formulate personalized learning resource recommendations. For example, if a learner has a low mastery level of a certain knowledge point, the recommendation engine will give priority to recommending relevant basic textbooks or exercises to help the learner strengthen the understanding and application in this area. For instance, suppose a learner's profile shows that he has strong logical understanding in the field of mathematics but is weak in geometry; at the same time, the knowledge point map indicates that geometric knowledge is the basis for understanding more advanced mathematical concepts. In this case, the learning material recommendation engine will analyze this data and recommend basic geometry tutorials, interactive geometry software, and video tutorials on problem-solving strategies related to geometry. These customized learning resources will improve the learner's knowledge mastery and interest in this area.

[0028] Through the above process, the learning material recommendation engine ensures that each learner can obtain customized learning resources that are most suitable for themselves, thereby maximizing learning efficiency and effectiveness. The application of this system not only improves the personalization of learning but also optimizes the overall learning experience and learning outcomes of learners through precise resource matching.

[0029] Obtain the AI parsing large model through the large model application service module. Based on the user interface module and the AI parsing large model, display the target learning path and the customized learning resources to the target user for parsing and learning, and obtain learning feedback data through the real-time feedback and evaluation module.

[0030] Exemplarily, the role of the large model application service module is to utilize advanced AI technologies to parse the large model through AI and process and present learning content in an intelligent manner, thereby providing a highly personalized and interactive learning environment for learners. The large model application service module is the core technical component of the system, undertaking intelligent parsing and generation tasks. This module deploys pre-trained large models based on the Transformer architecture, such as the GPT series, to process and understand a large amount of professional text data in the education field. These models are fine-tuned to be able to generate accurate AI parsing, including problem-solving steps, key point explanations, and related concept expansions, thus providing in-depth learning material understanding and generation capabilities. The fine-tuned model is the AI-parsed large model. The design of the large model application service module enables it to process large-scale data sets and utilize distributed training strategies to accelerate the pre-training process through multiple GPUs or TPUs. The trained AI-parsed large model has the ability to process complex data and generate learning paths and resources. For example, using information such as historical learning data, user interactions, and learning outcomes to optimize and train the AI model to ensure that the model can accurately understand and predict the needs of learners.

[0031] The obtained AI-parsed large model then works in collaboration with the user interface module to display the target learning path and customized learning resources to learners in a graphical and easy-to-interact manner. The user interface module, as an interactive platform, is responsible for presenting the generated target learning path and customized learning resources to learners in a user-friendly way. The design of the module is intuitive and easy to understand, and it can adjust the content display method in real time according to the operations and feedback of learners. For example, if the AI model analyzes that the learner prefers visual learning, the user interface may use more visual teaching resources such as charts and videos. The module usually includes a graphical progress bar, a course schedule, an interactive video player, and the function of dynamically adjusting the display of learning content. Through a clear visual design, learners can intuitively see their learning path and quickly understand their current learning stage, the next content, and the resources to be learned.

[0032] During the learning process, the AI-parsed large model continuously tracks and analyzes the progress of learners. When learners start to parse the content in the target learning path, the AI model will process the learners' input and feedback in real time and provide personalized learning suggestions and recommendations. For example, if learners perform poorly on a certain knowledge point, the AI model will automatically adjust the difficulty of the recommended resources and push more review materials or explanatory content suitable for that knowledge point.

[0033] The real-time feedback and evaluation module is responsible for collecting feedback data from learners, including learning progress, understanding level, knowledge mastery, and learning attitude, etc. These data are collected in real-time through the user interface and then transmitted to the AI parsing large model for processing and analysis. For example, after a learner completes a teaching unit, the system may require filling out a quick evaluation or automatically evaluate the learner's mastery of the unit content through algorithms. The collected feedback data is analyzed by the AI parsing large model and used to further optimize the learning path and resources. According to the learner's actual performance and feedback, the model may adjust the difficulty, format, or order of subsequent learning content to better meet the learner's needs. This process is iterative, and as more and more data is analyzed, the adjustment of the learning path and resources will become more and more precise. For example, if a learner performs excellently in a certain part of the learning, the system may automatically recommend them to higher-difficulty content; conversely, the system will rearrange easier content for review.

