Teaching content integration method and system based on knowledge graph, and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a teaching content integration method and system based on a knowledge graph, and a storage medium. The method comprises the following steps: performing cognitive difficulty labeling on multi-source teaching resources, and constructing a concept ontology library; fusing the semantic similarity and the cognitive span through a pre-dependency matrix algorithm to generate a pre-dependency graph; carrying out six-layer cognitive layer coding to obtain a multi-layer semantic matrix; the cognitive competence of the learner and the semantic matrix are dynamically planned and recombined, and a self-adaptive content sequence is generated; and constructing a timestamp fusion space, and obtaining a dynamic personalized learning path through a relationship attenuation model evolution graph. According to the method, the problems that in an existing teaching content integration method based on the knowledge graph, cognitive load quantification is missing, concept dependency relation mining is inaccurate, static knowledge representation is limited, and dynamic updating is lagged are solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, system and storage medium for integrating teaching content based on knowledge graphs. Background Technology

[0002] Existing methods for integrating teaching content primarily employ rule-based knowledge organization and static content recommendation strategies. Traditional methods organize teaching content through predefined syllabi and fixed relationships between knowledge points, utilize simple keyword matching or category tags to organize teaching resources, and recommend content based on learners' historical grades or basic information. These methods can, to some extent, provide basic classification and management of teaching resources, offering preliminary technical support for teaching activities.

[0003] However, the lack of an effective cognitive load quantification mechanism makes it impossible to accurately assess the cognitive pressure learners need to bear when learning specific concepts, leading to unreasonable arrangement of teaching content; secondly, the concept dependency mining is not precise enough, relying only on simple co-occurrence statistics or pre-set rules, which cannot identify the real pre-dependent relationships and cognitive spans between concepts; thirdly, the use of a single-level knowledge representation method cannot reflect the semantic differences and complexities of teaching concepts at different cognitive levels; and finally, the lack of a dynamic update mechanism means that once the knowledge graph is built, it remains static and cannot reflect changes in teaching content and feedback on learning outcomes in a timely manner.

[0004] Based on the above analysis, existing technologies lack a comprehensive technical solution that can simultaneously address cognitive load quantification, accurate dependency mining, multi-level semantic representation, and dynamic evolution. Because cognitive load cannot be accurately quantified, personalized content reorganization based on learners' cognitive abilities is impossible; inaccurate concept dependency mining further affects the quality of constructing multi-level semantic representations; the lack of multi-level semantic representation makes dynamic programming reorganization based on cognitive load lack an effective data foundation; and ultimately, the lack of a dynamic update mechanism prevents the entire teaching content integration system from adapting to changes in the timeliness of knowledge and the evolution of learners' personalized needs. Summary of the Invention

[0005] This application provides a knowledge graph-based method, system, and storage medium for integrating teaching content, which addresses the problems of insufficient cognitive load quantification, inaccurate concept dependency mining, limitations of static knowledge representation, and lagging dynamic updates in existing knowledge graph-based teaching content integration methods, thereby improving the personalization and adaptability of teaching content integration.

[0006] Firstly, this application provides a method for integrating teaching content based on knowledge graphs, the method comprising: The cognitive difficulty coefficient of multi-source teaching resources is labeled to obtain a teaching concept ontology containing cognitive attribute vectors; The semantic similarity and cognitive span of the teaching concept ontology are fused and calculated using the pre-dependency matrix algorithm to obtain a concept pre-dependency graph. Based on the aforementioned concept pre-dependency graph, the teaching concept is decomposed and encoded into a six-level cognitive hierarchy to obtain a multi-cognitive hierarchical semantic representation matrix. The learner's cognitive ability vector and the multi-cognitive hierarchical semantic representation matrix are recombined under cognitive load constraints through dynamic programming to obtain a cognitive load adaptive content sequence. Based on the cognitive load adaptive content sequence, an incremental update fusion space with timestamp markers is constructed. The knowledge graph is then processed through a relation strength decay model to obtain a dynamic personalized learning path.

[0007] Secondly, this application provides a knowledge graph-based teaching content integration system, which includes: The annotation module is used to annotate the cognitive difficulty coefficients of multi-source teaching resources to obtain a teaching concept ontology containing cognitive attribute vectors. The calculation module is used to perform semantic similarity and cognitive span fusion calculation on the teaching concept ontology library through the pre-dependency matrix algorithm to obtain the concept pre-dependency graph; The encoding module is used to perform six-level cognitive hierarchy decomposition and encoding processing on the teaching concept based on the concept pre-dependency graph to obtain a multi-cognitive hierarchy semantic representation matrix; The reorganization module is used to perform dynamic programming reorganization processing on the learner's cognitive ability vector and the multi-cognitive hierarchical semantic representation matrix under cognitive load constraints to obtain a cognitive load adaptive content sequence. The update module is used to construct an incremental update fusion space with timestamps based on the adaptive content sequence of the cognitive load, and to perform knowledge graph evolution processing through the relation strength decay model to obtain a dynamic personalized learning path.

[0008] Thirdly, a knowledge graph-based teaching content integration device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the knowledge graph-based teaching content integration device to execute the aforementioned knowledge graph-based teaching content integration method.

[0009] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned knowledge graph-based teaching content integration method.

[0010] The technical solution provided in this application achieves accurate fusion calculation of semantic similarity and cognitive span through a pre-dependency matrix algorithm. This algorithm can accurately identify the real dependencies between teaching concepts, overcoming the limitations of existing technologies that rely solely on simple co-occurrence statistics, and providing a reliable relational foundation for subsequent cognitive hierarchy decomposition. The six-layer cognitive hierarchy decomposition encoding process, based on Bloom's hierarchy of cognitive theories, semantically represents teaching concepts at six levels: memory, understanding, application, analysis, evaluation, and creation. This breaks through the limitations of traditional single-level knowledge representation, enabling the teaching content to be finely organized according to different cognitive complexities. The dynamic programming reorganization process under cognitive load constraints achieves content sequence optimization based on individual cognitive characteristics by matching learners' cognitive ability vectors with multi-cognitive-level semantic representation matrices, solving the key problem of existing technologies' inability to quantify cognitive load. The incremental update fusion space with timestamp marking, combined with a relation strength decay model, establishes a dynamic evolution mechanism for the knowledge graph, enabling the integration of teaching content to adapt to changes in the timeliness of knowledge and dynamic adjustments in learning needs.

[0011] The three-dimensional fusion computational features of the pre-dependency matrix algorithm significantly improve the accuracy of concept dependency mining, enabling the identification of implicit dependencies that traditional methods cannot discover, thus laying an accurate knowledge structure foundation for personalized learning path construction. The construction features of the multi-cognitive-level semantic representation matrix allow the same teaching concept to present differentiated semantic features at different cognitive levels, supporting more refined cognitive ability matching and load assessment. This multi-level representation capability is unattainable by traditional planar knowledge organization methods. The optimization features of the dynamic programming recombination algorithm under cognitive load constraints ensure that the generated learning sequence satisfies both the logical dependencies between concepts and the learner's cognitive capacity, achieving truly personalized teaching content arrangement. The timeliness adjustment features of the relation strength decay model enable the knowledge graph to reflect the dynamic changes in teaching knowledge, ensuring the timeliness and accuracy of teaching content integration. This dynamic evolution capability fills the technical gap in timeliness of existing static knowledge graphs, providing key technical support for the continuous optimization of intelligent education systems. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of one embodiment of the knowledge graph-based teaching content integration method in this application. Figure 2 This is a schematic diagram of one embodiment of the knowledge graph-based teaching content integration system in this application. Figure 3 This is a schematic block diagram of the teaching content integration device based on knowledge graphs in an embodiment of the present invention. Detailed Implementation

[0014] This application provides a method, system, and storage medium for integrating teaching content based on a knowledge graph. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the knowledge graph-based teaching content integration method in this application includes: Step S101: Perform cognitive difficulty coefficient annotation on the multi-source teaching resources to obtain a teaching concept ontology library containing cognitive attribute vectors; Step S102: Perform semantic similarity and cognitive span fusion calculation on the teaching concept ontology using the pre-dependency matrix algorithm to obtain the concept pre-dependency graph; Step S103: Based on the concept pre-dependency graph, the teaching concept is decomposed and encoded into a six-level cognitive hierarchy to obtain a multi-cognitive hierarchical semantic representation matrix; Step S104: Perform dynamic programming reorganization on the learner's cognitive ability vector and the multi-cognitive level semantic representation matrix under cognitive load constraints to obtain a cognitive load adaptive content sequence; Step S105: Construct an incremental update fusion space with timestamps based on the cognitive load adaptive content sequence, and perform knowledge graph evolution processing through a relation strength decay model to obtain a dynamic personalized learning path.

