Course reconstruction method and system based on knowledge graph

Abstract

The invention provides a course reconstruction method and system based on a knowledge graph, and relates to the technical field of intelligent education. The method comprises the following steps: collecting course resource data and student learning behavior data; constructing a knowledge graph fusing multiple courses; generating a personalized learning path based on the atlas and the behavior data; arranging experiment training tasks according to a path and collecting feedback data to update a graph state; and converting the student training result into a technical manuscript in a structured manner and accessing the technical manuscript to a management platform. Closed-loop optimization of teaching resource organization, path recommendation and achievement management is realized.

Description

Technical Field

[0001] The present invention relates to the field of smart education technology, and in particular to a course reconstruction method and system based on knowledge graphs. Background Art

[0002] Current higher education curriculum design relies heavily on faculty experience and lacks data support, making it difficult to precisely tailor individualized learning paths to students. Furthermore, the interconnectedness of course content is poorly structured, hindering students from developing interdisciplinary knowledge. While some educational platforms have introduced intelligent recommendations and online experimentation, these often exist as standalone modules, failing to achieve a closed-loop integration from course content understanding to practical training and ultimately, transformational learning outcomes.

[0003] With the development of artificial intelligence, natural language processing, and graph databases, the field of education is gradually evolving towards intelligence, personalization, and systematization. Knowledge graphs, as an important means of expressing structured knowledge, are increasingly becoming a core tool for reconstructing course structures, analyzing knowledge relationships, and recommending learning paths. Furthermore, the combination of virtual simulation and real equipment is widely used to enhance the immersiveness and practicality of experimental teaching. Intelligent teaching systems are trending towards new forms of multimodal data fusion, cross-platform collaboration, and industry-education integration and collaboration.

[0004] The existing technologies generally have the following problems: (1) The correlation structure between course contents is unclear, the teaching content is fragmented, and it is difficult to establish systematic cognition; (2) The personalized recommendation system lacks deep semantic analysis of learning behavior, and the accuracy of recommended content is insufficient; (3) Virtual simulation is disconnected from real training, and there is a lack of data closed loop; (4) The output of course results lacks automatic conversion and intellectual property coordination mechanisms, which restricts the in-depth development of industry-education integration. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a course reconstruction method and system based on knowledge graph, which realizes the closed-loop optimization of the educational process based on personalized learning and intelligent achievement management driven by knowledge graph.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A course reconstruction method based on knowledge graph, including:

[0008] Collect course resource data and students' learning behavior data;

[0009] Based on natural language processing technology, the course resource data is semantically analyzed to extract course knowledge points and logical relationships, and corresponding node and edge structures are established in the graph database to form a knowledge graph that integrates multiple courses;

[0010] Generate a personalized learning path based on the knowledge graph and student learning behavior data;

[0011] Arrange experimental training tasks according to the learning path; the process of the experimental training tasks includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph according to the feedback data;

[0012] Intelligently transform and manage students' training results.

[0013] Preferably, the course resource data includes teaching texts, experimental materials and professional literature; the learning behavior data includes answering records, knowledge point mastery rate and experimental feedback results.

[0014] Preferably, the course resource data is semantically parsed based on natural language processing technology to extract course knowledge points and logical relationships, and corresponding node and edge structures are established in a graph database to form a multi-course integrated knowledge graph, including:

[0015] Performing text preprocessing on the course resource data to generate a structured text sequence; the text preprocessing includes performing word segmentation, removing stop words, and performing part-of-speech tagging;

[0016] Performing semantic modeling on the structured text sequence based on a pre-trained language model, identifying course knowledge points, and assigning a semantic category label to each course knowledge point;

[0017] Based on contextual semantic dependencies, identifying logical associations between the course knowledge points; the logical associations include prerequisite relationships, inclusion relationships, or equivalence relationships;

[0018] In a graph database, the course knowledge points are used as graph nodes, and the logical association relationships are used as graph edges to establish a graph structure representation;

[0019] The established graph structure is subjected to node fusion and semantic alignment among multiple courses to generate the knowledge graph covering the contents of multiple courses.

[0020] Preferably, semantic modeling is performed on the structured text sequence based on a pre-trained language model to identify course knowledge points, and a semantic category label is assigned to each course knowledge point, including:

[0021] Encoding the structured text sequence into a context embedding vector sequence and inputting the sequence into the pre-trained language model to obtain a contextual semantic representation of each word or phrase;

[0022] Based on the contextual semantic representation, identifying text segments representing the course knowledge points, and taking each of the text segments as a candidate course knowledge point;

[0023] For each candidate course knowledge point, extract the corresponding contextual semantic representation and syntactic dependency features to construct a feature vector;

[0024] Input the feature vector into a multi-classification model and output corresponding semantic category labels; the semantic category labels include definitional concept class, operation skill class, application case class and symbolic expression class;

[0025] Each candidate course knowledge point is bound to the corresponding semantic category label to form a set of course knowledge points with semantic classification attributes.

