Knowledge association learning method and system based on knowledge graph and virtual reality

Abstract

The invention relates to the technical field of knowledge graph and virtual reality fusion, in particular to a knowledge association learning method and system based on a knowledge graph and virtual reality, and the method comprises the following steps: obtaining multi-source data including text data, image data, video data and three-dimensional model data; constructing an interdisciplinary knowledge graph based on the multi-source data; mapping the interdisciplinary knowledge graph into a virtual reality environment; generating a personalized learning path based on knowledge representation in a virtual reality environment; interactive knowledge exploration is realized in a virtual reality environment; updating the interdisciplinary knowledge graph based on the user interaction data; the virtual reality environment is synchronously updated to reflect the change of the interdisciplinary knowledge graph, and the understanding ability of a learner to complex concepts is greatly improved by visualizing an abstract knowledge structure in the virtual reality environment; and the learner can more easily find and understand the relation among different discipline knowledge.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph and virtual reality fusion technology, and more specifically, to a knowledge association learning method and system based on knowledge graph and virtual reality. Background Art

[0002] With the rapid development of science and technology and the explosive growth of knowledge, interdisciplinary learning has become an important trend in today's education field. The traditional single-discipline education model can no longer meet the needs of modern society for compound talents. However, the learning and association of interdisciplinary knowledge has always been a challenging problem.

[0003] In recent years, the application of knowledge graphs and virtual reality technology in the field of education has attracted widespread attention. As a structured knowledge representation method, knowledge graphs can effectively describe the complex relationships between concepts. For example, Christopoulos et al. explored the application potential of virtual reality in cultural education in their research (Christopoulos, A., Styliou, M., Ntalas, N., & Stylios, C. (2024). The Impact of Immersive Virtual Reality on Knowledge Acquisition and Adolescent Perceptions in Cultural Education. Information, 15, 261. The study showed that virtual reality technology can significantly improve students' knowledge acquisition ability and learning interest.

[0004] However, existing technical solutions still have many limitations. First, most knowledge graph-based learning systems lack intuitive knowledge presentation methods, making it difficult for learners to understand abstract concepts and complex relationships. Second, although virtual reality technology can provide an immersive learning experience, existing VR educational applications often lack systematic knowledge organization and intelligent learning guidance. In addition, the effective association and dynamic updating of interdisciplinary knowledge is also a problem that needs to be solved urgently.

[0005] The main problems of the existing technical solutions can be summarized as follows:

[0006] 1. The separation of knowledge representation and presentation: The abstractness of knowledge graphs and the intuitiveness of virtual reality have not been organically combined.

[0007] 2. Insufficient interdisciplinary knowledge linkage: There is a lack of effective mechanisms to establish and demonstrate connections between knowledge in different disciplines.

[0008] 3. Limited support for personalized learning: It is difficult to provide customized learning paths based on learners’ knowledge background and learning goals.

[0009] 4. Knowledge updating is out of touch with the learning environment: The integration of new knowledge is often delayed and cannot be reflected in the learning process in a timely manner.

[0010] 5. Poor learning interaction experience: Lack of natural and intuitive interaction methods affects learning effects and experience.

[0011] Therefore, there is an urgent need to provide a knowledge association learning method and system based on knowledge graph and virtual reality to address the shortcomings of existing technologies. Summary of the invention

[0012] In view of the above problems, the present invention proposes a knowledge association learning method and system based on knowledge graph and virtual reality. The method aims to solve key technical problems in interdisciplinary knowledge learning, including structured representation and intuitive presentation of knowledge, interdisciplinary knowledge association, personalized learning path generation, real-time knowledge update and natural interaction.

[0013] The present invention provides a knowledge association learning method based on knowledge graph and virtual reality, comprising the following steps: obtaining multi-source data, wherein the multi-source data includes text data, image data, video data and three-dimensional model data; constructing an interdisciplinary knowledge graph based on the multi-source data; mapping the interdisciplinary knowledge graph to a virtual reality environment; generating a personalized learning path based on the knowledge representation in the virtual reality environment; realizing interactive knowledge exploration in the virtual reality environment; updating the interdisciplinary knowledge graph based on user interaction data; and synchronously updating the virtual reality environment to reflect changes in the interdisciplinary knowledge graph.

[0014] Specifically, the steps of constructing an interdisciplinary knowledge graph include: first, processing the text data, including: using an improved two-way maximum matching algorithm for word segmentation, wherein the optimal word segmentation result is determined by calculating the product of the word sequence probability and the text generation probability; using a Transformer-based sequence labeling model for part-of-speech tagging, wherein a softmax function is used to predict labels for the Transformer output; using a graph attention network to construct a dependency syntactic tree, wherein the final syntactic structure is obtained by iteratively updating node representations and edge weights; then, processing the image data, normalizing the image data, and performing random data enhancement; using an improved EfficientNet-V2 model to extract image features, wherein an attention mechanism and multi-scale The invention relates to a method for extracting features of a three-dimensional model and a method for fusing features of the three-dimensional model. The method comprises the following steps: first, performing a multi-scale feature extraction and a multi-scale feature aggregation algorithm on the three-dimensional model data, and calculating the channel attention and spatial attention of the feature map; extracting features at different scales and fusing them through adaptive weights; processing the video data, extracting key frames using an improved scene boundary detection algorithm, wherein the optical flow estimation and the deep learning model are combined to calculate the difference between frames; finally, performing the same processing steps as the image data on the key frames; processing the three-dimensional model data, extracting geometric features using an improved PointNet++ model, wherein dynamic point cloud sampling and hierarchical feature aggregation are introduced, and the number of sampling points is adaptively adjusted according to the point cloud density; obtaining the final geometric feature representation through multi-level local feature extraction and global feature aggregation; designing a new texture encoding network to extract texture information, wherein multi-scale convolution and attention mechanism are used to process texture images.

[0015] Specifically, the step of constructing an interdisciplinary knowledge graph also includes: first, based on a context-aware synonym recognition algorithm, construct an interdisciplinary synonym dictionary and obtain dictionaries of various disciplines; calculate the context-aware embedding representation of words, combining the context information output by the BERT model and processed by the graph neural network; set a dynamic threshold and adjust the threshold according to the word frequency; identify synonyms and merge dictionaries based on the context-aware embedding and the dynamic threshold; then, use an improved semantic similarity calculation method to determine the similarity between interdisciplinary entities and calculate the cosine similarity of entity pairs; calculate the path similarity of entity pairs in the knowledge graph, considering the path length and type; calculate the structural similarity of entity pairs, based on the Jaccard similarity of entity neighbor nodes; weightedly fuse the above three similarities to obtain the final entity similarity; finally, use an iterative matching algorithm to achieve interdisciplinary entity alignment and initialize the alignment set; in each iteration, calculate the current threshold, where the threshold decays exponentially with the number of iterations; select entity pairs to be aligned based on the entity similarity and the current threshold; update the similarity matrix, considering the influence of aligned entities; repeat the above steps until the maximum number of iterations is reached or the termination condition is met.