[0016] It is understood that the executing entity of this application can be a knowledge graph-based teaching content integration system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0017] Specifically, the teaching text, video subtitles, and exercise bank are segmented into independent text segments based on knowledge point boundaries, with each segment corresponding to a main teaching concept. Next, keywords are extracted from the original text segments using TF-IDF weight calculation. The TF-IDF algorithm identifies important words in the text by calculating the product of word frequency and inverse document frequency, thus obtaining a list of core teaching concept identifiers. These concept identifiers are then input into a cognitive difficulty assessment model for quantification. This model outputs a difficulty coefficient and a learning time coefficient based on the concept's abstractness, prerequisite knowledge requirements, and learning complexity. Finally, a three-dimensional cognitive attribute vector containing the difficulty coefficient, learning time coefficient, and the number of prerequisite concepts is constructed based on these parameters, and then uniformly identified using a concept identifier mapping table.

[0018] Semantic similarity and cognitive span are fused using a pre-dependency matrix algorithm. The pre-dependency matrix is ​​an n×n square matrix, where n represents the total number of teaching concepts, and the matrix elements represent the pre-dependency strength between concepts. First, the teaching concepts are input into a Word2Vec model for vectorization. The Word2Vec model, trained via a neural network, maps words into high-dimensional dense vectors, capturing the semantic relationships between concepts. Then, a pre-dependency matrix is ​​constructed based on the set of concept semantic vectors. An initial pre-dependency strength matrix is ​​obtained through the fusion calculation of three dimensions: semantic similarity, temporal dependency, and cognitive span. Semantic similarity is calculated using cosine similarity, which is the cosine of the angle between two concept vectors. Temporal dependency is calculated based on the order in which concepts appear in the teaching content. Cognitive span is determined based on the difference in cognitive difficulty between concepts. The three-dimensional cognitive attribute vectors are input into a weight optimization algorithm for parameter adjustment, resulting in a combination of weight coefficients for the three dimensions. Finally, the initial pre-dependency strength matrix is ​​weighted and fused based on the weight coefficient combination. A threshold judgment algorithm is used to filter dependencies; when the dependency strength value exceeds a preset threshold, a pre-dependency relationship is determined between concepts.

[0019] A six-level cognitive hierarchy decomposition and encoding process is performed based on a concept pre-dependency graph. Based on Bloom's hierarchy of cognitive theories, teaching concepts are divided into six cognitive levels: memory, comprehension, application, analysis, evaluation, and creation. Each level represents different cognitive complexity and thinking requirements. The cognitive hierarchy determination algorithm identifies the main cognitive level of each concept by analyzing its lexical, grammatical, and contextual features. Lexical features include the complexity and specialization of the concept's vocabulary; grammatical features involve the syntactic structure of the concept's expression; and contextual features consider the concept's contextual information within the teaching materials. Then, hierarchical semantic encoding is performed on the teaching concepts, generating semantic vectors for each concept at each of the six cognitive levels, forming a set of hierarchical semantic vectors. A graph neural network models the semantic relationships between concepts through neighbor information aggregation, and an attention mechanism assigns different weights based on the importance of each concept. The updated hierarchical semantic vectors are organized into a multi-cognitive-level semantic representation matrix according to the hierarchical structure.

[0020] The learner cognitive ability vector and the multi-level cognitive semantic representation matrix are recombined using dynamic programming under cognitive load constraints. The learner cognitive ability vector, constructed based on historical learning data and cognitive test results, includes six dimensions: memory capacity, comprehension speed, application proficiency, analytical depth, evaluation accuracy, and innovation tendency. The matching degree calculation process involves matching the learner's six-dimensional ability characteristics with the six levels of cognitive semantic vectors in the multi-level cognitive semantic representation matrix layer by layer, forming an ability-demand pairing matrix. The Euclidean distance algorithm measures the degree of difference between learner abilities and concept demands, calculating a concept cognitive difference metric. A cognitive load mapping function converts the difference metric into a standardized cognitive load intensity value, and a sigmoid function performs a non-linear transformation to obtain a normalized cognitive load intensity coefficient. A time complexity factor adjusts the cognitive load in terms of the time dimension and is calculated through a weighted product with the concept learning duration. The dynamic programming algorithm optimizes the combination of teaching content under cognitive load constraints. The learning efficiency maximization objective function evaluates the sequence quality through a balance calculation between learning gains and cognitive load.

[0021] An incremental update fusion space based on timestamped markers is constructed using an adaptive content sequence for cognitive load. Timestamped markers record the creation time, update time, and access time of each teaching concept, forming a time-dimensional data tracking mechanism. The incremental update fusion space adopts a four-tuple structure, including four dimensions: concept set, relation set, time set, and update operation set. A spatial indexing algorithm is used for data organization and processing. A relation strength decay model calculates the timeliness of concept relations using an exponential decay function; the strength of relations between concepts gradually weakens over time, reflecting the timeliness characteristics of knowledge. Knowledge graph update processing updates the graph structure through a weighted fusion calculation of the strength of old and new relations and a conflict detection mechanism. The conflict detection mechanism identifies and resolves contradictions between new and existing knowledge. The A* search algorithm, combined with learner feedback data, performs path replanning, and a heuristic function estimates path costs based on concept difficulty and learner ability, generating dynamic personalized learning paths.

[0022] In one specific embodiment, the process of performing step S101 may specifically include the following steps: The teaching text, video subtitles, and exercise bank are processed by teaching text segmentation to obtain a set of original text fragments segmented according to the boundaries of knowledge points; Based on TF-IDF weight calculation, keyword extraction is performed on the original text fragment set to obtain a list of core teaching concept identifiers; The list of core teaching concept identifiers is input into the cognitive difficulty assessment model for quantitative calculation, resulting in concept attribute parameters including difficulty coefficient and learning time coefficient. Based on the conceptual attribute parameters, cognitive attribute vectors are constructed for the teaching concepts to obtain three-dimensional cognitive attribute vectors. Based on the concept identifier mapping table, the three-dimensional cognitive attribute vectors are uniformly identified to obtain a teaching concept ontology containing cognitive attribute vectors.

[0023] Specifically, the first step in labeling the cognitive difficulty coefficients of multi-source teaching resources is teaching text segmentation. This process performs a unified segmentation operation on three heterogeneous data sources: textbook text, video subtitles, and exercise banks. Teaching text segmentation employs a knowledge point boundary-based segmentation strategy. By identifying features such as paragraph titles, chapter separators, and keyword density variations, the location of knowledge point boundaries is determined, and continuous teaching text is segmented into independent text fragments according to the principle of semantic integrity. Each original text fragment corresponds to a complete knowledge point, including its definition, explanation, examples, and exercises, forming a set of original text fragments that serve as the basic data unit for subsequent processing.