[0026] Preferably, the established graph structure is subjected to node fusion and semantic alignment between multiple courses to generate the knowledge graph covering the contents of multiple courses, including:

[0027] uniformly encoding all the course knowledge points in the knowledge graphs of multiple courses to generate semantic vector representations of the course knowledge points;

[0028] Based on the similarity between the semantic vectors, identifying pairs of course knowledge points representing the same or similar concepts in multiple courses;

[0029] For each pair of course knowledge points that represent the same concept, merge them into a single course knowledge point node and retain the original course source information;

[0030] For each pair of course knowledge points that represent similar concepts, establish cross-course association edges and mark the similarity relationship type and similarity score;

[0031] The nodes and relationship edges of all fused and aligned course knowledge points are formed into a unified graph structure to form the knowledge graph covering multiple course contents.

[0032] Preferably, generating a personalized learning path based on the course knowledge graph and student learning behavior data includes:

[0033] Structural processing is performed on the student learning behavior data to extract the learning scores, correct answer rates, and experimental completion status corresponding to the course knowledge points, and form a course knowledge point mastery status table;

[0034] Identifying the course knowledge points that the student has mastered and the course knowledge points that the student has not mastered in the knowledge graph, and identifying the reachable course knowledge points to be learned based on the dependency relationships defined in the knowledge graph;

[0035] Based on the student's learning goals, starting from the currently mastered course knowledge points, a graph search-based path planning algorithm is used to calculate the shortest path to the course knowledge points to be learned;

[0036] The teaching resources corresponding to the course knowledge points involved in the shortest path are used as recommended content to construct a personalized learning path including course content, practice tasks and experimental projects.

[0037] Preferably, the shortest path D(v s ,v t ) is calculated as:

[0038]

[0039] in, For v s to v t The set of all feasible paths; (v i ,v j )∈P represents an edge in the path P, connecting the course knowledge point v i With v j ; R ij For the edge (v i ,v j ) represents the semantic closeness of the knowledge association relationship, the larger the value, the stronger the correlation; j For node v j The cognitive complexity index of the corresponding course knowledge point reflects the average time required for learning; σ j For students to target course knowledge points in historical behavior data j The control of volatility; It is the average of the mastery volatility of all the knowledge points in the courses to be learned.

[0040] Preferably, experimental training tasks are arranged according to the learning path; the process of the experimental training tasks includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph according to the feedback data, including:

[0041] Based on the course knowledge points involved in the learning path, matching the experimental training templates corresponding to the course knowledge points to generate an experimental training task set;

[0042] Loading the design scheme of the experimental training task set into the simulation platform, executing the virtual operation process, generating simulation operation results and performance indicators, and determining the feasibility of the design scheme;

[0043] Deploy the design solutions that are determined to be feasible to the physical hardware terminals for operation, and collect the real-time operation data generated by the physical hardware terminals as feedback data;

[0044] Processing and analyzing the feedback data to extract the performance characteristics and execution status of course knowledge points in a real execution environment;

[0045] The extracted performance characteristics and the execution status are mapped to the knowledge graph, and the graph edge weights or node states related to the course knowledge points are updated.

[0046] Preferably, students' training results are intelligently transformed and managed, including:

[0047] Extract the design plans, operation data and process documents generated by students in experimental training tasks to form a student training results data set;

[0048] Automatically analyze the content of the student training results dataset to identify the core technical structure, innovative point description and experimental verification basis;

[0049] Based on the established document template library and semantic generation model, the student training results dataset is structured and converted into technical documents; the technical documents include patent text drafts, course reports or academic paper drafts;

[0050] Conduct semantic classification and quality evaluation on the technical documents and determine the recommended output path;

[0051] Distribute technical documents to a preset achievement management platform; the achievement management platform includes an academic achievement library, an intellectual property declaration system, or a course evaluation archiving system, and records the submission status and version number.

[0052] A course reconstruction system based on knowledge graph, including:

[0053] Data collection unit, used to collect course resource data and students' learning behavior data;

[0054] A knowledge graph construction unit is used to perform semantic analysis on the course resource data based on natural language processing technology, extract course knowledge points and logical associations, and establish corresponding node and edge structures in the graph database to form a knowledge graph that integrates multiple courses;

[0055] A path generation unit, configured to generate a personalized learning path based on the knowledge graph and student learning behavior data;

[0056] An experimental design unit is used to arrange experimental training tasks according to the learning path; the process of the experimental training task includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph based on the feedback data;

[0057] The results management unit is used to intelligently transform and manage students' training results.

[0058] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0059] The present invention constructs a course knowledge graph that integrates the contents of multiple courses, dynamically generates personalized learning paths based on student learning behavior data, and realizes an integrated closed loop of "learning-training-evaluation" through simulation verification and physical hardware training. At the same time, the training results are structured and converted into technical documents and connected to the results management system, realizing intelligent collaboration from knowledge modeling, path guidance, ability training to results output, effectively improving the organizational efficiency of course content, the pertinence of teaching recommendations and the conversion rate of educational resources, and enhancing the systematicness and application ability of students' knowledge mastery. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in 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 paying any creative work.

[0061] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0062] Figure 2 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] The purpose of this invention is to provide a course reconstruction method and system based on knowledge graph, which realizes the closed-loop optimization of the educational process based on personalized learning and intelligent achievement management driven by knowledge graph.

