CN-122023070-A - Digital teacher knowledge modeling and content generation method based on artificial intelligence

Abstract

The invention discloses a digital teacher knowledge modeling and content generating method based on artificial intelligence, which comprises the following steps of constructing a teaching knowledge map and forming a capability map, generating a node vector, an edge vector and a path mode set, inputting the node vector, the edge vector and the path mode set into a Hopfield network to perform pre-training, outputting a corresponding teaching knowledge path sequence and association weight thereof based on teaching task information extracted from the teaching knowledge map and the capability map, generating a teaching content organization plan, generating a structured teaching content comprising teaching explanation texts, teaching example texts, interactive training tasks and evaluating question contents according to the teaching content organization plan, and performing incremental adjustment on the memory state of the Hopfield network and the external memory of a neural map smart machine. The invention integrates knowledge modeling, intelligent memory and content generation capability, and is suitable for intelligent education scenes.

Inventors

• JIANG FACHENG

Assignees

• 南京恒点信息技术有限公司

Dates

Publication Date

20260512

Application Date

20250923

Claims (7)

1. 1. The digital teacher knowledge modeling and content generating method based on artificial intelligence is characterized by comprising the following steps: Collecting teacher teaching data, carrying out structural processing on the teaching data, constructing a teaching knowledge graph, and forming a capability graph based on capability attribute relations among knowledge points in the teaching knowledge graph; embedding and encoding nodes and side relations in the teaching knowledge graph and the capability graph to generate node vectors, side vectors and path mode sets; Inputting the node vector, the edge vector and the path mode set into a Hopfield network, performing pre-training, initializing a memory state and constructing an association weight matrix, and storing the association weight matrix as a stable state according to a set energy function to form a mapping of a teaching knowledge path and the stable state; based on teaching task information extracted from a teaching knowledge spectrum and an energy spectrum, coding the teaching task information into an input vector, matching the input vector with a stored node vector and an edge vector, inputting the input vector into a Hopfield network, carrying out state iteration according to the mapping, and outputting a corresponding teaching knowledge path sequence and association weights thereof; Inputting the teaching knowledge path sequence and the association weight into the nerve image machine, and sequentially accessing the differentiable external memory by the controller, and executing reading and writing operations to generate a teaching content organization plan; Generating a structured teaching content comprising teaching explanation texts, teaching example texts, interactive exercise tasks and evaluating question contents according to a teaching content organization plan; User interaction data and learning feedback data are collected in the output process of teaching contents, the feedback data are used as updating conditions, and incremental adjustment is carried out on the memory state of the Hopfield network and the external memory of the nerve pattern-drawing machine.
2. 2. The method for modeling and generating content based on digital teacher knowledge based on artificial intelligence according to claim 1, wherein the structuring process comprises data cleaning, term normalization, sequence labeling, entity extraction, relationship construction and map generation.
3. 3. The method for modeling and generating digital teacher knowledge based on artificial intelligence according to claim 1, wherein the generation of the node vector, the edge vector and the path pattern set comprises the steps of extracting structural text features and semantic tag information based on each node in a teaching knowledge graph and a capability graph, performing feature aggregation through a graph neural network in combination with an adjacency relation to generate a node vector corresponding to a graph structure, classifying the edge relation connecting each node according to semantic types, generating the edge vector based on a preset relation embedding rule, enabling each edge to be associated with the node connected with each edge in a vector space, combining the nodes and the edge vector according to the topological sequence of continuous multi-hop paths in the graph to form a path level representation, and constructing the path level representation into the path pattern set.
4. 4. The method for modeling and generating content based on digital teacher knowledge based on artificial intelligence according to claim 1, characterized in that the formation of the mapping of the teaching knowledge path and the steady state includes: each teaching knowledge path in the path mode set is encoded into a path state vector with fixed dimension to form a path state vector set, and standardized preprocessing is carried out on the path state vector set, wherein the preprocessing comprises vector direction normalization and feature redundancy noise reduction operation; Loading the preprocessed path state vector set to the Hopfield network, initializing a memory state space of the network, distributing corresponding network state nodes for each path state vector, and establishing a state activation map; constructing a symmetrical association weight matrix based on semantic distance and local difference characteristics among path state vectors, wherein the weight matrix represents association strength among state nodes; the initialized network memory state space and the built symmetrical association weight matrix are input into a Hopfield network together, state evolution operation is executed, and state node values are adjusted according to the current network memory state and weight relation in each evolution process; Setting an energy function for measuring the stability of the current state of the network, integrally calculating the state energy of the network according to the current values of all state nodes and the corresponding association weights of the state nodes by the energy function, and taking the change trend of the energy function value as a criterion for judging whether evolution converges or not; In the continuous evolution process, when the falling rate of the energy function value is lower than a set threshold value and keeps stable and unchanged, judging that the Hopfield network reaches a stable state; Binding the stable states corresponding to the final convergence of each path state vector, and establishing a one-to-one mapping relation between the path states and the stable states.