[0024] Keyword extraction from the original text fragment set based on TF-IDF weight calculation is a crucial step in identifying core teaching concepts. The TF-IDF algorithm quantifies the importance of words in a specific document by calculating the product of word frequency and inverse document frequency (IVF). Term frequency reflects the frequency of a word's occurrence in a single text fragment, while IVF reflects the rarity of a word in the entire text fragment set. The algorithm first performs word segmentation and stop word removal on each text fragment. Then, it calculates the TF value for each word (the number of times the word appears in the current fragment divided by the total number of words in the fragment). Next, it calculates the IDF value (the total number of text fragments divided by the logarithm of the number of fragments containing the word). Finally, the TF value and IDF value are multiplied to obtain the TF-IDF weight. By setting a weight threshold, the words with the highest TF-IDF values ​​are selected as keywords for that text fragment. After concept identification and standardization, these keywords form a list of core teaching concept identifiers.

[0025] The core step in constructing concept attribute parameters is to input the list of core teaching concept identifiers into the cognitive difficulty assessment model for quantitative calculation. The cognitive difficulty assessment model is an evaluation algorithm based on multi-feature fusion. This model takes concept identifiers as input and calculates the cognitive difficulty of a concept by analyzing multiple dimensions of features, including lexical complexity, abstraction level, prerequisite knowledge requirements, and learning level. Lexical complexity is measured by indicators such as the number of syllables, character length, and density of technical terms in the concept vocabulary. Abstraction level is assessed through the concept's concreteness score and visualization level. Prerequisite knowledge requirements are determined by analyzing the concept's dependencies on other concepts. The learning level is divided according to the six levels of Bloom's Taxonomy of Cognition. The model inputs these feature values ​​into a neural network for weighted calculation and outputs concept attribute parameters containing a difficulty coefficient and a learning time coefficient. The difficulty coefficient reflects the cognitive complexity of the concept, and the learning time coefficient represents the average time required to master the concept.

[0026] Constructing cognitive attribute vectors for teaching concepts based on conceptual attribute parameters is a process of integrating multi-dimensional attribute information into a standardized vector representation. This process combines three parameters—difficulty coefficient, learning time coefficient, and the number of prerequisite concepts obtained through dependency analysis—into a three-dimensional vector structure. The vector construction process first normalizes the three parameters, mapping parameter values ​​of different dimensions to a standard range of zero to one. Then, the three normalized values ​​are organized into a vector form according to a fixed dimensional order. The first dimension of the three-dimensional cognitive attribute vector represents the normalized difficulty coefficient, the second dimension represents the normalized learning time coefficient, and the third dimension represents the normalized number of prerequisite concepts. The vector's magnitude reflects the overall cognitive complexity of the concept.

[0027] The final step in building a concept ontology is to uniformly identify 3D cognitive attribute vectors using a concept identifier mapping table. The concept identifier mapping table is a data structure containing information such as concept name, unique identifier, synonym list, and hierarchical relationships, used to solve the problem of inconsistent naming of the same concept from different data sources. The uniform identification process identifies different expressions pointing to the same concept through methods such as string matching, semantic similarity calculation, and expert rules, mapping them to a unified identifier. The process establishes an association between each 3D cognitive attribute vector and its corresponding unified identifier, while recording metadata information such as the concept's source, creation time, and update history. The data after uniform identification is organized according to the ontology structure, forming a teaching concept ontology containing concept entities, attribute relationships, and hierarchical structures.

[0028] In one specific embodiment, the process of performing step S102 may specifically include the following steps: The teaching concepts in the teaching concept ontology library are input into the Word2Vec model for vectorization processing to obtain a set of concept semantic vectors. A pre-dependency matrix is ​​constructed based on a set of concept semantic vectors, where the rows and columns of the matrix correspond to the teaching concepts, and the matrix element values ​​represent the pre-dependency strength between concepts. The initial pre-dependency strength matrix is ​​obtained by fusion calculation of three dimensions: semantic similarity, temporal dependency, and cognitive span. The three-dimensional cognitive attribute vector is input into the weight optimization algorithm for parameter adjustment, resulting in a combination of weight coefficients for semantic similarity, temporal dependence, and cognitive span. The initial pre-dependency strength matrix is ​​weighted and fused according to the combination of weight coefficients to obtain the weighted pre-dependency matrix. The weighted pre-dependency matrix is ​​input into a threshold determination algorithm to filter dependencies. When the dependency strength value exceeds a preset threshold, it is determined that there is a pre-dependency relationship between concepts, and a concept pre-dependency graph is obtained.

[0029] Specifically, the first step of the pre-dependency matrix algorithm is to input the teaching concepts from the teaching concept ontology into the Word2Vec model for vectorization. The Word2Vec model is a neural network-based word embedding technique that learns semantic representations of words by analyzing the contextual relationships of words in a large-scale teaching corpus. The model employs a skip-word model architecture, learning distributed representations of words by predicting the surrounding context words of the target word, mapping each teaching concept to a dense vector of fixed dimensions. The vectorization process first uses concept identifiers from the teaching concept ontology as input words. The model's embedding layer converts discrete concept identifiers into continuous numerical vectors, with each concept corresponding to a high-dimensional vector containing semantic information. These vectors reflect the semantic similarity between concepts; semantically similar concepts are closer in the vector space, thus forming a set of concept semantic vectors as the foundation for subsequent matrix construction.

[0030] Constructing a pre-dependency matrix based on a set of concept semantic vectors is the core step in establishing dependencies between concepts. The pre-dependency matrix is ​​a square matrix where rows and columns correspond to all concepts in the teaching concept ontology, and element values ​​represent the strength of the pre-dependency of a row concept on a column concept. The construction process determines the strength of dependencies between concepts through the fusion of three dimensions: semantic similarity, temporal dependency, and cognitive span. Semantic similarity is calculated by taking the cosine similarity of the semantic vectors of two concepts; cosine similarity equals the dot product of the two vectors divided by the product of their magnitudes, with a value ranging from -1 to +1. Temporal dependency is calculated based on the order in which concepts appear in teaching materials, determining temporal dependence by analyzing the positional relationships of concepts in the syllabus and textbook chapters. Cognitive span is calculated based on the difference in cognitive difficulty between concepts, measuring the difference in cognitive complexity by comparing the difficulty coefficients in the three-dimensional cognitive attribute vectors of two concepts. The results of the three dimensions are fused according to specific mathematical relationships to form an initial pre-dependency strength matrix.

[0031] Inputting the three-dimensional cognitive attribute vector into the weight optimization algorithm for parameter adjustment is a crucial step in determining the weight coefficients for the three dimensions. The weight optimization algorithm employs a gradient descent-based optimization strategy, learning the optimal weight combination by minimizing the error between the predicted dependency and the actual teaching sequence. The algorithm uses the three-dimensional cognitive attribute vector of the teaching concept as feature input, including values ​​for difficulty coefficient, learning time coefficient, and number of prior concepts. The optimization process quantifies the difference between the predicted result and the true label by constructing a loss function, which considers the accuracy of concept dependencies and the rationality of the learning path. The algorithm uses a backpropagation mechanism to update the weight parameters, gradually adjusting the weight coefficients of semantic similarity, temporal dependency, and cognitive span through multiple rounds of iterative training, ultimately converging to the weight coefficient combination that best fits the teaching pattern.

[0032] Weighted fusion calculation of the initial pre-dependency strength matrix based on weighted coefficient combinations is a crucial step in integrating multi-dimensional information. The weighted fusion calculation linearly sums the values ​​of semantic similarity, temporal dependency, and cognitive span according to coefficients obtained from a weight optimization algorithm. The fusion process calculates a weighted sum for each element in the matrix separately: multiplying the semantic similarity, temporal dependency, and cognitive span at that position by their respective weights, and then summing these three products to obtain the final dependency strength value at that position. Weighted fusion calculation ensures the reasonable integration of information from different dimensions, resolves potential biases in judging dependencies based on a single dimension, and forms a weighted pre-dependency matrix that comprehensively considers multiple factors.