[0065] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0066] Figure 1 A flow chart of the method provided in the embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides a course reconstruction method based on knowledge graph, comprising:

[0067] Step 100: Collect course resource data and student learning behavior data;

[0068] Step 200: Perform semantic analysis on the course resource data based on natural language processing technology, extract course knowledge points and logical relationships, and establish corresponding node and edge structures in the graph database to form a knowledge graph integrating multiple courses;

[0069] Step 300: Generate a personalized learning path based on the knowledge graph and student learning behavior data;

[0070] Step 400: Arrange experimental training tasks according to the learning path. The process of the experimental training tasks includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph based on the feedback data;

[0071] Step 500: Intelligently transform and manage students' training results.

[0072] Preferably, the course resource data includes teaching texts, experimental materials and professional literature; the learning behavior data includes answering records, knowledge point mastery rate and experimental feedback results.

[0073] In the present embodiment, the collection of course resource data adopts the mode of combining the present embodiment of natural language processing pre-processing module with document management to realize.First, the present embodiment accesses the course management platform through the educational resource interface, calls the course catalog and teaching plan, and automatically obtains the teaching text corresponding to the course, including the structured or unstructured documents such as lesson plan, PPT, teaching material chapter, question bank description. Secondly, the present embodiment connects the on-campus experimental platform, calls the experimental tasks and operating procedures configured under each course, extracts the experimental material content, including experimental step description, instrument parameter description and experimental report template. Finally, the present embodiment docks the professional academic resource database (such as CNKI, IEEE, Springer) subscribed by this school, automatically grabs relevant articles and teaching material excerpts according to course keywords (such as "Fourier transform", "feedback control"), extracts the professional literature paragraphs related to the course theme, forms a course resource data set of multi-source and multi-format fusion, as the input basis of knowledge map construction.

[0074] In this embodiment, students' learning behavior data are collected collaboratively through the on-campus intelligent teaching platform and the virtual-reality fusion experiment platform. For the online course learning process, this embodiment records the students' answer records and click behaviors in each class, and compares the answer results with the standard answers to calculate the accuracy of the answers for each course knowledge point. For offline or virtual experimental links, this embodiment records the design plans uploaded by students, the frequency of errors that occur during operation, the number of parameter debugging times, and other behaviors through the simulation platform, and collects the experimental report data submitted by students after the experiment. This embodiment maps the above data to course knowledge points to form a mastery rate table for each knowledge point for each student, and combines the achievement of experimental tasks to construct an experimental feedback result data set. All learning behavior data are indexed according to timestamps and student identities, and serve as dynamic inputs for personalized path recommendations and map updates.

[0075] Preferably, the course resource data is semantically parsed based on natural language processing technology to extract course knowledge points and logical relationships, and corresponding node and edge structures are established in a graph database to form a multi-course integrated knowledge graph, including:

[0076] Performing text preprocessing on the course resource data to generate a structured text sequence; the text preprocessing includes performing word segmentation, removing stop words, and performing part-of-speech tagging;

[0077] Performing semantic modeling on the structured text sequence based on a pre-trained language model, identifying course knowledge points, and assigning a semantic category label to each course knowledge point;

[0078] Based on contextual semantic dependencies, identifying logical associations between the course knowledge points; the logical associations include prerequisite relationships, inclusion relationships, or equivalence relationships;

[0079] In a graph database, the course knowledge points are used as graph nodes, and the logical association relationships are used as graph edges to establish a graph structure representation;

[0080] The established graph structure is subjected to node fusion and semantic alignment among multiple courses to generate the knowledge graph covering the contents of multiple courses.

[0081] In this embodiment, the first step in semantic analysis of course resource data is text preprocessing. The preprocessing objects include course teaching texts, experimental material descriptions, and professional literature paragraphs. The preprocessing process includes: first, performing standardization processing on the original text, unifying the character encoding format and removing non-content elements; then using a word segmentation tool (such as a dictionary-enhanced BERT word segmenter) to perform fine segmentation on the text; then removing stop words and retaining key words with academic significance or teaching functions; finally, performing part-of-speech tagging to record the part-of-speech information of each word in the context for subsequent semantic modeling. After the processing is completed, a structured text sequence is formed, and each element contains the word ontology, part-of-speech category, and context index position information as the input of the pre-trained language model.

[0082] Specifically, in this embodiment, on the basis of obtaining a structured text sequence, a pre-trained language model based on the Transformer architecture (such as BERT or RoBERTa) is used to perform semantic modeling on the input sequence. After the model outputs the context embedding vector of each word, the named entity recognition model is called to identify the continuous text segments representing the course knowledge points to generate a set of course knowledge points. Subsequently, based on syntactic dependency analysis and semantic dependency judgment, the logical association relationship between the course knowledge points is identified. Specifically, if there is an obvious time / learning sequence between two course knowledge points in the semantic path, it is marked as a "prerequisite relationship"; if a course knowledge point contains another knowledge point in the definition or expression, it is marked as an "inclusion relationship"; if two course knowledge points represent the same concept or formula form under different expressions, it is marked as an "equivalence relationship". The above results form a triple data structure, the content of which is "course knowledge point A-relationship type-course knowledge point B".

[0083] In this embodiment, after the extraction of course knowledge points and logical associations is completed, a graph database engine (such as Neo4j) is called to write the course knowledge points as graph nodes and the logical associations as graph edges to generate an initial graph structure. In order to achieve multi-course integration, the graph structures of multiple courses are further semantically aligned and node fused. The specific method includes: first encoding each course knowledge point as a semantic vector representation; then, based on semantic similarity calculation, identifying pairs of course knowledge points in different courses that express the same or highly similar concepts; for course knowledge points that represent the same concept, node merging is performed, and course label information is retained; for course knowledge points that represent similar concepts but have distinguishability, cross-course edges with similar relationships are established between them. Finally, all course graph structures are integrated to generate a unified multi-course fusion knowledge graph for use by the personalized learning path recommendation module.

