An individualized teaching system and method based on a knowledge graph framework

Abstract

This invention relates to the field of artificial intelligence knowledge graph technology, and more particularly to a personalized teaching system and method based on a knowledge graph framework, including a data acquisition module, a semantic feature module, a graph mapping module, and a sharing and distribution module. The data acquisition module receives teaching resource files. The semantic feature module extracts text content to generate a set of triples. The graph mapping module converts the triple set into entity nodes and relation edges, constructs a three-level graph topology, and applies external ontology rules to perform conditional inference to generate deep semantic dependencies. The sharing and distribution module applies query conditions to retrieve the three-level graph topology within the deep semantic dependencies and sends the target resource location identifier to downstream terminal nodes. This invention solves the technical problem of existing systems lacking semantic fusion and knowledge reasoning, making it difficult to reuse and share teaching resources.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence knowledge graph technology, and in particular to a personalized teaching system and method based on a knowledge graph framework. Background Technology

[0002] Currently, intelligent teaching systems commonly use relational database technology to store teaching content and exam questions, and record text materials and multimedia files according to a fixed two-dimensional table structure. Knowledge graphs, as a structured semantic knowledge base technology, can transform multi-source heterogeneous data into a mesh topology composed of nodes and edges. The integration of multi-source heterogeneous data aims to unify the processing of structured and unstructured files from different sources and in different formats. Nodes represent specific entity concepts, and edges represent the relationships between entity concepts. The computer performs conditional calculations on the mesh topology according to logical rules and derives new conclusions; simultaneously, the intelligent auxiliary teaching system extracts features based on the computational model and outputs customized learning path planning instructions to the user through the software terminal.

[0003] Because traditional information processing architectures lack deep semantic fusion mechanisms and logical reasoning capabilities, computer programs cannot parse the associated attributes of heterogeneous files. Independent storage hinders the interoperability of underlying data, ultimately creating a technical bottleneck that makes it difficult to reuse and share teaching resources. For example, server A establishes a computer network principles document library, while server B stores a network routing technology video library. Both underlying databases contain LAN protocol concepts. Since traditional data retrieval technologies rely solely on character comparison for surface-level matching, computer programs cannot deduce through semantic association rules that the two sets of data belong to the same underlying concept set. Client B cannot directly access the document resources stored on server A for synchronized practice; teaching and research staff must repeatedly enter digital assets and manually create new classification tags. This objectively isolated data ultimately prevents cross-platform sharing of teaching resources. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a personalized teaching system and method based on a knowledge graph framework. This invention solves the technical problem that traditional systems lack semantic fusion and knowledge reasoning, making it difficult to reuse and share teaching resources.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the present invention provides a personalized teaching system based on a knowledge graph framework, comprising: The data acquisition module receives external input, including text materials, multimedia video files, and teaching resource files such as test question banks; The semantic feature module is connected to the data acquisition module. It receives the teaching resource file, extracts the text content in the teaching resource file, parses the text content to obtain entity concepts and term fragments, calculates the association between the entity concepts, locates the starting position and character span of the term fragments in the text content, generates a set of triples based on the entity concepts, term fragments, association, starting position and character span, and outputs the set of triples. The graph mapping module is connected to the semantic feature module, receives the set of triples, converts the set of triples into entity nodes and relation edges, constructs a three-level graph topology, loads external ontology rules, applies the external ontology rules to perform conditional deduction on the entity nodes in the three-level graph topology, and extracts the deep semantic dependencies between the entity nodes. The shared distribution module is connected to the graph mapping module, receives the call instruction sent by the downstream terminal node, extracts the query conditions in the call instruction, applies the query conditions to retrieve the three-level graph topology structure within the deep semantic dependency relationship to obtain the matching target node, extracts the target resource location identifier recorded in the matching target node, and sends the target resource location identifier to the downstream terminal node.

[0006] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the semantic feature module includes: The concept extraction unit receives the teaching resource file, extracts the text content within the teaching resource file, loads an external domain knowledge ontology set, applies the external domain knowledge ontology set to perform word meaning recognition on the text content to obtain the entity concepts and the term fragments, and generates an initial entity dictionary. The coordinate calibration unit, connected to the concept extraction unit, receives the initial entity dictionary and the text content, traverses the initial entity dictionary, calculates the starting position and character span of the term fragments in the initial entity dictionary within the text content, parses the text content to output the association relationship between the entity concepts, and generates the triple set.

