Multidisciplinary knowledge point filling method and system based on ai adaptive algorithm

Abstract

The application provides a multi-disciplinary knowledge point filling method and system based on an AI adaptive algorithm, relates to the technical field of data processing, and comprises the following steps: constructing a multi-disciplinary knowledge graph and constructing a cross-disciplinary knowledge coupling network; performing learning level target analysis according to target section characteristics, and constructing a disciplinary target pyramid; introducing a disciplinary adaptive scoring channel to perform learning level scoring, and establishing a learning pyramid of each discipline; constructing a target weakness first graph network through multi-level weakness detection and cross-disciplinary weakness association reinforcement; and performing continuous closed-loop filling on a target learner according to a target weakness second graph network through global forgetting characteristic tracking compensation. The application solves the technical problem that, in the prior art, it is difficult to continuously track and compensate for the forgetting and knowledge gaps of learners, resulting in a scattered state of the filling strategy that only treats the symptoms and not the root cause, and further affecting the learning effect, improves the accuracy of knowledge point filling, and thus effectively improves the learning efficiency.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method and system for filling gaps in multidisciplinary knowledge based on AI adaptive algorithms. Background Technology

[0002] Currently, most existing knowledge gap-filling methods employ traditional single-subject assessment and filling strategies, typically targeting learners' current weaknesses. However, they lack continuous tracking and compensation for long-term forgetting and knowledge gaps. During the learning process, especially over long periods, learners may gradually forget previously acquired knowledge. The failure to provide timely feedback and correction for this forgetting leads to filling strategies that only address the symptoms, not the root cause, ultimately impacting overall learning effectiveness. Furthermore, existing technologies often lack the ability to identify the inherent connections between interdisciplinary knowledge points, making it difficult to accurately identify learners' knowledge gaps across different subjects during the filling process, thus affecting overall learning outcomes.

[0003] In summary, existing technologies suffer from the technical problem that the difficulty in continuously tracking and compensating for learners' forgetting and knowledge gaps leads to sporadic and ineffective remediation strategies, further impacting learning outcomes. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for filling in knowledge gaps in multiple disciplines based on AI adaptive algorithms, in order to solve the technical problem that existing technologies have difficulty in continuously tracking and compensating for learners' forgetting and knowledge gaps, resulting in a scattered state of filling gaps that only treats the symptoms and not the root cause, which further affects the learning effect.

[0005] To achieve the above objectives, this application provides a method and system for filling in knowledge gaps across multiple disciplines based on AI adaptive algorithms.

[0006] Firstly, this application provides a method for filling gaps in multidisciplinary knowledge points based on AI adaptive algorithms. This method is implemented through a system for filling gaps in multidisciplinary knowledge points based on AI adaptive algorithms. The method includes: constructing a multidisciplinary knowledge graph based on the target learner's multidisciplinary learning dataset; performing cross-disciplinary knowledge coupling feature analysis on the multidisciplinary knowledge graph to construct a cross-disciplinary knowledge coupling network; and performing learning level target analysis on the multidisciplinary knowledge graph based on the target learner's target learning stage characteristics to construct a cross-disciplinary knowledge coupling network. Construct a target pyramid for each subject; introduce a subject-adaptive scoring channel to score the learning level of the multi-disciplinary learning dataset, and establish a learning pyramid for each subject; perform multi-level weakness detection on the learning pyramids of each subject based on the target pyramids of each subject, and construct a weakness detection tower for each subject; strengthen the cross-disciplinary weakness association of the weakness detection towers of each subject based on the cross-disciplinary knowledge coupling network, and construct a first target weakness graph network; perform global forgetting characteristic tracking and compensation based on the first target weakness graph network to obtain a second target weakness graph network, and continuously perform closed-loop reinforcement for the target learner based on the second target weakness graph network.

[0007] Optionally, knowledge points are extracted from the multidisciplinary learning dataset to obtain knowledge point groups for each discipline; knowledge associations within each discipline are identified based on the knowledge point groups for each discipline to obtain knowledge associations within each discipline; and knowledge graph structure is reconstructed based on the knowledge associations within each discipline to generate the multidisciplinary knowledge graph.

[0008] Optionally, the dependency relationships between knowledge points of different disciplines are identified based on the multidisciplinary knowledge graph, and a cross-disciplinary knowledge point dependency vector set is constructed; the support relationships between knowledge points of different disciplines are identified based on the multidisciplinary knowledge graph, and a cross-disciplinary knowledge point support vector set is constructed; the transfer relationships between knowledge points of different disciplines are identified based on the multidisciplinary knowledge graph, and a cross-disciplinary knowledge point transfer vector set is constructed; the multi-relationship collaborative graph structure is sorted out based on the cross-disciplinary knowledge point dependency vector set, the cross-disciplinary knowledge point support vector set, and the cross-disciplinary knowledge point transfer vector set to generate the cross-disciplinary knowledge coupling network.

[0009] Optionally, based on the multidisciplinary knowledge graph, the knowledge graph of the Sth subject corresponding to the Sth subject is extracted, where S is a positive integer; based on the characteristics of the target learning stage, memory scoring targets are analyzed for each knowledge point in the knowledge graph of the Sth subject, and an Sth memory scoring target layer is constructed; based on the characteristics of the target learning stage, comprehension scoring targets are analyzed for each knowledge point in the knowledge graph of the Sth subject, and an Sth comprehension scoring target layer is constructed; based on the characteristics of the target learning stage, application scoring targets are analyzed for each knowledge point in the knowledge graph of the Sth subject, and an Sth application scoring target layer is constructed; based on the Sth memory scoring target layer, the Sth comprehension scoring target layer, and the Sth application scoring target layer, a pyramid mapping model is performed to generate the target pyramid of the Sth subject.

[0010] Optionally, based on the subject-adaptive scoring channel, the S-th subject-specific learning scoring model is activated; the S-th subject-specific learning dataset is extracted from the multi-disciplinary learning dataset, and knowledge point learning behavior features are identified based on the S-th subject-specific learning dataset to construct learning behavior feature vectors for each knowledge point; the learning behavior feature vectors for each knowledge point are input into the S-th learning scoring model to obtain learning scoring sequences for each knowledge point; learning level scoring clustering is performed based on the learning scoring sequences for each knowledge point to construct an S-th memory scoring layer, an S-th comprehension scoring layer, and an S-th application scoring layer; pyramid mapping modeling is performed based on the S-th memory scoring layer, the S-th comprehension scoring layer, and the S-th application scoring layer to generate a learning pyramid for the S-th subject.

[0011] Optionally, the subject-adaptive scoring channel includes multiple learning scoring models corresponding to multiple subjects, and the learning scoring indicators of each learning scoring model include memory score, comprehension score and application score.

[0012] Optionally, the S-th memory scoring layer is subjected to deviation detection based on the S-th memory scoring target layer to obtain the S-th memory scoring deviation sequence; based on the memory scoring deviation threshold, the S-th subject knowledge graph is used to locate memory weaknesses according to the S-th memory scoring deviation sequence to construct the S-th memory weakness layer; the S-th subject knowledge graph is used to locate comprehension weaknesses based on the S-th comprehension scoring target layer and the S-th comprehension scoring layer to construct the S-th comprehension weakness layer; the S-th subject knowledge graph is used to locate application weaknesses based on the S-th application scoring target layer and the S-th application scoring layer to construct the S-th application weakness layer; the S-th subject knowledge graph is used to analyze the correlation characteristics of the S-th memory weakness layer, the S-th comprehension weakness layer, and the S-th application weakness layer to construct the S-th weakness correlation layer; and the S-th subject weakness detection tower is generated by multi-level weakness tower fusion based on the S-th weakness correlation layer.

[0013] Optionally, a personal forgetting prediction model is constructed by performing deep learning on the target learner's historical knowledge mastery change dataset; the personal forgetting prediction model is then used to predict the forgetting of knowledge points in the multidisciplinary learning dataset to obtain forgetting curves for each knowledge point; the forgetting curves for each knowledge point are analyzed for intensity based on future time zones to determine the forgetting intensity of each knowledge point; based on forgetting intensity thresholds, risk detection is performed on the multidisciplinary knowledge graph according to the forgetting intensity of each knowledge point to obtain a forgetting risk knowledge graph; and the first target weakness graph is reconstructed by global forgetting characteristic compensation based on the forgetting risk knowledge graph to generate a second target weakness graph.

[0014] Optionally, based on the multidisciplinary learning dataset, multidisciplinary learning form preference analysis is performed to determine multiple subject learning preference characteristics; based on the multiple subject learning preference characteristics and the target weakness second graph network, multidisciplinary gap-filling decision is made to determine a global gap-filling strategy; when the target learner performs knowledge gap-filling learning according to the global gap-filling strategy, an updated learning dataset is generated synchronously; based on the updated learning dataset, weakness detection is performed to obtain an updated weakness graph network, and the target weakness second graph network is compared with the updated weakness graph network to generate a weakness comparison graph network; based on the weakness comparison graph network, gap-filling effect is evaluated to obtain gap-filling effect evaluation results, and the global gap-filling strategy is optimized based on the weakness comparison graph network and the gap-filling effect evaluation results.