[0084] Preferably, semantic modeling is performed on the structured text sequence based on a pre-trained language model to identify course knowledge points, and a semantic category label is assigned to each course knowledge point, including:

[0085] Encoding the structured text sequence into a context embedding vector sequence and inputting the sequence into the pre-trained language model to obtain a contextual semantic representation of each word or phrase;

[0086] Based on the contextual semantic representation, identifying text segments representing the course knowledge points, and taking each of the text segments as a candidate course knowledge point;

[0087] For each candidate course knowledge point, extract the corresponding contextual semantic representation and syntactic dependency features to construct a feature vector;

[0088] Input the feature vector into a multi-classification model and output corresponding semantic category labels; the semantic category labels include definitional concept class, operation skill class, application case class and symbolic expression class;

[0089] Each candidate course knowledge point is bound to the corresponding semantic category label to form a set of course knowledge points with semantic classification attributes.

[0090] In this embodiment, the process of encoding a structured text sequence into a contextual embedding vector sequence includes the following steps: first, the structured text sequence is tokenized and the words or phrases are mapped into subword units supported by the pre-trained language model; then the token sequence is input into a pre-trained language model based on the Transformer architecture (such as BERT, RoBERTa or ErlangBERT), and the semantic representation of each word or phrase in its context is obtained under the action of a multi-layer self-attention mechanism. The above semantic representation is a vector of fixed length, which is used to characterize the semantic role and semantic relevance of the word or phrase in a specific context, and provide basic semantic features for the recognition and classification of subsequent course knowledge points.

[0091] In this embodiment, based on the contextual semantic representation, a named entity recognition model is used to identify text fragments representing course knowledge points, and the judgment standard is: whether the fragment has a clear definition or skill description in multiple course resources. Each identified text fragment is used as a candidate course knowledge point. For each candidate course knowledge point, the representation vector of its center position is intercepted from the contextual semantic vector output by the pre-trained language model as the main semantic feature. At the same time, a syntactic dependency analysis tool (such as SpaCy or LTP) is called to construct a dependency syntax tree for the sentence containing the candidate course knowledge point, extract the subject-predicate relationship, attributive-adverbial-complement structure and other dependency edges between the knowledge point and its context words, and encode these structures into sparse vector representations as syntactic dependency relationship features. Finally, the semantic representation and syntactic features are spliced ​​to form a feature vector of the course knowledge point.

[0092] In this embodiment, the feature vector of the course knowledge point is input into the multi-classification model that has been trained. The model adopts a Softmax output layer structure, which can output the probability distribution of each course knowledge point belonging to different semantic categories. The semantic categories include four categories: definitional concept class (such as formulas, theorems, basic terms), operational skill class (such as experimental steps, calculation methods), application case class (such as engineering scenarios, actual tasks) and symbolic expression class (such as mathematical symbols, variable meanings). According to the category corresponding to the maximum probability, each course knowledge point is given a unique semantic category label. Subsequently, the semantic category label is bound to the corresponding course knowledge point to generate a set of course knowledge points with semantic classification attributes for subsequent map construction, learning path recommendation and content visualization.

[0093] Preferably, the established graph structure is subjected to node fusion and semantic alignment between multiple courses to generate the knowledge graph covering the contents of multiple courses, including:

[0094] uniformly encoding all the course knowledge points in the knowledge graphs of multiple courses to generate semantic vector representations of the course knowledge points;

[0095] Based on the similarity between the semantic vectors, identifying pairs of course knowledge points representing the same or similar concepts in multiple courses;

[0096] For each pair of course knowledge points that represent the same concept, merge them into a single course knowledge point node and retain the original course source information;

[0097] For each pair of course knowledge points that represent similar concepts, establish cross-course association edges and mark the similarity relationship type and similarity score;

[0098] The nodes and relationship edges of all fused and aligned course knowledge points are formed into a unified graph structure to form the knowledge graph covering multiple course contents.

[0099] In this embodiment, in order to realize the fusion of knowledge points between multiple courses, all course knowledge points in multiple course knowledge graphs are first uniformly encoded. The specific method is: extract the context paragraph corresponding to each course knowledge point in the original course resource data, input it into the same pre-trained language model as the graph construction stage, and perform semantic modeling on each course knowledge point. By obtaining the embedding vector of the knowledge point in the context and combining it with the average semantic vector of the adjacent nodes connected to it in the course graph, a semantic context enhanced representation is formed. The above representation is normalized and used as the semantic vector of the course knowledge point to calculate the semantic similarity between courses.

[0100] In this embodiment, all course knowledge points are compared in pairs based on the cosine similarity calculation results between semantic vectors. When the similarity between any pair of course knowledge points exceeds the preset threshold of 0.95 and the semantic category labels are consistent, they are determined to be course knowledge points representing the same concept; when the similarity is between 0.80 and 0.95, and the semantic categories are compatible (such as operation skills class and application case class), they are determined to be course knowledge points representing similar concepts. For course knowledge points representing the same concept, a node fusion operation is performed, a merge node is retained, and the original course identifier and merge history are recorded in the node. For course knowledge points representing similar concepts, a cross-course association edge is established between them, and the "synonymous relationship" type and similarity score are marked on the edge for subsequent recommendation and visualization.