[0034] Through the above steps, the collaborative work of the large model application service module and the user interface module not only enables learners to obtain the most suitable learning experience for their needs, but also continuously improves the teaching effect and learning efficiency through continuous feedback and optimization cycles.

[0035] Based on the backtracking mechanism adjustment module, the learning feedback data is dynamically analyzed to obtain a learning optimization strategy, and the target learning path and the customized learning resources are adjusted through the learning optimization strategy.

[0036] Specifically, the backtracking mechanism adjustment module receives learning feedback data from the real-time feedback and evaluation module. This data includes information such as the learner's learning progress, grades, wrong-question records, and understanding degree of learning content. By dynamically analyzing the learning feedback data, learning bottlenecks, progress lags, or understanding obstacles that may exist in the learning process of the learner are identified. For example, if a learner performs poorly in mastering a certain knowledge point, the feedback data will reflect the error frequency or abnormal learning duration on related questions. Subsequently, the backtracking mechanism adjustment module performs time series analysis on these data and evaluates the trend changes in the learning process through data processing algorithms to further determine which learning strategies have not achieved the expected results. Based on this analysis, the system generates a learning optimization strategy, which includes adjusting the difficulty in the learning path, the order of learning content, or increasing intensive practice for weak knowledge points. The goal of the learning optimization strategy is to help learners overcome existing learning obstacles through more precise adjustment of learning resources.

[0037] Based on the generated optimization strategy, the backtracking mechanism makes training backtracking adjustments to the target learning path and customized learning resources, that is, dynamically corrects the learning path and recommended learning resources according to the optimization strategy. For example, if a learner lags behind in the learning progress of a certain chapter, the system will suggest reviewing the basic content of that chapter or adding more practice materials, and adjust the learning path to ensure the coherence and efficiency of learning. This adjustment is a continuous and iterative optimization process. Whenever the learner provides new feedback data, the backtracking mechanism will re-analyze and further optimize the learning path and resources, so as to achieve a more personalized and efficient learning experience.

[0038] Through the cyclic feedback of this mechanism, it is ensured that each learner can continuously progress on the most suitable learning path for themselves and ultimately achieve the best learning effect.

[0039] Furthermore, constructing the learner portrait and knowledge point graph through the learning path planning module includes:

[0040] Collecting and obtaining the learning data set of the target user through the learning path planning module, where the learning data set includes basic information, learning history, performance records, and preference settings; using natural language processing technology to extract key features from the learning data set to obtain the key feature set of learning conditions, and the key feature set of learning conditions includes learning speed, knowledge point mastery level, and learning style; constructing the learner portrait according to the key feature set of learning conditions; using the graph database to store knowledge points and the relationships between knowledge points through the knowledge base application module to construct the knowledge point graph.

[0041] Furthermore, the learning path planning module starts the construction process by collecting the learning data set of the target user. The learning data set includes multi-dimensional data such as the basic information, learning history, performance records, and preference settings of the learner. This information provides the basis for subsequent data analysis and feature extraction. Among them, the basic information refers to the personal profile of the learner, including but not limited to name, age, gender, contact information, the school or educational institution where the learner is located, etc. The basic information helps the system maintain the uniqueness and personalization of records when processing learning data; the learning history refers to all learning activity records left by the learner during the use of the system, which includes the courses participated in, the modules completed, the interaction time, and past learning achievements, etc. By analyzing the learning history, the learning trajectory and progress of the learner can be understood; the performance records refer to the specific grades and performances obtained by the learner during the learning process, including the results of various tests, quizzes, exams, and other assessment activities. The performance records are an important data source for evaluating the learning effect and understanding degree of the learner, and can be used to analyze the learning effectiveness and identify learning difficulties; while the preference settings involve the personal choices and settings of the learner for the learning environment or teaching materials, which may include the preferred learning time, preferred learning methods (such as visual, auditory, or hands-on), interface configuration, notification settings, etc. The preference settings help to personalize the learning experience and make learning more in line with the personal habits and comfort of the learner.

