Teaching text generation method and system based on natural language processing

Abstract

The application provides a teaching text generation method and system based on natural language processing, and relates to the technical field of data processing, comprising: collecting and acquiring a target teaching knowledge base, performing entity recognition and relationship extraction, and constructing a target teaching knowledge graph; acquiring a target teaching input text, performing text structuring processing, and determining a target teaching text structure; presetting a teaching content rule library, performing content mapping and content filtering, and obtaining basic teaching text content; constructing a target student portrait based on learning behavior historical data, performing dynamic adjustment, and determining adaptive teaching text content; building a teaching template library, performing template matching and text injection, and generating target teaching text content. The application solves the technical problem that the teaching text generation method of the prior art usually relies on a templated question bank and fixed rules, and it is difficult to make flexible adjustments according to the learning progress and learning characteristics of each student, resulting in unsatisfactory teaching effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a teaching text generation method and system based on natural language processing. BACKGROUND

[0002] In today's education field, with the continuous development of information technology, intelligent teaching tools have gradually become an auxiliary tool for teachers, students, parents and other participants, especially in the aspects of homework design, correction, feedback and school-home interaction. Digital and intelligent means have increasingly become the key to improving education efficiency. Traditional teaching text generation methods usually rely on templated question banks and fixed correction rules. Although this method is efficient, it has weak response to personalized needs. Existing technologies often have difficulty in making flexible adjustments according to the learning progress and learning characteristics of each student when designing homework, resulting in a lack of sufficient flexibility and innovation in homework design for teachers, and further leading to the possibility that the content of homework and feedback may not fully meet the personalized learning needs of students, so that the teaching effect cannot be fully exerted and the learning improvement of students is not obvious. SUMMARY

[0003] The present application provides a teaching text generation method and system based on natural language processing, aiming to solve the technical problem that the existing teaching text generation method usually relies on templated question banks and fixed rules, and has difficulty in making flexible adjustments according to the learning progress and learning characteristics of each student, resulting in a lack of sufficient flexibility and innovation in teaching text design, and further leading to unsatisfactory teaching effect.

[0004] The first aspect of the present application provides a teaching text generation method based on natural language processing, which comprises: collecting a target teaching knowledge base, performing entity recognition and relationship extraction on the target teaching knowledge base, and constructing a target teaching knowledge graph; obtaining a target teaching input text, performing text structuring processing on the target teaching input text by using a natural language processing algorithm, and determining a target teaching text structure; presetting a teaching content rule library, performing content mapping and content filtering on the target teaching text structure based on the teaching content rule library by using the target teaching knowledge graph, and obtaining basic teaching text content; constructing a target student portrait based on learning behavior historical data, dynamically adjusting the basic teaching text content in combination with the target student portrait, and determining adaptive teaching text content; building a teaching template library, performing template matching and text injection on the adaptive teaching text content based on the teaching template library, and generating target teaching text content.

[0005] In a second aspect, the application discloses a natural language processing-based teaching text generation system, which is used for the natural language processing-based teaching text generation method, and comprises a relation extraction module, a structured processing module, a content filtering module, a dynamic adjustment module and a text injection module.

[0006] The one or more technical solutions provided in the application have at least the following beneficial effects:

[0007] By entity recognition and relation extraction on the target teaching knowledge base, the key concepts in the teaching content and their relationships are automatically extracted, which can efficiently convert the scattered teaching data into a structured knowledge graph, facilitating subsequent knowledge reasoning and content generation; through natural language processing algorithm for structured processing of the target teaching input text, the key information in the text can be identified and sorted into a standardized format, not only making the original text content more standardized, but also providing a clear framework for subsequent content mapping and generation, improving the understandability and operability of teaching content; through the pre-set teaching content rule library, the content in the target teaching text structure is automatically mapped to the corresponding concepts and terms in the teaching knowledge graph, ensuring that the generated teaching text content is consistent with the known teaching knowledge base and subject knowledge, combined with content filtering, which helps to generate more refined and teaching demand-oriented basic teaching text content; based on the learning behavior history data, the target student portrait can be constructed, which can reflect the student's knowledge mastery, learning style and interest preference, etc. Personalized information provides data support for subsequent teaching text adjustment, combined with the target student portrait, automatically adjusts the basic teaching text content to better meet the needs of specific students, significantly improving the individualization of teaching, and helping to improve the learning effect of students; by building a teaching template library, automatically selecting appropriate templates to fill and inject according to the adaptive teaching text content, ensuring that the generation of teaching text not only meets the teaching outline, but also efficiently generates structured and standardized teaching materials. After matching the appropriate template, automatically inject the teaching content into the template, complete the field replacement and dynamic generation. This automatic generation process not only reduces the workload of manual editing, but also ensures that the generated teaching text meets the expected teaching goals and subject requirements, greatly improving the efficiency and accuracy of teaching design.

[0008] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the following specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described in detail. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 The teaching text generation method based on natural language processing provided by the embodiment of the present application is shown in the flowchart.

[0010] Figure 2 The structure diagram of the teaching text generation system based on natural language processing provided by the embodiment of the present application is shown.

[0011] Explanation of reference signs: relation extraction module 10, structured processing module 20, content filtering module 30, dynamic adjustment module 40, text injection module 50. DETAILED DESCRIPTION

[0012] The embodiment of the present application provides a teaching text generation method and system based on natural language processing, and solves the technical problem that the teaching text generation method in the prior art usually depends on a templated question bank and fixed rules, is difficult to make flexible adjustment according to the learning progress and learning characteristics of each student, leads to lack of sufficient flexibility and innovation in teaching text design, and further leads to unsatisfactory teaching effect.

[0013] After introducing the basic principle of the present application, various non-limiting embodiments of the present application will be specifically introduced in combination with the drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0014] Embodiment 1

[0015] As shown in the Figure 1 The embodiment of the present application provides a teaching text generation method based on natural language processing, which comprises the following steps:

[0016] Collecting a target teaching knowledge base, performing entity recognition and relationship extraction on the target teaching knowledge base, and constructing a target teaching knowledge graph.

[0017] A large amount of teaching-related knowledge is obtained from multiple sources such as teaching materials, research papers, teaching documents and online education platforms, including theoretical knowledge, practical cases, teaching methods, etc. The collected teaching knowledge is integrated to obtain a target teaching knowledge base. Natural language processing technology is used to perform entity recognition on the text in the target teaching knowledge base, and important entities related to the subject in the teaching content are extracted. These entities constitute the basic unit of teaching knowledge. In addition to entity recognition, the relationship between entities is extracted, such as cause-and-effect relationship, inclusion relationship or similarity relationship, etc. Relationship extraction can be achieved through techniques such as dependency syntax analysis and semantic role labeling. The recognized entities and extracted relationships are organized into a graph structure, where the nodes represent entities and the edges represent relationships, and a target teaching knowledge graph is obtained.

[0018] A target teaching input text is obtained, and a natural language processing algorithm is used to perform text structuring processing on the target teaching input text to determine a target teaching text structure.

[0019] The target teaching input text refers to text data from students or teachers, which can come from textbooks, homework, discussions, and other forms of input. Text structuring processing is performed on the target teaching input text, which refers to converting unstructured text into information with clear structure, including: using syntax analysis algorithms to parse the text and identify the basic components of the sentence such as subject, predicate, and object; performing context semantic analysis on the text to understand the meaning of each sentence; and performing text blocking based on the semantic information of the text to divide long sentences or paragraphs into smaller semantic units. Through structured processing, the target teaching text structure is obtained, so that the input text can be matched with teaching goals, student needs, and other multi-dimensional dimensions to ensure that the generated teaching content has high personalization and precision.