[0101] In this embodiment, all the fused course knowledge point nodes and newly added cross-course relationship edges are uniformly imported into the graph database to construct a fused knowledge graph covering multiple course contents. During the import process, the merged nodes are uniquely numbered and managed, and the number of original knowledge points covered, the list of source course names, and the semantic categories are recorded in the node attributes; a special "course fusion" relationship type is set for the newly added relationship edges to distinguish them from the edges within the original course graph. The final generated fused knowledge graph has a unified namespace, a deduplicated knowledge point structure, and explicit cross-course semantic associations, and can serve as the core supporting data structure to support personalized path recommendations and cross-course knowledge transfer analysis.

[0102] Preferably, generating a personalized learning path based on the course knowledge graph and student learning behavior data includes:

[0103] Structural processing is performed on the student learning behavior data to extract the learning scores, correct answer rates, and experimental completion status corresponding to the course knowledge points, and form a course knowledge point mastery status table;

[0104] Identifying the course knowledge points that the student has mastered and the course knowledge points that the student has not mastered in the knowledge graph, and identifying the reachable course knowledge points to be learned based on the dependency relationships defined in the knowledge graph;

[0105] Based on the student's learning goals, starting from the currently mastered course knowledge points, a graph search-based path planning algorithm is used to calculate the shortest path to the course knowledge points to be learned;

[0106] The teaching resources corresponding to the course knowledge points involved in the shortest path are used as recommended content to construct a personalized learning path including course content, practice tasks and experimental projects.

[0107] In this embodiment, the process of structuring the student learning behavior data includes: extracting the student's answer records, grade, learning time and experimental results under each course module from the online teaching platform, and reorganizing the above data according to the knowledge point dimension based on the mapping relationship between the course knowledge points and the teaching content. Specifically, for each course knowledge point, the accuracy of the corresponding questions, the frequency of answers, and the completion score of the relevant experiments are calculated to construct a "course knowledge point-student behavior indicator" matrix. Based on this matrix, a normalized weighted model is applied to generate a mastery score for each course knowledge point, and a mastery threshold is set. Course knowledge points with scores higher than the threshold are marked as "mastered", and those with scores lower than the threshold are marked as "not mastered". Finally, a course knowledge point mastery status table is formed as the behavioral input basis for path recommendation.

[0108] In this embodiment, based on the course knowledge graph and the course knowledge point mastery status table, all "mastered" course knowledge point nodes are marked in the graph. Subsequently, the downstream adjacent nodes starting from these nodes are identified from the graph structure, and it is determined whether all their predecessor nodes have been mastered. If the condition is met, the node is marked as a "reachable course knowledge point to be learned". Then, based on the learning goals set by the students (such as mastering a chapter or solving a certain application problem), the target node set is determined. From the currently mastered node to the target node, a graph search-based shortest path algorithm (such as a weighted Dijkstra algorithm) is executed on the knowledge graph, and the path cost is calculated in combination with indicators such as semantic relevance, learning complexity, and mastery stability, and the shortest path set that meets the reachability constraint is finally output.

[0109] In this embodiment, the course knowledge points involved in the shortest path are used as a set of path nodes, and the teaching content, exercise sets and supporting experimental projects corresponding to each course knowledge point are retrieved from the teaching resource database to construct a resource recommendation list. The teaching content includes video explanations, text handouts and interactive courseware, the exercise sets include basic exercises and extension tasks, and the experimental projects include virtual simulation or physical operation training modules. According to the order of knowledge points in the path, the order of resource presentation is arranged according to the dependency relationship to form a personalized learning path plan with complete structure and complete resources. This plan can be completed sequentially by students on the learning terminal, and can also be used by teachers for differentiated teaching arrangements.

[0110] Preferably, the shortest path D(v s ,v t ) is calculated as:

[0111]

[0112] in, For v s to v t The set of all feasible paths; (v i ,v j )∈P represents an edge in the path P, connecting the course knowledge point v i With v j ; R ij For the edge (v i ,v j ) represents the semantic closeness of the knowledge association relationship, the larger the value, the stronger the correlation; j For node v j The cognitive complexity index of the corresponding course knowledge point reflects the average time required for learning; σ j For students to target course knowledge points in historical behavior data j The control of volatility; It is the average of the mastery volatility of all the knowledge points in the courses to be learned.

[0113] In this embodiment, the calculation of the shortest path is based on a comprehensive modeling of multiple factors such as the semantic closeness between course knowledge points in the knowledge graph, the cognitive complexity of the course knowledge points themselves, and the volatility of students' mastery of related knowledge points. Among them, the semantic closeness between course knowledge points is obtained by constructing the semantic vectors of the course knowledge point pairs and calculating the cosine similarity. The higher the value, the stronger the concept association. The cognitive complexity index of the course knowledge points is assessed by course design experts, and a standardized score is given based on factors such as the hierarchical position of the knowledge points in the teaching syllabus, the average learning time, and the general difficulty of students in mastering them. The volatility of students' mastery of course knowledge points is obtained by calculating the standard deviation of the scores of students on questions related to the knowledge points in previous tests, and then calculating the average value of the mastery volatility of all course knowledge points to be learned, which is used to evaluate the overall learning stability of students. During path planning, the algorithm dynamically calculates the total cost of the path from the mastered course knowledge points to the target knowledge points based on the above parameters, and selects the path with the lowest cost as the optimal learning path.