[0033] Inputting the weighted pre-dependency matrix into a threshold-based decision-making algorithm for dependency filtering is the decision-making step for determining the final concept dependencies. The threshold-based algorithm distinguishes between strong and weak dependencies by setting a preset threshold. When the dependency strength value of a matrix element exceeds the preset threshold, a pre-dependency relationship is determined between the corresponding concepts. The threshold setting is based on empirical data and statistical analysis results from teaching practice, typically selecting values ​​that filter out noisy relationships while retaining important dependencies. The decision-making process checks each element value in the weighted pre-dependency matrix one by one, marking positions exceeding the threshold as having a dependency relationship and positions below the threshold as having no dependency relationship. The binarized matrix after threshold filtering is converted into a graph structure representation, where concepts are nodes and dependencies are directed edges, ultimately forming a concept pre-dependency graph.

[0034] In one specific embodiment, the process of executing step S103 may specifically include the following steps: Based on Bloom's hierarchy of cognitive theories, the cognitive levels of teaching concepts are classified into six levels: memory, understanding, application, analysis, evaluation, and creation, resulting in a classification label for the cognitive levels of concepts. The teaching concepts in the concept pre-dependency graph are input into the cognitive level determination algorithm for main cognitive level identification. Through comprehensive analysis of lexical features, grammatical features and contextual features, the main cognitive level identifier of each concept is obtained. Based on the main cognitive level identifier, the teaching concepts are processed by hierarchical semantic encoding, and semantic vectors are generated for each concept at the six cognitive levels, resulting in a set of hierarchical semantic vectors for the concepts. The hierarchical semantic vector set of concepts is input into a graph neural network for neighbor information aggregation. The semantic relationship between concepts is modeled through an attention mechanism to obtain the updated hierarchical semantic vector. Based on the updated hierarchical semantic vectors, matrix organization processing is performed to arrange the six-layer cognitive semantic vectors of all concepts according to the hierarchical structure, resulting in a multi-cognitive hierarchical semantic representation matrix.

[0035] Specifically, classifying teaching concepts according to cognitive levels based on Bloom's hierarchy of cognitive theories is the theoretical foundation for the six-level cognitive hierarchy decomposition and coding. Bloom's hierarchy of cognitive theories divides the learning process into six progressive cognitive levels, each representing a different level of cognitive complexity and requirements. The memory level requires learners to recall and identify basic information such as facts, terms, and fundamental concepts; the comprehension level requires learners to interpret, summarize, and transform the meaning of information; the application level requires learners to apply learned knowledge to solve problems in new situations; the analysis level requires learners to decompose complex information and identify the relationships between its parts; the evaluation level requires learners to make judgments and assessments based on standards and criteria; and the creativity level requires learners to integrate information to generate new ideas or solutions. The classification process analyzes the content characteristics, learning objectives, and cognitive requirements of the teaching concepts, categorizing each concept into its corresponding cognitive level, establishing a mapping relationship between concepts and cognitive levels, and forming concept cognitive level classification labels as the basis for subsequent processing.

[0036] The core step in determining the primary cognitive level of a concept is inputting the teaching concepts from the concept pre-dependency graph into the cognitive level determination algorithm for primary cognitive level identification. The cognitive level determination algorithm analyzes the cognitive level attributes of a concept through multi-feature fusion. This algorithm receives teaching concepts from the concept pre-dependency graph as input data. Lexical feature analysis includes linguistic features such as the lexical complexity, terminology density, and abstraction level of the concept name, quantifying the linguistic complexity of the concept through indicators such as the number of syllables, character length, and word frequency statistics. Grammatical feature analysis considers syntactic features such as the syntactic structure, syntactic complexity, and modifier usage of the concept expression, extracting syntactic complexity information of the concept through natural language processing techniques such as syntactic analysis trees and dependency parsing. Contextual feature analysis focuses on the contextual information of the concept in the teaching materials, related concepts, and location of occurrence, determining the cognitive location of the concept by analyzing the distribution of surrounding vocabulary, semantic associations, and teaching sequences. The algorithm performs weighted fusion calculations of the three types of features, uses a machine learning classifier to predict the cognitive level of the concept, outputs the probability distribution of each concept across six cognitive levels, and selects the level with the highest probability as the primary cognitive level identifier for that concept.

[0037] Hierarchical semantic encoding of teaching concepts based on primary cognitive level identifiers is a key step in constructing multi-level semantic representations. Hierarchical semantic encoding generates corresponding semantic vector representations for each concept across all six cognitive levels, based on the primary cognitive level identifier. The encoding process first determines the semantic strength of the concept at the primary level based on its primary cognitive level identifier, and then calculates the semantic strength of the concept at other cognitive levels through inter-level cognitive correlation analysis. Inter-level cognitive correlation reflects the differences in the performance of the same concept at different cognitive levels: the memory level representation focuses on encoding factual information, the understanding level representation emphasizes the relationships between concepts, the application level representation highlights practical application scenarios, the analysis level representation focuses on the structural decomposition of concepts, the evaluation level representation reflects judgment criteria, and the creation level representation reflects innovative potential. The encoding algorithm generates fixed-dimensional dense vectors for each concept at each of the six cognitive levels. These vectors capture the semantic features of the concept at different cognitive levels, forming a set of hierarchical semantic vectors that serve as input data for the graph neural network.

[0038] The core algorithm for semantic relationship modeling is to input a set of hierarchical semantic vectors of concepts into a graph neural network for neighbor information aggregation. A graph neural network is a deep learning model specifically designed for graph-structured data. This network treats concepts as graph nodes and pre-existing dependencies between concepts as graph edges, propagating and aggregating information across the graph through a message-passing mechanism. Neighbor information aggregation updates the representation vector of each concept node by collecting information from its neighbors. The aggregation function performs a weighted sum or average of the semantic vectors of the neighboring nodes. An attention mechanism dynamically calculates the importance weights of different neighboring nodes during the aggregation process, determining which neighboring concepts are more important to the semantic representation of the current concept by learning attention scores. The attention score is calculated based on multiple factors such as semantic similarity between concepts, the strength of pre-existing dependencies, and differences in cognitive levels; a higher score indicates a greater influence of the corresponding neighboring concept on the current concept. The graph neural network iteratively updates the semantic vectors of all concepts through multiple rounds of message passing. In each iteration, the new representation of a concept incorporates information from its neighboring concepts, ultimately converging to a stable updated hierarchical semantic vector.

[0039] The final step in constructing a multi-cognitive-level semantic representation matrix is ​​to perform matrix organization processing based on the updated hierarchical semantic vectors. Matrix organization systematically arranges the six-layer cognitive semantic vectors of all concepts according to a hierarchical structure, forming a three-dimensional matrix structure. The first dimension of the matrix corresponds to the number of teaching concepts, the second dimension corresponds to the six cognitive levels, and the third dimension corresponds to the dimension of the semantic vectors. The organization process first determines the position of a concept in the matrix according to the identifier order, and then arranges the six-layer semantic vectors of each concept in the order of memory layer, understanding layer, application layer, analysis layer, evaluation layer, and creation layer. The value at each position in the matrix represents the semantic feature strength of the corresponding concept at a specific cognitive level, and the overall structure of the matrix reflects the complete semantic representation relationship of all concepts across multiple cognitive levels.