[0015] Secondly, this application also provides a multidisciplinary knowledge gap filling system based on AI adaptive algorithms, used to execute the multidisciplinary knowledge gap filling method based on AI adaptive algorithms as described in the first aspect. The multidisciplinary knowledge gap filling system based on AI adaptive algorithms includes: a cross-disciplinary knowledge coupling feature parsing module, used to construct a multidisciplinary knowledge graph based on the target learner's multidisciplinary learning dataset, and to perform cross-disciplinary knowledge coupling feature parsing on the multidisciplinary knowledge graph to construct a cross-disciplinary knowledge coupling network; a learning level target parsing module, used to perform learning level target parsing on the multidisciplinary knowledge graph based on the target learner's target learning stage characteristics to construct a target pyramid for each subject; and a learning level evaluation module. The system is divided into several modules: a subject-adaptive scoring channel for hierarchical scoring of the multi-disciplinary learning dataset, establishing learning pyramids for each subject; a multi-level weakness detection module for performing multi-level weakness detection on the learning pyramids of each subject based on the target pyramids of each subject, constructing a weakness detection tower for each subject; a cross-disciplinary weakness association reinforcement module for strengthening the cross-disciplinary weakness associations of the weakness detection towers of each subject based on the cross-disciplinary knowledge coupling network, constructing a first target weakness graph network; and a closed-loop gap filling module for tracking and compensating for global forgetting characteristics based on the first target weakness graph network, obtaining a second target weakness graph network, and continuously performing closed-loop gap filling for the target learner based on the second target weakness graph network.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: by using multidisciplinary knowledge graphs and interdisciplinary coupling networks, combined with learners' learning data, personalized weakness detection and knowledge gap filling are carried out. At the same time, by considering learners' forgetting curves, interdisciplinary root cause diagnosis, accurate positioning of cognitive levels, and predictive continuous intervention are achieved, ensuring the continuity and effectiveness of learning, improving the accuracy of knowledge gap filling, and thus effectively improving learning efficiency.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the multidisciplinary knowledge gap filling method based on AI adaptive algorithms used in this application.

[0019] Figure 2This is a schematic diagram of the structure of the multidisciplinary knowledge gap filling system based on AI adaptive algorithm in this application.

[0020] Figure labeling: Interdisciplinary knowledge coupling feature analysis module 11, learning level goal analysis module 12, learning level scoring module 13, multi-level weakness detection module 14, interdisciplinary weakness association reinforcement module 15, closed-loop gap filling module 16. Detailed Implementation

[0021] This application provides a multidisciplinary knowledge gap filling method and system based on AI adaptive algorithms. It addresses the technical problem in existing technologies where the difficulty in continuously tracking and compensating for learners' forgetting and knowledge gaps leads to fragmented, ineffective gap-filling strategies that merely address the symptoms, further impacting learning outcomes. By utilizing multidisciplinary knowledge graphs and interdisciplinary coupling networks, combined with learners' learning data, personalized weakness detection and knowledge gap filling are performed. Furthermore, by considering learners' forgetting curves, interdisciplinary root cause diagnosis, precise cognitive level localization, and predictive continuous intervention are achieved, ensuring the continuity and effectiveness of learning, improving the accuracy of knowledge gap filling, and thus effectively enhancing learning efficiency.

[0022] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0023] Example 1, please refer to the appendix. Figure 1 This application provides a method for filling gaps in multidisciplinary knowledge points based on AI adaptive algorithms. This method is applied to a system for filling gaps in multidisciplinary knowledge points based on AI adaptive algorithms. The specific steps of this method are as follows:

[0024] Based on the multidisciplinary learning dataset of the target learners, a multidisciplinary knowledge graph is constructed, and cross-disciplinary knowledge coupling feature analysis is performed on the multidisciplinary knowledge graph to construct a cross-disciplinary knowledge coupling network.

[0025] Furthermore, this application also includes the following steps: extracting knowledge points from the multidisciplinary learning dataset to obtain knowledge point groups for each discipline; identifying intra-disciplinary knowledge associations based on the intra-disciplinary knowledge point groups to obtain intra-disciplinary knowledge associations; and performing knowledge graph structured reconstruction on the intra-disciplinary knowledge point groups based on the intra-disciplinary knowledge associations to generate the multidisciplinary knowledge graph.

[0026] Specifically, the multidisciplinary learning dataset of the target learners undergoes preprocessing and in-depth analysis. Natural language processing technologies such as named entity recognition and keyword extraction are used to scan textbook texts, test question stems and explanations, study notes, and other content, automatically identifying the core knowledge terms involved. Simultaneously, combined with learning behavior data, the core knowledge terms are deduplicated, normalized, and categorized by subject, ultimately forming categorized knowledge point groups for each subject. The multidisciplinary learning dataset of the target learners is a collection of information such as learning behavior, learning progress, and test scores across multiple subjects, containing various types of data for analyzing their learning situation and weaknesses. The subject-specific knowledge point groups are collections of knowledge points grouped according to subject categories.

[0027] Based on the knowledge point groups of each discipline, this involves identifying intradisciplinary knowledge relationships. Within a single discipline, it analyzes and determines various logical relationships between knowledge points, including prerequisite / successor relationships, parent-child / part-whole relationships, parallel / similar relationships, and application relationships. A prerequisite / successor relationship is like learning knowledge A being a necessary foundation for learning knowledge B; a parent-child / part-whole relationship is like knowledge A being a subordinate concept or component of knowledge B; a parallel / similar relationship is like knowledge A and knowledge B being logically parallel or methodologically similar; and an application relationship is like knowledge A being a commonly used tool when solving problems involving knowledge B. This process generates an internal structural blueprint for each discipline describing the close connections between knowledge points, thus revealing the intradisciplinary knowledge relationship relationships for each discipline's knowledge point groups.

[0028] Based on the knowledge relationships within each discipline, a structured knowledge graph is reconstructed for each discipline's knowledge point groups. This involves formally storing and constructing all knowledge point entities within each discipline's knowledge point groups and the relationships defined in the knowledge relationships within each discipline. Each knowledge point becomes a node in the graph, with node attributes including metadata such as discipline, name, and description. Each edge represents a relationship type and has a weight, indicating the strength of the relationship. The multidisciplinary knowledge graph organizes its main structure by discipline, integrating all knowledge points and their internal relationships within the same graph framework. This not only displays the knowledge relationships within each discipline but also identifies learners' knowledge gaps and weaknesses across different disciplines through cross-disciplinary connections.

[0029] By extracting knowledge points and identifying interdisciplinary relationships, we can help understand learners' knowledge mastery in various subjects. Furthermore, through the structured reconstruction of knowledge graphs, we can effectively organize these knowledge points and their interrelationships, and accurately identify learners' knowledge weaknesses in various subjects.

[0030] Furthermore, this application also includes the following steps: identifying dependencies between knowledge points of different disciplines based on the multidisciplinary knowledge graph, and constructing a cross-disciplinary knowledge point dependency vector set; identifying support relationships between knowledge points of different disciplines based on the multidisciplinary knowledge graph, and constructing a cross-disciplinary knowledge point support vector set; identifying transfer relationships between knowledge points of different disciplines based on the multidisciplinary knowledge graph, and constructing a cross-disciplinary knowledge point transfer vector set; and organizing the multi-relationship collaborative graph structure based on the cross-disciplinary knowledge point dependency vector set, the cross-disciplinary knowledge point support vector set, and the cross-disciplinary knowledge point transfer vector set to generate the cross-disciplinary knowledge coupling network.

[0031] Specifically, this method involves analyzing the interdisciplinary knowledge coupling characteristics of a multidisciplinary knowledge graph, examining the relationships between knowledge points from different disciplines within the graph, and identifying the coupling characteristics between these knowledge points, including dependency relationships, supporting relationships, and transfer relationships. Based on the constructed multidisciplinary knowledge graph, three recognition channels are activated in parallel to identify dependency relationships, supporting relationships, and transfer relationships, respectively.

[0032] For each subject's knowledge points, we identify dependencies by analyzing their relevance to other subjects. For example, calculus and algebra in mathematics have a close dependency because learning calculus requires a certain level of algebraic knowledge. These relationships are transformed into vectors, forming a cross-disciplinary knowledge point dependency vector set. This vector set represents the strength and specific form of the dependency between knowledge points in different subjects. Cross-disciplinary knowledge point dependency means that a knowledge point in one subject is a mandatory prerequisite for learning a knowledge point in another subject. If knowledge point α in subject A is not mastered, knowledge point β in subject B cannot be correctly understood or applied; this is a strong sequential relationship.

[0033] Beyond dependencies, this involves identifying the supporting role of certain subject knowledge points on other subjects. For example, trigonometric function graphs and properties in mathematics support simple harmonic motion in physics, while molecular structure knowledge in chemistry provides the foundation for understanding the reaction mechanisms of organic chemistry. By identifying these supporting relationships, a cross-disciplinary knowledge point support vector set is generated. This vector set represents the strength and direction of the supporting relationship between knowledge points from different subjects. Cross-disciplinary knowledge point support relationships mean that a knowledge point in one subject provides important foundational background, a framework for understanding, or commonly used tools for knowledge points in another subject, but it is not absolutely mandatory. The lack of supporting knowledge may lead to difficulties in understanding or inefficiency, but it does not mean that learning is completely impossible.

[0034] In the process of knowledge transfer, knowledge points from one discipline can support the learning of other disciplines. For example, statistical methods in mathematics can be used for data processing in physics, and thermodynamics knowledge in engineering can be transferred to the calculation of reaction heat in chemistry. By analyzing the knowledge transferability between disciplines, a cross-disciplinary knowledge point transfer vector set is constructed to represent these transfer relationships and their potential impact. Cross-disciplinary knowledge point transfer relationships mean that the thinking methods, analytical models, or problem-solving strategies underlying knowledge points in different disciplines are highly similar or interconnected; mastering one helps to quickly understand and master another.

