An intelligent virtual auxiliary teaching knowledge demand analysis method

Abstract

This invention belongs to the field of virtual assisted teaching technology, and particularly relates to an intelligent virtual assisted teaching knowledge needs analysis method. It includes the following steps: extracting teaching keywords to establish a keyword database; extracting teaching corpora; defining element features, completing feature annotation, and completing element identification; extracting element pairs, establishing element classification features, extracting element relationships, and completing classification; establishing a historical assisted teaching knowledge graph; acquiring historical information, extracting historical teaching data, and outputting an element attention output matrix based on a multi-head self-attention module; element path and element extraction; this invention achieves intelligent and efficient chemical knowledge needs analysis based on historical data of training subjects, and is used to provide training subjects with recommendations for teaching content and teaching needs, as well as an automatic planning method for teaching courses.

Description

Technical Field

[0001] This invention belongs to the field of virtual-assisted teaching technology, and in particular relates to an intelligent virtual-assisted teaching knowledge demand analysis method. Background Technology

[0002] Virtual-assisted teaching technology has seen rapid development in training and teaching scenarios in the high-end equipment manufacturing industry in recent years. It can effectively simplify the teaching process and improve teaching effectiveness. It can significantly enhance students' understanding of equipment application scenarios and structural principles when they cannot access actual equipment. Unlike classroom teaching, virtual-assisted teaching mainly relies on students' active or passive operation to control the teaching process. Instructors often only act as external assistants. Compared to on-site guidance and targeted teaching planning, students undergoing virtual scenario training often find it difficult to clearly and effectively plan and manage their overall teaching knowledge points and needs. With the rapid iteration and development of virtual teaching technology and equipment, the choices and content faced by students are becoming increasingly complex. While this further improves the simulation capabilities of virtual reality teaching and enriches teaching content, it also further increases the difficulty of managing and planning students' teaching knowledge needs in more choices and situations. Summary of the Invention

[0003] The purpose of this invention is to provide a method for intelligent and efficient chemical knowledge demand analysis based on the historical data of trainees in various types of virtual assisted teaching scenarios, which can be used to provide trainees with recommendations for teaching content and teaching needs and automatic planning of teaching courses.

[0004] To achieve the above objectives, the present invention adopts the following technical solution.

[0005] A method for analyzing knowledge needs in intelligent virtual-assisted teaching includes the following steps:

[0006] Step a, Intelligent Virtual Assisted Teaching Data Extraction, refers to extracting teaching keywords to establish a keyword database, establishing an original sentence database based on teaching content materials, and extracting a teaching corpus;

[0007] Step b: Extract intelligent virtual-assisted teaching elements, define element features, obtain teaching corpus data, and divide it into element extraction training corpus and element extraction test corpus. Based on the aforementioned element feature definition, complete the feature annotation of the training set and test set, establish a CRF conditional random field model to complete element recognition, and obtain intelligent virtual-assisted teaching elements.

[0008] Step c: Intelligent virtual-assisted teaching element relationship extraction, including extracting element pairs and establishing element classification features; completing element relationship classification based on classification algorithms;

[0009] Step d: Knowledge subgraph acquisition. Construct a vertex set with all elements, establish an edge set with the relationships between all elements, and build a historical auxiliary teaching knowledge graph.

[0010] Step e: Intelligent virtual-assisted teaching evaluation sequence modeling

[0011] The system acquires historical information on intelligent virtual-assisted teaching and extracts historical teaching data. It uses word vector extraction technology to extract element word vectors, and then embeds them together with the element teaching quality comprehensive score encoded using one-hot encoding to obtain an embedding vector that integrates the historical evaluation of intelligent virtual-assisted teaching. Based on the multi-head self-attention module, the embedding vector is mapped to multiple subspaces for self-attention calculation, and the element attention output matrix is ​​output.

[0012] Step f: Element path and element extraction based on knowledge graph-based teaching path set ranking.

[0013] Based on the comprehensive evaluation results of teaching quality of elements, a set of teaching paths is extracted; a bidirectional long short-term memory network is used to obtain the path feature matrix; the element attention output matrix and the path feature matrix are used as the initial matrices, and their product is used as the element demand weight. Based on the ranking algorithm, the element path and corresponding element information that make the element demand weight the highest are determined.

