Intelligent courseware automatic classification method and system constructed by hierarchical semantic network
By integrating large language models and knowledge graphs into educational resource classification, a hierarchical semantic network is built, and combining machine learning and reinforcement learning, the problem of time-consuming manual operations and lack of deep understanding of automated classification in the existing technology is solved, and efficient and accurate courseware classification and personalized recommendations are achieved.
Patent Information
- Application Number
- CN202411837710.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-13
- Publication Date
- 2025-05-02
AI Technical Summary
The existing technology relies on manual operations in the classification of educational resources, which is time-consuming and labor-intensive and difficult to ensure the accuracy and consistency of classification. In addition, automated classification technology lacks in-depth understanding and cannot deeply explore the multi-dimensional attributes of courseware.
By introducing large language models and knowledge graphs, designing algorithms integrate semantic understanding and graph matching, building a hierarchical semantic network, realizing multi-level extraction of subjects, difficulty and applicable grades, combining machine learning and reinforcement learning, and dynamically adjusting the model based on user feedback and data.
It has achieved efficient and rapid classification of a large number of courseware, significantly improved the processing speed and accuracy and consistency of classification, met personalized needs, and enhanced the efficiency and comprehensiveness of educational resource management.
Smart Images
Figure CN119917908A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent courseware, and in particular to an automatic classification method and system for intelligent courseware constructed by hierarchical semantic web. Background Art
[0002] With the rapid development of Internet education, the trend of digitalization and networking of educational resources is obvious. A huge amount of courseware resources have been accumulated in the field of education, including documents, videos, audios, images and PPT presentations. The traditional manual classification method is not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and consistency of classification. The traditional teaching resource classification method relies on manual operation, such as teachers or administrators archiving and labeling courseware based on personal experience. The disadvantages of this method are obvious. On the one hand, manual processing of a large number of courseware is time-consuming and difficult to keep up with the speed of resource updates. On the other hand, subjective judgment may lead to different classification standards, affecting search efficiency and user experience, and may even miss some key information. Therefore, there is an urgent need for a new method that can integrate multi-dimensional information to achieve efficient and accurate classification.
[0003] However, the effective management and utilization of educational resources face severe challenges, including: traditional classification relies on manual operations, which is not only time-consuming and labor-intensive, but also easily affected by subjective factors, resulting in inconsistent classification standards, inconsistency and low efficiency; existing automated classification technologies mostly focus on surface labels or keyword matching, lack deep understanding, and are unable to deeply explore the multi-dimensional attributes of courseware, such as interdisciplinary nature, difficulty or applicable age, resulting in superficial and rough classification; they fail to effectively integrate the semantic and structural information of the content, have weak knowledge relevance, and are unable to reflect the relationship between the internal logic of the courseware and the external knowledge graph, limiting the deep understanding and personalized recommendation of educational content; static classification models cannot change with the update of educational resources and user behavior, lack a dynamic optimization mechanism, and reduce long-term use effects. Summary of the invention
[0004] In view of the above-mentioned problems, the present invention is proposed.
[0005] Therefore, the problem to be solved by the present invention is how to provide a method for automatically and efficiently processing a large number of courseware through the integration of a large language model and a knowledge graph, deeply understanding the content, realizing fast and accurate classification, and improving management efficiency and retrieval accuracy; designing an algorithm to integrate semantic understanding and graph matching, extracting subjects, difficulty and applicable grades at multiple levels, realizing deep classification, and meeting personalized needs; constructing a semantic network to form a hierarchical semantic network, in which nodes and edges represent courseware attributes and their relationships, promoting complex queries and recommendations, and enhancing educational content discovery and knowledge integration; introducing machine learning and reinforcement learning, dynamically adjusting the model according to user feedback and data, continuously optimizing classification, and ensuring the timeliness and adaptability of classification.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In the first aspect, an embodiment of the present invention provides a method for automatic classification of intelligent courseware based on a hierarchical semantic web, which includes obtaining and preprocessing courseware data, performing deep semantic extraction on the preprocessed courseware data using a large language model, and generating a high-dimensional semantic feature space; performing entity recognition and relationship extraction on the high-dimensional semantic feature space to construct a knowledge graph; performing multi-dimensional attribute extraction on the knowledge graph to obtain multi-dimensional attribute information; constructing and optimizing a hierarchical semantic web based on the multi-dimensional attribute information; constructing a dynamic feedback and online optimization mechanism and adjusting the hierarchical semantic web to achieve automatic classification of intelligent courseware.