[0035] The constructed interdisciplinary knowledge point dependency vector set, interdisciplinary knowledge point support vector set, and interdisciplinary knowledge point transfer vector set are used as three different types of edges. Based on a unified multidisciplinary knowledge graph node, these sets are fused and integrated to handle possible overlaps or conflicts, ultimately generating an interdisciplinary knowledge coupling network containing multiple types of edges. This interdisciplinary knowledge coupling network is a comprehensive graph network of multidisciplinary knowledge points and their interrelationships, demonstrating the dependencies, supports, and transfer relationships between different disciplines. It achieves a comprehensive understanding of the intrinsic connections between disciplines through the collaborative work of these dependencies, supports, and transfer relationships.

[0036] A single node pair may be connected by multiple edges of different types. For example, between mathematical vectors and physical forces, there are both dependency and transfer relationships. The structure analysis of multi-relationship collaborative graphs mainly requires relation normalization and conflict resolution. For instance, dependencies are assigned the highest priority, and dependency edges are retained when dependency and support relationships coexist. Simultaneously, relations of the same type from different identification channels are merged, and their overall weights are updated. An interdisciplinary knowledge coupling network is a network composed of knowledge points from different disciplines, where these knowledge points represent their dependency, support, or transfer relationships through edges. This network considers not only knowledge relationships within a discipline but also the interdisciplinary connections and knowledge transferability, forming a comprehensive knowledge structure graph used to identify connections and knowledge gaps between different disciplines.

[0037] For example, a multidisciplinary knowledge graph of student A, containing 200 nodes and over 300 intra-disciplinary relationship edges, along with the anonymized historical learning sequences and answer records of 10,000 students on the same learning platform, are used as the data source for relationship mining. Among students with low mastery of solving quadratic equations in mathematics, their average accuracy rate for physics-uniformly accelerated linear motion displacement calculation questions is only 31%, far lower than the overall average accuracy rate of 78%. Statistical tests show that this difference is extremely significant. Simultaneously, the expert rule base marks this as a strong dependency, thus generating a dependency vector: (Source: Mathematics - Solving quadratic equations, Target: Physics - Displacement calculation, Type: Dependency, Weight: 0.92). The semantic model calculates that the cosine similarity between mathematics-trigonometric function graphs and physics-simple harmonic motion is 0.86, which is extremely high. In the textbook text, they both appear in the chapter describing vibration using sine functions, generating a support vector: (Source: Mathematics - Trigonometric function graphs, Target: Physics - Simple harmonic motion, Type: Support, Weight: 0.80). The extraction module extracts the thought tags vector composition / geometric decomposition from cases of physics-force composition and parallelogram law, and vector operation / coordinate decomposition from mathematics-vector linear operations. The core thought matching degree of the two reaches 0.88. A transfer vector is generated: (Source: physics-force composition, Target: mathematics-vector operation, Type: transfer, Weight: 0.85). Hundreds of such interdisciplinary relationship vectors, such as the final identification of 45 dependency relationships, 60 support relationships, and 30 transfer relationships, are integrated with the original knowledge graph. After sorting, a coupled network containing 200 nodes and over 435 edges is generated. In this network, clicking on the mathematics-solving quadratic equation node shows that it emits multiple high-weight dependency edges pointing to multiple physics knowledge points, intuitively displaying its core hub status.

[0038] By constructing an interdisciplinary knowledge coupling network, the interdependence, support, and transfer relationships between knowledge points from different disciplines are clearly demonstrated, helping learners understand the intrinsic connections between disciplines, revealing the deep-seated relationships between disciplines, and improving learning efficiency and effectiveness.

[0039] Based on the target learning stage characteristics of the target learners, the learning level target of the multi-disciplinary knowledge graph is analyzed, and a target pyramid for each subject is constructed.

[0040] Furthermore, this application also includes the following steps: extracting the knowledge graph of the Sth subject corresponding to the multidisciplinary knowledge graph, where S is a positive integer; performing memory scoring target analysis on each knowledge point in the knowledge graph of the Sth subject according to the characteristics of the target learning stage, and constructing the Sth memory scoring target layer; performing comprehension scoring target analysis on each knowledge point in the knowledge graph of the Sth subject according to the characteristics of the target learning stage, and constructing the Sth comprehension scoring target layer; performing application scoring target analysis on each knowledge point in the knowledge graph of the Sth subject according to the characteristics of the target learning stage, and constructing the Sth application scoring target layer; performing pyramid mapping modeling based on the Sth memory scoring target layer, the Sth comprehension scoring target layer, and the Sth application scoring target layer to generate the target pyramid of the Sth subject.

[0041] Specifically, the system randomly extracts the knowledge graph corresponding to the Sth subject from a multi-disciplinary knowledge graph, where S is a positive integer. Subject extraction is performed based on different values ​​of S. For example, when S=1, the system extracts all knowledge points of mathematics and their relationships; when S=2, it extracts knowledge points of physics, and so on. In this way, the knowledge points and relationships of each subject are extracted independently, forming an independent subject knowledge graph.

[0042] The characteristics of the target learning stage are the specific requirements stipulated by the curriculum standards, mainstream teaching syllabus or specific training objectives corresponding to the learning stage of the target learner. They define the depth and breadth of mastery of each knowledge point that students should achieve by the end of the stage and serve as the basis for setting learning objectives.

[0043] Based on the characteristics of the target learning stage, a memory scoring target analysis is performed for each knowledge point in the knowledge graph of subject S. A memory scoring target refers to the memorization standard that learners need to achieve for each knowledge point. For example, in mathematics, the knowledge points learners need to remember may include formulas, theorems, or concepts. In physics, learners need to remember mechanics formulas, laws of electromagnetism, etc. By setting memory scoring targets for each knowledge point, the memory ability that learners should achieve at that stage is customized according to the characteristics of the learning stage. A basic target template is automatically matched based on the type of knowledge point, and then fine-tuned based on the fundamental position of that knowledge point in the entire graph. The stronger the fundamental position, the higher the required memory target score is usually. The S-level memory scoring target layer includes the memory scoring targets corresponding to each knowledge point within the knowledge graph of subject S.

[0044] Based on the characteristics of the target learning stage, an analysis of comprehension scoring objectives is conducted for each knowledge point in the knowledge graph of subject S. This analysis focuses on grasping the internal logic, interrelationships, causes, and significance of knowledge; that is, a deep analysis of the understanding of each knowledge point. The comprehension scoring objectives require learners not only to memorize knowledge points but also to understand their connotations and applications. The S-level comprehension scoring objective layer includes the comprehension scoring objectives corresponding to each knowledge point within the knowledge graph of subject S.

[0045] After completing the analysis of memorization and comprehension assessment objectives, the next step is to analyze the application assessment objectives for each knowledge point. Application assessment objectives require learners to apply their knowledge to real-world problems. For example, in mathematics, learners not only need to memorize mathematical formulas but also need to be able to use these formulas to solve practical problems. In physics, an application assessment objective might be to use learned principles of mechanics to analyze specific physical phenomena. Based on the characteristics of the target learning stage, an application assessment objective is set for each knowledge point to help learners improve their practical application abilities. Application assessment objectives are quantifiable targets for a specific knowledge point, requiring students to achieve a certain level of application in new contexts, focusing on the ability to use knowledge to solve problems, execute procedures, or make predictions. The Sth layer of application assessment objectives includes the application assessment objectives corresponding to each knowledge point within the Sth subject knowledge graph.

[0046] Based on the S-th memory scoring target layer, the S-th comprehension scoring target layer, and the S-th application scoring target layer, these targets are hierarchically represented and modeled using a pyramid mapping. Each subject's target pyramid contains three levels: the memory scoring target layer, the comprehension scoring target layer, and the application scoring target layer. For the i-th knowledge point T... i Its target state is defined as vector T. i =(M i U i A i In this pyramid, M represents memory score, U represents comprehension score, and A represents application score. The S-th layer of memory scores serves as the broad base, the S-th layer of comprehension scores as the middle layer, and the S-th layer of application scores as the narrow top layer, visually representing the progressive relationship of cognitive depth and the decreasing number of learners at each level. The S-th subject goal pyramid is a learning goal pyramid constructed for the S-th subject, with the pyramid's levels ordered according to different goals of memory, comprehension, and application.

[0047] For example, a high school mathematics knowledge map containing 120 knowledge points is extracted from Student A's multidisciplinary knowledge map, and the characteristics of the target learning stage are set according to the high school mathematics curriculum standards. Knowledge point k1 is the trigonometric function reduction formula. The curriculum standard analysis requires mastering the reduction formula, with a focus on application of key words. Quantitative objectives: The memorization scoring objective is to accurately recite six sets of reduction formulas. Based on its basic tool attribute, the objective is set at 92 points; the understanding scoring objective is to explain the geometric meaning of "odd changes, even remains the same, sign depends on quadrant," with an objective set at 85 points; the application scoring objective is to use the reduction formula to transform arbitrary angle trigonometric functions into acute angle trigonometric functions and evaluate them, with an objective set at 88 points. The target vector is T. k1 =(92,85,88). Knowledge point K2 is the dot product of planar vectors. The curriculum standard requires understanding its geometric meaning and being able to perform calculations. Target quantification: The memorization target is to remember the dot product formula a·b=|a||b|cosθ, with a target score of 90; the understanding target is to understand that the result is a scalar and explain its geometric meaning and projective product, with a target score of 87; the application target is to be able to use it to find angles, determine perpendicularity, and solve problems related to work calculations in physics. With a target score of 82, the target vector is T. k2 =(90, 87, 82). The above analysis was completed for each knowledge point. Finally, a high school mathematics target pyramid model was generated. Statistics show that among all 120 knowledge points, the average score for memorization was 89, the average score for understanding was 81, and the average score for application was 75. There are 45 knowledge points requiring an application target score ≥80, forming the crucial top part of the pyramid.