[0042] Next, using natural language processing (NLP) technology, the module deeply analyzes the collected data and extracts the key learning situation characteristics of the learner. These key characteristics include learning speed, knowledge point mastery, and learning style, etc. They are the core elements for constructing the learner profile. Natural language processing (NLP) is an interdisciplinary field of computer science, artificial intelligence, and linguistics. It is committed to enabling computers to understand, interpret, generate, and process the content of human language, including from processing and understanding the basic structure of words and phrases to more complex language understanding tasks such as sentiment analysis, language translation, semantic understanding, and dialogue systems. NLP technology enables computers to perform tasks such as speech recognition, natural language understanding, natural language generation, and machine translation. Through NLP technology, key learning situation characteristics such as learning speed, knowledge point mastery, and learning style can be extracted from various sources such as the learner's interactions and assignment submissions. These characteristics can then be used for the formulation and optimization of personalized learning paths to ensure that the learning resources highly match the actual needs of the learner. The set of key learning situation characteristics refers to a set of indicators that describe the learning behavior and performance of the learner extracted from the learning data. These characteristics are used to understand and optimize the learning process of the learner. Furthermore, a learner profile is constructed based on the set of key learning situation characteristics. The learner profile reflects the learning ability and preferences of the learner and provides customized parameters for personalized learning path planning.

[0043] In addition, the construction of the knowledge point graph is completed through the knowledge base application module. This module uses a graph database to store and manage knowledge points and their interrelationships, ensuring a structured and systematic expression of the knowledge system. Graph databases are very suitable for storing complex network structure data, such as the dependency relationships, prerequisite and successor conditions between knowledge points. Through such a structured representation, the learning path planning module can more accurately identify and recommend the learning sequence and path that learners should follow to ensure the coherence and systematicness of learning.

[0044] Generally speaking, through the above steps, the system can provide each learner with a highly personalized and efficient learning experience, significantly improving the quality and effectiveness of learning.

[0045] Furthermore, the construction of the knowledge point graph includes:

[0046] Obtain learning question bank data through the knowledge base application module, and construct a learning material knowledge base based on the learning question bank data; preprocess and represent the knowledge of the learning material knowledge base to obtain structured knowledge point data, where the structured knowledge point data includes entities, relationships, and attributes; use NLP technology and pre-trained models to perform vector conversion and storage on the structured knowledge point data to construct the knowledge point graph.

[0047] Optionally, obtain learning question bank data through the knowledge base application module. The learning question bank data contains a series of questions and answers about a specific learning field, and in-depth understanding of the learning content can be obtained through these data. The knowledge base application module extracts relevant knowledge information from the question bank and provides basic data for subsequent steps. These data include the knowledge points and their relationships involved by learners in different questions, providing the original materials for constructing the knowledge graph. Next, construct a learning material knowledge base according to the learning question bank data. This process involves organizing the data in the question bank into structured knowledge resources. By analyzing the content of each question and answer, the system can extract the core knowledge points from them and store them in the learning material knowledge base in a unified format. The learning material knowledge base not only stores knowledge points but also records the interrelationships between knowledge points, such as dependency relationships, sequence, etc.

[0048] Subsequently, preprocess and represent the knowledge base of learning materials. During this process, clean, normalize, and standardize the data in the knowledge base of learning materials to make it suitable for further analysis and storage. The preprocessing includes removing redundant information, eliminating noisy data, and transforming unstructured text information into structured data. Then, convert the knowledge points into a form that can be calculated and analyzed through a knowledge representation method system. Generally, these structured knowledge point data include entities, relationships, and attributes. An entity refers to the knowledge point itself, a relationship refers to the logical or dependency relationship between knowledge points, and an attribute describes the specific characteristics of each entity.

[0049] After the data preprocessing and knowledge representation are completed, use NLP technology and pre-trained models to perform vector conversion and storage on the structured knowledge point data. Natural Language Processing technology (NLP) and pre-trained models (such as GPT) can convert the structured knowledge point data into vector representations, that is, map each knowledge point to a set of digital vectors, which capture the semantics and relationships between knowledge points. Through this vectorization method, knowledge points can be stored and retrieved more efficiently, and it is also convenient for subsequent machine learning and reasoning operations. Finally, construct the knowledge point graph. The knowledge point graph forms a comprehensive knowledge network by integrating all structured knowledge point data, entities, relationships, and attributes. The nodes in the graph represent knowledge points, and the edges represent the associations or dependency relationships between knowledge points. This knowledge point graph provides rich semantic information and structured support for subsequent learning path planning, recommendation systems, etc., and can help the system accurately understand and recommend learning content.