[0020] A preset teaching content rule library is adopted to perform content mapping and content filtering on the target teaching text structure based on the teaching content rule library using the target teaching knowledge graph to obtain basic teaching text content.

[0021] The teaching content rule library is a collection of content rules, specifications, and logical frameworks for specific teaching fields, including knowledge point organization rules, teaching goal and ability requirement rules, teaching method rules, content update and specification rules, etc. Based on the target teaching text structure, its content is mapped to the corresponding content module in the teaching content rule library, for example, if the text describes "Newton's second law in physics", it is mapped to the "mechanics foundation" module in the "physics" field, and further refined to the knowledge point of "force and motion". According to the specifications of the teaching content rule library, the content of the target teaching text structure is filtered to remove irrelevant or non-compliant information, for example, if the text content is too complex and exceeds the student's cognitive level, it is simplified. Based on the mapping and filtering results, structured basic teaching text content is generated, which will strictly comply with the teaching goals and specifications.

[0022] Based on the learning behavior history data, a target student portrait is constructed, and the basic teaching text content is dynamically adjusted in combination with the target student portrait to determine the adaptive teaching text content.

[0023] The learning behavior history data includes the interactive behavior of the student, the homework score, the learning progress, the learning time, the test result, the participation in the classroom discussion, and the like, and can also include the performance of the student inside and outside the classroom, such as the frequency of accessing learning resources, the length of time for watching teaching videos, the interaction with other learners, and the like. The target student portrait is a personalized learning description generated based on the learning behavior history data. The portrait analysis dimensions include the knowledge mastery degree, the learning style, and the interest preference. The learning behavior history data is analyzed in multiple dimensions according to the portrait analysis dimensions, and the feature set of the student in different dimensions is obtained. The data clustering technology, such as K-means or hierarchical clustering, can be used to classify the student groups, divide different learning groups, construct the target student portrait, and better understand the learning mode of the student through clustering analysis, so as to provide a basis for subsequent dynamic adjustment. According to the target student portrait, the adaptive teaching text content is dynamically adjusted according to the characteristics of each student group, the adaptive teaching text content is determined, and the generation of personalized teaching content is realized.

[0024] The teaching template library is built, the adaptive teaching text content is matched with the template and the text is injected based on the teaching template library, and the target teaching text content is generated.

[0025] The teaching template library is built, the teaching template library is a database for storing multiple teaching templates, each teaching template defines a format for presenting teaching content, and the teaching template includes modules such as teaching objectives, knowledge point introduction, examples, exercises, and summaries. Among them, the teaching templates can be classified according to subjects, teaching methods, learning objectives, and student levels, such as basic teaching templates, interactive teaching templates, and advanced teaching templates.

[0026] According to the target student portrait and the difficulty, target, and the like of the teaching content, the most suitable teaching template is matched, for example, if the target student portrait shows that the group is a basic learner, a simple and clear basic teaching template is selected. According to the matched teaching template, the adaptive teaching text content is injected into the teaching template. This process replaces the fields and dynamically generates the content according to the structure of the teaching template, for example, the specific teaching content and cases in the "mechanics basics" module are inserted into the corresponding position in the template, and the depth and complexity of the content are dynamically adjusted according to the learning behavior of the student. The finally generated target teaching text content includes structured content, standardized teaching modules, and personalized content adapted to the needs of the student. These contents can provide knowledge point explanation, case analysis, and exercise training that meet the ability and interest of the student, and can effectively improve the teaching effect and the learning experience of the student.

[0027] Further, the construction of the target teaching knowledge graph includes:

[0028] According to the text data application standard, a data multi-level cleaning program is determined; the target teaching knowledge base is subjected to data cleaning processing according to the data multi-level cleaning program, and a standard teaching knowledge base is obtained; entity annotation and relation extraction are performed on the standard teaching knowledge base, and a teaching knowledge entity set and a knowledge entity relation set are obtained; a knowledge graph is constructed and verified and optimized based on the teaching knowledge entity set and the knowledge entity relation set, and the target teaching knowledge graph is constructed.