[0114] Preferably, arranging experimental training tasks according to the learning path includes:

[0115] Based on the course knowledge points involved in the learning path, matching the experimental training templates corresponding to the course knowledge points to generate an experimental training task set;

[0116] Loading the design scheme of the experimental training task set into the simulation platform, executing the virtual operation process, generating simulation operation results and performance indicators, and determining the feasibility of the design scheme;

[0117] Deploy the design solutions that are determined to be feasible to the physical hardware terminals for operation, and collect the real-time operation data generated by the physical hardware terminals as feedback data;

[0118] Processing and analyzing the feedback data to extract the performance characteristics and execution status of course knowledge points in a real execution environment;

[0119] The extracted performance characteristics and the execution status are mapped to the knowledge graph, and the graph edge weights or node states related to the course knowledge points are updated.

[0120] In this embodiment, based on the course knowledge points involved in the personalized learning path, experimental templates corresponding to each course knowledge point are retrieved from the preset experimental training template library, and the templates include input parameter descriptions, experimental goal definitions, design requirements and assessment indicators. The retrieved experimental templates are integrated to generate an experimental training task set, and the course parameters in the template are automatically filled in to form a set of design solutions to be run. The design solution set is loaded into the simulation platform. The simulation platform uses a virtual experimental environment based on a digital twin model to simulate the execution process of each solution in a typical working scenario, output voltage response, delay performance, resource consumption and other performance indicators, and compare the indicators with the preset standards. If there is no logical conflict in the operation process of the solution and the performance indicators meet the basic requirements, the solution is marked as feasible and enters the entity operation stage.

[0121] In the present embodiment, the feasible design scheme of simulation verification is converted into the executable code that meets the target hardware platform instruction format, and is deployed to the entity hardware terminal operation via Bluetooth, Wi-Fi or USB interface. Entity hardware terminal includes a microcontroller (such as STM32 or Arduino) for controlling execution, a peripheral module (such as ADC / DAC) and a sensor acquisition unit (such as temperature and humidity, voltage, current sensor) for signal processing. After the design scheme starts to be executed, all kinds of feedback data generated during the operation are collected in real time, including the indicators such as operation time, response time, current fluctuation, temperature rise. All feedback data are timestamp and course knowledge point label, and are cached in the local data buffer, waiting for upload analysis and processing.

[0122] In this embodiment, the feedback data collected by the physical hardware terminal is uploaded to the data processing module, and filtering, noise reduction and standardization processing are performed, and the performance feature vectors of each course knowledge point in the real execution environment are extracted, such as average response speed, energy efficiency ratio and stability score. At the same time, the execution status of the course knowledge point is judged in combination with the training log, including whether the first run is successful, whether it has undergone a debugging process, whether the target indicators are achieved, etc. The corresponding course knowledge point node is searched in the course knowledge graph, and the above performance characteristics and execution status are mapped to the node. The weight value of the relevant edge is adjusted according to the performance result (such as reducing the dependence on a certain previous knowledge point) or the mastery status score of the node is updated (such as improving the mastery confidence), so that the graph structure can dynamically reflect the student's real learning performance and enhance the adaptability of the graph in subsequent path recommendations.

[0123] Preferably, students' training results are intelligently transformed and managed, including:

[0124] Extract the design plans, operation data and process documents generated by students in experimental training tasks to form a student training results dataset;

[0125] Automatically analyze the content of the student training results dataset to identify the core technical structure, innovative point description and experimental verification basis;

[0126] Based on the established document template library and semantic generation model, the student training results dataset is structured and converted into technical documents; the technical documents include patent text drafts, course reports or academic paper drafts;

[0127] Conduct semantic classification and quality evaluation on the technical documents and determine the recommended output path;

[0128] Distribute technical documents to a preset achievement management platform; the achievement management platform includes an academic achievement library, an intellectual property declaration system, or a course evaluation archiving system, and records the submission status and version number.

[0129] In this embodiment, after students complete the experimental training task, the process outputs related to the task are automatically collected, including design files (such as circuit diagrams, control logic codes), operation data (such as sensor records, output waveforms, error logs), and process documents written by students (such as experimental records, debugging notes, and experimental conclusion paragraphs). The above data are archived with course knowledge points as indexes, integrated into a unified data structure, and a student training results data set is constructed. Each data set entry contains a course knowledge point identifier, data type label, timestamp, generation method, and format standard, which serves as the basic input for subsequent content parsing and semantic modeling.

[0130] In this embodiment, a text parsing model based on deep semantic extraction is used to parse the content of the student training results data set. The parsing process includes three contents: first, identifying the technical structure, such as hardware block diagram, algorithm module division, input and output logic flow; second, identifying the description of the innovation point, and judging whether there are extension modules or optimization strategies that are not in the textbook template in the design plan; third, extracting the experimental verification basis, including successful operation screenshots, performance curve data or experimental effect descriptions. Subsequently, the semantic generation model (such as T5, BART) and the document template engine are called to perform semantic filling and template mapping on the parsed structure and content, and automatically generate technical manuscripts. The manuscript types include patent text drafts, course reports or academic paper drafts, and the output is in standardized DOCX or LaTeX format.