[0007] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the three-level knowledge graph topology is divided into three hierarchical dimensions, and the knowledge graph mapping module includes: The directory receiving unit receives the course directory structure file input from outside; The first-level unit is connected to the directory receiving unit, receives the course directory structure file, and parses the course directory structure file to extract the course roster, the list of preceding courses, and the list of subsequent courses. The second-level unit connects to the first-level unit, receives the course roster, and extracts the set of chapter modules belonging to the course roster; The third-level unit connects the second-level unit and the semantic feature module, receives the chapter module set and the triple set, splits the chapter module set to generate independent knowledge point modules, and associates the triple set with the independent knowledge point modules.

[0008] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the third-level unit is configured with: The node recording submodule receives the independent knowledge point module, reads the name text and description text contained in the independent knowledge point module, and stores the name text and description text into the memory. The connection submodule connects to the node record submodule, receives the independent knowledge point module, reads the set of triples associated with the independent knowledge point module, applies the set of triples to calculate the preceding knowledge point association and subsequent knowledge point association of the independent knowledge point module, and generates the relation edge. The resource binding submodule connects to the connection submodule, receives the independent knowledge point module and the relationship edge, loads an external media library, extracts the multimedia video segments contained in the multimedia video file that matches the independent knowledge point module from the external media library, and the exercise set contained in the test question bank, generates resource access addresses pointing to the multimedia video segments and the exercise set, and merges the resource access addresses into the relationship edge.

[0009] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the graph mapping module further includes: The rule loading unit reads the externally input resource description framework file and extracts the set of logical judgment conditions within the resource description framework file; The condition calculation unit is connected to the rule loading unit, receives the set of logical judgment conditions, applies the set of logical judgment conditions to scan the entity nodes in the three-level graph topology, and filters out isolated entity nodes that have not established the relationship edge. The dependency generation unit is connected to the condition calculation unit, receives the isolated entity nodes, compares the entity concepts contained in the isolated entity nodes, constructs implicit semantic relationship edges between two isolated entity nodes with the same entity concepts, and outputs the deep semantic dependency relationship based on the implicit semantic relationship edges.

[0010] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the sharing and distribution module includes: The instruction parsing unit receives the call instruction, extracts the query conditions within the call instruction, and parses the query conditions to generate a vector of words to be queried; The semantic conversion unit, connected to the instruction parsing unit, receives the query word vector and compiles the query word vector into a standard query semantic string; The node retrieval unit, connected to the semantic conversion unit, receives the standardized query semantic string, inputs the standardized query semantic string as the query condition into the three-level graph topology, compares the entity nodes according to the deep semantic dependency relationship, outputs the matching target node, reads the target resource location identifier in the matching target node, and sends the target resource location identifier to the downstream terminal node.

[0011] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the instruction parsing unit includes: The interface receiving subunit receives the call instruction sent by the downstream terminal node and extracts the original input text in the call instruction as the query condition; The pattern judgment subunit is connected to the interface receiving subunit, receives the original input text, extracts the natural language text in the original input text, performs word segmentation processing to generate a first vocabulary vector, extracts the structured query code in the original input text and obtains the search keywords in the structured query code to generate a second vocabulary vector; The feature summarization subunit is connected to the pattern judgment subunit, receives the first word vector and the second word vector, and merges the first word vector and the second word vector into the word vector to be queried.

[0012] Furthermore, in the personalized teaching system based on a knowledge graph framework described in this invention, the concept extraction unit is further configured with: The address assignment component receives the text content and assigns a network access address code to the text content. A statement segmentation component is connected to the address assignment component, receives the text content and the network access address code, and segments the text content into independent statement fragments. The annotation mapping component is connected to the statement segmentation component, receives the independent statement fragments, establishes a semantic annotation set corresponding to the independent statement fragments, extracts the words to be matched in the independent statement fragments, applies the external domain knowledge ontology set to match the words to be matched to obtain word meaning codes, stores the word meaning codes in the semantic annotation set, and merges the independent statement fragments with the semantic annotation set to output the initial entity dictionary.