[0008] As a preferred solution of the intelligent courseware automatic classification method constructed by the hierarchical semantic web described in the present invention, the acquisition of courseware data and preprocessing refers to multimodal data extraction of educational courseware, including text, images, video and audio, to form an original data set.
[0009] As a preferred solution of the intelligent courseware automatic classification method constructed by the hierarchical semantic network described in the present invention, the deep semantic extraction of the preprocessed courseware data using a large language model includes a BERT deep bidirectional encoder and a GPT generative model; the BERT deep bidirectional encoder includes the following steps: using the BERT model to perform deep text analysis, optimizing the model parameters θ through pre-training and fine-tuning processes to maximize the loss function; introducing a regularization term based on the original loss function, and using Focal Loss to solve the category imbalance problem, assuming that the category weight is ω c , then the new loss function is as follows:
[0010]
[0011] Among them, L class_imbalance is the loss of using Focal Loss to deal with the category imbalance problem; β and λ are the coefficients of the regularization term; the GPT generative model includes the following steps: the GPT model is applied to the text generation task, and the loss function is optimized by maximizing the probability of the next word under a given historical word order; the learning rate is adjusted using cosine annealing by introducing an adaptive learning rate scheduling strategy, and the specific formula is as follows:
[0012]
[0013] Among them, η anneal (t) is the learning rate that changes with the training round t; Ω(θ GPT ) is the regularization term of model complexity.
[0014] As a preferred solution of the intelligent courseware automatic classification method constructed by the hierarchical semantic web of the present invention, wherein: the construction of the knowledge graph includes entity recognition, relationship extraction, and graph reasoning and completion; the entity recognition refers to the use of the BiLSTM-CRF model to optimize the objective function, and the specific formula is as follows:
[0015]
[0016] The relationship extraction refers to the introduction of the Transformer model, combined with multi-task learning, and optimization of the loss function. The specific formula is as follows:
[0017] L RE =λ 1 L start +λ 2 L end +λ 3 L type
[0018] Among them, L start and L end is the entity boundary classification loss; L type is the relationship type classification loss; the graph reasoning and completion refers to the use of the GNN model for knowledge reasoning, and the model update rules are as follows:
[0019]
[0020] in, is the normalized adjacency matrix; W (l) is the weight matrix; σ is the activation function.
[0021] As a preferred solution of the intelligent courseware automatic classification method constructed by the hierarchical semantic web of the present invention, the multi-dimensional attribute extraction includes attribute fusion model and attribute dimension optimization; the attribute fusion model refers to constructing an attribute fusion model based on the attention mechanism, combining the large language model and knowledge graph information, and the optimization objectives are as follows:
[0022]
[0023] Among them, α i is the attention weight; β is the integration weight; the attribute dimension optimization includes subject classification, difficulty judgment, age / grade classification and attribute optimization; the subject classification refers to classification using a multi-layer perceptron, and the specific formula of the loss function is as follows:
[0024]
[0025] The difficulty judgment refers to the judgment using SVR, and the specific formula of the loss function is as follows:
[0026]
[0027] The age / grade classification refers to the use of random forest to optimize the Gini index, the specific formula is as follows:
[0028]
[0029] The specific formula for the attribute optimization is as follows:
[0030]
[0031] in, is the exponential moving average of the smoothed squared gradient; η is the learning rate; ∈ is a small constant to prevent division by zero.