[0131] In this embodiment, after the technical manuscript is generated, a text classifier is called to perform semantic classification on the manuscript to identify its adapted achievement type label, such as "experimental report type", "course evaluation type" or "patentable conversion type". The quality assessment model is then used to perform multi-dimensional scoring on the manuscript content. The scoring indicators include structural integrity, language clarity and innovation contribution, and the recommended output path is determined based on the scoring results. Finally, the technical manuscript is uploaded to the corresponding achievement management platform, including: an academic achievement library (such as a school-level learning archive platform), an intellectual property declaration system (such as a patent service platform connected to the Academic Affairs Office), or a course evaluation archiving system (for comprehensive evaluation of teachers). The platform submission record includes the achievement number, submitter information, version number and archiving timestamp to ensure that the achievement is included in the manageable education resource pool under intelligent evaluation.

[0132] Corresponding to the above method, such as Figure 2 As shown, this embodiment discloses a course reconstruction system based on knowledge graph, including:

[0133] Data collection unit, used to collect course resource data and students' learning behavior data;

[0134] A knowledge graph construction unit is used to perform semantic analysis on the course resource data based on natural language processing technology, extract course knowledge points and logical associations, and establish corresponding node and edge structures in the graph database to form a knowledge graph that integrates multiple courses;

[0135] A path generation unit, configured to generate a personalized learning path based on the knowledge graph and student learning behavior data;

[0136] An experimental design unit is used to arrange experimental training tasks according to the learning path; the process of the experimental training task includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph based on the feedback data;

[0137] The results management unit is used to intelligently transform and manage students' training results.

[0138] The beneficial effects of the present invention are as follows:

[0139] This invention constructs a knowledge graph that integrates multiple courses based on natural language processing technology, transforms the originally scattered teaching texts, experimental materials and literature content into a structured course knowledge point network, establishes clear knowledge dependency relationships and conceptual connections, effectively solves the problems of fragmentation of existing teaching content and separation between courses, and improves the organizational efficiency of teaching resources and the logical coherence across courses.

[0140] The present invention combines students' answer records, mastery status and experimental feedback to dynamically model the mastery status in the course knowledge graph, and generates the shortest learning path based on the graph search algorithm, realizing personalized learning task recommendations based on students' actual abilities, avoiding the "one-size-fits-all" teaching schedule arrangement, and improving the pertinence and efficiency of learning.

[0141] This method arranges experimental training tasks based on learning paths. By combining virtual verification on a simulation platform with physical hardware terminal operation, it forms a closed-loop "design-verification-operation-feedback" process, ensuring that students not only master the knowledge but also acquire engineering operational capabilities. Feedback data can be used to update the knowledge graph in real time, enabling the dynamic accumulation and cyclical optimization of teaching data.

[0142] Based on the experimental training results, the present invention automatically extracts technical structures and innovative elements, and combines semantic generation models with document template libraries to transform student achievements into standardized patent drafts, course reports or academic paper manuscripts, and connects them to intellectual property platforms or achievement archiving systems, thereby improving the ability to integrate industry and education and promoting the knowledge-based, structured and sustainable use of teaching achievements.

[0143] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0144] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A course reconstruction method based on knowledge graph, characterized by: include: Collect course resource data and students' learning behavior data; Based on natural language processing technology, the course resource data is semantically analyzed to extract course knowledge points and logical relationships, and corresponding node and edge structures are established in the graph database to form a knowledge graph that integrates multiple courses; Generate a personalized learning path based on the knowledge graph and student learning behavior data; Arrange experimental training tasks according to the learning path; The process of the experimental training task includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph according to the feedback data; Intelligently transform and manage students' training results.

2. The course reconstruction method based on knowledge graph according to claim 1 is characterized in that: The course resource data includes teaching texts, experimental materials and professional literature; the learning behavior data includes answer records, knowledge point mastery rate and experimental feedback results.

3. The course reconstruction method based on knowledge graph according to claim 1 is characterized in that: Based on natural language processing technology, the course resource data is semantically parsed to extract course knowledge points and logical relationships, and corresponding node and edge structures are established in the graph database to form a multi-course integrated knowledge graph, including: Performing text preprocessing on the course resource data to generate a structured text sequence; the text preprocessing includes performing word segmentation, removing stop words, and performing part-of-speech tagging; Performing semantic modeling on the structured text sequence based on a pre-trained language model, identifying course knowledge points, and assigning a semantic category label to each course knowledge point; Based on contextual semantic dependencies, identifying logical associations between the course knowledge points; the logical associations include prerequisite relationships, inclusion relationships, or equivalence relationships; In a graph database, the course knowledge points are used as graph nodes, and the logical association relationships are used as graph edges to establish a graph structure representation; The established graph structure is subjected to node fusion and semantic alignment among multiple courses to generate the knowledge graph covering the contents of multiple courses.