[0050] Through the above steps, the knowledge point graph not only contains rich learning resources in terms of content but also reflects the internal connections between knowledge points in terms of structure, thus providing strong support for personalized learning.

[0051] Furthermore, the determination of the target learning path includes:

[0052] Obtain a learning path planning algorithm through the learning path planning module; use the learning path planning algorithm to perform learning path planning and comparison on the learner profile and the knowledge point graph to obtain an initial learning path; train a learning effect prediction model using a machine learning model, evaluate the initial learning path based on the learning effect prediction model to obtain learning effect prediction information; optimize and adjust the initial learning path based on the learning effect prediction information to determine the target learning path.

[0053] In a specific embodiment, the learning path planning module obtains a learning path planning algorithm. The learning path planning module uses a series of algorithms to determine the optimal learning path for the learner. In this step, the learning path planning module generates a preliminary learning route for the learner by calling a preset learning path planning algorithm based on the information in the learner profile and the knowledge point graph. The learning path planning algorithm formulates a personalized learning path framework by considering multiple factors, such as the learner's current ability, mastered knowledge points, learning style, etc. Next, the learning path planning algorithm is used to perform learning path planning and comparison on the learner profile and the knowledge point graph to obtain an initial learning path. In this step, the learning path planning algorithm compares and matches the data in the learner profile (such as learning progress, mastered knowledge points, learning style, etc.) with the structured knowledge points in the knowledge point graph to generate a preliminary learning path. This path will show the order of knowledge points that the learner needs to learn and the learning objectives at each stage, ensuring that the knowledge points in the learning process progress step by step without missing key content.

[0054] Subsequently, a learning effect prediction model is obtained through training with a machine learning model. To further optimize the learning path and predict its effect, the system uses a machine learning model to train the potential effect of the learning path. This prediction model analyzes past learning behaviors based on historical learning data and the learner's personalized information, and then predicts the effects of different learning paths. Through continuous training, the learning effect prediction model can more accurately evaluate the adaptability of the learning path to the learner and its final effect. After obtaining the learning effect prediction model, the initial learning path is evaluated based on the learning effect prediction model to obtain learning effect prediction information. At this time, the learning effect prediction model will predict the effect of this path in actual learning according to the structure of the initial learning path and the characteristics of the learner. The evaluation content includes the learner's mastery of each knowledge point, learning efficiency, progress arrangement, etc. The learning effect prediction information provides quantitative data support for subsequent optimization, enabling the path planning to be adjusted and optimized according to the prediction results.

[0055] Finally, based on the learning effect prediction information, the initial learning path is optimized and adjusted to determine the target learning path. By analyzing the learning effect prediction information, the system optimizes and adjusts the initial learning path to improve learning efficiency and effect. The optimization process includes adjusting the order of learning content, modifying the difficulty of learning tasks, rearranging learning time, etc., to ensure that the learner can learn at the most suitable pace and in the most suitable way for themselves, so as to achieve the best learning effect.

[0056] Through the above steps, the finally determined target learning path not only meets the personalized needs of learners, but also can be dynamically adjusted according to the prediction information to ensure the smooth progress of the learning process. This process, through precise planning and intelligent optimization, helps learners efficiently master the required knowledge and improve learning effectiveness.

[0057] Furthermore, the obtaining of customized learning resources includes:

[0058] Performing collaborative filtering analysis on the learner profile through the learning material recommendation engine to obtain learning preference prediction features; combining a pre-trained language model to perform semantic content analysis on the knowledge point graph to obtain learning material semantic features; performing content matching recommendation on the learning material semantic features and the learning preference prediction features through a recommendation system based on specific rules to obtain the customized learning resources.