[0032] As a preferred solution of the intelligent courseware automatic classification method constructed by the hierarchical semantic web of the present invention, wherein: the construction of the hierarchical semantic web and optimization includes the definition of the hierarchical semantic web, node clustering and weight allocation and optimization strategy; the definition of the hierarchical semantic web refers to the construction of a hierarchical semantic web G = V, E, W (, wherein the edge weight W is calculated according to the attribute correlation, and the specific formula is as follows:
[0033]
[0034] The node clustering and weight allocation include: clustering adopts agglomerative hierarchical algorithm, and updating the distance between clusters according to the Lance-Williams formula; edge weight is based on attribute joint frequency or mutual information; the optimization strategy includes the following steps: introducing heuristic pruning and online learning strategies, dynamically adjusting the network structure and weight based on user feedback and system performance evaluation, the specific formula is as follows:
[0035]
[0036] The introduction of the Meta-Learning mechanism enables the model to quickly generalize from previous tasks to new tasks. The specific formula is as follows:
[0037]
[0038] Among them, β is the parameter that controls the update step size of meta-learning.
[0039] As a preferred solution of the intelligent courseware automatic classification method constructed by the hierarchical semantic web described in the present invention, the dynamic feedback and online optimization mechanism includes online learning and feedback collection, reinforcement learning mechanism and knowledge graph dynamic optimization; the online learning and feedback collection refers to real-time processing of feedback data, updating model parameters θ, and optimizing model performance based on user behavior; the reinforcement learning mechanism refers to designing a reward mechanism to adjust the strategy π θ, optimize the cumulative reward R; the dynamic optimization of the knowledge graph refers to adjusting the graph according to feedback information through the graph evolution algorithm GEA, and updating the rules to consider model contribution and user feedback.
[0040] On the second aspect, in order to further solve the security problems existing in intelligent courseware, the embodiment of the present invention provides an intelligent courseware automatic classification system constructed by a hierarchical semantic network, which includes: a data processing module, which is used to obtain and preprocess courseware data, and use the BERT deep bidirectional encoder and the GPT generative model to perform deep semantic extraction to generate a high-dimensional semantic feature space; a knowledge graph module, which is used to perform entity recognition, relationship extraction, graph reasoning and completion on the high-dimensional semantic feature space to construct a knowledge graph; an attribute extraction module, which is used to extract multi-dimensional attributes of the knowledge graph to obtain multi-dimensional attribute information; a network construction module, which is used to construct a hierarchical semantic network according to the multi-dimensional attribute information, and optimize the hierarchical semantic network using an optimization strategy; a feedback optimization module, which is used to construct a dynamic feedback and online optimization mechanism and adjust the hierarchical semantic network to achieve automatic classification of intelligent courseware.
[0041] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for automatic classification of intelligent courseware constructed by a hierarchical semantic web as described in the first aspect of the present invention.
[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for automatic classification of intelligent courseware constructed by a hierarchical semantic web as described in the first aspect of the present invention.
[0043] Beneficial effects of the present invention: The present invention realizes efficient and rapid classification of a large number of courseware by introducing automated processing procedures and large language models, which not only significantly improves the processing speed, but also can easily expand the scope of application to cover a wider range of educational content, thereby enhancing the efficiency and comprehensiveness of educational resource management; comprehensively uses multiple classification dimensions of subject, difficulty and learning stage, and utilizes the deep understanding ability of the large language model to ensure that each courseware can be refined and accurately classified, improves the pertinence and practicality of the classification, and meets the personalized retrieval needs of different user groups; by designing dynamic feedback and online optimization mechanisms, the system can continuously self-learn and adjust according to user behavior, market trends and the update of educational resources; by constructing a hierarchical knowledge graph and semantic network, it not only optimizes the organizational structure of educational resources, but also promotes the integration and discovery of interdisciplinary knowledge, providing users with an efficient and convenient retrieval platform; the advanced technologies of deep learning and knowledge graphs, especially the application of large language models, not only improve the intelligence level of the system, but also realize accurate analysis of educational content through deep semantic understanding, demonstrating the powerful efficiency improvement brought by technical collaboration. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0045] Figure 1 This is an overall flow chart of the intelligent courseware automatic classification method constructed by the hierarchical semantic web in Example 1.
[0046] Figure 2 This is a schematic diagram of the structure of the computer device in Example 3. DETAILED DESCRIPTION
[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0050] Example 1
[0051] Reference Figure 1 , which is the first embodiment of the present invention, and provides an automatic classification method for intelligent courseware constructed based on a hierarchical semantic web.