[0016] Specifically, the steps of mapping the interdisciplinary knowledge graph into the virtual reality environment include: first, using a cross-modal attention fusion network to fuse text features, image features and three-dimensional features, and perform modality-specific encoding on each modality feature; calculating the cross-modal attention matrix, in which a scaled dot product attention mechanism is used; performing information aggregation to weightedly fuse features of different modalities; using a gating mechanism to dynamically adjust the ratio of original features and aggregated features; obtaining the final fused feature representation through a multi-layer perceptron; then, based on the knowledge distillation method, migrating the pre-trained knowledge graph embedding to the student model to obtain the pre-trained knowledge graph embedding and fused features; obtaining the student model through a multi-layer perceptron network, respectively. The outputs of the student model and the teacher model are calculated; the knowledge distillation loss is calculated, and the KL divergence is used to measure the difference between the outputs of the student model and the teacher model; the temperature parameter is dynamically adjusted, where the temperature parameter decays exponentially with the training progress; the knowledge embedding is mapped to the virtual space representation through a multi-layer perceptron network; finally, a consistency constraint is adopted to ensure the consistency of the virtual object and the knowledge representation, and a projection function is defined to project the knowledge embedding to the virtual space; the mean square error between the virtual object representation and the projected knowledge embedding is calculated as the consistency loss; the overall loss function is obtained by combining the knowledge distillation loss, the consistency loss and the downstream task loss; and a high-quality mapping of the knowledge representation to the virtual object is achieved by optimizing the overall loss function.

[0017] Specifically, the steps of generating a personalized learning path include: first, evaluating the user's current knowledge status, and constructing a user knowledge vector based on the user's historical learning data and test results; using a knowledge graph structure to calculate the integrity and relevance scores of the user's knowledge; setting a knowledge mastery threshold, and marking knowledge points with scores higher than the threshold as mastered; then, based on the user's knowledge status and learning goals, performing a knowledge gap analysis to obtain a learning goal knowledge set; calculating a gap set, including knowledge points that have not been mastered and knowledge points with a mastery level lower than a threshold; ranking the knowledge points in the gap set by importance, taking into account the centrality and relevance of the knowledge points; finally, using an adaptive learning path generation algorithm, taking into account the difficulty and importance of the knowledge points, generating an optimal learning sequence, initializing a priority queue, and adding the knowledge points in the gap set to the queue; selecting the knowledge point with the highest priority from the priority queue; and determining whether the learning requirements are met.

[0018] Specifically, the steps for implementing interactive knowledge exploration include: first, using an improved gesture recognition algorithm to implement natural gesture control and obtain user hand gesture data; applying a spatiotemporal attention mechanism to extract the spatiotemporal features of hand gestures; using a long short-term memory network to model temporal features; performing gesture classification through a multi-layer perceptron; then, using a multi-scale convolution and attention pooling voice control algorithm to process voice commands and obtain audio signals; calculating the spectrum of the audio signal; applying multi-scale convolution to extract features, using convolution kernels of different sizes; using an attention pooling mechanism to aggregate features and dynamically adjust the importance of different features; recognizing commands through a Transformer network; finally, based on user attention data, implementing dynamic rendering and adaptive detail level adjustment, obtaining user attention data and a set of virtual objects; calculating an attention-guided importance map to map the user's attention distribution to a virtual environment; for each virtual object, calculating an adaptive detail level based on its score in the importance map; constructing a rendering priority queue to sort virtual objects according to the importance score; in a rendering loop, rendering according to time budget and priority to ensure that key objects are rendered in a timely manner.

[0019] Specifically, it also includes the steps of realizing multi-user collaborative learning: first, based on user ability evaluation and task requirements, dynamically assign collaborative roles and obtain user sets and task requirements; evaluate the ability of each user, considering the knowledge mastery and historical performance; calculate the matching degree between the user and the role, and use cosine similarity to measure the matching degree between the user ability and the role requirements; use the Hungarian algorithm to perform optimal role assignment and maximize the overall matching degree; then, use the task decomposition algorithm to divide the collaborative task into subtasks, obtain the collaborative tasks and the assigned roles; construct a task graph based on the dependencies and complexity of the tasks; use the graph partitioning algorithm to divide the task graph into subgraphs, each subgraph corresponds to a subtask; assign the most suitable role to each subtask, considering the role ability and subtask requirements; finally, adopt an improved state synchronization mechanism to ensure consistency among multiple users, and each user broadcasts its state regularly; the server collects all user states and performs consistency checks; if a conflict is detected, use a conflict resolution strategy based on timestamps and priorities; broadcast the resolved state to all users to achieve state synchronization.

[0020] Specifically, the steps of updating the interdisciplinary knowledge graph include: first, extracting relevant information from user interaction data, using a knowledge extraction algorithm to identify new entities and relationships, and preprocessing the user interaction data, including text cleaning and standardization; using a named entity recognition model to identify potential new entities; using a relationship extraction model to identify potential new relationships between entities; performing preliminary filtering on the extracted entities and relationships to remove low-confidence candidates; then, performing fact verification to ensure the reliability of the new knowledge, and comparing the extracted new knowledge with an external knowledge base for verification; calculating the confidence score of the new knowledge, considering multiple factors such as the confidence of the extraction model and external verification results; setting a confidence threshold to screen new knowledge with high reliability; finally, using an incremental update strategy to efficiently update the knowledge graph, and checking the consistency of the verified new knowledge with the existing knowledge graph; if a conflict is found, using a conflict resolution strategy based on confidence and timeliness; integrating conflict-free new knowledge into the existing knowledge graph; and updating the index structure of the knowledge graph to adapt to the newly added entities and relationships.

[0021] Specifically, the step of synchronously updating the virtual reality environment includes: first, detecting the difference between the knowledge graph and the current virtual environment representation, and comparing the updated knowledge graph with the knowledge representation in the virtual environment; identifying newly added, deleted and modified nodes and edges; calculating the scope and importance of each change; then, based on the change classification and priority allocation, formulating an update plan, and assigning priority to each change according to the change type and importance; considering the user's current focus, adjusting the priority of the change; generating a final update sequence to ensure that important changes are given priority; then, using an incremental rendering method to update the virtual environment frame by frame and preload the resources required for the update; building a rendering queue and sorting it according to the priority of the update plan; in each frame, performing a limited number of update operations; dynamically adjusting the update speed to keep the rendering frame rate stable; finally, realizing adaptive adjustment of the user's perspective and monitoring the impact of environmental changes on the user's current perspective; calculating the optimal perspective transition path to ensure visual continuity; and dynamically adjusting the perspective transition speed according to the user's attention and the importance of environmental changes.