[0014] For a further improvement or preferred implementation of the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method, the comprehensive score of the teaching quality of the specific elements is normalized using a ten-point system.

[0015] A further improvement or preferred implementation of the aforementioned intelligent virtual assisted teaching knowledge demand analysis method, wherein the keywords in step a are extracted based on the intelligent virtual assisted teaching outline materials, and the keywords refer to the names of knowledge points, key equipment or key operations that the teaching objects need to understand in the process of intelligent virtual assisted teaching; the original sentence database is established from the sentences in the intelligent virtual assisted teaching content flow materials, and the teaching corpus is obtained by removing stop words from the original sentence database text, truncating long sentences, and performing word segmentation based on the teaching keywords in the keyword database.

[0016] In a further improved or preferred implementation of the aforementioned intelligent virtual assisted teaching knowledge demand analysis method, step b includes the following part-of-speech features: nouns (including nouns, numerals, and classifiers) and pronouns used to represent the operation objects, devices, or knowledge points involved in intelligent virtual assisted teaching; predicates (verbs, adjectives), relational words (conjunctions, prepositions), and adjunctive words (distinguishing words, adverbs) used to represent the operation content or interaction path of intelligent virtual assisted teaching; and auxiliary words (particles, modal particles, interjections, and onomatopoeic words) that generally do not have specific meanings; the part-of-speech features are represented by different codes or symbols.

[0017] The positional features of elements are described using four positional relationships, which are defined as the first element at the beginning of the sentence, the last element at the end of the sentence, the middle element in the middle of the sentence, and the independent element that exists as a sentence on its own. The part-of-speech features of elements are represented by different codes or identifiers.

[0018] In a further improvement or preferred implementation of the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method, in step b, the element features include element part-of-speech features, element position features, element marker features, and element nesting features; element part-of-speech features refer to the textual part-of-speech of the element; element marker features refer to the presence of marker attributes between the element and element objects in the teaching corpus; element nesting features refer to the ability of the element to be further segmented from element objects in the teaching corpus.

[0019] Further improvements or preferred implementation methods for the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method further divide the element characteristics into knowledge point characteristics, equipment characteristics, and teaching skill characteristics.

[0020] In a further improvement or preferred implementation of the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method, step c, extracting element pairs, refers to: acquiring a teaching corpus, removing statements that do not contain element objects or contain only a single element object, and extracting any two element objects from the remaining statements that contain multiple element objects to form element pairs.

[0021] A further improvement or preferred implementation of the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method includes the following: In step c, the element classification features include element pair features, element pair attribute features, and element relationship features; element pair features include element pair length, element pair distance, and element pair type; element pair attribute features include: the element pair's identity in the sentence, the average distance between the element pair and the core dimension, etc.; element relationship features include: dependency relationship, which means that element object A is the basis for the existence of another element object B; extension relationship, which means that the appearance of element object A will lead to element object B; inclusion relationship, which means that element object A is a part of element object B; parallel relationship, which means that element object A and element object B are essentially equal; and other relationships, which means that there is no relationship between the two elements or that do not belong to other types of relationships mentioned above.

[0022] Further improvements or preferred implementation methods for the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method, therefore step e specifically includes:

[0023] Acquire historical information on intelligent virtual-assisted teaching and extract historical teaching data; including: elements from the knowledge graph involved in the history of intelligent virtual-assisted teaching. Comprehensive score of teaching quality for each specific element ;

[0024] Word vectors are extracted using word vector extraction techniques and then jointly embedded with the element of teaching quality comprehensive score, which is encoded using one-hot encoding, to obtain an embedding vector that integrates intelligent virtual-assisted teaching historical evaluation. , ,in This refers to the element number, where m is the total number of elements.

[0025] The embedding vector is based on a multi-head self-attention module. Self-attention computation is performed by mapping to multiple subspaces, and the query vector, key vector, and value matrix of each attention module are embedded using subspace vectors. Replacement, self-attention weights To determine the intensity of element requirements during intelligent virtual-assisted teaching, the output of the multi-head self-attention module represents the element requirements.