[0052] The existing methods for managing and utilizing educational resources mainly have the following problems: traditional classification relies on manual operations, which is not only time-consuming and labor-intensive, but also easily affected by subjective factors, resulting in inconsistent classification standards, inconsistency and low efficiency; existing automated classification technologies mostly focus on surface labels or keyword matching, lack deep understanding, and are unable to deeply explore the multi-dimensional attributes of courseware, such as interdisciplinary nature, difficulty or applicable age, resulting in superficial and rough classification; they fail to effectively integrate the semantic and structural information of the content, have weak knowledge relevance, and are unable to reflect the relationship between the internal logic of the courseware and the external knowledge graph, limiting the deep understanding and personalized recommendation of educational content; static classification models cannot change with the update of educational resources and user behavior, lack a dynamic optimization mechanism, and reduce long-term use effects.
[0053] The present application provides an effective solution to the above-mentioned problems. Next, a plurality of embodiments will be combined to elaborate on how to implement the automatic classification method of intelligent courseware constructed by the hierarchical semantic web.
[0054] Figure 1 The overall flow chart of the intelligent courseware automatic classification method constructed by hierarchical semantic web is shown, including:
[0055] S1: Obtain courseware data and preprocess it, use a large language model to perform deep semantic extraction on the preprocessed courseware data, and generate a high-dimensional semantic feature space.
[0056] Preferably, obtaining courseware data and performing preprocessing refers to extracting multimodal data from the educational courseware, including text, images, videos, and audios, to form an original data set.
[0057] Specifically, a large language model is used to perform deep semantic extraction on the preprocessed courseware data, including the BERT deep bidirectional encoder and the GPT generative model.
[0058] Furthermore, the BERT deep bidirectional encoder includes the following steps: using the BERT model for deep text analysis, optimizing the model parameters θ through pre-training and fine-tuning processes to maximize the loss function. The specific formula of the loss function is as follows:
[0059] L BERT =L MLM +αL NSP
[0060] Among them, L MLM is the masked language model loss; L NSP is the prediction loss for the next sentence; α is the balance coefficient, which ensures that the model can understand the context of a single sentence and capture the relationship between sentences.
[0061] A regularization term is introduced on the basis of the original loss function to prevent overfitting, and Focal Loss is used to solve the problem of class imbalance, especially for long-tail distribution data sets. Assuming that the class weight is ω c , then the new loss function is as follows:
[0062]
[0063] Among them, L class_imbalance is the loss for using Focal Loss to deal with the problem of class imbalance; β and λ are the coefficients of the regularization term, which control the impact of class imbalance and model complexity respectively.
[0064] Furthermore, the GPT generative model includes the following steps: The GPT model is applied to the text generation task, and the loss function is optimized by maximizing the probability of the next word under a given historical word order. The specific formula is as follows:
[0065]
[0066] Among them, x t is the tth word in the sequence; θ GPT is the model parameter; T is the sequence length.
[0067] In the GPT model, by introducing an adaptive learning rate scheduling strategy and using cosine annealing to adjust the learning rate, the model convergence speed and final performance are improved. The specific formula is as follows:
[0068]
[0069] Among them, η anneal (t) is the learning rate that changes with the training round t; Ω)θ GPT ) is a regularization term for the model complexity, which can be a weight decay term or other complexity measure.
[0070] Preferably, the use of the BERT model can deeply understand the text context in the courseware, capture complex language structure and meaning, and thus improve the accuracy of content analysis; the GPT model enhances the system's ability to generate high-quality, context-related educational materials, such as automatically generating teaching summaries or auxiliary explanations, making educational resources richer and more diverse.
[0071] S2: Perform entity recognition and relationship extraction on high-dimensional semantic feature space to build a knowledge graph.
[0072] Preferably, constructing a knowledge graph includes entity recognition, relationship extraction, and graph reasoning and completion.
[0073] Specifically, entity recognition refers to using the BiLSTM-CRF model to optimize the objective function. The specific formula is as follows:
[0074]
[0075] Specifically, relation extraction refers to introducing the Transformer model, combining multi-task learning, and optimizing the loss function. The specific formula is as follows:
[0076] L RE =λ 1 L start +λ 2 L end +λ 3 L type
[0077] Among them, L start and L end is the entity boundary classification loss; L type Classification loss for relation types.