[0040] In one specific embodiment, the process of executing step S104 may specifically include the following steps: Based on learners' historical learning data and cognitive test results, a six-dimensional cognitive ability feature is constructed, including memory capacity, comprehension speed, application proficiency, analytical depth, evaluation accuracy, and innovation tendency, resulting in a learner's cognitive ability vector; The matching degree of learners’ cognitive ability vectors and multi-cognitive level semantic representation matrices is calculated. Through the difference analysis between conceptual cognitive needs and learners’ abilities, a set of conceptual cognitive load coefficients is obtained. A cognitive load constraint function is constructed based on the set of conceptual cognitive load coefficients. The constraint condition that the total cognitive load does not exceed the learner's tolerance threshold is set. The teaching content combination optimization is carried out through dynamic programming algorithm to obtain a set of candidate content sequences that meet the constraint conditions. The candidate content sequence set is input into the learning efficiency maximization objective function for sequence evaluation. By balancing the learning benefits and cognitive load, the optimal content sequence ranking result is obtained. Based on the optimal content sequence sorting results, the teaching concepts are reorganized temporally to ensure that the prerequisite dependencies are satisfied and the cognitive load is evenly distributed, thus obtaining a cognitive load adaptive content sequence.

[0041] Specifically, multi-source data fusion is used to comprehensively characterize learners' cognitive abilities. Historical learning data includes records of learners' performance in past learning activities, covering behavioral data such as answer accuracy, learning time, knowledge mastery, and learning path selection. This data reflects learners' performance in actual learning processes. Cognitive test results come from specially designed cognitive ability assessment tests, including standardized assessment tools such as memory tests, comprehension tests, and application ability tests. The construction process of the six-dimensional cognitive ability characteristics maps raw data to six cognitive dimensions. Memory capacity is calculated by analyzing learners' performance in memory-related tasks, including a comprehensive evaluation of indicators such as information retention time, memory accuracy, and memory capacity. Comprehension speed is calculated based on the time and depth of understanding required for learners to comprehend new concepts. Application proficiency is assessed through learners' performance in solving practical problems, with in-depth analysis examining learners' ability to decompose complex problems. Evaluation accuracy measures the accuracy of learners' judgments and evaluations. Innovation tendency reflects learners' performance tendencies in creative tasks. The construction process normalizes the raw data of each dimension, unifying data of different dimensions into the range of zero to one, and finally forming a six-dimensional learner cognitive ability vector.

[0042] The key step in quantifying cognitive load is calculating the matching degree between learners' cognitive ability vectors and multi-level cognitive semantic representation matrices. This matching degree calculation determines learning difficulty by comparing learners' cognitive abilities with the cognitive needs of the taught concepts. This process compares the six dimensions of the learner's cognitive ability vector with the corresponding semantic vectors of the cognitive levels in the multi-level cognitive semantic representation matrix. The matching degree calculation uses a vector similarity algorithm, measuring the degree of matching by calculating the Euclidean distance between the learner's ability vector and the concept's need vector. A smaller distance indicates a better match between the learner's ability and the concept's need, while a larger distance indicates a gap in ability. The difference analysis between concept cognitive needs and learners' abilities compares the ability needs of each concept for each concept, analyzing the gap between the intensity of need for each concept at the six cognitive levels and the learner's corresponding ability. The results of the difference analysis are converted into a cognitive load coefficient, which reflects the amount of cognitive resources a learner needs to invest when learning a specific concept. A larger difference indicates a higher cognitive load, and a smaller difference indicates a lower cognitive load. The calculation process organizes the cognitive load coefficients of all concepts into a set, providing quantitative load assessment data for subsequent optimization algorithms.

[0043] Constructing a cognitive load constraint function based on a set of concept cognitive load coefficients is a constraint setting step in dynamic programming optimization. The cognitive load constraint function is a mathematical function that takes the sequence of teaching content as input and outputs the total cognitive load corresponding to that sequence. The constraint function is constructed based on the principle of cognitive load accumulation, meaning the total cognitive load of the learning sequence equals the sum of the cognitive load coefficients of all concepts in the sequence. The constraint condition sets the total cognitive load to not exceed a learner's tolerance threshold, which is determined comprehensively based on factors such as the learner's cognitive ability level, learning time constraints, and fatigue level. Dynamic programming is an algorithm used to solve optimization problems. This algorithm decomposes the complex problem of combining teaching content into multiple subproblems and constructs the global optimum by progressively solving the optimal solutions to the subproblems. During the search process, the algorithm maintains a state transition table, recording the optimal choices in different states, and calculates the cost of transitioning from one state to another using state transition equations. The optimization process, under the premise of satisfying the cognitive load constraints, searches for the sequence combination containing the most effective teaching content, generating multiple candidate content sequence sets that satisfy the constraints.

[0044] The optimal sequence selection step involves inputting the candidate content sequence set into the learning efficiency maximization objective function for sequence evaluation. The learning efficiency maximization objective function is an evaluation function that measures the ratio between the learning effectiveness of the teaching content sequence and the cognitive load. The objective function design considers a balance between learning gains and cognitive load. Learning gains are calculated by evaluating factors such as the importance of concepts, knowledge coverage, and skill enhancement potential in the sequence, while cognitive load is obtained from the aforementioned cognitive load coefficient set. The balance calculation employs a weighted scoring mechanism, using learning gains as a positive factor and cognitive load as a negative factor, balancing their influence by setting appropriate weight coefficients. The sequence evaluation process calculates the objective function value for each sequence in the candidate content sequence set; a higher function value indicates better learning efficiency. The evaluation process also considers quality factors such as sequence completeness, logical coherence, and difficulty gradient. Through multi-dimensional evaluation, it ensures that the selected sequences are both efficient and reasonable, ultimately outputting the optimal content sequence ranking results based on the objective function values.

[0045] The key step in generating the final learning path is to reorder the teaching concepts based on the optimal content sequence. This reordering process determines the learning order of the teaching concepts based on the optimal content sequence, and it must simultaneously satisfy two constraints: pre-dependencies between concepts and a uniform distribution of cognitive load. Pre-dependencies ensure that learners have mastered all the pre-concepts of a given concept before learning it, avoiding learning difficulties due to a lack of foundational knowledge. A uniform distribution of cognitive load requires the reasonable allocation of concepts with high cognitive loads in the learning sequence, avoiding cognitive overload caused by learning too many high-load concepts in a short period. The reordering process uses a topological sorting algorithm combined with a load balancing strategy. First, it determines the basic order of concepts based on pre-dependencies. Then, while satisfying the dependencies, it adjusts the positions of the concepts to balance the distribution of cognitive load. The adjustment process optimizes the load distribution through strategies such as swapping adjacent concept positions, inserting buffer concepts, and segmenting high-load concept groups, ultimately generating an adaptive content sequence that satisfies both logical dependencies and a reasonable load distribution.

[0046] In one specific embodiment, the process of executing step S105 may specifically include the following steps: The six-dimensional ability features in the learner's cognitive ability vector are matched with the six-layer cognitive semantic vectors in the multi-cognitive level semantic representation matrix to obtain the ability-demand pairing matrix. The cognitive demand intensity of each teaching concept is calculated based on the ability-demand matching matrix. The difference between learners' abilities and concept demands is measured by the Euclidean distance algorithm to obtain the concept cognitive difference metric. A cognitive load mapping function is constructed based on the conceptual cognitive difference metric. The difference metric is converted into a standardized cognitive load intensity value. The normalized cognitive load intensity coefficient is obtained by nonlinear transformation through the sigmoid function. The normalized cognitive load intensity coefficient and the concept learning time are weighted and multiplied to calculate the time-weighted cognitive load coefficient. A time complexity factor is introduced to adjust the cognitive load in the time dimension. All teaching concepts are organized and processed in a set based on time-weighted cognitive load coefficients, and then indexed and arranged according to concept identifiers to obtain a set of concept cognitive load coefficients.