[0048] For other subjects in the multidisciplinary knowledge graph, repeat the aforementioned steps to obtain the subject-specific goal pyramids for those subjects. Integrate the S-th subject-specific goal pyramid with the corresponding pyramids for other subjects to obtain the S-th subject-specific goal pyramid. Through this pyramid structure, the learning objectives for each subject are hierarchically processed, starting from basic memorization and progressing to understanding and application, helping learners master knowledge step by step. Based on the characteristics of the target learning stage and considering the age, knowledge level, and learning progress of different learners, the goal requirements for each level are adjusted to provide personalized learning paths.

[0049] An adaptive subject-based scoring channel is introduced to score the learning levels of the multi-disciplinary learning dataset, and a learning pyramid for each subject is established.

[0050] Furthermore, this application also includes the following steps: activating the S-th learning scoring model corresponding to the S-th subject according to the subject-adaptive scoring channel; extracting the S-th subject learning dataset according to the multi-disciplinary learning dataset, and identifying knowledge point learning behavior features based on the S-th subject learning dataset to construct learning behavior feature vectors for each knowledge point; inputting the learning behavior feature vectors for each knowledge point into the S-th learning scoring model to obtain a learning scoring sequence for each knowledge point; performing learning level scoring clustering based on the learning scoring sequences for each knowledge point to construct an S-th memory scoring layer, an S-th comprehension scoring layer, and an S-th application scoring layer; and performing pyramid mapping modeling based on the S-th memory scoring layer, the S-th comprehension scoring layer, and the S-th application scoring layer to generate a learning pyramid for the S-th subject.

[0051] Furthermore, this application also includes the following steps: the subject-adaptive scoring channel includes multiple learning scoring models corresponding to multiple subjects, and the learning scoring indicators of each learning scoring model include memory scoring, comprehension scoring and application scoring.

[0052] Specifically, the subject-adaptive scoring channel is a standardized cognitive state assessment pipeline for multiple disciplines. For each discipline, it takes the multimodal learning behavior data of the target learner as input, processes it through a dedicated AI scoring model, and outputs a quantitative score for the learner at the three cognitive levels of memory, comprehension, and application, corresponding to each knowledge point in the subject's knowledge graph. The subject-adaptive scoring channel includes multiple learning scoring models for multiple disciplines, that is, multiple scoring models for different disciplines, each with its own specific evaluation indicators.

[0053] The subject-adaptive scoring channel adopts a microservice architecture of subject routing and model invocation. The subject router routes the data flow and requests to the corresponding subject-specific scoring model based on the input subject identifier, such as mathematics or physics. The subject-specific scoring model library is a collection of pre-trained models, each a neural network model optimized for a specific subject's knowledge system and cognitive characteristics. All models share the same input / output interface, i.e., a three-layer scoring, but their internal parameters and feature extraction logic vary by subject. Standardized input / output adapters are responsible for transforming the raw heterogeneous data into a unified tensor format required by the models and decoding the model's vector output into a specific knowledge point scoring table.

[0054] The input data is for the target knowledge point. The model receives learner behavior data packets related to the knowledge point and organized in a time sequence, including: text sequences, i.e., the question stems, student answers, and explanations of related exercises; behavior sequences, i.e., multiple encounters with learning resources related to the knowledge point, such as watching videos, doing exercises, and reviewing notes, and their metadata, such as duration [120s], correctness [True], number of attempts [1], and interactive hotspots [repeatedly dragging the video progress bar at 90%]; and contextual features, i.e., the neighbor node information of the knowledge point in the knowledge graph and the timestamps of the learning events. The feature extraction module includes a text encoding submodule, a behavior encoding submodule, and a graph context encoding submodule. The text encoding submodule uses the last hidden state of a lightweight pre-trained language model (such as MiniLM) as the text feature vector; the behavior encoding submodule uses a one-dimensional convolutional neural network combined with an attention mechanism to encode behavior sequences into behavior feature vectors. This network can capture behavior patterns, such as the effective pattern of watching a video first, then doing the questions, and then correcting the answers, and the ineffective pattern of guessing directly; the graph context encoding submodule utilizes the embeddings of nodes and their one-hop neighbors in the knowledge graph, and aggregates them through a graph attention network to obtain graph structure feature vectors.

[0055] The text feature vectors, behavior feature vectors, and graph structure feature vectors obtained from the text encoding submodule, behavior encoding submodule, and graph context encoding submodule are concatenated and then dimensionality-reduced and deeply fused through a fully connected layer to obtain a comprehensive feature vector. The cognitive state inference layer is a Transformer decoder layer containing a multi-head self-attention mechanism and a feedforward neural network, using only 1-2 layers to maintain efficiency. The comprehensive feature vector serves as the initial query. Simultaneously, the model has a built-in set of learnable cognitive prototype vectors (memory, understanding, and application) as keys and values. Through attention calculation, the comprehensive feature vector interacts and is weighted with these three cognitive prototypes to infer the learner's distance or matching degree to the ideal cognitive state in each of the three dimensions, outputting three preliminary cognitive state latent vectors. Three parallel fully connected networks share the bottom layer, but their output layers are independent. Each outputs a scalar score, which is scaled to the 0-1 range by a sigmoid function and then linearly mapped to the 0-100 score interval, serving as the final memory score, understanding score, and application score.

[0056] Collect learners' learning data, including homework scores, exam scores, learning time, and error logs. Each data sample requires triple annotation by experts or high-confidence rules. Extract valuable learning behavior features from this data, such as learning frequency, accuracy, and homework completion time. Select an appropriate scoring model structure, such as a multimodal feature extractor, based on the characteristics of each subject. Construct the structure of the multimodal feature extractor, with parameters including: text feature vector dimension 768, behavior sequence encoding CNN convolution kernel size [3,5], graph attention network head number 4, cognitive prototype vector dimension 128, fusion layer fully connected unit number 512, and initial training learning rate 1e. -4 Using the AdamW optimizer, the learning rate for online learning is 1e. -6 .

[0057] The scoring model is trained using historical learning data to optimize its parameters. During training, the model learns how to predict learners' scores in memory, comprehension, and application based on their learning behaviors. The model is validated using methods such as cross-validation to ensure its generalization ability across different learner groups.

[0058] AI adaptive algorithms play a crucial role throughout the model training and scoring process. Their main function is to automatically adjust model parameters based on learners' learning data, enabling continuous model optimization to adapt to the scoring requirements of different subjects. By acquiring learners' data in real time, AI adaptive algorithms evaluate the effectiveness of the current scoring model and fine-tune its parameters. The structure of the scoring model is adjusted based on learner feedback. For example, if a learner performs poorly on memory scoring, the model might optimize the scoring process by increasing the weight of memory training tasks. The scoring standards for each subject are dynamically adjusted based on the learner's learning progress and grade level. For instance, for elementary school, there might be a greater emphasis on memory scoring, while for high school, there might be a greater emphasis on comprehension and application scoring.

[0059] The assessment model for each subject is structured hierarchically according to memory assessment, comprehension assessment, and application assessment. Memory assessment primarily evaluates whether learners can accurately recall and recall subject knowledge points; in mathematics, it assesses whether students have memorized basic mathematical formulas. Comprehension assessment evaluates whether learners understand the deeper meaning of knowledge points. For example, can students clearly explain the meaning and application of a physical law? Application assessment evaluates whether students can apply their knowledge to practical problems. For example, can students use learned mathematical formulas to solve real-world engineering problems?

[0060] The subject-adaptive scoring channel activates the scoring model corresponding to the S-th subject, that is, it selects and activates the most suitable learning scoring model for each subject. For example, if S=1 represents the mathematics subject, then a scoring model designed for mathematics will be selected. At the same time, the learning dataset corresponding to the S-th subject is extracted from the multi-subject learning dataset and processed by time sorting, outlier cleaning, and session segmentation.

[0061] The learning behavior feature identification for the S-subject learning dataset involves extracting all raw behavioral logs related to each knowledge point in the S-subject knowledge graph from the dataset. Then, using predefined feature engineering rules or algorithms, quantitative indicators reflecting the learner's learning process and state at that knowledge point are calculated, such as exposure frequency, accuracy, response time stability, and resource interaction depth. This yields a learning behavior feature vector for each knowledge point. The knowledge point learning behavior feature vector is a numerical representation of the learner's behavioral characteristics when learning each knowledge point. For example, learning time, error rate, number of repetitions, and completion rate when learning a particular knowledge point are represented as vectors and used as input to the model.

[0062] The learning behavior feature vectors of each knowledge point are input into the S-th learning scoring model. Each vector is processed through a forward propagation network, and three scores are output in parallel. All outputs are collected and associated by knowledge point ID to form a learning scoring sequence for each knowledge point. Each knowledge point learning scoring sequence includes a memory score, a comprehension score, and an application score for each knowledge point.

[0063] Once the learning score sequence for all knowledge points is completed, learning level score clustering is performed based on the learning score sequence for each knowledge point, creating three empty sets to store memory, comprehension, and application scores, respectively. The score sequence is traversed, storing the memory score for each knowledge point in the memory set, the comprehension score in the comprehension set, and the application score in the application set, ultimately obtaining the S-th memory score layer, the S-th comprehension score layer, and the S-th application score layer. These three layers are then merged, and for each knowledge point, the corresponding scores are extracted from each layer and combined to form an actual score vector. The set of actual score vectors for all knowledge points constitutes the S-th subject learning pyramid. The S-th subject learning pyramid is structurally isomorphic to the S-th subject target pyramid. The bottom layer of the pyramid contains knowledge points with strong memory capabilities, the middle layer contains knowledge points with strong comprehension capabilities, and the top layer contains knowledge points that can be applied.