[0078] Specifically, graph reasoning and completion refers to the use of GNN models (such as RGCN) for knowledge reasoning. The model update rules are as follows:
[0079]
[0080] in, is the normalized adjacency matrix; W (l( is the weight matrix; σ is the activation function, which realizes the iterative update of node features.
[0081] It should be noted that for the GNN model, especially RGCN, the aggregation of node features is enhanced by introducing the attention mechanism to improve the model's ability to distinguish important neighbors. The specific formula is as follows:
[0082]
[0083] Among them, ⊙ is element-wise multiplication; Attention is a function used to calculate the attention weight based on the features of the central node and its neighboring layers.
[0084] Preferably, through entity recognition and relationship extraction technology, the system can accurately extract key information from a large amount of courseware, such as course topics, difficulty levels or applicable grades, and then achieve refined classification of educational content, which is crucial for building personalized learning paths and meeting the learning needs of different students.
[0085] S3: Extract multi-dimensional attributes from the knowledge graph to obtain multi-dimensional attribute information.
[0086] Preferably, multi-dimensional attribute extraction includes attribute fusion model and attribute dimension optimization.
[0087] Specifically, the attribute fusion model refers to building an attribute fusion model based on the attention mechanism, combining the large language model and knowledge graph information, and the optimization objectives are as follows:
[0088]
[0089] Among them, α i is the attention weight; β is the integration weight, through the attention weight α i and the integration weight β balances the contribution of information from different sources.
[0090] Furthermore, the attribute dimension optimization includes subject classification, difficulty judgment, age / grade classification, and attribute optimization.
[0091] Specifically, subject classification refers to classification using a multi-layer perceptron, and the specific formula of the loss function is as follows:
[0092]
[0093] Specifically, difficulty judgment refers to the use of SVR for judgment, where the loss function is the mean square error. The specific formula is as follows:
[0094]
[0095] Specifically, age / grade classification refers to using random forest to optimize the Gini index. The specific formula is as follows:
[0096]
[0097] Specifically, attribute optimization refers to the use of more advanced optimization algorithms such as Adamax or RMSprop. These algorithms are more efficient than basic gradient descent when dealing with sparse gradients or non-stationary objective functions. The specific formula is as follows:
[0098]
[0099] in, is the exponential moving average of the smoothed squared gradient; η is the learning rate; ∈ is a small constant to prevent division by zero.
[0100] Preferably, the attribute fusion model and attribute dimension optimization strategy ensure a comprehensive evaluation of educational resources from multiple perspectives. Combined with the online learning mechanism based on user feedback, the system can dynamically adjust the recommendation strategy and continuously optimize the matching degree and personalization of learning resources.
[0101] S4: Construct and optimize a hierarchical semantic web based on multi-dimensional attribute information.
[0102] Preferably, constructing and optimizing a hierarchical semantic web includes defining a hierarchical semantic web, clustering nodes, assigning weights, and optimizing strategies.
[0103] Specifically, the definition of hierarchical semantic network refers to constructing a hierarchical semantic network G = 9V, E, W), where the edge weight W is calculated based on the attribute correlation, such as cosine similarity, and the specific formula is as follows:
[0104]
[0105] Specifically, node clustering and weight allocation include: clustering using an agglomerative hierarchical algorithm, and updating inter-cluster distances according to the Lance-Williams formula.
[0106] Edge weights are based on attribute joint frequencies or mutual information, enabling efficient representation of the network.
[0107] Furthermore, the optimization strategy includes the following steps: introducing heuristic pruning and online learning strategies, dynamically adjusting the network structure and weights based on user feedback and system performance evaluation. The specific formula is as follows:
[0108]
[0109] For the optimization strategy of hierarchical semantic network, in addition to heuristic pruning and online learning, the meta-learning mechanism is also introduced to enable the model to quickly generalize from previous tasks to new tasks, especially in network structure adjustment, such as using the Reptile algorithm to update parameters. The specific formula is as follows:
[0110]
[0111] Among them, β is the parameter that controls the update step size of meta-learning. In this way, the model can learn how to better adapt to new data or structural changes that may be encountered in the future in each adjustment.