[0022] The system for implementing the method includes: a multi-source data processing module for preprocessing multimodal input data; a knowledge graph construction module for generating and maintaining an interdisciplinary knowledge graph; a virtual reality mapping module for mapping the knowledge graph to a three-dimensional virtual environment; a learning path generation module for creating a personalized learning plan; an interactive exploration module for realizing natural interaction between users and virtual environments; a collaborative learning module for supporting multi-user joint learning; and a knowledge update module for dynamically updating knowledge graphs and virtual environments. The system interacts with users through virtual reality devices and uses distributed computing resources for data processing and rendering optimization.

[0023] The core of the present invention is to deeply integrate knowledge graphs with virtual reality technology to create an intelligent, personalized, immersive learning environment. Through innovative technologies such as multi-source data processing, interdisciplinary knowledge graph construction, virtual reality mapping, personalized learning path generation, interactive knowledge exploration, and dynamic update mechanisms, the present invention achieves the following beneficial effects:

[0024] 1. Improve the depth of knowledge understanding: By visualizing abstract knowledge structures in a virtual reality environment, learners' ability to understand complex concepts is greatly improved.

[0025] 2. Enhance interdisciplinary knowledge association: Based on the structured representation of knowledge graphs and the spatial presentation of virtual reality, learners can more easily discover and understand the connections between knowledge in different disciplines.

[0026] 3. Optimize learning efficiency: Personalized learning path generation algorithm and interactive knowledge exploration mechanism enable learners to locate and absorb key knowledge faster.

[0027] 4. Improve learning experience: The immersive virtual reality environment and natural interaction methods greatly enhance the fun and participation of learning.

[0028] 5. Realize real-time updating of knowledge: Innovative incremental update strategy ensures that learning content is always kept up to date, improving the timeliness and relevance of learning.

[0029] 6. Support collaborative learning: The design of multi-user virtual environment provides new possibilities for interdisciplinary team collaboration.

[0030] In summary, this invention provides a new paradigm for interdisciplinary knowledge learning by innovatively integrating knowledge graphs and virtual reality technology. It not only solves the key problems existing in the existing technology, but also points out the direction for the development of future educational technology. With the further optimization and application promotion of technology, this invention is expected to play an important role in improving learning effects, promoting innovative thinking and cultivating compound talents, and provide strong support for meeting the challenges of the knowledge economy era. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a logic block diagram of the core steps of the present invention.

[0032] Figure 2 Construct an algorithm logic diagram for the knowledge graph of the present invention.

[0033] Figure 3 This is a logic block diagram of the virtual reality mapping algorithm of the present invention.

[0034] Figure 4Generate an algorithm logic block diagram for the personalized learning path of the present invention.

[0035] Figure 5 This is the logic block diagram of the interactive knowledge exploration algorithm of the present invention.

[0036] Figure 6 This is the logic block diagram of the knowledge graph update algorithm of the present invention.

[0037] Figure 7 This is a logic block diagram of the virtual environment synchronization update algorithm of the present invention.

[0038] Figure 8 It is a logic block diagram of the system of the present invention.

[0039] Fig. 9 It is the conditional judgment and loop logic block diagram of the present invention. DETAILED DESCRIPTION

[0040] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0041] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0042] The present invention provides a knowledge association learning method and system based on knowledge graph and virtual reality. The method realizes the effective association and intuitive presentation of interdisciplinary knowledge by integrating knowledge graph and virtual reality technology, and provides learners with an immersive knowledge exploration experience.

[0043] Please refer to Figure 1-9 Specifically, the method of the present invention comprises the following steps:

[0044] Step S1: Obtain multi-source data. The present invention first obtains multi-source data including text data, image data, video data and three-dimensional model data. These data sources can be teaching materials, academic papers, online course videos, virtual laboratory models, etc. Preferably, in one embodiment of the present invention, the acquisition of text data adopts web crawler technology to capture literature materials in related fields from preset academic websites and digital libraries. Image and video data are obtained by cooperating with educational institutions to obtain authorized multimedia teaching resources. The three-dimensional model data can be purchased from a professional 3D model library, or it can be generated by field experts according to teaching needs.

[0045] Step S2: Construct an interdisciplinary knowledge graph. Based on the obtained multi-source data, the present invention adopts an innovative interdisciplinary knowledge graph construction method. First, the text data is processed. Word segmentation is performed by an improved two-way maximum matching algorithm, which determines the optimal word segmentation result by calculating the product of the word sequence probability P(w|d) and the text generation probability P(d|w). Practice shows that when the dictionary size is between 100,000 and 1 million, the word segmentation accuracy of the algorithm can reach more than 97%. Next, a Transformer-based sequence labeling model is used for part-of-speech tagging, in which the softmax function is used to predict the label of the Transformer output. In order to improve the labeling accuracy, we used focal loss during model training, which effectively solved the label imbalance problem and made the F1 score of the part-of-speech tagging reach 0.95. Finally, a dependency syntactic tree is constructed using a graph attention network, in which the final syntactic structure is obtained by iteratively updating the node representation and edge weight. Experience shows that when the number of iterations is set to 5-10 times, a good balance can be achieved between efficiency and accuracy.

[0046] Then, the image data is processed. First, the image data is normalized and random data enhancement is performed. Data enhancement methods include random cropping, rotation, flipping, and color jittering. The number of enhancement operations is set to 2-4, and the enhancement intensity is controlled between 0.1-0.3, which can effectively improve the generalization ability of the model. Next, the improved EfficientNet-V2 model is used to extract image features, in which the attention mechanism and multi-scale feature fusion are introduced. Specifically, we calculate the channel attention and spatial attention of the feature map. The channel attention uses the SE (Squeeze-and-Excitation) module, and the spatial attention uses the non-local module. Features are extracted at different scales and fused through adaptive weights, where the weights are normalized by the softmax function. This improvement has increased the Top-1 accuracy of image classification by 2.5 percentage points compared to the original EfficientNet-V2, reaching 87.5%.

[0047] For video data, the present invention uses an improved scene boundary detection algorithm to extract key frames. The algorithm combines optical flow estimation and deep learning models to calculate the difference between frames. Specifically, we first use the TVL1 algorithm to calculate the optical flow between adjacent frames, and then use the pre-trained ResNet-18 model to extract frame-level features. The weighted sum of the optical flow amplitude and the feature difference is calculated as the difference between frames, where the weight ratio is determined to be 3:7 through cross-validation. When the difference exceeds a preset threshold (the empirical value is 0.15), it is judged as a scene boundary. This method makes the F1 score of scene boundary detection reach 0.85, which is significantly better than the traditional color histogram-based method. For the extracted key frames, we perform the same processing steps as the image data.