[0059] Specifically, performing collaborative filtering analysis on the learner profile through the learning material recommendation engine to obtain learning preference prediction features. The learning material recommendation engine is a data-driven intelligent system that uses the collaborative filtering analysis method to predict the learning resources that learners may like in the future through the historical behavior and preference data of learners. Collaborative filtering analysis is based on the following assumption: If learner A has similar preferences to learner B for certain learning resources, then A may be interested in other resources that B likes. Through this method, the recommendation engine can predict personalized learning preference features according to the interests and learning styles of learners. The learning preference prediction features include the learning methods, learning content preferences, learning time allocation, etc. of learners, and these features are the basic data for subsequent resource recommendation.

[0060] Next, combining a pre-trained language model to perform semantic content analysis on the knowledge point graph to obtain learning material semantic features. Pre-trained language models (such as GPT, etc.) can effectively understand the deep meaning of language through pre-training on a large amount of text data. In this stage, the pre-trained model is used to perform semantic analysis on each knowledge point in the knowledge point graph to extract its semantic features. Each knowledge point in the knowledge point graph not only has its own label, but also contains the context information and related concepts of this knowledge point, and these information help the model understand the core content of each knowledge point and its relationship with other knowledge points. Through semantic content analysis, the system can extract more accurate and hierarchical semantic features from the knowledge point graph, providing richer information for subsequent resource recommendation.

[0061] Then, a recommendation system based on specific rules performs content matching and recommendation on the semantic features of the learning materials and the predicted features of the learning preferences to obtain the customized learning resources. In this step, the system uses specific rules in the recommendation system to match the semantic features of the learning materials and the predicted features of the learners' preferences. The rules of the recommendation system are usually based on experience and algorithm models. For example, if a learner shows interest in a certain knowledge area, the system will give priority to recommending learning materials in that area; or when the learner's preferences highly match the semantic features of some learning resources, the system will push these resources to the learner. Through this content matching, learners can obtain learning resources that best meet their current needs and interests, such as textbooks, exercise questions, video tutorials, etc.

[0062] In summary, the acquisition of customized learning resources is achieved through in-depth analysis of the learner portrait and the knowledge point map, and by combining advanced semantic processing and recommendation algorithms to achieve accurate learning resource recommendation. In this way, learners can find the most suitable content among rich learning resources, thereby improving learning efficiency and effectiveness.

[0063] Furthermore, the method further includes:

[0064] Obtain the training sub-task information of the training, design a proprietary knowledge data set according to the training sub-task information; perform data preprocessing based on the proprietary knowledge data set to obtain task sample data; perform training and fine-tuning on the AI parsing large model based on the task sample data.

[0065] Further, obtain the training sub-task information. The training sub-task information refers to the detailed requirements and goals for a specific learning task, such as learning the concepts or skills of a specific subject. This information is collected from the database of the education system or directly from educators. For example, if the task is to help students understand the basics of machine learning, then the sub-task information will include relevant knowledge points such as the main concepts of machine learning, basic algorithms, and practical applications. Then, design a proprietary knowledge data set according to the training sub-task information. The proprietary knowledge data set is a data set specially constructed to meet specific learning tasks, including the text of relevant textbooks, questions and answers, and the transcribed text of teaching videos. Designing such a data set requires determining which content is directly related to the sub-task, and then screening and compiling this content from a wide range of educational resources. For example, to train a model for parsing machine learning concepts, the data set may include machine learning-related content extracted from textbooks, scientific research articles, and online courses.

[0066] After that, data preprocessing is performed based on the proprietary knowledge dataset. Data preprocessing is an essential step in data analysis, including cleaning data (removing irrelevant information), standardizing formats (unifying text formats, handling data missing problems, etc.), and data transformation (such as converting text into a format that the model can process). The preprocessed data is the task sample data, which is organized and provides a suitable input format for the subsequent model training. For example, text data may be converted into word vectors or other types of numerical representations. Finally, the AI parsing large model is trained and fine-tuned based on the task sample data. Fine-tuning refers to further specialized training using task sample data on the basis of a well-trained general AI model, enabling the model to better adapt to specific applications. During the fine-tuning process, the model will learn to identify and process data features and patterns related to specific learning tasks, thereby optimizing its performance in actual educational scenarios. For example, using task sample data of machine learning basic knowledge to fine-tune the AI model to make it more accurate in processing and answering queries related to machine learning.