[0112] The optimal hierarchical semantic web construction and optimization strategy, especially the introduction of meta-learning and reinforcement learning, enables the system to have the ability to self-learn and adapt to new situations, ensuring that the knowledge graph can timely reflect the latest educational content and learning trends, and provide a knowledge structure that is more in line with actual needs.
[0113] S5: Build dynamic feedback and online optimization mechanisms and adjust the hierarchical semantic web to achieve automatic classification of intelligent courseware.
[0114] Preferably, the dynamic feedback and online optimization mechanism includes online learning and feedback collection, reinforcement learning mechanism and knowledge graph dynamic optimization.
[0115] Specifically, online learning and feedback collection refers to real-time processing of feedback data, updating model parameters θ, and optimizing model performance based on user behavior.
[0116] Specifically, the reinforcement learning mechanism refers to the design of a reward mechanism to adjust the strategy π θ , optimize the cumulative reward R, the specific formula is as follows:
[0117]
[0118] Specifically, dynamic optimization of knowledge graph refers to adjusting the graph according to feedback information through the graph evolution algorithm GEA, and updating the rules to consider model contribution and user feedback. The specific formula is as follows:
[0119]
[0120] It should be noted that in the reinforcement learning part, more complex policy gradient methods are used, such as Proximal Policy Optimization (PPO) or Actor-Critid architecture. These methods perform better in terms of stability and data efficiency. The specific formula is as follows:
[0121]
[0122] in, is the estimate of the advantage function; α is the learning rate. This method not only takes into account the action value when updating the strategy, but also limits the amplitude of the strategy update, thereby avoiding performance degradation caused by large fluctuations.
[0123] In summary, the present invention realizes efficient and rapid classification of a large number of courseware by introducing automated processing procedures and large language models, which not only significantly improves the processing speed, but also can easily expand the scope of application to cover a wider range of educational content, thereby enhancing the efficiency and comprehensiveness of educational resource management; comprehensively uses multiple classification dimensions such as subject, difficulty and learning stage, and utilizes the deep understanding ability of the large language model to ensure that each courseware can be refined and accurately classified, improves the pertinence and practicality of the classification, and meets the personalized retrieval needs of different user groups; by designing dynamic feedback and online optimization mechanisms, the system can continuously self-learn and adjust according to user behavior, market trends and the update of educational resources; by constructing a hierarchical knowledge graph and semantic web, it not only optimizes the organizational structure of educational resources, but also promotes the integration and discovery of interdisciplinary knowledge, providing users with an efficient and convenient retrieval platform; it integrates the advanced technologies of deep learning and knowledge graphs, especially the application of large language models, which not only improves the intelligence level of the system, but also realizes the accurate analysis of educational content through deep semantic understanding, demonstrating the powerful performance improvement brought by technical collaboration.
[0124] Embodiment 2 is an embodiment of the present invention, which provides an intelligent courseware automatic classification system constructed by a hierarchical semantic web, including: a data processing module, which is used to obtain and preprocess courseware data, and use the BERT deep bidirectional encoder and the GPT generative model to perform deep semantic extraction to generate a high-dimensional semantic feature space; a knowledge graph module, which is used to perform entity recognition, relationship extraction, graph reasoning and completion on the high-dimensional semantic feature space to construct a knowledge graph; an attribute extraction module, which is used to extract multi-dimensional attributes of the knowledge graph to obtain multi-dimensional attribute information; a network construction module, which is used to construct a hierarchical semantic web according to the multi-dimensional attribute information, and optimize the hierarchical semantic web using an optimization strategy; a feedback optimization module, which is used to construct a dynamic feedback and online optimization mechanism and adjust the hierarchical semantic web to achieve automatic classification of intelligent courseware.
[0125] Embodiment 3 is an embodiment of the present invention, which is different from the previous embodiment in that:
[0126] like Figure 2As shown, if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0127] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0128] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.
[0129] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0130] Example 4 is an embodiment of the present invention, which provides an automatic classification method for intelligent courseware constructed by a hierarchical semantic network. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.