5. 5. The method for modeling and generating content of digital teacher knowledge based on artificial intelligence according to claim 1, characterized in that the output of the corresponding teaching knowledge path sequence and its associated weights includes: The method comprises the steps of carrying out structural definition on teaching task information, wherein the teaching task information comprises target teaching knowledge points, current student learning states, student response feedback characteristics, target teaching difficulty intervals, teaching scene identifiers and time window parameters, and coding the teaching task information to generate teaching task input vectors; Performing similarity calculation on the teaching task input vector and each path state vector in the path state vector set, wherein the similarity calculation is based on vector direction cosine values, main feature dimension matching ratios and teaching sequence constraint consistency, and a path matching score set is obtained; the path matching score sets are subjected to path screening, the path state vectors with the scores ranked at the front are selected according to the score ranking to form candidate path state vector sets, the corresponding path matching scores are normalized, and the association weight sets of the candidate paths are generated; The teaching task input vector is used as the initial network memory state input of the Hopfield network, the state node corresponding to each path state vector in the candidate path state vector set is loaded into the network state space, and the state node is marked as an initial activation state node set; Carrying out state evolution control on the association weight set of the candidate path, wherein the state evolution operation adopts an asynchronous updating mode, sequentially selecting state nodes according to the association weight sequence in each round to carry out state updating, and transmitting an updating result to the state nodes directly connected with the association weight matrix in the Hopfield network after the updating is finished; After each state evolution operation is finished, recording the change amplitude between the current state vector and the previous state vector of the Hopfield network, and simultaneously recording the change amplitude between the current energy function value and the previous energy function value of the Hopfield network, and judging that the Hopfield network reaches a stable state if the two change amplitudes in a plurality of continuous state evolution rounds are lower than a set threshold; Matching the state node set in the active state in the final steady state of the Hopfield network with the established one-to-one mapping relation between the path state vector and the steady state, outputting a teaching knowledge path sequence, and sequencing the teaching knowledge path sequence according to the association weight set to form an ordered teaching knowledge path sequence corresponding to teaching task information and the association weight set thereof.
6. 6. The method for modeling and generating content based on digital teacher knowledge based on artificial intelligence according to claim 1, characterized in that the generation of the teaching content organization plan includes: inputting a teaching knowledge path sequence and an association weight set thereof into a nerve graphing machine, initializing a controller state vector, an address pointer vector and an external differentiable memory matrix, wherein the external differentiable memory matrix is divided into a content slot area, a structure slot area and a path control index area; For each path in the teaching knowledge path sequence, extracting a structural identifier, a knowledge point label and a teaching context type, and generating a defined access candidate address set through a path control index; The controller sequentially receives each path vector, calls the current controller state and the input vector to calculate a reading head addressing vector, generates a reading weight vector based on the similarity with each address vector in the candidate address set, and executes weighted reading operation in a content slot area to obtain teaching content representation corresponding to the path vector; The controller generates write-in control parameters based on the current input path vector and the read teaching content representation in a combined way, wherein the write-in control parameters comprise a write weight vector, an erasure vector, an addition vector and a structure write-in vector, and the structure write-in vector describes the task type, the knowledge point number and the sequence in the path sequence of the current path; the controller sequentially accesses the content slot area and the structure slot area in the candidate address set range, performs double-channel writing operation, completes content updating according to the writing weight vector in the content slot area, writes teaching structure information corresponding to the structure writing vector in the structure slot area, and records path index position information; after each time of read-write operation, the controller updates the state vector and the address pointer; After all the paths are processed, the controller extracts path indexes, structure labels and teaching sequences from the structure groove area, and reconstructs a teaching content organization structure according to the original path sequence by combining the segment coding contents in the content groove area to generate a teaching content organization plan; The tutorial organization plan is stored in an output buffer of an external differentiable memory matrix.
7. 7. The method for modeling and generating digital teacher knowledge based on artificial intelligence according to claim 1, wherein the process of collecting user interaction data and learning feedback data in the output process of teaching content and incrementally adjusting the memory state of the Hopfield network and the external differentiable memory matrix of the neuro-graphic machine by using the learning feedback data as the update condition comprises the following steps: Collecting user interaction data and learning feedback data in the output process of teaching contents, wherein the learning feedback data comprises student question text, question answering results, question answering time consumption, operation paths and interaction behavior records; Coding the learning feedback data to construct a feedback vector, wherein the feedback vector comprises teaching deviation degree, knowledge point mastering estimated value and path adjustment factor; Performing feature comparison on the feedback vector and the original teaching task input vector to generate a task correction vector; inputting a task correction vector into a Hopfield network, taking the task correction vector as an initial memory state in the Hopfield network, executing an incremental state evolution operation, activating a newly-added state node associated with the task correction vector, and updating an association weight matrix in the Hopfield network; Synchronously inputting a feedback vector into the neural turing machine, and accessing a structural slot area and a content slot area in the external differentiable memory matrix by a controller in the neural turing machine according to the task correction vector, positioning an address position to be updated and generating a write-in control parameter; The controller performs an incremental write operation in the structure slot region and the content slot region, including generating a write weight vector, an erase vector, and an add vector, and performs a content overwrite operation only on the structure tag location associated with the feedback vector, with the remaining regions maintaining the memory state unchanged.