[0026] Obtain the attention matrix Subspace attention output Finally, the attention outputs of all subspaces are concatenated to obtain the element attention output matrix. , This is the subspace weight matrix.

[0027] A further improvement or preferred implementation of the aforementioned intelligent virtual-assisted teaching knowledge demand analysis method, wherein step f specifically includes:

[0028] Based on the comprehensive score of teaching quality of each element, the learning progress of each element is defined. These represent the levels of mastery of key knowledge points: high, medium, and low; according to the learning progress. The preferred order is to select teaching elements as the starting point of the path. After determining a certain element as the starting point, related elements are extracted and added to the path nodes in the order of dependency, extension, inclusion, and parallel relationships. If no next element can be found, the remaining elements are selected according to the learning progress. The order of selection is preferred to choose the next path starting element, and the above process is repeated until all elements have been added to the path to obtain a complete teaching path set. The teaching time is controlled by adjusting the number of path nodes.

[0029] Preset teaching duration The following teaching path As input, a bidirectional long short-term memory network is used to train the teaching path through sequence encoding, and the hidden layer state results are extracted. Max pooling is used to extract the most salient features of each teaching path. The path feature matrix is ​​obtained. ;

[0030] Element attention output matrix and path feature matrix These are the initial matrices, and their product. As the factor demand weight, based on the Bayesian personalized ranking algorithm, the factor path and corresponding factor information that give the highest factor demand weight are determined. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the main process of the intelligent virtual-assisted teaching knowledge needs analysis method;

[0032] Figure 2 This is a schematic diagram of the corpus and knowledge graph creation and processing flow;

[0033] Figure 3 This is a schematic diagram of a multi-head self-attention layer module. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0035] Historical supplementary teaching data refers to the recorded data of students' corresponding teaching content and teaching quality evaluation during the intelligent virtual supplementary teaching process. One of the objectives of this invention is to provide a solution that can achieve optimal learning efficiency while meeting the requirements of the intelligent virtual supplementary teaching process, based on long-term and complex historical teaching status analysis. Figure 1 As shown below, the specific steps will be explained in detail.

[0036] Step a: Extraction of intelligent virtual-assisted teaching data

[0037] like Figure 2As shown, based on the intelligent virtual-assisted teaching content requirements of relevant subjects, teaching keywords are extracted from teaching materials such as textbooks and syllabi to establish a keyword database. The teaching keywords refer to the names of knowledge points, key equipment, or key operations that the learners need to understand during the intelligent virtual-assisted teaching process. Based on the intelligent virtual-assisted teaching content flow of relevant subjects, an original sentence database is established using sentences from virtual-assisted teaching content materials such as teaching content descriptions, auxiliary teaching materials, operation process texts, and auxiliary teaching instruction texts. Stop words in the original sentence database text are removed, long sentences are truncated, and word segmentation is performed based on the teaching keywords in the keyword database to obtain a teaching corpus.

[0038] Step b: Extraction of intelligent virtual-assisted teaching elements

[0039] To determine the specific knowledge points and teaching needs related elements in the teaching process, as well as their correlation with students' historical learning data, we first define element features, including element part-of-speech features, element position features, element marker features, and element nesting features.

[0040] Element part-of-speech features refer to the textual part-of-speech of elements, including: nouns (including nouns, numerals, and classifiers) and pronouns used to represent the objects, devices, or knowledge points involved in intelligent virtual assisted teaching; predicates (verbs, adjectives), relational words (conjunctions, prepositions), and adverbs (distinguishing words, adverbs) used to represent the operation content or interaction path of intelligent virtual assisted teaching; and auxiliary words (particles, modal particles, interjections, and onomatopoeic words) that usually do not have specific meanings. Element part-of-speech features are represented by different codes or symbols.

[0041] Element positional features refer to the positional relationship of elements in sentences within the teaching corpus. Four positional relationships are used for description, which are defined as the first element at the beginning of the sentence, the last element at the end of the sentence, the middle element in the middle of the sentence, and an independent element existing as a sentence on its own. Element part-of-speech features are represented by different codes or markers.