[0064] For example, let's construct a high school physics learning pyramid for Student A. Activate the high school physics learning scoring model and extract relevant physics records from Student A's data over the past three months: 320 answer records, 45 video viewing records, and 15 virtual experiment operation records. Taking Newton's Second Law as an example, extract the following records: 12 answers, 8 correct; 2 videos viewed, 100% and 80% completion rates respectively; 1 virtual experiment performed successfully. Calculate the following features: frequency of contact = 15 times, most recent contact = 2 days ago, average accuracy = 67%, first attempt accuracy = 1 (i.e., correct on the first try), response time stability = 1.5 seconds, average video completion rate = 90%, experiment success rate = 1. After normalization, generate a 25-dimensional feature vector, such as [0.75, 0.92, 0.67, 1.0, 0.85, 1.0, ...]. Input this feature vector into the physics scoring model, and output a memory score of 88, a comprehension score of 82, and an application score of 75. Similarly, all 80 knowledge points in physics are processed. Assuming all outputs are statistically analyzed, the average scores are: 85 for memorization, 78 for comprehension, and 70 for application. A scoring layer is constructed: the memorization layer contains 80 items, such as Newton's First Law (90), Newton's Second Law (88), the concept of work (86), ...; the comprehension layer contains items such as Newton's First Law (85), Newton's Second Law (82), the concept of work (80), ...; and the application layer contains items such as Newton's First Law (80), Newton's Second Law (75), the concept of work (72), .... The three layers of data are then merged. The vector for Newton's Second Law in this learning pyramid is (88, 82, 75). The entire pyramid shows that student A performs relatively well in the memorization aspect of physics (85 points) but is weakest in the application aspect (70 points), presenting a typical pyramid shape, consistent with the current learning situation of insufficient application ability development.

[0065] Similarly, for other subjects, repeat the above steps to obtain the subject learning pyramids corresponding to other subject datasets in the multi-disciplinary learning dataset. Combine the learning pyramid of subject S with the subject learning pyramids corresponding to other subject datasets to obtain the learning pyramids for each subject. Through pyramid mapping modeling, the learning objectives of each subject are clearly displayed, from basic memorization to in-depth application, helping learners gradually master knowledge. By scoring and clustering each knowledge point of the learner, personalized learning paths are developed for learners, strengthening weaknesses and consolidating strengths. Through the learning scoring sequence, the learner's mastery of each knowledge point is accurately assessed, providing comprehensive learning feedback from memorization to understanding to application.

[0066] Based on the target pyramids of each subject, a multi-level weakness detection is performed on the learning pyramids of each subject, thus constructing a weakness detection pyramid for each subject.

[0067] Furthermore, this application also includes the following steps: performing deviation detection on the S-th memory scoring layer according to the S-th memory scoring target layer to obtain the S-th memory scoring deviation sequence; based on the memory scoring deviation threshold, locating memory weaknesses in the S-th subject knowledge graph according to the S-th memory scoring deviation sequence to construct the S-th memory weakness layer; locating comprehension weaknesses in the S-th subject knowledge graph according to the S-th comprehension scoring target layer and the S-th comprehension scoring layer to construct the S-th comprehension weakness layer; locating application weaknesses in the S-th subject knowledge graph according to the S-th application scoring target layer and the S-th application scoring layer to construct the S-th application weakness layer; analyzing the correlation characteristics of the S-th memory weakness layer, the S-th comprehension weakness layer, and the S-th application weakness layer according to the S-th subject knowledge graph to construct the S-th weakness correlation layer; and performing multi-level weakness tower fusion on the S-th memory weakness layer, the S-th comprehension weakness layer, and the S-th application weakness layer according to the S-th weakness correlation layer to generate the S-th subject weakness detection tower.

[0068] Specifically, the S-th memory scoring target layer is a part of the S-th subject target pyramid, and the S-th memory scoring layer is a part of the S-th subject learning pyramid; that is, it is a part of the subject target pyramid and subject learning pyramid corresponding to the S-th subject. Based on the S-th memory scoring target layer, bias detection is performed on the S-th memory scoring layer to obtain the S-th memory scoring bias sequence, which is the deviation between the learner's actual score and the target score.

[0069] A pre-defined memory scoring deviation threshold is used to determine whether a knowledge point constitutes a weakness at the memory cognitive level. The S-th memory scoring deviation sequence is compared with the memory scoring deviation threshold. If the memory scoring deviation is greater than or equal to the memory scoring deviation threshold, the corresponding knowledge point is a memory weakness and is added to the S-th memory weakness layer.

[0070] Similarly, for the understanding layer and application layer in the target pyramid and learning pyramid of each subject, repeat the above process to screen out the weaknesses of each layer, thereby obtaining the Sth layer of understanding weaknesses and the Sth layer of application weaknesses, which include the weaknesses of each subject in each cognitive dimension.

[0071] Based on the knowledge graph of subject S, the association characteristics of the S-th memory weakness layer, S-th comprehension weakness layer, and S-th application weakness layer are analyzed, including cross-layer associations and intra-layer associations, to analyze the relationships between weaknesses. That is, by utilizing the inherent logical relationships between knowledge points in the S-th subject knowledge graph, such as prerequisites, successors, parent-child categories, etc., the interrelationships of weak knowledge points in different weakness layers are analyzed. For a knowledge point of an application weakness, its direct prerequisite knowledge points in the subject graph are checked to see if they happen to appear in the comprehension or memory weakness layer. If so, a possible causal edge is established from the lower-level weakness to the higher-level weakness. Within the same weakness layer, the graph-based close connections exist between weak knowledge points, such as belonging to the same knowledge module. If so, these weaknesses are aggregated into a weakness cluster, indicating that they may require collaborative remediation. All analysis results are stored together in the S-th weakness association layer to understand the interrelationships between weaknesses.

[0072] Based on the relationship blueprint provided by the Sth Weakness Association Layer, a multi-level weakness pyramid fusion is performed on the Sth Memory Weakness Layer, the Sth Understanding Weakness Layer, and the Sth Application Weakness Layer. That is, using the relationships revealed by the Sth Weakness Association Layer as the framework, the three independent weakness layers are integrated three-dimensionally to construct a structured set of weaknesses that reflects the causal, hierarchical, and evolutionary relationships between weaknesses. All knowledge points in the Sth Memory Weakness Layer are placed at the bottom of the pyramid. This represents the weakness in the knowledge foundation. Knowledge points in the Sth Understanding Weakness Layer are placed in the middle layer. Simultaneously, based on the data from the association layer, understanding weaknesses that are not directly caused by bottom-level memory weaknesses are marked with prominent symbols as independent cognitive obstacles. Knowledge points in the Sth Application Weakness Layer are placed at the top. Based on the causal relationships resolved in the association layer, clear connections are drawn between the bottom, middle, and top layers of the pyramid. For example, if the application weakness K at the top of the pyramid... top The analysis revealed that the weakness in understanding the tower body was K_ mid This leads to, and K_ mid Furthermore, due to Taji's memory weakness K base As a result, the system will establish a path from K base →K mid →K top The path of weakness transmission is the key chain of gaps that needs to be addressed first. The resulting weakness detection tower for the Sth subject is a visual and computable structure that not only lists all weaknesses but, more importantly, reveals the hierarchical relationships and causal chains between them, indicating where to start repairing them most effectively.

[0073] By using multi-level deviation detection and correlation analysis, we can comprehensively identify the weak knowledge points of target learners at each level, and develop personalized learning paths based on students' specific weaknesses, thereby improving learning efficiency.

[0074] Based on the interdisciplinary knowledge coupling network, the interdisciplinary weakness detection towers are strengthened through interdisciplinary weakness association, and a first graph network of target weaknesses is constructed.

[0075] Specifically, the weaknesses detection towers for each discipline and the interdisciplinary knowledge coupling network are aligned at each node to ensure that every weakness node in each discipline's weakness detection tower can find a corresponding knowledge point node in the interdisciplinary knowledge coupling network. Starting from the knowledge point corresponding to the weakness node, a one-hop traversal is performed along the dependency, support, and transfer relationship edges in the interdisciplinary knowledge coupling network to find all knowledge points from other disciplines that are directly connected to it. It is then checked whether these knowledge points from other disciplines, connected through coupling relationships, also appear in the weakness detection tower of their respective disciplines.

[0076] Record the relationship type between these two linked weaknesses, such as dependency, support, or migration. Calculate the comprehensive weight of this interdisciplinary weakness association based on the strength of the original relationship in the coupled network and the severity (bias value) of each weakness. For example, if the original dependency strength is 0.92, the mathematical weakness bias is 18, and the physical weakness bias is 15, the comprehensive weight might be calculated as 0.92*(0.5*18 / 15+0.5*15 / 15), and then normalized. A higher weight indicates a tighter association between the two weaknesses and a greater need for collaborative processing.

[0077] All identified interdisciplinary weakness pairs (edges) are collected and merged with all weakness nodes (nodes) in the weakness detection towers of all disciplines to construct a unified graph. Each node represents a specific weakness, with attributes including discipline, knowledge point, cognitive level, and deviation value. Edges are divided into two types: intra-discipline edges come from the weakness association layer of the original weakness detection tower, representing intra-discipline causal relationships; interdisciplinary edges are newly generated edges, representing interdisciplinary dependency, support, or migration relationships, and are accompanied by type and comprehensive weight. After construction, network analysis is performed on the entire graph to calculate the interdisciplinary centrality of each weak node. Weak nodes that connect multiple disciplines and are on multiple interdisciplinary association paths are identified as pivotal weaknesses or root weaknesses. The final generated target weakness first graph network is a panoramic weakness strategy map that integrates the weakness detection towers of all disciplines and uses interdisciplinary coupling relationships as bridges to connect weaknesses that were originally scattered in various disciplines into an organic whole, thereby revealing pivotal or root weaknesses that affect multiple disciplines.