[0048] Finally, the three-dimensional model data is processed. The present invention uses an improved PointNet++ model to extract geometric features, in which dynamic point cloud sampling and hierarchical feature aggregation are introduced. The specific steps are: first, the number of sampling points is adaptively adjusted according to the point cloud density, and the number of sampling points n satisfies the equation n=min(1000,max(100,10*log(N))), where N is the number of points in the original point cloud. This dynamic sampling strategy not only ensures the accuracy of feature extraction, but also controls the computational complexity. Then, the final geometric feature representation is obtained through multi-level local feature extraction and global feature aggregation. At each level, we use the k-nearest neighbor algorithm to construct a local neighborhood (the k value is usually set to 16 or 32), and apply the PointNet structure to extract local features. Global features are obtained through maximum pooling operations. This hierarchical feature extraction method enables the model to capture local details and global structural information at the same time. In addition, we also designed a new texture encoding network to extract texture information, in which multi-scale convolution and attention mechanisms are used to process texture images. Specifically, we use a multi-branch structure similar to the Inception module, use 1x1, 3x3, and 5x5 convolution kernels to extract features in parallel, and then adaptively fuse features of different scales through the channel attention mechanism. This improvement has enabled the accuracy of 3D model classification to reach 94%, an increase of 3 percentage points over the original PointNet++.

[0049] Step S3: Mapping the interdisciplinary knowledge graph into the virtual reality environment. This step is one of the core innovations of the present invention, which realizes the transformation of abstract knowledge into concrete virtual objects. First, we use a cross-modal attention fusion network to fuse text features, image features, and three-dimensional features.

[0050] Specifically, for each modality feature, we perform modality-specific encoding:

[0051] E i =MLP i (F i),i∈{text,image,3d}

[0052] Among them, E i Represents the encoded features, MLP i is a multilayer perceptron designed for each modality, F i is the original feature, then we calculate the cross-modal attention matrix:

[0053]

[0054] Here, A ij is the attention matrix, Q i and K j are the query matrix and key matrix respectively, and d is the feature dimension. It is used for scaling to prevent the dot product result from being too large. Then, we perform information aggregation and weighted fusion of features from different modalities:

[0055]

[0056] Among them, F i,agg is the aggregated feature, W V is the weight of the value matrix.

[0057] Next, we use a gating mechanism to dynamically adjust the ratio of original features to aggregated features:

[0058] G i =σ(W G [E i ; F i,arg ])

[0059] F i,gated =G i E i +(1-G i )F i,agg

[0060] Here, G i is the gate value, W G is a learnable parameter, and [;] represents feature concatenation.

[0061] Finally, through the multi-layer perceptron, the final fusion feature representation is obtained:

[0062] F fused =MLP([F text,gated ; F image,gated ; F 3d,gated ])

[0063] In practical applications, for example, when learning molecular biology, text features may come from descriptions in scientific papers, image features may be cell images under a microscope, and 3D features may be structural models of protein molecules. Through this fusion, learners can see the text description, 2D image, and 3D structure of molecules simultaneously in a virtual environment, greatly enhancing their understanding of complex concepts.

[0064] S4: Generate personalized learning paths based on knowledge representation in virtual reality environment

[0065] Based on historical learning data and test results, we construct the user knowledge vector K user Then, using the knowledge graph structure, the completeness and relevance scores of the user’s knowledge are calculated. The completeness score is defined as:

[0066]

[0067] Among them, K total It is the collection of all knowledge points in the knowledge graph.

[0068] The relevance score is measured by calculating the average shortest path length of the mastered knowledge points in the knowledge graph:

[0069]

[0070] Next, we set the knowledge mastery threshold θ mastery (usually 0.7), and mark the knowledge points with scores above the threshold as mastered.

[0071] Then, based on the user's knowledge status and learning goals, we conduct knowledge gap analysis. Get the learning goal knowledge set T and calculate the gap set:

[0072]

[0073] The knowledge points in the Gap are sorted by importance, and the importance score is calculated as follows:

[0074] Importance(k i )=λ1·Centrality(k i )+λ2·Relevance(k i ,T)

[0075] Among them, λ1 and λ2 are weight coefficients, satisfying λ1+λ2=1. Centrality(k i ) represents the centrality of the knowledge point in the graph, which is calculated using the PageRank algorithm. i ,T) represents the relevance to the learning target and is calculated using cosine similarity.

[0076] Finally, we use an adaptive learning path generation algorithm to generate the optimal learning sequence by considering the difficulty and importance of knowledge points. The core steps of the algorithm are as follows:

[0077] First, initialize the priority queue Q = PriorityQueue(), and then for each knowledge point

[0078] k i ∈Gap:difficulty=CalculateDifficulty(k i ,K user )

[0079] importance=Importance(k i )Q.push((k i ,difficulty·importance))

[0080] Next, initialize the learning path Path = [], when Q is not empty and the length of Path is less than the preset maximum length max path _length when k next =Q.pop(), if k can be learned next , k next Join Path to update user knowledge status, K user =UpdateKnowledgeState(K user ,k next ), for next Related knowledge points related , calculate the new priority priority = CalculatePriority(k related ,K user ,T),(k related ,priority) is added to Q; finally, the generated learning path Path is returned.

[0081] In practical applications, for example, a student studying artificial intelligence wants to master the topic of "deep reinforcement learning". The system will first evaluate the student's mastery of prior knowledge such as "neural networks", "supervised learning", "Markov decision process", etc. Then, based on the student's knowledge status and learning goals, the system may generate a personalized learning path from "policy gradient method" to Q-learning and then to Deep Q-Network.

[0082] S5: Enable interactive knowledge exploration in a virtual reality environment.

[0083] This step aims to provide an immersive learning experience and take full advantage of the advantages of virtual reality technology. First, we use an improved gesture recognition algorithm to achieve natural gesture control. The core steps of the algorithm are as follows: First, obtain the user's hand posture data HandPose; then, apply the spatiotemporal attention mechanism to extract the spatiotemporal features of the hand posture: A = softmax(W2·tanh(W1·F+b1)+b2), where F is the extracted feature, W1, W2, b1, b2 are learnable parameters; use the long short-term memory network (LSTM) to model the temporal features: h t =LSTM(A t FeatureExtractor(HandPose t ),h t-1 ) where h t is the hidden state at time t, FeatureExtractor is the feature extraction function; finally, gesture classification is performed through a multi-layer perceptron: G = softmax (MLP (h final )), where h final is the final hidden state of the LSTM and MLP is a multi-layer perceptron.

[0084] Next, we use the voice control algorithm of multi-scale convolution and attention pooling to process the voice command. The specific steps are as follows: First, obtain the audio signal AudioSignal, and then calculate the spectrum of the audio signal: Spectrogram = STFT (AudioSignal), where STFT is short-time Fourier transform, and apply multi-scale convolution to extract features: F multi =[Conv i (Spectrogram)fori∈[3,5,7]], where Conv i Represents convolution operations of different sizes, followed by the use of attention pooling to aggregate features:

[0085] α=softmax(W a tanh(W f ·F multi )); F pooled =∑(α i ·F i ),

[0086] Where W a and W f is a learnable weight matrix, Transformer network recognition command: C = Softmax (Transformer (F pooled )).