[0042] Element marker features refer to the marker attributes that exist between elements and element objects in the teaching corpus. That is, the appearance of a certain element will lead to another element with a certain probability. For example, in the teaching process, when "flipping the XX switch" is involved, "flipping" is the marker word for the knowledge point of XX switch. The value of element marker features is 0 or 1, where 0 represents false and 1 represents true.

[0043] In particular, to meet the more personalized needs of the intelligent virtual-assisted teaching recommendation process, the various element characteristics can be constructed in the form of multiple tags. For example, the element marker characteristics can be further divided into knowledge point marker characteristics, equipment marker characteristics, and teaching skill marker characteristics.

[0044] The nested feature of elements refers to the fact that elements can be further segmented from element objects in the teaching corpus, such as "electric motor" and "magnetic" in "permanent magnet motor"; the nested feature of elements takes the value of 0 or 1, where 0 represents false and 1 represents true;

[0045] Acquire teaching corpus data and divide it into element extraction training corpus and element extraction test corpus. Based on the aforementioned element feature definition, complete the feature annotation of the training set and test set, establish a CRF conditional random field model to complete element recognition, and obtain intelligent virtual auxiliary teaching elements.

[0046] Step c: Extraction of Relationships Between Intelligent Virtual-Assisted Teaching Elements

[0047] Obtain the teaching corpus, remove statements that do not contain element objects or contain only a single element object, and extract any two element objects from the remaining statements that contain multiple element objects to form an element pair;

[0048] Establish element classification features consisting of element-to-feature characteristics, element-to-attribute characteristics, and element-relationship characteristics.

[0049] Feature pair characteristics include feature pair length, feature pair distance, and feature pair type;

[0050] Feature pair attribute characteristics include: the identity characteristics of the feature pair in the sentence, the average distance between the feature pair and the core dimension, etc.

[0051] The characteristics of element relationships include: dependency relationship, which means that element object A is the basis for the existence of another element object B; derivative relationship, which means that the appearance of element object A will lead to element object B; inclusion relationship, which means that element object A is a part of element object B; parallel relationship, which means that element object A and element object B are essentially equal; other relationships, which means that there is no relationship between the two elements or other types of relationships that do not belong to the aforementioned relationships.

[0052] Element relation extraction is performed by acquiring teaching corpus data and dividing it into element relation training corpus and element relation test corpus. Feature annotation of the training and test sets is completed based on element features. The element relation classification training of the training set is completed based on support vector machine classification algorithm or K nearest neighbor classification algorithm. The training model is tested using the element relation test corpus.

[0053] Step d: Knowledge subgraph acquisition, constructing a vertex set from all elements. Establish edge sets based on the relationships between all elements. Establish a knowledge graph for history-assisted teaching ;

[0054] Step e: Intelligent virtual-assisted teaching evaluation sequence modeling

[0055] Acquire historical information on intelligent virtual-assisted teaching and extract historical teaching data; including: elements from the knowledge graph involved in the history of intelligent virtual-assisted teaching. Comprehensive score of teaching quality for each specific element ;

[0056] Word vectors are extracted using word vector extraction techniques and then jointly embedded with the element of teaching quality comprehensive score, which is encoded using one-hot encoding, to obtain an embedding vector that integrates intelligent virtual-assisted teaching historical evaluation. , ;

[0057] in This refers to the element number, where m is the total number of elements.

[0058] like Figure 3 As shown, the multi-head self-attention module will embed the vector The query vectors are mapped to multiple subspaces for self-attention computation, and each attention module's query vector is used for this purpose. Key vector and value matrix Using subspace embedding vectors Replacement, self-attention weights To determine the intensity of element requirements during intelligent virtual-assisted teaching, the output of the multi-head self-attention module represents the element requirements.

[0059] Obtain the attention matrix Solve for the subspace attention output Finally, the attention outputs of all subspaces are concatenated to obtain the element attention output matrix. , This is the subspace weight matrix;

[0060] Step f: Element path and element extraction based on knowledge graph-based teaching path set ranking.

[0061] The order of the teaching path and the types of specific elements within the path affect the content of the recommended knowledge points and the progress of mastering these elements during the teaching process. Priority analysis and sorting are required.