[0047] Specifically, the fundamental step in establishing the relationship between ability and demand is to perform a layer-by-layer correspondence matching process between the six-dimensional ability features in the learner's cognitive ability vector and the six-layer cognitive semantic vectors in the multi-cognitive-level semantic representation matrix. This layer-by-layer correspondence matching process compares the learner's cognitive ability with the cognitive demand of the taught concept one-to-one through dimension correspondence. This process matches the learner's cognitive ability vector's memory capacity dimension with the memory layer vector of the multi-cognitive-level semantic representation matrix, the comprehension speed dimension with the comprehension layer vector, the application proficiency dimension with the application layer vector, the analysis depth dimension with the analysis layer vector, the evaluation accuracy dimension with the evaluation layer vector, and the innovation tendency dimension with the creativity layer vector. The matching process pairs the semantic vector of each taught concept at the six cognitive levels with the learner's corresponding ability values ​​in the corresponding dimensions, forming an ability-demand pairing matrix as input data for difference calculation. The ability-demand pairing matrix is ​​a two-dimensional data structure. The rows of the matrix correspond to the taught concepts, and the columns correspond to the six cognitive levels. Each matrix element contains the learner's ability value in the corresponding cognitive dimension and the demand intensity value of that concept at the corresponding cognitive level. The pairing matrix intuitively reflects the correspondence between learner abilities and the demands of each concept.

[0048] Calculating the intensity of cognitive demand for each teaching concept based on the ability-demand matching matrix is ​​the core step in quantifying learning difficulty. This calculation determines the learner's difficulty level for a specific concept by analyzing each row of data in the ability-demand matching matrix, calculating the intensity of cognitive demand for each teaching concept separately. The Euclidean distance algorithm is a mathematical method for measuring the distance between two points in a multidimensional space. This algorithm treats the learner's six-dimensional ability characteristics and the six-dimensional cognitive demands of the teaching concept as two points in a six-dimensional space, measuring the degree of matching between ability and demand by calculating the straight-line distance between the two points. The distance calculation process first calculates the difference between the learner's ability value and the concept demand value in each dimension, then squares each of the six differences, sums the six squared values, and finally takes the square root of the sum to obtain the Euclidean distance value. In the measurement process, a larger distance value indicates a greater difference between the learner's ability and the concept demand, and a higher difficulty in learning the concept; a smaller distance value indicates a better match and lower learning difficulty. Each teaching concept is calculated using the Euclidean distance algorithm to obtain a corresponding concept cognitive difference metric, which quantifies the degree of cognitive challenge learners face when learning a specific concept.

[0049] Constructing a cognitive load mapping function based on conceptual cognitive difference measures is a crucial step in converting these differences into standardized load coefficients. The cognitive load mapping function is a mathematical transformation function that maps the original difference measures to a standardized cognitive load intensity range. The construction of this mapping function is based on the principle of the non-linear relationship between cognitive load and learning difficulty in cognitive psychology; that is, the increase in cognitive load is not a simple linear relationship with the difference in difficulty. The sigmoid function is an S-shaped curve function with excellent non-linear transformation characteristics, capable of mapping input values ​​of any range to an output range of zero to one. Non-linear transformation processing uses the sigmoid function to transform the conceptual cognitive difference measures. During the transformation, smaller difference measures result in lower load coefficients after the function transformation, and larger difference measures result in higher load coefficients. The S-shaped characteristic of the transformation curve ensures a smooth transition and reasonable distribution of load coefficients. This transformation process solves the potential extreme value problem and dimension inconsistency problem of the original difference measures. The output normalized cognitive load intensity coefficients are uniformly distributed within the standard range of zero to one, facilitating subsequent calculations and comparisons.

[0050] Introducing a time dimension is a crucial step in calculating the weighted product of the normalized cognitive load intensity coefficient and the concept learning time. This weighted product calculation comprehensively assesses learners' cognitive load levels by combining cognitive intensity with time factors, taking into account the cognitive pressure learners experience per unit of time. Concept learning time refers to the average time learners require to master a specific concept; this data comes from teaching practice statistics and cognitive experiment measurements. The time complexity factor is an adjustment parameter reflecting the degree of influence of time pressure on cognitive load; its value is based on research in cognitive psychology regarding the relationship between time pressure and cognitive performance. The weighted product calculation multiplies the normalized cognitive load intensity coefficient by the concept learning time, and then performs a second product operation with the time complexity factor. The result reflects the learner's overall cognitive load level under specific time constraints. This time dimension adjustment ensures the comprehensiveness and accuracy of cognitive load assessment, avoiding the limitations of considering only cognitive intensity while ignoring time factors, resulting in a time-weighted cognitive load coefficient that more closely reflects actual learning conditions.

[0051] The final step in constructing the cognitive load coefficient set is to organize all teaching concepts in a aggregated manner based on time-weighted cognitive load coefficients. This aggregated organization process organizes and stores the dispersed time-weighted cognitive load coefficients according to a unified data structure, ensuring data orderliness and accessibility. The concept identifier is a unique identifier for each teaching concept, determined during the construction phase of the teaching concept ontology, possessing uniqueness and immutability. The indexing sorts the time-weighted cognitive load coefficients according to the lexicographical or numerical order of the concept identifiers, establishing a mapping relationship between concept identifiers and their corresponding load coefficients. The organization process also includes data integrity checks and outlier handling, ensuring that each concept has a corresponding load coefficient with values ​​within a reasonable range. The processing result forms a concept cognitive load coefficient set, which stores concept identifiers and their corresponding time-weighted cognitive load coefficients in key-value pairs. The set structure supports fast querying and batch processing operations.

[0052] In one specific embodiment, the process of performing step S106 may specifically include the following steps: Each teaching concept in the cognitive load adaptive content sequence is timestamped, and the creation time, update time, and access time of the concept are recorded to obtain a timestamped concept sequence. A four-tuple fusion space structure is constructed based on the concept sequence marked by timestamps, which includes four dimensions: concept set, relation set, time set, and update operation set. Data is organized and processed through a spatial indexing algorithm to obtain the incremental update fusion space. The concept relationships in the concept pre-dependency graph are input into the relationship strength decay model for timeliness calculation. The relationship strength is adjusted by time weighting through the exponential decay function to obtain the time decay relationship strength matrix. The knowledge graph is updated by performing incremental update fusion space processing based on the relation strength matrix of time decay. The evolved knowledge graph structure is obtained through weighted fusion calculation of the strength of old and new relations and conflict detection mechanism. Based on the evolved knowledge graph structure, the path optimization process is performed on the cognitive load adaptive content sequence. The path replanning is carried out by combining the A* search algorithm with learner feedback data to obtain a dynamic personalized learning path.

[0053] Specifically, timestamping each teaching concept in the cognitive load adaptive content sequence is a fundamental step in constructing a time-sensitive knowledge graph. Timestamping tracks the lifecycle of each teaching concept by attaching time information. This process records three key time information points to comprehensively reflect the time-sensitive characteristics of the concept: First, the creation time record shows the point in time the concept was first added to the knowledge graph, marking its initial establishment within the teaching system. Second, the update time record shows the latest point in time when the concept's content or attributes were modified, reflecting the dynamic changes in the concept's information. Third, the access time record shows the point in time the concept was most recently consulted or used by learners, reflecting the concept's actual application frequency and importance. The timestamping process establishes associations between the three timestamps and the corresponding teaching concepts, forming a concept data structure containing time-dimensional information. The timestamps are standardized using a unified time format to ensure consistency and comparability of time information from different sources. The processing result generates a timestamped concept sequence, which adds time-dimensional information to the original cognitive load adaptive content sequence. Each concept not only contains cognitive attributes and dependencies but also complete time tracking information.

[0054] Constructing a four-tuple fusion spatial structure based on timestamped concept sequences is a core step in establishing multi-dimensional knowledge representation. The four-tuple fusion spatial structure is a multi-dimensional data organization model that comprehensively describes the state and evolution of a knowledge graph through four different sets. The concept set contains all teaching concepts and their attribute information; each concept element carries a cognitive attribute vector and timestamp information. The relation set stores various relationships between concepts, including pre-dependencies, semantic similarity, cognitive hierarchy, and other types of concept connections. The time set records all time-related information in the knowledge graph, including concept timestamps, relation timeliness, update time series, and other time-dimensional data. The update operation set records all modification operations performed on the knowledge graph, including operation types such as concept addition, relation updates, and attribute modifications, along with their execution times. The spatial indexing algorithm is an efficient data organization method that supports rapid data querying and update operations by establishing a multi-dimensional index structure. Data organization processing arranges and stores the four-dimensional data according to the index structure, establishing multiple index relationships such as concept identifiers, relation types, time intervals, and operation types. The incremental update fusion space supports dynamic data insertion, deletion, and modification operations, maintaining data consistency and integrity.