[0013] Furthermore, the personalized teaching system based on a knowledge graph framework described in this invention also includes: The attribute reading module is connected to the shared distribution module, receives the target resource location identifier, locates the corresponding called resource in the external resource library according to the target resource location identifier, and extracts the difficulty feature value and knowledge coverage feature value of the called resource. The path planning module is connected to the attribute reading module, receives the difficulty feature value and the knowledge coverage feature value, sorts and calculates the difficulty feature value and the knowledge coverage feature value to generate a learning path sequence, generates a learning order ranking list based on the learning path sequence, and sends the learning order ranking list to the downstream terminal node.

[0014] Secondly, the present invention provides a personalized teaching method based on a knowledge graph framework, applied to the personalized teaching system based on the knowledge graph framework as described above, comprising: Step 1: Receive external input including text materials, multimedia video files, and teaching resource files such as test question banks; Step 2: Extract the text content from the teaching resource file, parse the text content to obtain entity concepts and terminology fragments, calculate the association between the entity concepts, locate the starting position and character span of the terminology fragments in the text content, and generate a set of triples based on the entity concepts, terminology fragments, association, starting position, and character span. Step 3: Convert the set of triples into entity nodes and relation edges, construct a three-level graph topology, load external ontology rules, apply the external ontology rules to perform conditional deduction on the entity nodes in the three-level graph topology, and extract the deep semantic dependencies between the entity nodes. Step 4: Receive the invocation instruction sent by the downstream terminal node, extract the query conditions in the invocation instruction, apply the query conditions to retrieve the three-level graph topology structure within the deep semantic dependency relationship to obtain the matching target node, extract the target resource location identifier recorded in the matching target node, and send the target resource location identifier to the downstream terminal node.

[0015] Beneficial effects of this invention: The knowledge graph framework-based personalized teaching system of this invention utilizes a semantic feature module to extract entity concepts and terminology fragments from text content, generating a set of triples. The graph mapping module converts the triple set into entity nodes and relation edges, constructing a three-level graph topology. This three-level graph topology breaks away from the independent storage of traditional databases, integrating multi-source heterogeneous data into a network structure, thus solving the technical problem of the difficulty in reusing and sharing teaching resources.

[0016] The graph mapping module of this invention loads external ontology rules to perform conditional deduction on entity nodes, extracting deep semantic dependencies between entity nodes. The sharing and distribution module retrieves the three-level graph topology structure based on these deep semantic dependencies. The knowledge deduction mechanism overcomes the limitations of surface character matching, enabling computer programs to identify underlying concept sets through semantic association rules, thus improving the retrieval accuracy of target resource location identifiers.

[0017] The attribute reading module of this invention extracts the difficulty feature value and knowledge coverage feature value of the called resource based on the target resource location identifier. The path planning module sorts and calculates the difficulty feature value and knowledge coverage feature value to generate a learning path sequence. The system generates a learning order list based on the learning path sequence and sends it to downstream terminal nodes, providing front-end users with a precisely customized learning plan. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on the accompanying drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a personalized teaching method based on a knowledge graph framework according to the present invention. Detailed Implementation

[0020] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In a first aspect, the present invention provides a personalized teaching system based on a knowledge graph framework, comprising: The data acquisition module receives external input, including text materials, multimedia video files, and teaching resource files such as test question banks; The semantic feature module is connected to the data acquisition module. It receives the teaching resource file, extracts the text content in the teaching resource file, parses the text content to obtain entity concepts and term fragments, calculates the association between the entity concepts, locates the starting position and character span of the term fragments in the text content, generates a set of triples based on the entity concepts, term fragments, association, starting position and character span, and outputs the set of triples. The graph mapping module is connected to the semantic feature module, receives the set of triples, converts the set of triples into entity nodes and relation edges, constructs a three-level graph topology, loads external ontology rules, applies the external ontology rules to perform conditional deduction on the entity nodes in the three-level graph topology, and extracts the deep semantic dependencies between the entity nodes. The shared distribution module is connected to the graph mapping module, receives the call instruction sent by the downstream terminal node, extracts the query conditions in the call instruction, applies the query conditions to retrieve the three-level graph topology structure within the deep semantic dependency relationship to obtain the matching target node, extracts the target resource location identifier recorded in the matching target node, and sends the target resource location identifier to the downstream terminal node.