[0062] First, based on the comprehensive score of the teaching quality of each element, the learning progress of each element is defined. These represent the levels of mastery of key knowledge points: high, medium, and low; according to the learning progress. The preferred order is to select teaching elements as the starting point of the path. After determining a certain element as the starting point, related elements are extracted and added to the path nodes in the order of dependency, extension, inclusion, and parallel relationships. If no next element can be found, the remaining elements are selected according to the learning progress. The order of selection is preferred to choose the next path starting element, and the above process is repeated until all elements have been added to the path to obtain a complete teaching path set. The teaching time is controlled by adjusting the number of path nodes.

[0063] Secondly, the pre-set teaching duration The following teaching path As input, a bidirectional long short-term memory network is used to train the teaching path through sequence encoding, and the hidden layer state results are extracted. Max pooling is used to extract the most salient features of each teaching path. The path feature matrix is ​​obtained. ;

[0064] Finally, the feature attention output matrix is ​​used. and path feature matrix These are the initial matrices, and their product. As the factor demand weight, based on the ranking algorithm, the factor path and corresponding factor information that give the highest factor demand weight are determined.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for analyzing knowledge needs in intelligent virtual-assisted teaching, characterized in that, Includes the following steps: Step a, Intelligent Virtual Assisted Teaching Data Extraction, refers to extracting teaching keywords to establish a keyword database, establishing an original sentence database based on teaching content materials, and extracting a teaching corpus; Step b: Extract intelligent virtual-assisted teaching elements, define element features, obtain teaching corpus data, and divide it into element extraction training corpus and element extraction test corpus. Based on the aforementioned element feature definition, complete the feature annotation of the training set and test set, establish a CRF conditional random field model to complete element recognition, and obtain intelligent virtual-assisted teaching elements. Step c: Extraction of intelligent virtual-assisted teaching element relationships, including extracting element pairs and establishing element classification features; Element relationship classification is performed based on a classification algorithm; Step d: Knowledge subgraph acquisition. Construct a vertex set with all elements, establish an edge set with the relationships between all elements, and build a historical auxiliary teaching knowledge graph. Step e: Intelligent virtual-assisted teaching evaluation sequence modeling The system acquires historical information on intelligent virtual-assisted teaching and extracts historical teaching data. It uses word vector extraction technology to extract element word vectors, and then embeds them together with the element teaching quality comprehensive score encoded using one-hot encoding to obtain an embedding vector that integrates the historical evaluation of intelligent virtual-assisted teaching. Based on the multi-head self-attention module, the embedding vector is mapped to multiple subspaces for self-attention calculation, and the element attention output matrix is ​​output. Step f: Element path and element extraction based on knowledge graph-based teaching path set ranking. Based on the comprehensive score of teaching quality of each element, a set of teaching paths is extracted; a bidirectional long short-term memory network is used to obtain the path feature matrix. Using the element attention output matrix and the path feature matrix as initial matrices, and their product as element demand weights, the element path and corresponding element information that result in the highest element demand weights are determined based on a ranking algorithm.

2. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, The comprehensive evaluation of teaching quality for the specific elements is normalized using a ten-point scale.

3. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, In step a, the keywords are extracted from the intelligent virtual-assisted teaching outline materials. Keywords refer to the names of knowledge points, key equipment, or key operations that the learners need to understand during the intelligent virtual-assisted teaching process. The original sentence database is established from the sentences in the intelligent virtual-assisted teaching content flow materials. The teaching corpus is obtained by removing stop words from the original sentence database text, truncating long sentences, and performing word segmentation based on the teaching keywords in the keyword database.

4. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, In step b, the part-of-speech features of the elements include: nouns (including nouns, numerals, and classifiers) and pronouns used to represent the objects, devices, or knowledge points involved in intelligent virtual assisted teaching; predicates (verbs, adjectives), relational words (conjunctions, prepositions), and adverbs (distinguishing words, adverbs) used to represent the operation content or interaction path of intelligent virtual assisted teaching; and auxiliary words (auxiliary words, modal particles, interjections, and onomatopoeic words) that generally do not have specific meanings; the part-of-speech features of the elements are represented by different codes or symbols; The positional features of elements are described using four positional relationships, which are defined as the first element at the beginning of the sentence, the last element at the end of the sentence, the middle element in the middle of the sentence, and the independent element that exists as a sentence on its own. The part-of-speech features of elements are represented by different codes or identifiers.

5. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, In step b, the element features include element part-of-speech features, element location features, element marker features, and element nesting features; element part-of-speech features refer to the textual part-of-speech of the element; element marker features refer to the presence of marker attributes between the element and the element objects in the teaching corpus; element nesting features refer to the fact that the element can be further segmented from the element objects in the teaching corpus.

6. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 5, characterized in that, The key features are further divided into knowledge point features, equipment features, and teaching skill features.

7. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, Step c, extracting element pairs, refers to: acquiring the teaching corpus, removing statements that do not contain element objects or contain only a single element object, and extracting any two element objects from the remaining statements containing multiple element objects to form element pairs; acquiring the teaching corpus data and dividing it into an element relationship training corpus and an element relationship test corpus; completing feature annotation for the training and test sets based on element features; completing element relationship classification training and optimization for the training set based on a support vector machine classification algorithm or a K-nearest neighbor classification algorithm; and using the element relationship test corpus to test the trained model.

8. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, The feature classification features in step c include feature pair features, feature pair attribute features, and feature relationship features; feature pair features include feature pair length, feature pair distance, and feature pair type. Element pair attribute characteristics include: the identity characteristics of the element pair in the statement, the average distance between the element pair and the core dimension, etc.; element relationship characteristics include: dependency relationship, which means that element object A is the basis for the existence of another element object B; derivative relationship, which means that the appearance of element object A will lead to element object B; inclusion relationship, which means that element object A is a part of element object B; parallel relationship, which means that element object A and element object B are essentially equal; other relationships, which means that there is no relationship between the two elements or other types of relationships that do not belong to the aforementioned relationships.

9. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, Therefore, step e specifically includes: Acquire historical information on intelligent virtual-assisted teaching and extract historical teaching data; including: elements from the knowledge graph involved in the history of intelligent virtual-assisted teaching. Comprehensive score of teaching quality for each specific element ; Word vectors are extracted using word vector extraction techniques and then jointly embedded with the element of teaching quality comprehensive score, which is encoded using one-hot encoding, to obtain an embedding vector that integrates intelligent virtual-assisted teaching historical evaluation. , ,in This refers to the element number, where m is the total number of elements. The embedding vector is based on a multi-head self-attention module. Self-attention computation is performed by mapping to multiple subspaces, and the query vector, key vector, and value matrix of each attention module are embedded using subspace vectors. Replacement, self-attention weights To determine the intensity of element requirements during intelligent virtual-assisted teaching, the output of the multi-head self-attention module represents the element requirements. Obtain the attention matrix Subspace attention output Finally, the attention outputs of all subspaces are concatenated to obtain the element attention output matrix. , This is the subspace weight matrix.

10. The intelligent virtual-assisted teaching knowledge demand analysis method according to claim 1, characterized in that, Step f specifically includes: Based on the comprehensive score of teaching quality of each element, the learning progress of each element is defined. These represent the levels of mastery of key knowledge points: high, medium, and low; according to the learning progress. The preferred order is to select teaching elements as the starting point of the path. After determining a certain element as the starting point, related elements are extracted and added to the path nodes in the order of dependency, extension, inclusion, and parallel relationships. If no next element can be found, the remaining elements are selected according to the learning progress. The order of selection is preferred to choose the next path starting element, and the above process is repeated until all elements have been added to the path to obtain a complete teaching path set. The teaching time is controlled by adjusting the number of path nodes. Preset teaching duration The following teaching path As input, a bidirectional long short-term memory network is used to train the teaching path through sequence encoding, and the hidden layer state results are extracted. Max pooling is used to extract the most salient features of each teaching path. The path feature matrix is ​​obtained. ; Element attention output matrix and path feature matrix These are the initial matrices, and their product. As the factor demand weight, based on the Bayesian personalized ranking algorithm, the factor path and corresponding factor information that give the highest factor demand weight are determined.

Images