[0078] By strengthening interdisciplinary weaknesses through association and constructing a first-level network of target weaknesses, we identify students' weaknesses across multiple subjects and reveal the connections between disciplines, avoiding single-subject remediation strategies. By linking weaknesses across different subjects, we ensure the coherence and depth of the remediation process, helping students gradually improve across multiple subjects.

[0079] Based on the first target weakness graph network, global forgetting characteristics are tracked and compensated to obtain the second target weakness graph network, and the target learner is continuously closed-looped to fill gaps based on the second target weakness graph network.

[0080] Furthermore, this application also includes the following steps: performing deep learning based on the target learner's historical knowledge mastery change dataset to construct a personal forgetting prediction model; calling the personal forgetting prediction model to predict knowledge point forgetting on the multidisciplinary learning dataset to obtain forgetting curves for each knowledge point; performing intensity analysis on the forgetting curves of each knowledge point based on the future time zone to determine the forgetting intensity of each knowledge point; performing risk detection on the multidisciplinary knowledge graph based on the forgetting intensity threshold and the forgetting intensity of each knowledge point to obtain a forgetting risk knowledge graph; and performing global forgetting characteristic compensation reconstruction on the first target weakness graph based on the forgetting risk knowledge graph to generate the second target weakness graph.

[0081] Specifically, a dataset of historical knowledge mastery changes for target learners is collected; that is, sequential data showing the change in mastery scores for each knowledge point over a past period. Each data point typically includes a knowledge point ID, a timestamp, and the mastery score at that time. A deep time series model is used for training. The model's input is a historical mastery sequence for a knowledge point, the feature vector of that knowledge point, and the learner's personal features; the output is a predicted mastery sequence for that knowledge point over a future period. The training objective is to minimize the error between the predicted sequence and the actually observed subsequent mastery sequence. After training, a personal forgetting prediction model for that learner is obtained.

[0082] To implement a personal forgetting prediction model, a deep learning-based neural network model needs to be designed, including an input layer, hidden layers, and an output layer. The input layer will accept multiple features, including knowledge point mastery, the learner's historical learning data, and the learner's behavioral data. Hidden layers: Multiple fully connected layers are designed to extract deep-level patterns from the input data. Neurons in each layer are processed by activation functions. Commonly used activation functions include ReLU to introduce non-linear relationships, enabling the model to capture complex patterns. Time series layer: Considering that forgetting is time-related, the model should include an LSTM to process time series data, helping the model learn the time dependencies related to forgetting in historical learning data. The output layer will generate a forgetting prediction value for each knowledge point, typically a forgetting curve describing the forgetting trend of the knowledge point over a future time period. The historical knowledge mastery change dataset needs to be cleaned to ensure data accuracy and completeness. The dataset includes the learner's mastery of each knowledge point and learning behavior data, such as review frequency and incorrect answer records. The data is standardized or normalized to ensure that the data input into the model is within a reasonable range. The dataset is divided into training and validation sets. For example, 80% of the data can be used for training, and 20% for validating model performance. Backpropagation is used to optimize network weights to minimize the loss function. The loss function can be measured using mean squared error (MSE) to determine the error between the model's predictions and the actual values. The model is optimized by adjusting hyperparameters such as the learning rate, batch size, and number of training epochs. The initial learning rate is typically set to 0.001 and fine-tuned experimentally. A learning rate that is too large may lead to unstable convergence; a rate that is too small may result in slow training. The batch size is typically set to 32 or 64, depending on the size of the training dataset. The number of training epochs is typically set to 50-100, but should be adjusted based on the model's convergence. The model is evaluated using a validation set, calculating metrics such as accuracy and recall to ensure the model's predictive performance.

[0083] This study uses a personal forgetting prediction model to predict knowledge point forgetting in a multi-disciplinary learning dataset. Starting with the learner's current mastery score for each knowledge point, a forgetting curve of length T days (where T is a positive integer) is generated for each knowledge point in the multi-disciplinary knowledge graph. Based on future time zones, the average slope of each curve within that time period is calculated, representing the forgetting intensity. The forgetting intensity is the absolute value of the average slope of the forgetting curve within the future time zone, quantifying the expected rate of forgetting for that knowledge point during that period. The calculation formula can be simplified to: (Initial Mastery - Ending Mastery) / Time Zone Length. A larger value indicates faster forgetting and a higher risk. The knowledge point forgetting curve is very long, corresponding to multiple future times, with each future time zone representing a portion of that time.

[0084] The forgetting intensity of each knowledge point is compared with a preset forgetting intensity threshold. If the forgetting intensity is greater than or equal to the preset threshold, the knowledge point is marked as a knowledge point at risk of forgetting, and its forgetting intensity value is added as a node attribute to the multidisciplinary knowledge graph, generating a forgetting risk knowledge graph that shows which parts of the knowledge system are most likely to deteriorate in the coming week. For example, the forgetting intensity threshold is set to 0.2. If the absolute value of the slope of the forgetting curve of a certain knowledge point is 0.3 in the future time zone, exceeding the threshold, then the knowledge point will be marked as a knowledge point at risk of forgetting; if the slope of the forgetting curve of another knowledge point is 0.1, below the threshold, then the forgetting intensity of that knowledge point is low and it will not be marked as a knowledge point at risk of forgetting. The forgetting risk knowledge graph is formed by marking all nodes identified as knowledge points at risk of forgetting and their risk intensities on the basis of the original multidisciplinary knowledge graph, indicating which learned knowledge may decline due to forgetting and become new weaknesses in the future.

[0085] Based on a knowledge graph of forgetting risk, a global forgetting characteristic compensation reconstruction is performed on the first target weakness network. This involves integrating forward-looking information on forgetting risk—a temporal dimension—into the static first target weakness network. By adjusting the weights of weakness nodes and adding risk nodes, the network can simultaneously reflect current weaknesses and potential future weaknesses. The second target weakness network not only includes existing interdisciplinary weakness associations but also incorporates knowledge points with high forgetting risk. It also provides early warnings and weight reinforcement for situations where existing weaknesses may be exacerbated by forgetting risk, thereby enabling more forward-looking learning intervention planning.

[0086] Personal forgetting prediction models can accurately predict what knowledge learners may forget in the future and provide corresponding review suggestions. By accurately predicting forgetting points, ineffective learning is reduced, helping students focus their time on reviewing key knowledge that is easily forgotten.

[0087] Furthermore, this application also includes the following steps: performing multi-disciplinary learning form preference analysis based on the multi-disciplinary learning dataset to determine the learning preference characteristics of multiple disciplines; making multi-disciplinary gap-filling decisions based on the learning preference characteristics of multiple disciplines and the target weakness second graph network to determine a global gap-filling strategy; simultaneously generating an updated learning dataset when the target learner performs knowledge gap-filling learning according to the global gap-filling strategy; performing weakness detection based on the updated learning dataset to obtain an updated weakness graph network, and comparing the target weakness second graph network with the updated weakness graph network to generate a weakness comparison graph network; evaluating the gap-filling effect based on the weakness comparison graph network to obtain a gap-filling effect evaluation result, and optimizing the global gap-filling strategy based on the weakness comparison graph network and the gap-filling effect evaluation result.

[0088] Specifically, based on a multidisciplinary learning dataset, the learning styles of learners across different subjects are analyzed to extract the learning preference characteristics for each subject. Learning style preference analysis determines which learning methods learners tend to use in which subjects and identifies their preference characteristics. Based on these subject-specific preference characteristics and the previously generated target weakness second graph network, multidisciplinary gap-filling decisions are made to obtain a global gap-filling strategy, formulate a comprehensive gap-filling learning plan, determine which knowledge points should be prioritized for review, and how to improve learning effectiveness through effective learning methods.

[0089] A global gap-filling strategy is pushed to the target learners. As the target learners fill in the knowledge gaps according to the global gap-filling strategy, their learning progress is tracked in real time and an updated learning dataset is generated. This dataset includes new learning records after the target learners have filled in the gaps according to the global gap-filling strategy, reflecting the learners' learning progress and improvement in the gap-filling process.

[0090] Based on the updated learning dataset, weakness detection is performed through the aforementioned process to obtain an updated weakness graph network, similar to the process of obtaining the target weakness second graph network. The target weakness second graph network is compared with the updated weakness graph network to generate a weakness comparison graph network, showing which knowledge points have been improved through gap-filling learning and which still have problems. The evaluation results of the weakness comparison graph network help determine the effectiveness of the gap-filling strategy and measure the effect of gap-filling.

[0091] The effectiveness of the gap-filling strategy is evaluated by comparing the library's weaknesses with the graph network, confirming the overall effectiveness of the gap-filling strategy. The strategy is then adjusted and optimized based on the evaluation results. For example, if the weakness in the Mathematics Hub is successfully addressed, the three related Physical weaknesses improve simultaneously, achieving an overall improvement rate of 40%. However, two new application weaknesses caused by insufficient understanding are discovered. Feedback optimization is initiated: if the improvement rate of a certain type of resource, such as video A, for a corresponding weakness is significantly higher than the average, the recommendation weight of this type of resource is increased in the next round of strategy; if two related weaknesses learned in a specific order (A then B) both improve, while the reverse order is less effective, this order rule is strengthened; if a weakness shows no improvement despite learning from recommended resources, the knowledge point is marked as needing a change in teaching method, and a completely different type of resource is recommended next time, such as changing from explanatory videos to interactive exploration.