[0029] The quality and format of the data in the target teaching knowledge base can be relatively chaotic, and therefore the standardization of the text data needs to be ensured through data cleaning. A series of text data application standards are defined to uniformly and standardize the processing of the data, such as data integrity standards, data consistency standards, and data accuracy standards, so as to determine the data multi-level cleaning program. For example, redundant content, noise information, and irrelevant text are removed, spelling is checked and corrected, text format is standardized, such as unified punctuation, abbreviation specification, and case specification, semantic content in the text is annotated and cleaned, polysemous words and ambiguities are identified and corrected, and common words that have no substantial meaning for semantic understanding are removed.

[0030] The data cleaning processing procedure is performed on the target teaching knowledge base according to the data multi-level cleaning program, and the standard teaching knowledge base after cleaning processing conforms to a unified format, the content is accurate, and there is no redundant or irrelevant information.

[0031] The purpose of entity annotation is to identify important entities in the text and assign corresponding categories to them. A rule-based annotation method or a deep learning model is used for entity recognition. According to the results of entity recognition, a teaching knowledge entity set is obtained. The purpose of relation extraction is to identify the relationship between different entities in the text, such as the relationship between “Newton's Law” and “mechanics”. A pattern matching-based method or a deep learning model can be used for relation extraction. The goal of relation extraction is to determine which entities have a specific type of relationship. Through relation extraction, a knowledge entity relation set is obtained.

[0032] Based on the obtained teaching knowledge entity set and knowledge entity relation set, a knowledge graph is constructed. The knowledge graph represents the structured relationship between knowledge through nodes and edges. Each entity is a node, and the edge between each pair of associated entities represents their relationship. Further, the construction of the knowledge graph not only generates nodes and edges, but also includes an optimization process, which includes removing redundant information, optimizing the representation of nodes and relationships, cleaning inaccurate entities and relationships, etc. After optimization, the target teaching knowledge graph is obtained, which has high accuracy and can provide strong support for subsequent teaching content generation.

[0033] Further, the obtaining the teaching knowledge entity set and the knowledge entity relationship set comprises:

[0034] An entity standardization system is built, entity recognition and labeling are performed on the standard teaching knowledge base according to the entity standardization system, and an entity labeled sample set is obtained; the entity labeled sample set is proportionally divided and labeled using a BERT network structure for training, and an entity labeling model is obtained; the standard teaching knowledge base is labeled based on the entity labeling model, and a teaching knowledge entity set is obtained; an entity relationship rule template is obtained, the entity relationship rule template is used for relationship extraction and template matching on the teaching knowledge entity set, and a knowledge entity relationship set is obtained.

[0035] In the teaching knowledge base, entities can be divided into different categories, such as subject knowledge points, mathematical theorems, and physical formulas. The task of the entity standardization system is to clearly define these entity categories and assign standardized labels to each category to avoid confusion between different categories. According to the entity standardization system, existing semi-automatic labeling tools are used to perform entity recognition and labeling on the standard teaching knowledge base. After labeling, an entity labeled sample set is formed, which contains labeled entity information and its classification. This entity labeled sample set is used to train and evaluate subsequent entity labeling models.

[0036] The BERT network structure is a pre-training language model based on the Transformer architecture. Its main advantage is that it can capture the context information in the text and understand the context of the word through bidirectional training. The entity labeled sample set is proportionally divided into training and validation data sets, and the common division ratio is 80% for training and 20% for validation. This division can ensure model validation during training and prevent overfitting. During the labeling training process, the training data set is used to train the BERT network structure, and the cross-entropy loss function is used to calculate the difference between the predicted label and the actual label of the model output. The model continuously adjusts the parameters to minimize the loss function through multiple rounds of training, and finally reaches a state that can effectively perform entity labeling. After training is completed, the performance of the model is evaluated using the validation data set, and the model is further optimized based on the evaluation results. Finally, an entity labeling model is obtained, which can automatically identify various entities in the teaching knowledge base.