[0022] When receiving teaching resource files, including text materials, multimedia video files, and test question banks, the input port monitors the file stream write status through an asynchronous input / output interface. When a file stream is detected, the file parsing engine is invoked to classify and buffer the files according to media type. During the data processing path of the semantic feature module, the data acquisition module is connected to extract the text content contained in the teaching resource files.

[0023] Natural language processing algorithms are used to perform word segmentation and part-of-speech tagging on the text content, obtaining a set of entity concepts containing noun sequences and a set of term fragments containing technical vocabulary strings. A co-occurrence frequency matrix is ​​calculated between entity concepts, and a relationship is established between two entity concepts whose co-occurrence frequency is greater than a preset association threshold.

[0024] In the location operation, term fragments are retrieved from the text content using a string matching algorithm. The byte offset of the first matching character is recorded as the starting position, and the total number of bytes occupied by the term fragment is calculated as the character span. Subsequently, according to the resource description framework standard, the entity concept, term fragment, relationship, starting position, and character span are concatenated into a set of triplets consisting of subject, predicate, and object, and then output.

[0025] Upon receiving the set of triples, the graph mapping module instantiates the subject and object as entity nodes recorded in the graph database, and instantiates the predicate as relational edges connecting the entity nodes, thereby initializing a three-level graph topology in memory. External ontology rules written in a network ontology language are loaded, and a forward chaining inference algorithm is executed using an inference engine to deduce conditions for the entity nodes contained in the three-level graph topology, generating deep semantic dependencies that include class inheritance and equivalence attribute constraints.

[0026] The shared distribution module connects to the graph mapping module and receives invocation commands from downstream terminal nodes. Regular expressions are used to extract the query conditions encapsulated in the invocation commands, which are then converted into graph query language commands. Within the deep semantic dependencies, the adjacency matrix of the three-level graph topology is traversed, and the matching target node with the highest cosine similarity of its feature vectors is output. The Uniform Resource Locator (URL) stored in the matching target node is extracted as the target resource location identifier, and this identifier is sent to downstream terminal nodes via a Hypertext Transfer Protocol (HTTP) response message.

[0027] The concept extraction unit receives teaching resource files and extracts the text content. It loads an external domain knowledge ontology set and calls ontology files in the format of an integrated medical language system or a clinical concept terminology set. When applying the external domain knowledge ontology set to perform semantic recognition on the text content, it calls a named entity recognition model to calculate the label probability distribution of each character in the text sequence. Sequence segments with predicted probabilities exceeding a set confidence threshold are extracted to obtain entity concepts and terminology segments. The extraction results are stored in a hash table as key-value pairs to generate an initial entity dictionary.

[0028] The coordinate calibration unit connects to the concept extraction unit, receiving the initial entity dictionary and text content. It iterates through the initial entity dictionary using an iterator to extract the key-value pairs from the hash table. A sliding window search is performed on the character array formed by the text content, recording the index of the first character of the term fragment as its starting position and counting the number of characters contained in the term fragment as its character span. A dependency parsing tree is invoked to extract the verb-object structure carried in the text content, outputting the relationships between entity concepts. The index values, character counts, and relationships are packaged into a structure, generating a set of triples.

[0029] In the construction phase of dividing the three-level graph topology structure into three hierarchical dimensions, the data flow within the system architecture of the graph mapping module manifests as a multi-level tree-like instantiation process. The directory receiving unit receives externally input data in the form of a course directory structure file.

[0030] The first-level unit connects to the directory receiving unit, and uses a data parser to deserialize the course directory structure file. It extracts the course roster field from the top-level key-value pair records, and recursively reads the dependent attribute array to extract the list of preceding and subsequent courses, thus completing the creation of the first-level knowledge unit node.

[0031] The second-level unit connects to the first-level unit, which receives the course roster. It retrieves the child node array bound to the course roster, extracts the chapter module set defined in the child node array, and constructs the second-level knowledge unit hierarchy.

[0032] The third-level unit connects the second-level unit and the semantic feature module, receiving the chapter module set and the triple set. A clustering algorithm is used to decompose the text feature vectors contained in the chapter module set, generating independent knowledge point modules corresponding to independent concepts. Based on the globally unique identifiers of the nodes contained in the triple set, the attribute information attached to the triple set is inserted into the attribute table associated with the independent knowledge point module; this completes the operation of associating the triple set with the independent knowledge point module, forming a three-level knowledge unit hierarchy.