[0087] Finally, we implement dynamic rendering and adaptive detail level adjustment based on user attention data. The specific steps are as follows:

[0088] First, obtain the user attention data A and the virtual object set VirtualObjects; then calculate the importance map of attention guidance: ImportanceMap = GaussianBlur (A), where GaussianBlur is a Gaussian blur operation; for each virtual object obj, calculate the adaptive detail level: LOD (obj) = BaseLOD (obj) · (1 + λ · NormalizedImportance (obj)), where BaseLOD is the basic detail level of the object, λ is the adjustment factor, and NormalizedImportance is the normalized importance score;

[0089] Construct the rendering priority queue:

[0090] RenderQueue=PriorityQueue(VirtualObjects,key=NormalizedImportance)

[0091] In the rendering loop, render according to the time budget and priority: For obj in RenderQueue: If TimeRemaining()>0: RenderWithLOD(obj,LOD(obj)), Otherwise: break. In practical applications, such as learning astronomy, learners can use gestures to rotate and scale the universe model, and use voice commands such as "show the solar system" to quickly locate the area of ​​interest. The system dynamically adjusts the rendering quality according to the learner's gaze point to ensure that the details of the area of ​​interest are clearly visible while maintaining the overall rendering performance.

[0092] S6: Update the interdisciplinary knowledge graph based on user interaction data.

[0093] This step realizes the dynamic optimization of the knowledge graph, enabling it to continuously adapt to new knowledge structures and user needs. First, relevant information is extracted from user interaction data, and new entities and relationships are identified using knowledge extraction algorithms. The specific steps are as follows: First, the user interaction data is preprocessed, including text cleaning and standardization, and a named entity recognition model is used to identify potential new entities: Entities = NER (ProcessedText), where NER is a sequence labeling model based on BERT; a relation extraction model is used to identify potential new relationships between entities: Relations = Relation Extraction (Entities, Processed Text), and a PCNN+ATT model is used for relation extraction; then the extracted entities and relationships are preliminarily filtered:

[0094] FilteredKnowledge = Filter(Entities, Relations, threshold = 0.75), where threshold is the confidence threshold; then, conduct fact verification to ensure the reliability of the new knowledge. The specific steps are:

[0095] First, the extracted new knowledge is compared and verified with the external knowledge base:

[0096] VerifiedKnowledge=ExternalVerification(FilteredKnowledge,ExternalKB)

[0097] Then calculate the confidence score of the new knowledge:

[0098] Confidence=w1·ModelConf+w2·ExternalVerification+w3·UserFeedback

[0099] Among them, w1, w2, and w3 are weight coefficients, ModelConf is the confidence of the model output, ExternalVerification is the external verification result, and UserFeedback is the user feedback; set the confidence threshold to filter new knowledge with high reliability:

[0100] ReliableKnowledge={k∣k∈VerifiedKnowledge and k.Confidence>0.8}

[0101] Finally, an incremental update strategy is adopted to efficiently update the knowledge graph. The specific steps are: Check the consistency of the verified new knowledge with the existing knowledge graph:

[0102] Conflicts=ConsistencyCheck(ReliableKnowledge,ExistingKG)

[0103] If conflicts are found, use a conflict resolution strategy based on confidence and timeliness:

[0104] ResolvedKnowledge=ConflictResolution(Conflicts,ReliableKnowledge,ExistingKG)

[0105] Integrate new knowledge into the existing knowledge graph without conflict:

[0106] UpdatedKG=MergeKnowledge(ExistingKG,ResolvedKnowledge)

[0107] Finally, update the index structure of the knowledge graph:

[0108] UpdatedIndex=UpdateIndex(UpdatedKG)

[0109] In practical applications, for example, when multiple learners are discussing the impact of climate change on the ecosystem, the system may extract new conceptual relationships from their conversations, such as the causal relationship between "ocean acidification" and "coral bleaching". The system will verify this relationship and integrate it into the existing environmental science knowledge graph, thereby enriching learning resources.

[0110] S7: Synchronously update the VR environment to reflect changes in the interdisciplinary knowledge graph.

[0111] This step ensures that the virtual learning environment is consistent with the latest knowledge structure. First, we detect the difference between the knowledge graph and the current virtual environment representation. The specific steps are as follows: First, compare the updated knowledge graph with the knowledge representation in the virtual environment:

[0112] Diff=CompareGraphs(UpdatedKG,VirtualEnvironment.KnowledgeRepresentation)

[0113] Then identify newly added, deleted, and modified nodes and edges:

[0114] AddedNodes,DeletedNodes,ModifiedNodes=ClassifyChanges(Diff)

[0115] AddedEdges,DeletedEdges,ModifiedEdges=ClassifyEdges(Diff)

[0116] Finally, calculate the impact and importance of each change. For change change among all changes:

[0117] change.ImpactRange=CalculateImpactRange(change,UpdatedKG)

[0118] change.Importance=CalculateImportance(change,UpdatedKG)

[0119] Among them, CalculateImpactRange uses a graph traversal algorithm to calculate the number of nodes affected by the change, and CalculateImportance is evaluated based on the centrality and connectivity of the nodes.

[0120] Next, based on the change classification and priority assignment, we develop an update plan. The specific steps are:

[0121] Assign a priority to each change based on its type and importance:

[0122] Priority=w1·Importance+w2·ImpactRange+w3·ChangeType

[0123] Among them, w1, w2, and w3 are weight coefficients, usually set to w1 = 0.4, w2 = 0.3, and w3 = 0.3. The value of ChangeType is: 1.0 for adding, 0.8 for modifying, and 0.6 for deleting. Consider the user's current focus and adjust the priority of the change:

[0124] ifchange.AffectsUserFocus:change.Priority*=1.5

[0125] Finally, the final update sequence is generated:

[0126] UpdateSequence=SortByPriority(AllChanges)

[0127] We then use incremental rendering to update the virtual environment frame by frame. The specific steps are: preload the resources needed for the update:

[0128] PreloadResources(UpdateSequence)

[0129] Build the render queue:

[0130] RenderQueue=BuildRenderQueue(UpdateSequence)

[0131] In each frame, a finite number of update operations are performed:

[0132] for frame in Frames:

[0133] UpdatesForFrame=RenderQueue.GetUpdatesForFrame(frame,TimeRemaining)

[0134] ApplyUpdates(VirtualEnvironment,UpdatesForFrame)

[0135] RenderFrame(VirtualEnvironment)

[0136] Finally, dynamically adjust the update speed to keep the rendering frame rate stable:

[0137] TargetFPS=90

[0138] UpdateBudget=AdjustUpdateBudget(CurrentFPS,TargetFPS)

[0139] Among them, AdjustUpdateBudget uses a PID controller to adjust the update budget to ensure that the frame rate is stable at around 90fps.

[0140] Finally, we implement adaptive adjustment of the user's perspective. The specific steps are:

[0141] Monitor the impact of environmental changes on the user's current perspective:

[0142] ImpactedObjects=GetImpactedObjects(UserViewport,UpdatedObjects)

[0143] Calculate the best perspective transformation path:

[0144] OptimalPath=CalculateViewTransition(CurrentView,ImpactedObjects)

[0145] Bezier curves are used here to generate a smooth camera path.