Description

Digital teacher knowledge modeling and content generation method based on artificial intelligence Technical Field The invention relates to the technical field of artificial intelligence, in particular to a digital teacher knowledge modeling and content generation method based on artificial intelligence. Background Along with the development of intelligent education, virtual teachers and personalized learning systems, digital teachers are built by using artificial intelligence to assist teaching, and the digital teachers become key directions for improving teaching efficiency and teaching adaptability. Currently, mainstream intelligent teaching methods generally rely on rule-based knowledge graphs or pre-trained language models for content pushing and answering responses. The method mainly carries out teaching content matching through a static knowledge structure, and lacks associative ability and a dynamic organization mechanism, so that response is stiff and adaptability is poor when the method faces non-standardized problems, cross-subject connection or student personality differences. On one hand, the existing knowledge modeling method mostly adopts an explicit graph structure or semantic embedded matching mode, lacks the capability of carrying out structural memory and associative expression on a teacher complex knowledge system, cannot actively recall a complete teaching path chain from incomplete input, and on the other hand, a traditional content generation system is difficult to effectively fuse teaching task contexts, student states and teacher styles, lacks a generation mechanism with memory read-write control, and causes content presentation fragmentation and difficult to dynamically adapt to a teaching process. Especially under the continuous interaction scene, the current system is difficult to adjust the generation strategy in real time according to student feedback, and the real teaching evolution and self-adaptation capability cannot be realized. Therefore, how to provide a digital teacher knowledge modeling and content generation method based on artificial intelligence is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a digital teacher knowledge modeling and content generation method based on artificial intelligence, which comprehensively utilizes a structured knowledge graph construction, an associative memory network and a differentiable memory generation mechanism, and describes the associative recall of a teaching knowledge path through a Hopfield network in detail, and the dynamic organization and generation process of teaching contents is completed through an improved neural graphics machine. According to the embodiment of the invention, the digital teacher knowledge modeling and content generating method based on artificial intelligence comprises the following steps: collecting teacher teaching data, wherein the teaching data comprises teaching outline texts, classroom explanation records, homework design contents, student answer feedback and evaluation result data, carrying out structural processing on the teaching data, constructing a teaching knowledge graph comprising teaching knowledge points, teaching strategies, teaching examples and evaluation rules, and forming a capability graph based on capability attribute relations among all knowledge points in the teaching knowledge graph; embedding and encoding nodes and side relations in the teaching knowledge graph and the capability graph to generate node vectors, side vectors and path mode sets; Inputting the node vector, the edge vector and the path mode set into a Hopfield network, performing pre-training, initializing a memory state and constructing an association weight matrix, and storing the association weight matrix as a stable state according to a set energy function to form a mapping of a teaching knowledge path and the stable state; based on teaching task information extracted from a teaching knowledge spectrum and an energy spectrum, coding the teaching task information into an input vector, matching the input vector with a stored node vector and an edge vector, inputting the input vector into a Hopfield network, carrying out state iteration according to the mapping, and outputting a corresponding teaching knowledge path sequence and association weights thereof; Inputting the teaching knowledge path sequence and the association weight into the nerve image machine, and sequentially accessing the differentiable external memory by the controller, and executing reading and writing operations to generate a teaching content organization plan; Generating a structured teaching content comprising teaching explanation texts, teaching example texts, interactive exercise tasks and evaluating question contents according to a teaching content organization plan; User interaction data and learning feedback data are collected in the output process of teaching contents,