[0131] In order to demonstrate the performance comparison between the traditional classification method and the intelligent courseware automatic classification method constructed by the hierarchical semantic web proposed in the present invention, this example selects three common traditional classification methods, including rule-based classification, naive Bayes classification and support vector machine (SVM), and compares them with the hierarchical semantic web method. The specific data are shown in Table 1, where the accuracy rate represents the proportion of the number of correctly classified courseware to the total number of courseware, the recall rate represents the proportion of the correctly classified courseware that truly belongs to the category, the F1 score represents the harmonic mean of the precision and recall rate, and comprehensively measures the performance of the classification model. The training time represents the time required for the model to be trained from the beginning to the completion, and the update flexibility represents the ability of the model to adapt to new data or change the classification criteria. The higher the value, the faster the model can adapt to changes.
[0132] Table 1 Comparative experimental data of automatic classification methods for educational courseware
[0133]
[0134] As can be seen from the above table, the hierarchical semantic web construction method shows the best performance in accuracy, recall and F1 score, indicating that it can understand and classify educational courseware more accurately and provide higher quality classification results; although the training time of the hierarchical semantic web construction method is longer, which reflects its complex model structure and computing requirements, its update flexibility score is the highest. Once the model training is completed, it can quickly adapt to new classification needs or user feedback, which improves the response speed and adaptability of the system in the long run; in contrast, although rule-based classification is fast to train, it lacks flexibility and learning ability, and is far inferior to other machine learning methods and hierarchical semantic web methods in classification effect; Naive Bayes and SVM provide a good balance point, but they are still inferior to hierarchical semantic web in classification performance, especially when dealing with complex semantics and relationships; The intelligent courseware automatic classification method constructed by hierarchical semantic web is significantly superior to traditional methods in classification accuracy and dynamic adaptability. Although the initial investment, such as training time, is large, the long-term benefits and intelligence level are significantly improved.
[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A method for automatic classification of intelligent courseware based on hierarchical semantic web, characterized in that: include: Obtain courseware data and preprocess it, use a large language model to perform deep semantic extraction on the preprocessed courseware data, and generate a high-dimensional semantic feature space; Performing entity recognition and relationship extraction on the high-dimensional semantic feature space to construct a knowledge graph; Performing multi-dimensional attribute extraction on the knowledge graph to obtain multi-dimensional attribute information; Constructing a hierarchical semantic web according to the multi-dimensional attribute information and optimizing it; A dynamic feedback and online optimization mechanism is constructed and the hierarchical semantic network is adjusted to realize automatic classification of intelligent courseware.
2. The method for automatically classifying intelligent courseware based on hierarchical semantic web construction as claimed in claim 1, characterized in that: The obtaining of courseware data and preprocessing thereof refers to extracting multimodal data from educational courseware, including text, images, videos, and audio, to form an original data set.
3. The method for automatic classification of intelligent courseware constructed by hierarchical semantic web as claimed in claim 2, characterized in that: The use of a large language model to perform deep semantic extraction on the preprocessed courseware data includes a BERT deep bidirectional encoder and a GPT generative model; The BERT deep bidirectional encoder includes the following steps: Use the BERT model for deep text analysis and optimize the model parameters θ through pre-training and fine-tuning to maximize the loss function; A regularization term is introduced on the basis of the original loss function, and FocalLoss is used to solve the category imbalance problem. Assuming that the category weight is ω c , then the new loss function is as follows: Among them, L class_imbalance is the loss of using FocalLoss to deal with the class imbalance problem; β and λ are the coefficients of the regularization term; The GPT generative model includes the following steps: The GPT model is applied to text generation tasks and optimizes the loss function by maximizing the probability of the next word given a historical word order; By introducing an adaptive learning rate scheduling strategy, cosine annealing is used to adjust the learning rate. The specific formula is as follows: Among them, η anneal (t) is the learning rate that changes with the training round t; Ω(θ GPT ) is the regularization term of model complexity.