[0033] In the specific configuration steps of the third-level unit, the node record submodule receives the independent knowledge point module. When reading the name text and description text contained in the independent knowledge point module, the text stream is encoded and converted, and the name text and description text are stored in the preset data table of the relational database in string form.

[0034] The connection submodule records the nodes that receive independent knowledge point modules and reads the set of triples associated with each independent knowledge point module. Using the dependency direction identifiers contained in the triple set, it calculates the preceding and subsequent knowledge point associations of each independent knowledge point module. It extracts the subject whose predicate direction points to the current receiving module from the triple set to generate the preceding knowledge point association, and extracts the object whose subject is the current receiving module to generate the subsequent knowledge point association; thus, it inserts directed edge data into the graph database boundary table to generate relational edges.

[0035] The resource binding submodule connects to the connection submodule, receiving independent knowledge point modules and relational edges. It loads an external media library and performs a structured query language match search in the metadata table of the external media library. It extracts multimedia video segments from multimedia video files whose names match the names of the independent knowledge point modules, as well as practice question sets from the test question bank. It generates Uniform Resource Locators (URLs) pointing to the multimedia video segments and practice question sets as resource access addresses, and merges these resource access addresses as edge attribute fields into the relational edges, completing the association binding operation.

[0036] The graph mapping module's computation of topological isolated points involves a rigorous logical scan. The rule loading unit calls the file reading stream to load the externally input resource description framework file and extracts the set of logical judgment conditions contained in the resource description framework file. Specifically, it extracts a constraint matrix containing equivalence, subset, or exclusion logic.

[0037] The condition calculation unit and the connection rule loading unit receive a set of logical judgment conditions. When scanning the entity nodes contained in the three-level graph topology using the set of logical judgment conditions, the adjacency list stored in the graph database is traversed to calculate the sum of the in-degree and out-degree of each entity node; node objects with a total degree of zero are filtered out, and isolated entity nodes without established relationship edges are output.

[0038] The dependency generation unit and connection condition calculation unit receive isolated entity nodes. When comparing the entity concepts contained in the isolated entity nodes, the word embedding vectors carried by the two isolated entity nodes are extracted and cosine similarity is calculated. When the similarity score is greater than the set semantic threshold, it is determined that they have the same entity concept. A graph edge insertion algorithm is executed between two isolated entity nodes with the same entity concept to fill the memory topology matrix with values ​​to construct implicit semantic relationship edges. Based on the implicit semantic relationship edges, the graph export interface is called to output the deep semantic dependency relationship.

[0039] For the shared distribution module, its internal data flow logic is as follows: The instruction parsing unit extracts the query conditions carried by the call instruction, performs word segmentation and dimensionality reduction and word frequency weight calculation on the query conditions, and generates a vector of the query vocabulary consisting of floating-point numbers.

[0040] The semantic conversion unit connects to the instruction parsing unit to receive the query term vector. It inputs the query term vector into a pre-compiled semantic mapping template, converts the floating-point features into the standard syntax format required by the protocol, and compiles it into a standardized query semantic string.

[0041] The node retrieval unit connects to the semantic transformation unit to receive the standardized query semantic string. This string is then used as the query condition input to the query engine endpoint configured in the three-level graph topology. Based on deep semantic dependencies, a subgraph isomorphic matching algorithm is executed in the graph database, comparing the attribute fields of entity nodes with the standardized query semantic string to calculate a matching score. The group of nodes whose matching scores reach a preset lower limit is output as the target nodes. The target resource location identifier stored in the target node is read, and a data response packet containing the target resource location identifier is sent to downstream terminal nodes.

[0042] When the parameter extraction process of the instruction parsing unit is executed, the interface receiving subunit starts a listening service to receive the call instructions sent by downstream terminal nodes. By parsing the data payload loaded in the network request body, the original input text entered by the user in the call instruction is extracted as query conditions.

[0043] The pattern judgment subunit connects to the interface receiving subunit to receive the raw input text. A character encoding range detection mechanism extracts the natural language text covered by the raw input text. The natural language text is then input into a Hidden Markov Model for word segmentation, generating a first vocabulary vector composed of word sequences. Simultaneously, a regular expression matching mechanism extracts the structured query code attached to the raw input text. Operators located between selection and condition keywords within the structured query code are extracted to generate search keywords, which are then summarized into a second vocabulary vector.