[0055] Introducing the concept relationships in the concept pre-dependency graph into the relationship strength decay model for time-sensitivity calculation is a key step in introducing the time factor. The relationship strength decay model is a time-based modeling method that dynamically adjusts the strength of relationships between concepts based on the principle of knowledge timeliness. The decay model considers the time-sensitivity characteristics of pedagogical knowledge, meaning that the importance and relevance of certain concept relationships change over time. The exponential decay function is a mathematical function that simulates the change in relationship strength over time through an exponential decay law. The decay function uses the difference between the current time and the relationship establishment time as an input variable and outputs a corresponding decay coefficient to adjust the original relationship strength. Time weight adjustment calculates the adjusted relationship strength by multiplying the original relationship strength by the decay coefficient, considering multiple time factors such as relationship establishment time, most recent update time, and access frequency. The calculation process performs decay calculations on each relationship edge in the concept pre-dependency graph, determining the corresponding decay degree based on the relationship's time attributes. The processing result generates a time-decayed relationship strength matrix, which reflects the distribution of inter-concept relationship strength after considering the time factor.

[0056] Updating the knowledge graph based on the time-decaying relation strength matrix through incremental updates to the fusion space is a crucial step in achieving dynamic evolution. This knowledge graph update process maintains its timeliness and accuracy by integrating new relation strength information, involving the fusion of old and new data and conflict resolution mechanisms. The weighted fusion calculation of old and new relation strengths employs a weighted average method, merging the new values ​​in the time-decaying relation strength matrix with the original values ​​in the incremental update fusion space. The fusion calculation considers factors such as data timeliness, reliability, and source authority to determine weight allocation, ensuring the fusion result retains historical information while reflecting the latest changes. A conflict detection mechanism identifies inconsistencies and contradictions between old and new data, judging the type and severity of conflicts through rule-based reasoning and expert knowledge. The detection mechanism includes various types of consistency verification, such as numerical conflict detection, logical conflict detection, and semantic conflict detection. Conflict resolution strategies employ different methods depending on the conflict type, including data overlay, weighted averaging, expert arbitration, and user confirmation. The update process ultimately generates the evolved knowledge graph structure, which integrates the time-adjusted relation information and the consistent data after conflict resolution.

[0057] The final step in generating dynamic learning paths is to optimize the cognitive load-adaptive content sequence based on the evolved knowledge graph structure. Path optimization adapts the learning sequence to the dynamic changes in the knowledge graph by replanning the learning sequence, considering factors such as evolved concept relationships, updated cognitive load, and learner feedback. The A* search algorithm is a heuristic path search algorithm that guides the search process through an evaluation function, reducing the search space while ensuring the optimal path is found. The search algorithm uses the learning objective as the endpoint and the current knowledge state as the starting point, selecting the search direction by evaluating the estimated cost from each intermediate node to the endpoint. The heuristic function estimates the path cost by combining factors such as the cognitive load of concepts, prior dependencies, and learning benefits. The function design ensures that the estimated value does not exceed the actual cost to guarantee the algorithm's optimality. Learner feedback data includes learning effect evaluations, learning preference surveys, and cognitive ability tests, reflecting the learner's actual learning status and personalized needs. Path replanning, combined with learner feedback data, adjusts the original cognitive load-adaptive content sequence, optimizing the learning order of concepts, adjusting the distribution of cognitive load, and adding or deleting related concepts. The replanning process ensures that the new learning path satisfies both the pre-requirement dependencies of the concept and adapts to the learner's cognitive characteristics and feedback requirements, ultimately generating a dynamic and personalized learning path.

[0058] The above describes the knowledge graph-based teaching content integration method in the embodiments of this application. The following describes the knowledge graph-based teaching content integration system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the knowledge graph-based teaching content integration system in this application includes: The annotation module is used to annotate the cognitive difficulty coefficients of multi-source teaching resources to obtain a teaching concept ontology containing cognitive attribute vectors. The calculation module is used to perform semantic similarity and cognitive span fusion calculation on the teaching concept ontology library through the pre-dependency matrix algorithm to obtain the concept pre-dependency graph; The encoding module is used to perform six-level cognitive hierarchy decomposition and encoding processing on the teaching concept based on the concept pre-dependency graph to obtain a multi-cognitive hierarchy semantic representation matrix; The reorganization module is used to perform dynamic programming reorganization processing on the learner's cognitive ability vector and the multi-cognitive hierarchical semantic representation matrix under cognitive load constraints to obtain a cognitive load adaptive content sequence. The update module is used to construct an incremental update fusion space with timestamps based on the adaptive content sequence of the cognitive load, and to perform knowledge graph evolution processing through the relation strength decay model to obtain a dynamic personalized learning path.

[0059] above Figure 2The knowledge graph-based teaching content integration system in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The knowledge graph-based teaching content integration device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0060] Reference Figure 3 This invention also provides a knowledge graph-based teaching content integration device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the knowledge graph-based instructional content integration device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computational and control capabilities. The memory of the knowledge graph-based instructional content integration device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the knowledge graph-based instructional content integration device stores the data corresponding to this embodiment. The network interface of the knowledge graph-based instructional content integration device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0061] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the knowledge graph-based teaching content integration device to which the present invention is applied.

[0062] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the knowledge graph-based teaching content integration method.

[0063] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a knowledge graph-based teaching content integration device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for integrating teaching content based on knowledge graphs, characterized in that, The method includes: The cognitive difficulty coefficient of multi-source teaching resources is labeled to obtain a teaching concept ontology containing cognitive attribute vectors; The semantic similarity and cognitive span of the teaching concept ontology are fused and calculated using the pre-dependency matrix algorithm to obtain a concept pre-dependency graph. Based on the aforementioned concept pre-dependency graph, the teaching concept is decomposed and encoded into a six-level cognitive hierarchy to obtain a multi-cognitive hierarchical semantic representation matrix. The learner's cognitive ability vector and the multi-cognitive hierarchical semantic representation matrix are recombined under cognitive load constraints through dynamic programming to obtain a cognitive load adaptive content sequence. Based on the cognitive load adaptive content sequence, an incremental update fusion space with timestamp markers is constructed. The knowledge graph is then processed through a relation strength decay model to obtain a dynamic personalized learning path.

2. The method for integrating teaching content based on knowledge graphs according to claim 1, characterized in that, The process of labeling multi-source teaching resources with cognitive difficulty coefficients yields a teaching concept ontology containing cognitive attribute vectors, including: The teaching text, video subtitles, and exercise bank are processed by teaching text segmentation to obtain a set of original text fragments segmented according to the boundaries of knowledge points; Based on TF-IDF weight calculation, keyword extraction is performed on the original text fragment set to obtain a list of core teaching concept identifiers; The list of core teaching concept identifiers is input into the cognitive difficulty assessment model for quantitative calculation to obtain concept attribute parameters including difficulty coefficient and learning time coefficient. Based on the aforementioned conceptual attribute parameters, cognitive attribute vectors are constructed for the teaching concepts to obtain three-dimensional cognitive attribute vectors. The three-dimensional cognitive attribute vectors are uniformly identified based on the concept identifier mapping table to obtain a teaching concept ontology containing the cognitive attribute vectors.