4. The course reconstruction method based on knowledge graph according to claim 3 is characterized in that: The structured text sequence is semantically modeled based on the pre-trained language model to identify course knowledge points and assign semantic category labels to each course knowledge point, including: Encoding the structured text sequence into a context embedding vector sequence and inputting the sequence into the pre-trained language model to obtain a contextual semantic representation of each word or phrase; Based on the contextual semantic representation, identifying text segments representing the course knowledge points, and taking each of the text segments as a candidate course knowledge point; For each candidate course knowledge point, extract the corresponding contextual semantic representation and syntactic dependency features to construct a feature vector; Input the feature vector into a multi-classification model and output corresponding semantic category labels; the semantic category labels include definitional concept class, operation skill class, application case class and symbolic expression class; Each candidate course knowledge point is bound to the corresponding semantic category label to form a set of course knowledge points with semantic classification attributes.

5. The course reconstruction method based on knowledge graph according to claim 3 is characterized in that: The established graph structure is subjected to node fusion and semantic alignment between multiple courses to generate the knowledge graph covering the contents of multiple courses, including: uniformly encoding all the course knowledge points in the knowledge graphs of multiple courses to generate semantic vector representations of the course knowledge points; Based on the similarity between the semantic vectors, identifying pairs of course knowledge points representing the same or similar concepts in multiple courses; For each pair of course knowledge points that represent the same concept, merge them into a single course knowledge point node and retain the original course source information; For each pair of course knowledge points that represent similar concepts, establish cross-course association edges and mark the similarity relationship type and similarity score; The nodes and relationship edges of all fused and aligned course knowledge points are formed into a unified graph structure to form the knowledge graph covering multiple course contents.

6. The course reconstruction method based on knowledge graph according to claim 1 is characterized in that: Generate a personalized learning path based on the course knowledge graph and student learning behavior data, including: Structural processing is performed on the student learning behavior data to extract the learning scores, correct answer rates, and experimental completion status corresponding to the course knowledge points, and form a course knowledge point mastery status table; Identifying the course knowledge points that the student has mastered and the course knowledge points that the student has not mastered in the knowledge graph, and identifying the reachable course knowledge points to be learned based on the dependency relationships defined in the knowledge graph; Based on the student's learning goals, starting from the currently mastered course knowledge points, a graph search-based path planning algorithm is used to calculate the shortest path to the course knowledge points to be learned; The teaching resources corresponding to the course knowledge points involved in the shortest path are used as recommended content to construct a personalized learning path including course content, practice tasks and experimental projects.

7. The course reconstruction method based on knowledge graph according to claim 6 is characterized in that: The shortest path D(v s ,v t ) is calculated as: in, For v s to v t The set of all feasible paths; (v i ,v j )∈P represents an edge in the path P, connecting the course knowledge point v i With v j ; R ij For the edge (v i ,v j ) represents the semantic closeness of the knowledge association relationship, the larger the value, the stronger the correlation; j For node v j The cognitive complexity index of the corresponding course knowledge point reflects the average time required for learning; σ j For students to target course knowledge points in historical behavior data j The control of volatility; It is the average of the mastery volatility of all the knowledge points in the courses to be learned.

8. The course reconstruction method based on knowledge graph according to claim 1 is characterized in that: Arrange experimental training tasks according to the learning path; The process of the experimental training task includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph according to the feedback data, including: Based on the course knowledge points involved in the learning path, matching the experimental training templates corresponding to the course knowledge points to generate an experimental training task set; Loading the design scheme of the experimental training task set into the simulation platform, executing the virtual operation process, generating simulation operation results and performance indicators, and determining the feasibility of the design scheme; Deploy the design solutions that are determined to be feasible to the physical hardware terminals for operation, and collect the real-time operation data generated by the physical hardware terminals as feedback data; Processing and analyzing the feedback data to extract the performance characteristics and execution status of course knowledge points in a real execution environment; The extracted performance characteristics and the execution status are mapped to the knowledge graph, and the graph edge weights or node states related to the course knowledge points are updated.

9. The course reconstruction method based on knowledge graph according to claim 1 is characterized in that: Intelligently transform and manage students' training results, including: Extract the design plans, operation data and process documents generated by students in experimental training tasks to form a student training results data set; Automatically analyze the content of the student training results dataset to identify the core technical structure, innovative point description and experimental verification basis; Based on the established document template library and semantic generation model, the student training results dataset is structured and converted into technical documents; the technical documents include patent text drafts, course reports or academic paper drafts; Conduct semantic classification and quality evaluation on the technical documents and determine the recommended output path; Distribute technical documents to a preset achievement management platform; the achievement management platform includes an academic achievement library, an intellectual property declaration system, or a course evaluation archiving system, and records the submission status and version number.

10. A course reconstruction system based on knowledge graph, characterized by: include: Data collection unit, used to collect course resource data and students' learning behavior data; A knowledge graph construction unit is used to perform semantic analysis on the course resource data based on natural language processing technology, extract course knowledge points and logical associations, and establish corresponding node and edge structures in the graph database to form a knowledge graph that integrates multiple courses; A path generation unit, configured to generate a personalized learning path based on the knowledge graph and student learning behavior data; an experimental design unit, configured to arrange experimental training tasks according to the learning path; The process of the experimental training task includes: executing the design plan corresponding to the course knowledge point on the simulation platform for virtual operation; deploying the design plan that has passed the simulation verification to the physical hardware terminal for physical operation; collecting feedback data during the physical operation process, and updating the edge weight or node status in the knowledge graph according to the feedback data; The results management unit is used to intelligently transform and manage students' training results.

Images