[0067] Through the above steps, an AI model can be effectively customized for specific educational tasks, improving its accuracy and applicability in actual applications, and thus better serving the needs of education.

[0068] Furthermore, obtaining the learning optimization strategy includes:

[0069] Based on the backtracking mechanism adjustment module, the learning feedback data is arranged in chronological order to obtain learning time series feedback data; progress analysis and result prediction are performed on the learning time series feedback data to obtain learning progress information and learning result prediction performance; based on the learning progress information and learning result prediction performance, dynamic analysis of the learning strategy is carried out to obtain the learning optimization strategy.

[0070] Specifically, based on the backtracking mechanism adjustment module, the learning feedback data is arranged in chronological order to obtain learning time series feedback data. Learning feedback data is various information generated by learners during the learning process, such as learning progress, exam scores, homework performance, online interaction records, etc. The first step of the backtracking mechanism adjustment module is to arrange this data in chronological order to form a series of learning time series feedback data. This time series data can help the system understand the progress of learners and reflect their learning performance at each time node. For example, if a learner performs well in the early stage of learning but encounters difficulties in the later stage, the time series data can clearly show this change trend.

[0071] Next, progress analysis and outcome prediction are performed on the learning time series feedback data to obtain learning progress information and learning outcome prediction performance. By analyzing the learning time series data, the system can identify the learner's learning progress, that is, the completion status of the learner in the entire learning path, whether the learner has completed each task according to the predetermined schedule. At the same time, the system will also perform outcome prediction, predicting the possible grades or understanding levels that the learner may achieve in the subsequent learning based on the learning progress and performance. The analysis at this stage provides detailed feedback on the learner's current state, enabling a clear view of the learner's performance at each stage and predicting their future learning effects. For example, if the learner progresses slowly in mastering a certain knowledge point, the system can predict that the learner's learning outcome in this area may lag behind.

[0072] Then, based on the learning progress information and learning outcome prediction performance, dynamic analysis of learning strategies is carried out to obtain the learning optimization strategy. After obtaining the learning progress information and learning outcome prediction performance, the system will perform dynamic analysis of learning strategies based on these data. The purpose of dynamic analysis is to continuously evaluate and optimize learning strategies during the learning process to ensure that the selection of the learning path and learning resources matches the actual situation of the learner. This analysis will adjust the learning plan according to the learner's current performance. For example, if the learner has a low mastery level in a certain chapter, the system may recommend more review materials or extend the learning time for this part to help the learner make up for the knowledge gap. The learning optimization strategy generated in this process is a dynamically adjusted learning path and learning resource allocation strategy, ensuring that the learner is always on the most suitable learning trajectory during the changing learning process.

[0073] Through the above steps, the learning optimization strategy can be precisely adjusted based on the learner's real-time feedback, ensuring that the learning content and progress are always consistent with the learner's needs, thereby improving learning efficiency and the final outcome. The core of this process lies in providing an intelligent learning path optimization solution based on the learner's current state through the analysis of time series data and outcome prediction, significantly enhancing the personalization and effectiveness of the learning process.

[0074] Through the technical solution of the above embodiment, a training method based on an AI large model provided by this application solves the technical problems existing in the existing education system, such as insufficient applicability and accuracy of learning resources, poor timeliness and personalization of learning feedback, and inability to effectively adjust the learning path, which in turn leads to low learning efficiency and unguaranteed teaching quality. It achieves the technical effects of personalized learning path planning, customized learning resource recommendation, and dynamic optimization based on real-time feedback, thereby improving learning efficiency and effect and meeting the personalized needs of different learners.

[0075] Embodiment 2. Based on the same inventive concept as the learning method based on the AI large model in the foregoing embodiment, as Figure 2 shown, this application provides a learning system based on the AI large model, and the system includes:

[0076] A knowledge graph construction unit 11, configured to construct a learner portrait and a knowledge point graph through a learning path planning module, a database, and a knowledge base application module.

[0077] A path planning unit 12, configured to use the learning path planning module to perform learning path planning and optimization on the learner portrait and the knowledge point graph, and determine a target learning path.