[0044] The feature summarization subunit connection mode determination subunit receives the first word vector and the second word vector. It performs an array concatenation operation to merge the first word vector and the second word vector into the same one-dimensional array, and fills the end of the array with check bits to merge them into the word vector to be queried.

[0045] Furthermore, the additional component processing mechanism of the concept extraction unit includes: an address assignment component calling a generation algorithm to receive text content; and assigning a fixed-length network access address code to the text content to achieve a unique index of the physical storage location of the text content.

[0046] The statement segmentation component connects to the address assignment component, which receives the text content and the network access address code. Based on the encoding rules set for punctuation marks, it performs regular expression segmentation on the text content, dividing it into an array of strings composed of independent statement fragments.

[0047] The annotation mapping component connects to the statement segmentation component, receiving an array of independent statement fragments. It initializes the semantic annotation set corresponding to each independent statement fragment in memory; it extracts the words to be matched from each independent statement fragment, consisting of nouns and verbs. Applying an external domain knowledge ontology set, it calculates edit distance to match the words to be matched, obtaining the concept encoding features of the nodes in the ontology's concept library to form semantic codes. It calls the database write interface to store the semantic codes in the semantic annotation set, merges the independent statement fragments with semantic annotation sets in the cache, and outputs the initial entity dictionary to the subsequent computation layer.

[0048] The data flow between the attribute reading module and the path planning module manifests as collaborative computational steps. The attribute reading module connects to the shared distribution module to receive the target resource location identifier. Based on the target resource location identifier, it resolves the server address and file path, locating the corresponding called resource in the external resource repository. It extracts the metadata header file bound to the called resource, reading pre-defined difficulty feature values ​​representing the level of cognitive difficulty, and knowledge coverage feature values ​​representing the number of concepts involved.

[0049] The path planning module connects to the attribute reading module to receive difficulty feature values ​​and knowledge coverage feature values. It constructs a fitness evaluation function that includes both difficulty and knowledge coverage feature values; a greedy algorithm is applied to monotonically increase the difficulty and knowledge coverage feature values, generating a learning path sequence ordered by resource identifiers. Based on the learning path sequence, an output data structure is encapsulated to generate a learning order ranking list containing time-series labels. Finally, the network communication interface is called to send the learning order ranking list to downstream terminal nodes.

[0050] Secondly, please refer to Figure 1This invention provides a personalized teaching method based on a knowledge graph framework, applied to the personalized teaching system based on the knowledge graph framework as described above, comprising: Step 1: Receive external input including text materials, multimedia video files, and teaching resource files such as test question banks; Step 2: Extract the text content from the teaching resource file, parse the text content to obtain entity concepts and terminology fragments, calculate the association between the entity concepts, locate the starting position and character span of the terminology fragments in the text content, and generate a set of triples based on the entity concepts, terminology fragments, association, starting position, and character span. Step 3: Convert the set of triples into entity nodes and relation edges, construct a three-level graph topology, load external ontology rules, apply the external ontology rules to perform conditional deduction on the entity nodes in the three-level graph topology, and extract the deep semantic dependencies between the entity nodes. Step 4: Receive the invocation instruction sent by the downstream terminal node, extract the query conditions in the invocation instruction, apply the query conditions to retrieve the three-level graph topology structure within the deep semantic dependency relationship to obtain the matching target node, extract the target resource location identifier recorded in the matching target node, and send the target resource location identifier to the downstream terminal node.

[0051] In this embodiment of the invention, during the operation of a personalized teaching system based on a knowledge graph framework, the system first activates the data acquisition module. The data acquisition module receives externally input teaching resource files, which include text materials, multimedia video files, and a test question bank. Subsequently, the semantic feature module connects to the data acquisition module and receives the aforementioned teaching resource files, extracting the text content contained within them. The semantic feature module parses the text content to obtain entity concepts and terminology fragments, while simultaneously mining the relationships between entity concepts. The semantic feature module locates the starting position and character span of terminology fragments within the text content, and generates a set of triples based on entity concepts, terminology fragments, relationships, starting positions, and character spans, which is then output downstream.

[0052] The graph mapping module connects to the semantic feature module and receives a set of triples. It converts these triples into entity nodes and relation edges, thereby constructing a three-level graph topology within the system. The graph mapping module further loads external ontology rules and applies these rules to logically deduce the entity nodes contained within the three-level graph topology, extracting the hidden deep semantic dependencies between them.