[0146] Dynamically adjust the perspective conversion speed based on the importance of user attention and environmental changes:

[0147] Speed=BaseSpeed*(1+α*ChangeImportance-β*AttentionLevel)

[0148] Where BaseSpeed ​​is the base conversion speed (usually set to 5 degrees / second), α=0.5, β=0.3, ChangeImportance is the normalized importance of the change, and AttentionLevel is the user's attention level (estimated by eye movement data).

[0149] In practical applications, for example, when an important interdisciplinary concept "application of quantum computing in drug design" is added to the knowledge graph, the system will create a new visual node in the virtual environment and establish connections with the existing "quantum computing" and "drug design" nodes. If the user is currently learning related content, the system will smoothly adjust the perspective to introduce the new content into the user's field of vision while maintaining the coherence of the overall learning experience.

[0150] This dynamic update mechanism ensures that the learning environment always reflects the latest knowledge status, providing learners with a learning experience that keeps pace with the times. It not only presents new research results and interdisciplinary connections in a timely manner, but also intelligently adjusts the content presentation method according to learners' interests and focus points, thereby maximizing learning effects.

[0151] In order to realize the above-mentioned knowledge association learning method based on knowledge graph and virtual reality, the present invention also provides a corresponding system. The system is an intelligent learning platform integrating multiple advanced technologies, aiming to create an immersive and interactive interdisciplinary learning environment for learners.

[0152] The core architecture of the system includes multiple key modules that work together to support the entire knowledge-related learning ecosystem. First, the multi-source data processing module 1 is responsible for preprocessing and preliminary analysis of text, images, videos, and 3D model data from different channels. This module uses advanced natural language processing and computer vision technologies to effectively extract valuable information from multimodal data.

[0153] Next, the knowledge graph construction module 2 uses innovative algorithms to construct an interdisciplinary knowledge graph based on multi-source data. This module can not only identify and extract key concepts and relationships in various disciplines, but also discover implicit knowledge connections between different disciplines through advanced association analysis algorithms. It is worth mentioning that this module adopts a dynamic update mechanism, which can continuously optimize and expand the structure of the knowledge graph based on the latest research results and learning feedback.

[0154] Virtual reality mapping module 3 is another innovation of this system. It is responsible for converting the abstract knowledge graph structure into a concrete virtual three-dimensional environment. This module uses advanced 3D modeling and rendering technology to present complex knowledge structures in an intuitive and vivid way. For example, knowledge nodes of different disciplines may be represented as unique virtual objects, and the associations between knowledge may be visualized as dynamic links connecting these objects.

[0155] The learning path generation module 4 is the intelligent brain of the system. It generates personalized learning paths based on the user's learning history, knowledge background and learning goals, combined with the structure of the knowledge graph, using advanced machine learning algorithms. This module not only considers the logical relationship between knowledge points, but also dynamically adjusts the learning sequence according to the learner's real-time feedback to ensure the efficiency and pertinence of the learning process.

[0156] Interactive Exploration Module 5 provides learners with a variety of interactive methods. It integrates a variety of advanced human-computer interaction technologies such as gesture recognition, voice control, and eye tracking, allowing learners to freely explore knowledge structures in a virtual environment. The design concept of this module is to make the learning process more natural and intuitive, thereby increasing the immersion and participation of learning.

[0157] Collaborative Learning Module 6 supports multiple users to learn and communicate in a virtual environment at the same time. It uses real-time network synchronization technology to ensure that operations and status between different users can be shared instantly. This module provides an ideal platform for cross-regional and cross-disciplinary team collaborative learning, which helps promote the exchange of knowledge and the collision of innovative thinking.

[0158] The knowledge update module 7 is the key to ensuring the continued effectiveness of the system. It can automatically capture the user's learning behavior and feedback, combined with the update of external knowledge sources, to adjust and expand the knowledge graph in real time. The work of this module ensures that the learning content is always up-to-date and relevant, allowing the system to adapt to the rapidly changing knowledge environment.

[0159] The entire system runs on a high-performance distributed computing framework. In order to handle large-scale data and complex computing tasks, the system adopts an architecture that combines cloud computing and edge computing. This design not only ensures the system's responsiveness and stability, but also provides good scalability, which can easily cope with the growth of user scale and the addition of new functions.

[0160] In actual applications, learners enter a virtual learning environment by wearing a VR headset. The system first evaluates the learner's knowledge status and then presents a personalized knowledge structure in the virtual space. Learners can interact with the environment through natural gestures and voice commands, such as "zooming in" on a knowledge node to view detailed information, or "connecting" two seemingly unrelated concepts to explore the relationship between them.

[0161] A typical application scenario of the system is interdisciplinary project learning. For example, in an AI medical project involving biology, computer science, and ethics, learners can intuitively see the knowledge structure of these three disciplines and the intersections between them in a virtual environment. The system will recommend the optimal learning path based on project requirements and learners' backgrounds. During the learning process, learners can view relevant experimental data, code examples, or ethical cases at any time, which are integrated in the virtual environment in an interactive form.

[0162] In general, this system provides an unprecedented intelligent platform for interdisciplinary learning by intelligently integrating cutting-edge technologies such as knowledge graphs, virtual reality, and artificial intelligence. It not only solves the problems of knowledge fragmentation and single learning experience in traditional learning methods, but also provides a powerful tool for cultivating innovative and compound talents. With the continuous advancement of technology and the innovation of educational concepts, such a system is expected to play an increasingly important role in the future education field, promoting changes in educational methods and innovations in talent training models.

[0163] In order to verify the superiority of the "knowledge association learning method and system based on knowledge graph and virtual reality" of the present invention, we designed a set of comparative experiments. The following will introduce the settings, test indicators, detection methods and experimental results of the embodiments and comparative examples in detail.

[0164] Example 1: A complete implementation of the method of the present invention, including multi-source data processing, interdisciplinary knowledge graph construction, virtual reality mapping, personalized learning path generation, interactive knowledge exploration and dynamic update mechanism.

[0165] Comparative Example 1: Traditional knowledge graph-based learning system, without virtual reality environment.

[0166] Comparative Example 2: A learning system based on virtual reality, but without knowledge graph and intelligent recommendation.

[0167] Comparative Example 3: A learning system that simply combines knowledge graphs and virtual reality lacks the core innovations of the present invention, such as cross-modal fusion, personalized recommendations, and dynamic updates.

[0168] We selected the following five key indicators to evaluate the performance of the system:

[0169] 1. Depth of knowledge: Assess learners’ understanding of complex concepts through standardized tests.

[0170] 2. Learning efficiency: records the time required for learners to master specific knowledge.

[0171] 3. Knowledge association ability: assesses learners' ability to establish cross-disciplinary knowledge connections.

[0172] 4. Learning experience: Assess learners’ immersion and satisfaction through questionnaires.

[0173] 5. Knowledge updating speed: measures the time it takes for the system to integrate new knowledge and reflect it in the learning environment.