[0078] A learning customization unit 13, configured to analyze the learner portrait and the knowledge point graph by using a learning material recommendation engine to obtain customized learning resources.

[0079] An analysis and learning unit 14, configured to obtain an AI analysis large model through a large model application service module, and based on a user interface module and the AI analysis large model, display the target learning path and the customized learning resources to a target user for analysis and learning, and obtain learning feedback data through a real-time feedback and evaluation module.

[0080] A dynamic analysis unit 15, configured to perform dynamic analysis on the learning feedback data based on a backtracking mechanism adjustment module to obtain a learning optimization strategy, and perform learning backtracking adjustment on the target learning path and the customized learning resources through the learning optimization strategy.

[0081] Furthermore, the knowledge graph construction unit 11 is further configured to perform the following steps:

[0082] Collect and obtain the learning data set of the target user through the learning path planning module, where the learning data set includes basic information, learning history, performance records, and preference settings; use natural language processing technology to extract key features from the learning data set to obtain a key feature set of the learning situation, where the key feature set of the learning situation includes learning speed, knowledge point mastery level, and learning style; construct the learner portrait according to the key feature set of the learning situation; use a graph database to store knowledge points and the mutual relationships between knowledge points through the knowledge base application module to construct the knowledge point graph.

[0083] Furthermore, the knowledge graph construction unit 11 is further configured to perform the following steps:

[0084] Obtain learning question bank data through the knowledge base application module, and construct a learning material knowledge base according to the learning question bank data; preprocess and represent the knowledge of the learning material knowledge base to obtain structured knowledge point data, where the structured knowledge point data includes entities, relationships, and attributes; use NLP technology and pre-trained models to perform vector conversion and storage on the structured knowledge point data to construct the knowledge point graph.

[0085] Furthermore, the path planning unit 12 is also used to perform the following steps:

[0086] Obtain a learning path planning algorithm through the learning path planning module; use the learning path planning algorithm to perform learning path planning and comparison on the learner portrait and the knowledge point graph to obtain an initial learning path; use a machine learning model to train and obtain a learning effect prediction model, and evaluate the initial learning path based on the learning effect prediction model to obtain learning effect prediction information; optimize and adjust the initial learning path based on the learning effect prediction information to determine the target learning path.

[0087] Furthermore, the learning customization unit 13 is also used to perform the following steps:

[0088] Perform collaborative filtering analysis on the learner portrait through the learning material recommendation engine to obtain learning preference prediction features; combine a pre-trained language model to perform semantic content analysis on the knowledge point graph to obtain learning material semantic features; perform content matching and recommendation on the learning material semantic features and the learning preference prediction features based on a recommendation system with specific rules to obtain the customized learning resources.

[0089] Furthermore, the system further includes a model fine-tuning unit, which is used to obtain training sub-task information, design a proprietary knowledge data set according to the training sub-task information; perform data preprocessing based on the proprietary knowledge data set to obtain task sample data; perform training and fine-tuning on the AI analysis large model based on the task sample data.

[0090] Furthermore, the dynamic analysis unit 15 is also used to perform the following steps:

[0091] Arrange the learning feedback data in chronological order based on the backtracking mechanism adjustment module to obtain learning time series feedback data; perform progress analysis and result prediction on the learning time series feedback data to obtain learning progress information and learning result prediction performance; perform dynamic analysis of learning strategies based on the learning progress information and learning result prediction performance to obtain the learning optimization strategy.

[0092] Through the foregoing detailed description of a learning method based on an AI large model in this specification, those skilled in the art can clearly know a learning system based on an AI large model in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For related parts, reference may be made to the description in the method section.

[0093] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A training method based on AI big model, characterized in that: The method comprises: Build learner portraits and knowledge point maps through learning path planning modules and database and knowledge base application modules; Using a learning path planning module to plan and optimize the learning path of the learner portrait and the knowledge point map to determine the target learning path; Analyze the learner portrait and the knowledge point map using a learning material recommendation engine to obtain customized learning resources; Obtaining an AI analysis big model through a big model application service module, presenting the target learning path and the customized learning resources to target users for analysis and learning based on a user interface module and the AI ​​analysis big model, and obtaining learning feedback data through a real-time feedback and evaluation module; Based on the backtracking mechanism adjustment module, the learning feedback data is dynamically analyzed to obtain a learning optimization strategy, and the target learning path and the customized learning resources are adjusted through backtracking through the learning optimization strategy.