[0053] When the system provides services to end users, the sharing and distribution module connects to the graph mapping module and receives invocation commands from downstream terminal nodes. The sharing and distribution module extracts the query conditions contained in the invocation commands and uses these conditions to retrieve the three-level graph topology within deep semantic dependencies. The sharing and distribution module obtains matching target nodes through node feature comparison, extracts the target resource location identifier recorded within the matching target node, and finally sends the target resource location identifier to the downstream terminal nodes. Upon receiving the target resource location identifier, the downstream terminal nodes can locate and access the corresponding teaching resource files.

Claims

1. A personalized teaching system based on a knowledge graph framework, characterized in that, include: The data acquisition module receives external input, including text materials, multimedia video files, and teaching resource files such as test question banks; The semantic feature module is connected to the data acquisition module. It receives the teaching resource file, extracts the text content in the teaching resource file, parses the text content to obtain entity concepts and term fragments, calculates the association between the entity concepts, locates the starting position and character span of the term fragments in the text content, generates a set of triples based on the entity concepts, term fragments, association, starting position and character span, and outputs the set of triples. The graph mapping module is connected to the semantic feature module, receives the set of triples, converts the set of triples into entity nodes and relation edges, constructs a three-level graph topology, loads external ontology rules, applies the external ontology rules to perform conditional deduction on the entity nodes in the three-level graph topology, and extracts the deep semantic dependencies between the entity nodes. The shared distribution module is connected to the graph mapping module, receives the call instruction sent by the downstream terminal node, extracts the query conditions in the call instruction, applies the query conditions to retrieve the three-level graph topology structure within the deep semantic dependency relationship to obtain the matching target node, extracts the target resource location identifier recorded in the matching target node, and sends the target resource location identifier to the downstream terminal node.

2. The personalized teaching system based on a knowledge graph framework according to claim 1, characterized in that, The semantic feature module includes: The concept extraction unit receives the teaching resource file, extracts the text content within the teaching resource file, loads an external domain knowledge ontology set, applies the external domain knowledge ontology set to perform word meaning recognition on the text content to obtain the entity concepts and the term fragments, and generates an initial entity dictionary. The coordinate calibration unit, connected to the concept extraction unit, receives the initial entity dictionary and the text content, traverses the initial entity dictionary, calculates the starting position and character span of the term fragments in the initial entity dictionary within the text content, parses the text content to output the association relationship between the entity concepts, and generates the triple set.

3. The personalized teaching system based on a knowledge graph framework according to claim 2, characterized in that, The three-level graph topology is divided into three hierarchical dimensions, and the graph mapping module includes: The directory receiving unit receives the course directory structure file input from outside; The first-level unit is connected to the directory receiving unit, receives the course directory structure file, and parses the course directory structure file to extract the course roster, the list of preceding courses, and the list of subsequent courses. The second-level unit connects to the first-level unit, receives the course roster, and extracts the set of chapter modules belonging to the course roster; The third-level unit connects the second-level unit and the semantic feature module, receives the chapter module set and the triple set, splits the chapter module set to generate independent knowledge point modules, and associates the triple set with the independent knowledge point modules.

4. The personalized teaching system based on a knowledge graph framework according to claim 3, characterized in that, The third-level unit is configured with: The node recording submodule receives the independent knowledge point module, reads the name text and description text contained in the independent knowledge point module, and stores the name text and description text into the memory. The connection submodule connects to the node record submodule, receives the independent knowledge point module, reads the set of triples associated with the independent knowledge point module, applies the set of triples to calculate the preceding knowledge point association and subsequent knowledge point association of the independent knowledge point module, and generates the relation edge. The resource binding submodule connects to the connection submodule, receives the independent knowledge point module and the relationship edge, loads an external media library, extracts the multimedia video segments contained in the multimedia video file that matches the independent knowledge point module from the external media library, and the exercise set contained in the test question bank, generates resource access addresses pointing to the multimedia video segments and the exercise set, and merges the resource access addresses into the relationship edge.