[0092] Personalized learning paths are developed based on learners' learning habits, preferences, and weaknesses. These paths are dynamically adjusted based on real-time feedback to ensure each learner focuses their learning and improvement on the areas where they need it most. Through precise weakness detection and optimized remediation strategies, learners' overall learning outcomes are effectively improved, forgetting is reduced, and knowledge mastery is enhanced.

[0093] In summary, the multidisciplinary knowledge gap filling method based on AI adaptive algorithms provided in this application has the following technical effects: by using multidisciplinary knowledge graphs and interdisciplinary coupling networks, combined with learners' learning data, personalized weakness detection and knowledge gap filling are performed. At the same time, by considering learners' forgetting curves, interdisciplinary root cause diagnosis, accurate positioning of cognitive levels, and predictive continuous intervention are achieved, ensuring the continuity and effectiveness of learning, improving the accuracy of knowledge gap filling, and thus effectively improving learning efficiency.

[0094] Example 2: Based on the same inventive concept as the multidisciplinary knowledge gap filling method based on AI adaptive algorithms in Example 1, this application also provides a multidisciplinary knowledge gap filling system based on AI adaptive algorithms. Please refer to the appendix. Figure 2 The AI-adaptive algorithm-based multidisciplinary knowledge gap filling system includes: a cross-disciplinary knowledge coupling feature analysis module 11, used to construct a multidisciplinary knowledge graph based on the target learner's multidisciplinary learning dataset, and to perform cross-disciplinary knowledge coupling feature analysis on the multidisciplinary knowledge graph to construct a cross-disciplinary knowledge coupling network; a learning level target analysis module 12, used to perform learning level target analysis on the multidisciplinary knowledge graph based on the target learner's target learning stage characteristics, and to construct a target pyramid for each subject; and a learning level scoring module 13, used to introduce a subject-adaptive scoring channel to perform learning level scoring on the multidisciplinary learning dataset. The system includes a scoring system to establish learning pyramids for each subject; a multi-level weakness detection module 14, used to perform multi-level weakness detection on the learning pyramids of each subject based on the target pyramids of each subject, and to construct a weakness detection tower for each subject; an interdisciplinary weakness association reinforcement module 15, used to perform interdisciplinary weakness association reinforcement on the weakness detection towers of each subject based on the interdisciplinary knowledge coupling network, and to construct a first target weakness graph network; and a closed-loop gap filling module 16, used to perform global forgetting characteristic tracking and compensation based on the first target weakness graph network, to obtain a second target weakness graph network, and to perform continuous closed-loop gap filling for the target learner based on the second target weakness graph network.

[0095] Furthermore, the interdisciplinary knowledge coupling feature parsing module 11 in the multidisciplinary knowledge gap filling system based on AI adaptive algorithm is also used to: extract knowledge points from the multidisciplinary learning dataset to obtain knowledge point groups of each discipline; identify intradisciplinary knowledge associations based on the intradisciplinary knowledge point groups to obtain intradisciplinary knowledge associations; and perform knowledge graph structured reconstruction of the intradisciplinary knowledge point groups based on the intradisciplinary knowledge associations to generate the multidisciplinary knowledge graph.

[0096] Furthermore, the interdisciplinary knowledge coupling feature analysis module 11 in the multidisciplinary knowledge gap filling system based on AI adaptive algorithms is also used for: identifying the dependency relationships between knowledge points of different disciplines based on the multidisciplinary knowledge graph, and constructing an interdisciplinary knowledge point dependency vector set; identifying the support relationships between knowledge points of different disciplines based on the multidisciplinary knowledge graph, and constructing an interdisciplinary knowledge point support vector set; identifying the transfer relationships between knowledge points of different disciplines based on the multidisciplinary knowledge graph, and constructing an interdisciplinary knowledge point transfer vector set; and sorting out the multi-relationship collaborative graph structure based on the interdisciplinary knowledge point dependency vector set, the interdisciplinary knowledge point support vector set, and the interdisciplinary knowledge point transfer vector set to generate the interdisciplinary knowledge coupling network.

[0097] Furthermore, the learning level target parsing module 12 in the multidisciplinary knowledge point gap filling system based on the AI ​​adaptive algorithm is also used for: extracting the knowledge graph of the Sth subject corresponding to the Sth subject according to the multidisciplinary knowledge graph, where S is a positive integer; performing memory scoring target parsing on each knowledge point in the knowledge graph of the Sth subject according to the characteristics of the target learning stage, and constructing the Sth memory scoring target layer; performing comprehension scoring target parsing on each knowledge point in the knowledge graph of the Sth subject according to the characteristics of the target learning stage, and constructing the Sth comprehension scoring target layer; performing application scoring target parsing on each knowledge point in the knowledge graph of the Sth subject according to the characteristics of the target learning stage, and constructing the Sth application scoring target layer; and performing pyramid mapping modeling based on the Sth memory scoring target layer, the Sth comprehension scoring target layer, and the Sth application scoring target layer to generate the target pyramid of the Sth subject.

[0098] Furthermore, the learning level scoring module 13 in the multidisciplinary knowledge gap filling system based on AI adaptive algorithms is also used for: activating the Sth learning scoring model corresponding to the Sth subject according to the subject adaptive scoring channel; extracting the Sth subject learning dataset according to the multidisciplinary learning dataset, and identifying knowledge point learning behavior features based on the Sth subject learning dataset to construct learning behavior feature vectors for each knowledge point; inputting the learning behavior feature vectors for each knowledge point into the Sth learning scoring model to obtain a learning scoring sequence for each knowledge point; performing learning level scoring clustering based on the learning scoring sequences for each knowledge point to construct an Sth memory scoring layer, an Sth comprehension scoring layer, and an Sth application scoring layer; and performing pyramid mapping modeling based on the Sth memory scoring layer, the Sth comprehension scoring layer, and the Sth application scoring layer to generate a learning pyramid for the Sth subject.

[0099] Furthermore, the learning level scoring module 13 in the multidisciplinary knowledge gap filling system based on AI adaptive algorithm is also used for: the subject adaptive scoring channel includes multiple learning scoring models corresponding to multiple subjects, and the learning scoring indicators of each learning scoring model include memory score, comprehension score and application score.

[0100] Furthermore, the multi-level weakness detection module 14 in the AI ​​adaptive algorithm-based multidisciplinary knowledge gap filling system is also used for: performing deviation detection on the Sth memory scoring layer according to the Sth memory scoring target layer to obtain the Sth memory scoring deviation sequence; based on the memory scoring deviation threshold, performing memory weakness localization on the Sth subject knowledge graph according to the Sth memory scoring deviation sequence to construct the Sth memory weakness layer; and performing comprehension weakness localization on the Sth subject knowledge graph according to the Sth comprehension scoring target layer and the Sth comprehension scoring layer to construct the Sth comprehension weakness layer. The process involves: understanding the weakness layer; locating application weaknesses in the S-th subject knowledge graph based on the S-th application scoring target layer and the S-th application scoring layer, and constructing the S-th application weakness layer; analyzing the correlation characteristics of the S-th memory weakness layer, the S-th understanding weakness layer, and the S-th application weakness layer based on the S-th subject knowledge graph, and constructing the S-th weakness association layer; and performing multi-level weakness tower fusion of the S-th memory weakness layer, the S-th understanding weakness layer, and the S-th application weakness layer based on the S-th weakness association layer to generate the S-th subject weakness detection tower.

[0101] Furthermore, the closed-loop knowledge gap filling module 16 in the multidisciplinary knowledge gap filling system based on AI adaptive algorithms is also used for: performing deep learning based on the target learner's historical knowledge mastery change dataset to construct a personal forgetting prediction model; calling the personal forgetting prediction model to perform knowledge point forgetting prediction on the multidisciplinary learning dataset to obtain the forgetting curve of each knowledge point; performing intensity analysis on the forgetting curve of each knowledge point based on the future time zone to determine the forgetting intensity of each knowledge point; performing risk detection on the multidisciplinary knowledge graph based on the forgetting intensity threshold and the forgetting intensity of each knowledge point to obtain a forgetting risk knowledge graph; and performing global forgetting characteristic compensation and reconstruction on the first target weakness graph based on the forgetting risk knowledge graph to generate the second target weakness graph.

[0102] Furthermore, the closed-loop knowledge gap filling module 16 in the AI ​​adaptive algorithm-based multidisciplinary knowledge gap filling system is also used for: performing multidisciplinary learning form preference analysis based on the multidisciplinary learning dataset to determine multiple subject learning preference characteristics; making multidisciplinary knowledge gap filling decisions based on the multiple subject learning preference characteristics and the target weakness second graph network to determine a global knowledge gap filling strategy; synchronously generating an updated learning dataset when the target learner performs knowledge gap filling learning according to the global knowledge gap filling strategy; performing weakness detection based on the updated learning dataset to obtain an updated weakness graph network, and comparing the target weakness second graph network with the updated weakness graph network to generate a weakness comparison graph network; evaluating the knowledge gap filling effect based on the weakness comparison graph network to obtain a knowledge gap filling effect evaluation result, and performing feedback optimization of the global knowledge gap filling strategy based on the weakness comparison graph network and the knowledge gap filling effect evaluation result.