4. The method for automatic classification of intelligent courseware constructed by hierarchical semantic web as claimed in claim 3, characterized in that: The construction of the knowledge graph includes entity recognition, relationship extraction, and graph reasoning and completion; The entity recognition refers to using the BiLSTM-CRF model to optimize the objective function. The specific formula is as follows: The relationship extraction refers to the introduction of the Transformer model, combined with multi-task learning, and optimization of the loss function. The specific formula is as follows: L RE =λ1L start +λ2L end +λ3L type Among them, L start and L end is the entity boundary classification loss; L type Classify the loss for the relation type; The graph reasoning and completion refers to the use of the GNN model for knowledge reasoning. The model update rules are as follows: in, is the normalized adjacency matrix; W (l) is the weight matrix; σ is the activation function.
5. The method for automatic classification of intelligent courseware constructed by hierarchical semantic web as claimed in claim 4, characterized in that: The multi-dimensional attribute extraction includes attribute fusion model and attribute dimension optimization; The attribute fusion model refers to building an attribute fusion model based on the attention mechanism, combining the large language model and knowledge graph information, and the optimization objectives are as follows: Among them, α i is the attention weight; β is the integration weight; The attribute dimension optimization includes subject classification, difficulty judgment, age / grade classification and attribute optimization; The subject classification refers to classification using a multi-layer perceptron, and the specific formula of the loss function is as follows: The difficulty judgment refers to the judgment using SVR, and the specific formula of the loss function is as follows: The age / grade classification refers to the use of random forest to optimize the Gini index, the specific formula is as follows: The specific formula for the attribute optimization is as follows: in, is the exponential moving average of the smoothed squared gradient; η is the learning rate; ∈ is a small constant to prevent division by zero.
6. The method for automatic classification of intelligent courseware constructed by hierarchical semantic web as claimed in claim 5, characterized in that: The construction and optimization of the hierarchical semantic network includes the definition of the hierarchical semantic network, node clustering and weight allocation and optimization strategy; The hierarchical semantic network definition refers to constructing a hierarchical semantic network G = (V, E, W), where the edge weight W is calculated based on the attribute correlation. The specific formula is as follows: The node clustering and weight allocation include: Clustering uses an agglomerative hierarchical algorithm, and the inter-cluster distances are updated according to the Lance-Williams formula; Edge weights are based on attribute joint frequencies or mutual information; The optimization strategy includes the following steps: Heuristic pruning and online learning strategies are introduced to dynamically adjust the network structure and weights based on user feedback and system performance evaluation. The specific formula is as follows: The introduction of the Meta-Learning mechanism enables the model to quickly generalize from previous tasks to new tasks. The specific formula is as follows: Among them, β is the parameter that controls the update step size of meta-learning.
7. The method for automatic classification of intelligent courseware constructed by hierarchical semantic web as claimed in claim 6, characterized in that: The dynamic feedback and online optimization mechanism includes online learning and feedback collection, reinforcement learning mechanism and knowledge graph dynamic optimization; The online learning and feedback collection refers to real-time processing of feedback data, updating model parameters θ, and optimizing model performance based on user behavior; The reinforcement learning mechanism refers to the design of a reward mechanism to adjust the strategy π θ , optimize the cumulative reward R; The dynamic optimization of the knowledge graph refers to adjusting the graph according to feedback information through the graph evolution algorithm GEA, and updating the rules to consider model contribution and user feedback.
8. An intelligent courseware automatic classification system based on hierarchical semantic web construction, based on the intelligent courseware automatic classification method based on hierarchical semantic web construction according to any one of claims 1 to 7, characterized in that: include, The data processing module is used to obtain and preprocess courseware data, use the BERT deep bidirectional encoder and the GPT generative model to perform deep semantic extraction, and generate a high-dimensional semantic feature space; The knowledge graph module is used to perform entity recognition, relationship extraction, graph reasoning and completion on high-dimensional semantic feature space to build a knowledge graph; The attribute extraction module is used to extract multi-dimensional attributes from the knowledge graph to obtain multi-dimensional attribute information; The network construction module is used to construct a hierarchical semantic network based on multi-dimensional attribute information and optimize the hierarchical semantic network using optimization strategies; Feedback optimization module is used to build dynamic feedback and online optimization mechanism and adjust the hierarchical semantic network to realize automatic classification of intelligent courseware.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for automatic classification of intelligent courseware constructed by hierarchical semantic web as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for automatically classifying intelligent courseware constructed by a hierarchical semantic web as described in any one of claims 1 to 7 are implemented.