5. The personalized teaching system based on a knowledge graph framework according to claim 4, characterized in that, The map mapping module also includes: The rule loading unit reads the externally input resource description framework file and extracts the set of logical judgment conditions within the resource description framework file; The condition calculation unit is connected to the rule loading unit, receives the set of logical judgment conditions, applies the set of logical judgment conditions to scan the entity nodes in the three-level graph topology, and filters out isolated entity nodes that have not established the relationship edge. The dependency generation unit is connected to the condition calculation unit, receives the isolated entity nodes, compares the entity concepts contained in the isolated entity nodes, constructs implicit semantic relationship edges between two isolated entity nodes with the same entity concepts, and outputs the deep semantic dependency relationship based on the implicit semantic relationship edges.

6. The personalized teaching system based on a knowledge graph framework according to claim 5, characterized in that, The shared distribution module includes: The instruction parsing unit receives the call instruction, extracts the query conditions within the call instruction, and parses the query conditions to generate a vector of words to be queried; A semantic conversion unit, connected to the instruction parsing unit, receives the query word vector and compiles the query word vector into a standard query semantic string; The node retrieval unit, connected to the semantic conversion unit, receives the standardized query semantic string, inputs the standardized query semantic string as the query condition into the three-level graph topology, compares the entity nodes according to the deep semantic dependency relationship, outputs the matching target node, reads the target resource location identifier in the matching target node, and sends the target resource location identifier to the downstream terminal node.

7. The personalized teaching system based on a knowledge graph framework according to claim 6, characterized in that, The instruction parsing unit includes: The interface receiving subunit receives the call instruction sent by the downstream terminal node and extracts the original input text in the call instruction as the query condition; The pattern judgment subunit is connected to the interface receiving subunit, receives the original input text, extracts the natural language text in the original input text, performs word segmentation processing to generate a first vocabulary vector, extracts the structured query code in the original input text and obtains the search keywords in the structured query code to generate a second vocabulary vector; The feature summarization subunit is connected to the pattern judgment subunit, receives the first word vector and the second word vector, and merges the first word vector and the second word vector into the word vector to be queried.

8. The personalized teaching system based on a knowledge graph framework according to claim 7, characterized in that, The concept extraction unit is also configured with: The address assignment component receives the text content and assigns a network access address code to the text content. A statement segmentation component is connected to the address assignment component, receives the text content and the network access address code, and segments the text content into independent statement fragments. The annotation mapping component is connected to the statement segmentation component, receives the independent statement fragments, establishes a semantic annotation set corresponding to the independent statement fragments, extracts the words to be matched in the independent statement fragments, applies the external domain knowledge ontology set to match the words to be matched to obtain word meaning codes, stores the word meaning codes in the semantic annotation set, and merges the independent statement fragments with the semantic annotation set to output the initial entity dictionary.

9. The personalized teaching system based on a knowledge graph framework according to claim 8, characterized in that, Also includes: The attribute reading module is connected to the shared distribution module, receives the target resource location identifier, locates the corresponding called resource in the external resource library according to the target resource location identifier, and extracts the difficulty feature value and knowledge coverage feature value of the called resource. The path planning module is connected to the attribute reading module, receives the difficulty feature value and the knowledge coverage feature value, sorts and calculates the difficulty feature value and the knowledge coverage feature value to generate a learning path sequence, generates a learning order ranking list based on the learning path sequence, and sends the learning order ranking list to the downstream terminal node.

10. A personalized teaching method based on a knowledge graph framework, applied to the personalized teaching system based on a knowledge graph framework as described in any one of claims 1 to 9, characterized in that, include: Step 1: Receive external input including text materials, multimedia video files, and teaching resource files such as test question banks; Step 2: Extract the text content from the teaching resource file, parse the text content to obtain entity concepts and terminology fragments, calculate the association between the entity concepts, locate the starting position and character span of the terminology fragments in the text content, and generate a set of triples based on the entity concepts, terminology fragments, association, starting position, and character span. Step 3: Convert the set of triples into entity nodes and relation edges, construct a three-level graph topology, load external ontology rules, apply the external ontology rules to perform conditional deduction on the entity nodes in the three-level graph topology, and extract the deep semantic dependencies between the entity nodes. Step 4: Receive the invocation instruction sent by the downstream terminal node, extract the query conditions in the invocation instruction, apply the query conditions to retrieve the three-level graph topology structure within the deep semantic dependency relationship to obtain the matching target node, extract the target resource location identifier recorded in the matching target node, and send the target resource location identifier to the downstream terminal node.

Images