3. The method for integrating teaching content based on knowledge graphs according to claim 2, characterized in that, The step involves performing semantic similarity and cognitive span fusion calculations on the teaching concept ontology using a pre-dependency matrix algorithm to obtain a concept pre-dependency graph, including: The teaching concepts in the teaching concept ontology library are input into the Word2Vec model for vectorization processing to obtain a set of concept semantic vectors. Based on the set of concept semantic vectors, a pre-dependency matrix is ​​constructed, where the rows and columns of the matrix correspond to the teaching concepts, and the matrix element values ​​represent the pre-dependency strength between concepts. The initial pre-dependency strength matrix is ​​obtained through the fusion calculation of three dimensions: semantic similarity, temporal dependency, and cognitive span. The three-dimensional cognitive attribute vector is input into a weight optimization algorithm for parameter adjustment to obtain a combination of weight coefficients for semantic similarity, temporal dependence, and cognitive span. The initial pre-dependency strength matrix is ​​weighted and fused according to the weight coefficient combination to obtain the weighted pre-dependency matrix. The weighted pre-dependency matrix is ​​input to the threshold determination algorithm for dependency filtering. When the dependency strength value exceeds the preset threshold, it is determined that there is a pre-dependency relationship between concepts, and a concept pre-dependency map is obtained.

4. The method for integrating teaching content based on knowledge graphs according to claim 1, characterized in that, The process of performing a six-level cognitive hierarchy decomposition and encoding of the teaching concept based on the concept pre-dependency graph yields a multi-cognitive hierarchy semantic representation matrix, including: Based on Bloom's hierarchy of cognitive theories, the cognitive levels of teaching concepts are classified into six levels: memory, understanding, application, analysis, evaluation, and creation, resulting in a classification label for the cognitive levels of concepts. The teaching concepts in the concept pre-dependency graph are input into the cognitive level determination algorithm for main cognitive level identification. Through comprehensive analysis of lexical features, grammatical features and contextual features, the main cognitive level identifier of each concept is obtained. Based on the main cognitive level identifier, the teaching concepts are subjected to hierarchical semantic encoding processing, and semantic vectors are generated for each concept at the six cognitive levels, resulting in a set of hierarchical semantic vectors for the concepts. The hierarchical semantic vector set of the concepts is input into a graph neural network for neighbor information aggregation. The semantic relationships between concepts are modeled through an attention mechanism to obtain the updated hierarchical semantic vectors. Based on the updated hierarchical semantic vectors, matrix organization processing is performed to arrange the six-layer cognitive semantic vectors of all concepts according to the hierarchical structure, resulting in a multi-cognitive hierarchical semantic representation matrix.

5. The method for integrating teaching content based on knowledge graphs according to claim 1, characterized in that, The step of performing dynamic programming reorganization processing on the learner's cognitive ability vector and the multi-cognitive level semantic representation matrix under cognitive load constraints to obtain a cognitive load adaptive content sequence includes: Based on learners' historical learning data and cognitive test results, a six-dimensional cognitive ability feature is constructed, including memory capacity, comprehension speed, application proficiency, analytical depth, evaluation accuracy, and innovation tendency, resulting in a learner's cognitive ability vector; The learner's cognitive ability vector and the multi-cognitive level semantic representation matrix are matched. By analyzing the difference between conceptual cognitive needs and learner abilities, a set of conceptual cognitive load coefficients is obtained. Based on the set of cognitive load coefficients, a cognitive load constraint function is constructed. The constraint condition that the total cognitive load does not exceed the learner's tolerance threshold is set. The teaching content combination optimization is performed through dynamic programming algorithm to obtain a set of candidate content sequences that meet the constraint condition. The candidate content sequence set is input into the learning efficiency maximization objective function for sequence evaluation. The optimal content sequence ranking result is obtained by balancing the learning benefits and cognitive load. Based on the optimal content sequence sorting result, the teaching concepts are reorganized in a temporal sequence to ensure that the prerequisite dependencies are satisfied and the cognitive load is evenly distributed, thus obtaining a cognitive load adaptive content sequence.

6. The method for integrating teaching content based on knowledge graphs according to claim 5, characterized in that, The process of calculating the matching degree between the learner's cognitive ability vector and the multi-cognitive hierarchical semantic representation matrix, and obtaining a set of conceptual cognitive load coefficients through the difference analysis between conceptual cognitive needs and learner abilities, includes: The six-dimensional ability features in the learner's cognitive ability vector are matched with the six-layer cognitive semantic vectors in the multi-cognitive hierarchical semantic representation matrix to obtain the ability-demand pairing matrix. Based on the ability-demand matching matrix, the cognitive demand intensity of each teaching concept is calculated, and the difference between learners' abilities and concept demands is measured by the Euclidean distance algorithm to obtain a concept cognitive difference metric. A cognitive load mapping function is constructed based on the cognitive difference metric value. The difference metric value is converted into a standardized cognitive load intensity value. The normalized cognitive load intensity coefficient is obtained by nonlinear transformation through the sigmoid function. The normalized cognitive load intensity coefficient and the concept learning time are weighted and multiplied to calculate the time-weighted cognitive load coefficient. A time complexity factor is introduced to adjust the cognitive load in the time dimension. Based on the time-weighted cognitive load coefficient, all teaching concepts are aggregated and organized, and then indexed and arranged according to concept identifiers to obtain a set of concept cognitive load coefficients.

7. The method for integrating teaching content based on knowledge graphs according to claim 1, characterized in that, The incremental update fusion space based on the cognitive load adaptive content sequence, constructed with timestamp markers, undergoes knowledge graph evolution processing through a relation strength decay model to obtain a dynamic personalized learning path, including: Each teaching concept in the cognitive load adaptive content sequence is timestamped, and the creation time, update time, and access time of the concept are recorded to obtain a timestamped concept sequence. Based on the concept sequence marked by the timestamp, a four-tuple fusion space structure is constructed, which includes four dimensions: concept set, relation set, time set, and update operation set. Data is organized and processed through a spatial indexing algorithm to obtain an incremental update fusion space. The concept relationships in the aforementioned concept pre-dependency graph are input into the relationship strength decay model for timeliness calculation. The relationship strength is adjusted by time weighting through the exponential decay function to obtain the time-decayed relationship strength matrix. The incremental update fusion space is updated with knowledge graph based on the relation strength matrix with time decay. The evolved knowledge graph structure is obtained by weighted fusion calculation of old and new relation strengths and conflict detection mechanism. Based on the evolved knowledge graph structure, the cognitive load adaptive content sequence is subjected to path optimization processing. The path is replanned by combining the A* search algorithm with learner feedback data to obtain a dynamic personalized learning path.

8. A knowledge graph-based teaching content integration system, characterized in that, For implementing the knowledge graph-based teaching content integration method as described in any one of claims 1-7, the knowledge graph-based teaching content integration system comprises: The annotation module is used to annotate the cognitive difficulty coefficients of multi-source teaching resources to obtain a teaching concept ontology containing cognitive attribute vectors. The calculation module is used to perform semantic similarity and cognitive span fusion calculation on the teaching concept ontology library through the pre-dependency matrix algorithm to obtain the concept pre-dependency graph; The encoding module is used to perform six-level cognitive hierarchy decomposition and encoding processing on the teaching concept based on the concept pre-dependency graph to obtain a multi-cognitive hierarchy semantic representation matrix; The reorganization module is used to perform dynamic programming reorganization processing on the learner's cognitive ability vector and the multi-cognitive hierarchical semantic representation matrix under cognitive load constraints to obtain a cognitive load adaptive content sequence. The update module is used to construct an incremental update fusion space with timestamps based on the adaptive content sequence of the cognitive load, and to perform knowledge graph evolution processing through the relation strength decay model to obtain a dynamic personalized learning path.

9. A knowledge graph-based teaching content integration device, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the knowledge graph-based teaching content integration method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by a processor, the processor performs the knowledge graph-based teaching content integration method as described in any one of claims 1 to 7.

Images