[0174] During the experiment, we invited 200 college students from different disciplines to participate in a 4-week learning experiment. The participants were randomly divided into 4 groups, each with 50 people, and used the learning system of Example 1 and three comparative examples respectively. Before and after the experiment, all participants took a unified knowledge level test and questionnaire survey.

[0175] The following is a detailed table of experimental results:

[0176] Evaluation Metrics Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 Depth of knowledge understanding 92.5 78.3 75.6 83.7 Learning efficiency 1.2 2.5 2.8 1.8 Knowledge association ability 8.7 6.2 5.8 7.1 Learning Experience 4.8 3.5 4.2 4 Knowledge update speed 5 60 120 30

[0177] It can be seen from the experimental results that Example 1 of the present invention performs best in all evaluation indicators.

[0178] The specific analysis is as follows:

[0179] 1. Depth of knowledge comprehension: Example 1 scored significantly higher than other solutions, reaching 92.5 points. This shows that the present invention greatly improves learners' ability to understand complex concepts by combining the structured representation of knowledge graphs and the intuitive presentation of virtual reality. In particular, the advantages of the present invention are particularly obvious in the visualization of abstract concepts.

[0180] 2. Learning efficiency: Example 1 only takes 1.2 hours on average to master a knowledge point, which is much faster than other solutions. This is due to the personalized learning path generation algorithm and interactive knowledge exploration mechanism of the present invention, which enables learners to locate and absorb key knowledge more quickly.

[0181] 3. Knowledge association ability: Example 1 scored 8.7 points, far ahead of other solutions. This fully demonstrates the excellent performance of the present invention in promoting interdisciplinary knowledge association. Through the combination of interdisciplinary knowledge graph and virtual reality environment, learners can more easily discover the connection between knowledge in different fields.

[0182] 4. Learning experience: Example 1 received a high score of 4.8, reflecting the learners’ high recognition of the system. The immersiveness of the virtual reality environment, the structured presentation of the knowledge graph, and the intelligent learning guidance together create a high-quality learning experience.

[0183] 5. Knowledge update speed: Example 1 can complete the integration and presentation of new knowledge in just 5 minutes. This near-real-time update capability ensures the timeliness and relevance of learning content. In comparison, the update speed of other solutions is significantly lagging behind.

[0184] It is worth noting that although Comparative Example 3 also combines knowledge graph and virtual reality technology, due to the lack of the core innovations of the present invention (such as cross-modal fusion, personalized recommendation and dynamic update mechanism), its performance is still significantly lower than that of Example 1. This further proves the unique value of the present invention in technology integration and innovation.

[0185] In summary, Example 1 is undoubtedly the best implementation scheme. It not only achieves the best results in all indicators, but more importantly, it demonstrates a new knowledge learning paradigm. By intelligently integrating knowledge graphs and virtual reality technology, the present invention creates a dynamic, interactive, and personalized learning environment, which effectively solves the limitations of traditional learning methods in terms of interdisciplinary knowledge association and abstract concept understanding.

[0186] These experimental results fully demonstrate the significant advantages of the present invention in improving learning effects, enhancing knowledge association, optimizing learning experience, etc. It not only provides a new direction for the current development of educational technology, but also provides a valuable reference for the design of future intelligent learning systems. With the further optimization and promotion of technology, the present invention is expected to play an important role in promoting educational innovation and promoting interdisciplinary learning.

[0187] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A knowledge association learning method based on knowledge graph and virtual reality, characterized in that: The following steps are involved: Obtaining multi-source data, wherein the multi-source data includes text data, image data, video data, and three-dimensional model data; Based on the multi-source data, construct an interdisciplinary knowledge graph; Mapping the interdisciplinary knowledge graph into a virtual reality environment; Generate a personalized learning path based on the knowledge representation in the virtual reality environment; implement interactive knowledge exploration in the virtual reality environment; update the interdisciplinary knowledge graph based on user interaction data; and synchronously update the virtual reality environment to reflect changes in the interdisciplinary knowledge graph.

2. The method according to claim 1, characterized in that The steps of constructing an interdisciplinary knowledge graph include: first, processing the text data, including: using an improved two-way maximum matching algorithm for word segmentation, wherein the optimal word segmentation result is determined by calculating the product of the word sequence probability and the text generation probability; using a Transformer-based sequence labeling model for part-of-speech tagging, wherein a softmax function is used to predict labels for Transformer output; using a graph attention network to construct a dependency syntactic tree, wherein the final syntactic structure is obtained by iteratively updating node representations and edge weights; then, processing the image data, normalizing the image data, and performing random data enhancement; using an improved EfficientNet-V2 model to extract image features, wherein an attention mechanism and multi-scale features are introduced. The invention relates to a method for extracting features of a video image by performing feature fusion, calculating the channel attention and spatial attention of the feature map; extracting features at different scales and fusing them through adaptive weights; processing the video data, extracting key frames using an improved scene boundary detection algorithm, wherein optical flow estimation and a deep learning model are combined to calculate the difference between frames; finally, performing the same processing steps as the image data on the key frames; processing the three-dimensional model data, extracting geometric features using an improved PointNet++ model, wherein dynamic point cloud sampling and hierarchical feature aggregation are introduced, and the number of sampling points is adaptively adjusted according to the point cloud density; obtaining the final geometric feature representation through multi-level local feature extraction and global feature aggregation; designing a new texture encoding network to extract texture information, wherein multi-scale convolution and attention mechanisms are used to process texture images.

3. The method according to claim 1, characterized in that The step of constructing an interdisciplinary knowledge graph also includes: first, constructing an interdisciplinary synonym dictionary based on a context-aware synonym recognition algorithm and obtaining dictionaries of various disciplines; calculating the context-aware embedding representation of words, combining the context information output by the BERT model and processed by the graph neural network; setting a dynamic threshold and adjusting the threshold according to the word frequency; identifying synonyms and merging dictionaries based on the context-aware embedding and the dynamic threshold; then, using an improved semantic similarity calculation method to determine the similarity between interdisciplinary entities and calculate the cosine similarity of entity pairs; calculating the path similarity of entity pairs in the knowledge graph, considering the path length and type; calculating the structural similarity of entity pairs, based on the Jaccard similarity of entity neighbor nodes; weighted fusion of the above three similarities to obtain the final entity similarity; finally, using an iterative matching algorithm to achieve interdisciplinary entity alignment and initialize the alignment set; in each iteration, calculating the current threshold, wherein the threshold decays exponentially with the number of iterations; selecting entity pairs to be aligned based on the entity similarity and the current threshold; updating the similarity matrix, considering the influence of aligned entities; repeating the above steps until the maximum number of iterations is reached or the termination condition is met.