[0103] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The multidisciplinary knowledge point filling method and specific examples based on AI adaptive algorithms in the foregoing embodiment one are also applicable to the multidisciplinary knowledge point filling system based on AI adaptive algorithms in this embodiment. Through the foregoing detailed description of the multidisciplinary knowledge point filling method based on AI adaptive algorithms, those skilled in the art can clearly understand the multidisciplinary knowledge point filling system based on AI adaptive algorithms in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0104] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0105] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A multidisciplinary knowledge gap filling method based on AI adaptive algorithms, characterized in that, include: Based on the multidisciplinary learning dataset of the target learners, a multidisciplinary knowledge graph is constructed, and cross-disciplinary knowledge coupling feature analysis is performed on the multidisciplinary knowledge graph to construct a cross-disciplinary knowledge coupling network. Based on the target learning stage characteristics of the target learners, the multi-disciplinary knowledge graph is analyzed for learning level objectives, and a target pyramid for each subject is constructed. A subject-adaptive scoring channel is introduced to score the learning level of the multi-disciplinary learning dataset and to establish a learning pyramid for each subject. Based on the target pyramids of each subject, a multi-level weakness detection is performed on the learning pyramids of each subject to construct a weakness detection pyramid for each subject. Based on the interdisciplinary knowledge coupling network, the interdisciplinary weakness detection towers are strengthened through interdisciplinary weakness association, and a first graph network of target weaknesses is constructed. Based on the first target weakness graph network, global forgetting characteristics are tracked and compensated to obtain the second target weakness graph network, and the target learner is continuously closed-looped to fill gaps based on the second target weakness graph network. Based on the first target weakness graph network, global forgetting characteristic tracking and compensation are performed to obtain the second target weakness graph network, including: Based on the target learner's historical knowledge mastery change dataset, a personal forgetting prediction model is constructed using deep learning. The personal forgetting prediction model is invoked to predict the forgetting of knowledge points in the multidisciplinary learning dataset, and the forgetting curve of each knowledge point is obtained. Based on the future time zone, the forgetting curves of each knowledge point are analyzed to determine the forgetting intensity of each knowledge point; Based on the forgetting intensity threshold, risk detection is performed on the multidisciplinary knowledge graph according to the forgetting intensity of each knowledge point to obtain a forgetting risk knowledge graph. Based on the forgetting risk knowledge graph, the first network of the target weakness is reconstructed by global forgetting characteristic compensation to generate the second network of the target weakness.

2. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, The multidisciplinary knowledge graph is subjected to cross-disciplinary knowledge coupling feature analysis to construct a cross-disciplinary knowledge coupling network, including: Based on the multidisciplinary knowledge graph, the dependency relationships between knowledge points of different disciplines are identified, and a cross-disciplinary knowledge point dependency vector set is constructed. Based on the multidisciplinary knowledge graph, the supporting relationships between knowledge points of different disciplines are identified, and a cross-disciplinary knowledge point support vector set is constructed. Based on the multidisciplinary knowledge graph, the transfer relationships between knowledge points of different disciplines are identified, and a cross-disciplinary knowledge point transfer vector set is constructed. Based on the interdisciplinary knowledge point dependency vector set, the interdisciplinary knowledge point support vector set, and the interdisciplinary knowledge point transfer vector set, a multi-relationship collaborative graph structure is sorted out to generate the interdisciplinary knowledge coupling network.

3. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, Based on the target learners' characteristics at their target learning stages, the multidisciplinary knowledge graph is analyzed to determine learning level objectives, and a target pyramid for each subject is constructed, including: Based on the multidisciplinary knowledge graph, extract the knowledge graph of the Sth subject corresponding to the Sth subject, where S is a positive integer; Based on the characteristics of the target learning stage, memory scoring targets are analyzed for each knowledge point in the knowledge graph of the Sth subject, and the Sth memory scoring target layer is constructed. Based on the characteristics of the target learning stage, the understanding and scoring objectives of each knowledge point in the knowledge graph of the Sth subject are analyzed to construct the Sth understanding and scoring objective layer. Based on the characteristics of the target learning stage, the application scoring target is analyzed for each knowledge point in the knowledge graph of the Sth subject, and the Sth application scoring target layer is constructed. Based on the Sth memory scoring target layer, the Sth comprehension scoring target layer, and the Sth application scoring target layer, a pyramid mapping model is performed to generate the Sth subject target pyramid.

4. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, A subject-adaptive scoring channel is introduced to score the learning hierarchy of the multi-disciplinary learning dataset, and a learning pyramid for each subject is established, including: Based on the subject-adaptive scoring channel, activate the Sth learning scoring model corresponding to the Sth subject; Extract the Sth subject learning dataset from the multi-disciplinary learning dataset, and identify the learning behavior features of each knowledge point based on the Sth subject learning dataset to construct the learning behavior feature vector of each knowledge point. Input the learning behavior feature vectors of each knowledge point into the Sth learning scoring model to obtain the learning scoring sequence of each knowledge point; Based on the learning score sequence of each knowledge point, learning level score clustering is performed to construct the Sth memory score layer, the Sth comprehension score layer and the Sth application score layer. Based on the Sth memory scoring layer, the Sth comprehension scoring layer, and the Sth application scoring layer, a pyramid mapping model is performed to generate the Sth subject learning pyramid.

5. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, Based on the target pyramids of each subject, a multi-level weakness detection is performed on the learning pyramids of each subject, constructing a weakness detection tower for each subject, including: Based on the target layer of the Sth memory scoring layer, bias detection is performed on the Sth memory scoring layer to obtain the Sth memory scoring bias sequence; Based on the memory scoring deviation threshold, the Sth subject knowledge graph is used to locate memory weaknesses according to the Sth memory scoring deviation sequence, and the Sth memory weakness layer is constructed. Based on the Sth comprehension scoring target layer and the Sth comprehension scoring layer, the Sth subject knowledge graph is used to locate comprehension weaknesses, and the Sth comprehension weakness layer is constructed. Based on the Sth application scoring target layer and the Sth application scoring layer, the Sth subject knowledge graph is used to locate application weaknesses, and the Sth application weakness layer is constructed. Based on the Sth subject knowledge graph, the Sth memory weakness layer, the Sth comprehension weakness layer, and the Sth application weakness layer are analyzed for their correlation characteristics to construct the Sth weakness correlation layer; Based on the Sth weakness association layer, a multi-level weakness tower fusion is performed on the Sth memory weakness layer, the Sth understanding weakness layer, and the Sth application weakness layer to generate the Sth subject weakness detection tower.

6. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, Based on the target weaknesses, the second graph network continuously performs closed-loop remediation on the target learner, including: Based on the multidisciplinary learning dataset, perform multidisciplinary learning form preference analysis to determine the learning preference characteristics of multiple disciplines; Based on the learning preference characteristics of the multiple disciplines and the target weakness second graph network, a multi-disciplinary gap-filling decision is made to determine the global gap-filling strategy; When the target learner performs knowledge gap filling learning according to the global gap filling strategy, the updated learning dataset is generated synchronously. Weakness detection is performed based on the updated learning dataset to obtain an updated weakness graph network. The target weakness second graph network is then compared with the updated weakness graph network to generate a weakness comparison graph network. The effectiveness of the vulnerability patching is evaluated based on the vulnerability comparison network, and the evaluation results are obtained. The global vulnerability patching strategy is then optimized based on the vulnerability comparison network and the evaluation results.

7. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, Based on the target learners' multidisciplinary learning dataset, a multidisciplinary knowledge graph is constructed, including: Based on the multidisciplinary learning dataset, knowledge points are extracted to obtain knowledge point groups for each discipline. Based on the knowledge point groups of each discipline, the knowledge associations within each discipline are identified to obtain the knowledge association relationships within each discipline; Based on the knowledge relationships within each discipline, the knowledge point groups of each discipline are reconstructed into a knowledge graph structure to generate the multidisciplinary knowledge graph.

8. The multidisciplinary knowledge gap filling method based on AI adaptive algorithm as described in claim 1, characterized in that, The subject-adaptive scoring channel includes multiple learning scoring models corresponding to multiple subjects. The learning scoring indicators of each learning scoring model include memory score, comprehension score and application score.

9. A multidisciplinary knowledge gap filling system based on AI adaptive algorithms, characterized in that, The steps for implementing the multi-disciplinary knowledge gap filling method based on AI adaptive algorithm according to any one of claims 1 to 8, wherein the multi-disciplinary knowledge gap filling system based on AI adaptive algorithm comprises: The interdisciplinary knowledge coupling feature parsing module is used to construct a multidisciplinary knowledge graph based on the multidisciplinary learning dataset of the target learner, and to perform interdisciplinary knowledge coupling feature parsing on the multidisciplinary knowledge graph to construct an interdisciplinary knowledge coupling network. The learning level target analysis module is used to analyze the learning level target of the multi-disciplinary knowledge graph according to the target learning stage characteristics of the target learner, and construct the target pyramid of each subject. The learning level scoring module is used to introduce a subject-adaptive scoring channel to score the learning level of the multi-subject learning dataset and establish a learning pyramid for each subject. A multi-level weakness detection module is used to perform multi-level weakness detection on the learning pyramids of each subject based on the target pyramids of each subject, and to construct a weakness detection pyramid for each subject. The interdisciplinary weakness association reinforcement module is used to perform interdisciplinary weakness association reinforcement on the weakness detection towers of each discipline according to the interdisciplinary knowledge coupling network, and construct the first graph network of target weaknesses; The closed-loop gap-filling module is used to perform global forgetting characteristic tracking and compensation based on the first target weakness graph network to obtain the second target weakness graph network, and to perform continuous closed-loop gap-filling for the target learner based on the second target weakness graph network.

Images