2. A training method based on AI big model as claimed in claim 1, characterized in that: The learning path planning module is used to construct a learner portrait and a knowledge point map, including: Acquire a learning data set of the target user through a learning path planning module, wherein the learning data set includes basic information, learning history, achievement records and preference settings; Using natural language processing technology to extract key features of the learning data set to obtain a key feature set of learning conditions, wherein the key feature set of learning conditions includes learning speed, knowledge point mastery and learning style; Constructing the learner portrait according to the key feature set of the learning situation; The knowledge base application module uses a graph database to store knowledge points and relationships between knowledge points to construct the knowledge point graph.

3. A training method based on AI big model as claimed in claim 2, characterized in that: The constructing of the knowledge point graph includes: Acquire learning question and answer question bank data through the knowledge base application module, and construct a learning material knowledge base based on the learning question and answer question bank data; Preprocessing and knowledge representation are performed on the learning material knowledge base to obtain structured knowledge point data, wherein the structured knowledge point data includes entities, relationships and attributes; The structured knowledge point data is vectorized and stored using NLP technology and a pre-trained model to construct the knowledge point graph.

4. The AI ​​big model-based learning method according to claim 1, characterized in that: Determining the target learning path includes: Acquire a learning path planning algorithm through the learning path planning module; Using the learning path planning algorithm to plan and compare the learner portrait and the knowledge point map to obtain an initial learning path; Using a machine learning model to train a learning effect prediction model, evaluating the initial learning path based on the learning effect prediction model, and obtaining learning effect prediction information; The initial learning path is optimized and adjusted based on the learning effect prediction information to determine the target learning path.

5. The AI ​​big model-based learning method according to claim 1, characterized in that: Obtaining customized learning resources includes: Perform collaborative filtering analysis on the learner profile through the learning material recommendation engine to obtain learning preference prediction features; Performing semantic content analysis on the knowledge point graph in combination with a pre-trained language model to obtain semantic features of the learning material; The recommendation system based on specific rules performs content matching recommendation on the semantic features of the learning materials and the learning preference prediction features to obtain the customized learning resources.

6. The AI ​​big model-based learning method according to claim 1, characterized in that: The method comprises: Acquire student training subtask information, and design a proprietary knowledge data set based on the student training subtask information; Performing data preprocessing based on the proprietary knowledge data set to obtain task sample data; The AI ​​analysis model is trained and fine-tuned based on the task sample data.

7. The AI ​​big model-based learning method according to claim 1, characterized in that: The learning optimization strategy includes: Arranging the learning feedback data in chronological order based on the backtracking mechanism adjustment module to obtain learning time series feedback data; Performing progress analysis and outcome prediction on the learning time series feedback data to obtain learning progress information and learning outcome prediction performance; Based on the learning progress information and the predicted performance of learning outcomes, a dynamic analysis of the learning strategy is performed to obtain the learning optimization strategy.

8. A training system based on AI big model, characterized by: A system for implementing a learning method based on an AI big model as described in any one of claims 1 to 7, the system comprising: The knowledge graph construction unit is used to construct learner portraits and knowledge point graphs through the learning path planning module and the database and knowledge base application module; A path planning unit, configured to use a learning path planning module to plan and optimize the learning path for the learner portrait and the knowledge point map, and determine a target learning path; A learning customization unit, used to analyze the learner portrait and the knowledge point map using a learning material recommendation engine to obtain customized learning resources; The parsing learning unit is used to obtain the AI ​​parsing big model through the big model application service module, present the target learning path and the customized learning resources to the target user for parsing learning based on the user interface module and the AI ​​parsing big model, and obtain learning feedback data through the real-time feedback and evaluation module; A dynamic analysis unit is used to dynamically analyze the learning feedback data based on the backtracking mechanism adjustment module to obtain a learning optimization strategy, and to perform training and backtracking adjustments on the target learning path and the customized learning resources through the learning optimization strategy.

Images