4. The method according to claim 1, characterized in that The steps of mapping the interdisciplinary knowledge graph into the virtual reality environment include: first, using a cross-modal attention fusion network to fuse text features, image features and three-dimensional features, and performing modality-specific encoding on each modality feature; calculating a cross-modal attention matrix, wherein a scaled dot product attention mechanism is used; performing information aggregation to weightedly fuse features of different modalities; using a gating mechanism to dynamically adjust the ratio of original features and aggregated features; obtaining the final fused feature representation through a multi-layer perceptron; then, based on a knowledge distillation method, migrating the pre-trained knowledge graph embedding to a student model to obtain the pre-trained knowledge graph embedding and fused features; and obtaining the student model through a multi-layer perceptron network. and the output of the teacher model; calculate the knowledge distillation loss and use the KL divergence to measure the difference between the output of the student model and the teacher model; dynamically adjust the temperature parameter, where the temperature parameter decays exponentially with the training progress; map the knowledge embedding to the virtual space representation through a multi-layer perceptron network; finally, adopt a consistency constraint to ensure the consistency of the virtual object and the knowledge representation, define a projection function, and project the knowledge embedding to the virtual space; calculate the mean square error between the virtual object representation and the projected knowledge embedding as the consistency loss; combine the knowledge distillation loss, the consistency loss and the downstream task loss to obtain the overall loss function; and achieve high-quality mapping of knowledge representation to virtual objects by optimizing the overall loss function.

5. The method according to claim 1, characterized in that The steps of generating a personalized learning path include: first, evaluating the user's current knowledge status, and constructing a user knowledge vector based on the user's historical learning data and test results; using a knowledge graph structure to calculate the integrity and relevance scores of the user's knowledge; setting a knowledge mastery threshold, and marking knowledge points with scores higher than the threshold as mastered; then, based on the user's knowledge status and learning goals, performing a knowledge gap analysis to obtain a learning goal knowledge set; calculating a gap set, including knowledge points that have not been mastered and knowledge points with a mastery level lower than a threshold; ranking the knowledge points in the gap set by importance, taking into account the centrality and relevance of the knowledge points; finally, using an adaptive learning path generation algorithm, taking into account the difficulty and importance of the knowledge points, generating an optimal learning sequence, initializing a priority queue, and adding the knowledge points in the gap set to the queue; selecting the knowledge point with the highest priority from the priority queue; and judging whether the learning is satisfied.

6. The method according to claim 1, characterized in that The steps for implementing interactive knowledge exploration include: first, using an improved gesture recognition algorithm to implement natural gesture control and obtain user hand gesture data; applying a spatiotemporal attention mechanism to extract the spatiotemporal features of hand gestures; using a long short-term memory network to model temporal features; performing gesture classification through a multi-layer perceptron; then, using a multi-scale convolution and attention pooling voice control algorithm to process voice commands and obtain audio signals; calculating a spectrogram of the audio signal; applying multi-scale convolution to extract features, using convolution kernels of different sizes; using an attention pooling mechanism to aggregate features and dynamically adjust the importance of different features; recognizing commands through a Transformer network; finally, based on user attention data, implementing dynamic rendering and adaptive detail level adjustment to obtain user attention data and a set of virtual objects; calculating an attention-guided importance map to map the user's attention distribution to a virtual environment; for each virtual object, calculating an adaptive detail level based on its score in the importance map; constructing a rendering priority queue to sort virtual objects according to the importance score; in a rendering loop, rendering is performed according to a time budget and priority to ensure that key objects are rendered in a timely manner.

7. The method according to claim 1, characterized in that It also includes steps to achieve multi-user collaborative learning: first, based on user ability evaluation and task requirements, dynamically assign collaborative roles to obtain user sets and task requirements; evaluate the ability of each user, taking into account knowledge mastery and historical performance; calculate the matching degree between the user and the role, and use cosine similarity to measure the matching degree between the user's ability and the role requirements; use the Hungarian algorithm to perform optimal role allocation and maximize the overall matching degree; then, use the task decomposition algorithm to divide the collaborative task into subtasks, obtain the collaborative tasks and the assigned roles; construct a task graph based on the dependency and complexity of the task; use the graph partitioning algorithm to divide the task graph into subgraphs, each subgraph corresponding to a subtask; Assign the most suitable role to each subtask, considering the role capabilities and subtask requirements; Finally, an improved state synchronization mechanism is adopted to ensure consistency among multiple users, and each user broadcasts its status regularly; The server collects all user states and performs consistency checks. If a conflict is detected, it uses a conflict resolution strategy based on timestamps and priorities. The resolved state is broadcast to all users to achieve state synchronization.

8. The method according to claim 1, characterized in that The steps of updating the interdisciplinary knowledge graph include: first, extracting relevant information from user interaction data, using a knowledge extraction algorithm to identify new entities and relationships, and preprocessing the user interaction data, including text cleaning and standardization; using a named entity recognition model to identify potential new entities; using a relationship extraction model to identify potential new relationships between entities; preliminarily filtering the extracted entities and relationships to remove low-confidence candidates; then, conducting fact verification to ensure the reliability of the new knowledge, and comparing the extracted new knowledge with an external knowledge base for verification; calculating the confidence score of the new knowledge, considering multiple factors such as the confidence of the extraction model and external verification results; setting a confidence threshold to screen new knowledge with high reliability; finally, adopting an incremental update strategy to efficiently update the knowledge graph, and checking the consistency of the verified new knowledge with the existing knowledge graph; if a conflict is found, using a conflict resolution strategy based on confidence and timeliness; integrating the conflict-free new knowledge into the existing knowledge graph; updating the index structure of the knowledge graph to adapt to the newly added entities and relationships.

9. The method according to claim 1, characterized in that: The steps of synchronously updating the virtual reality environment include: first, detecting the difference between the knowledge graph and the current virtual environment representation, and comparing the updated knowledge graph with the knowledge representation in the virtual environment; identifying the newly added, deleted and modified nodes and edges; calculating the impact scope and importance of each change; then, based on the change classification and priority allocation, formulating an update plan, and assigning priority to each change according to the change type and importance; considering the user's current focus, adjusting the priority of the change; generating a final update sequence to ensure that important changes are processed first; then, using an incremental rendering method to update the virtual environment frame by frame and preload the resources required for the update; building a rendering queue and sorting it according to the priority of the update plan; in each frame, performing a limited number of update operations; dynamically adjusting the update speed to keep the rendering frame rate stable; finally, realizing adaptive adjustment of the user's perspective and monitoring the impact of environmental changes on the user's current perspective; calculating the optimal perspective conversion path to ensure visual continuity; and dynamically adjusting the perspective conversion speed according to the user's attention and the importance of environmental changes.

10. A system for implementing the method according to any one of claims 1 to 9, characterized in that: include: Multi-source data processing module, used to pre-process multi-modal input data; A knowledge graph construction module is used to generate and maintain interdisciplinary knowledge graphs; a virtual reality mapping module is used to map knowledge graphs to three-dimensional virtual environments; Learning path generation module for creating personalized learning plans; Interactive exploration module, used to enable natural interaction between users and virtual environments; Collaborative learning module, used to support multi-user learning; A knowledge update module is used to dynamically update the knowledge graph and virtual environment; wherein the system interacts with users through virtual reality devices and uses distributed computing resources for data processing and rendering optimization.

Images