[0037] The trained entity labeling model is used to process each text in the standard teaching knowledge base, and the entity types in the text are labeled. The labeled text contains multiple entities and their corresponding category labels. These entities constitute a teaching knowledge entity set.

[0038] The goal of relation extraction is to extract the logical relationships between entities from a set of teaching knowledge entities. Common relationship types include definition relationships, causal relationships, belonging relationships, and dependency relationships. Entity relationship rule templates are predefined patterns used to extract relationships between different entities. These templates can be static or dynamically learned through machine learning methods, such as dependency syntax analysis in natural language processing to understand the grammatical relationships between entities. In a labeled set of teaching knowledge entities, predefined entity relationship rule templates are applied to identify and extract relationships between entities, i.e., through template matching to extract corresponding relationships. After relation extraction, the generated knowledge entity relationship set contains entity pairs and their relationship types.

[0039] Further, the determination of the target teaching text structure includes:

[0040] The natural language processing algorithm is used to perform denoising processing and format conversion on the target teaching input text to obtain a standard teaching input text. PCFG is used to perform syntax analysis and semantic recognition on the standard teaching input text to obtain teaching input text semantic information. A preset text blocking rule is used to perform blocking processing on the standard teaching input text based on the teaching input text semantic information to obtain a teaching text blocking set. NER is used to perform entity extraction and relationship classification on the teaching text blocking set, and the target teaching text structure is output in a structured manner.

[0041] Denoising processing includes removing irrelevant content, spelling and grammar correction, removing stop words, etc. Specifically, the target teaching input text may contain unnecessary or irrelevant content, such as advertisements, copyright statements, etc. These contents usually affect the quality of text processing and need to be removed, leaving only the main text. The target teaching input text may contain spelling errors or unstandardized grammar, which affects subsequent processing. Spelling errors are corrected through spelling check. Stop words are common words that do not contribute to semantic understanding and often interfere with text analysis results, so they need to be removed. Format conversion, i.e., converting the original text into a standardized structure, such as converting free text into JSON, XML, or other structured formats suitable for computer processing. After denoising and format conversion, the standard teaching input text is uniform in format, free of redundant content, and suitable for further processing, providing high-quality input data for subsequent semantic recognition and syntax analysis.

[0042] PCFG is a probability-based context-free grammar that assigns probabilities to grammar rules to represent the likelihood of each rule generating text in a certain context. PCFG is widely used in syntax analysis, especially in handling complex text grammatical structures, providing better predictive capabilities.

[0043] The goal of syntactic analysis is to convert the input natural language text into a syntactic tree that shows the structural relationships between words. PCFG infers the most likely syntactic structure based on probability values. Specifically, the standard teaching input text is processed into lexical units through tokenization, and PCFG is applied to these lexical units for syntactic analysis, generating multiple possible syntactic trees. The nodes of the syntactic tree represent words or phrases, and the edges represent the relationships between words. Finally, the one with the highest probability is selected as the optimal structure.

[0044] Based on syntactic analysis, semantic recognition is performed. Specifically, by analyzing the syntactic tree, semantic roles such as subject, predicate, and object are identified, and the relationships between the components are understood. For example, in the sentence "Newton's second law is the foundation of physics," "Newton's second law" is the subject, "is" is the predicate, and "the foundation of physics" is the object. This helps to understand the semantic structure of the sentence and ultimately obtain the semantic information of the teaching input text, providing basic data for subsequent teaching content generation and processing.

[0045] Text chunking rules are used to divide the text into multiple units with independent meanings based on the semantic information and structure of the teaching text. These units can be divided based on topics, concepts, chapters, paragraphs, sentences, or grammatical structures, etc. The design of the rules should be based on the characteristics of the teaching content, such as for textbook or lesson plan text, the following rules can be used for chunking: divide according to different teaching topics or chapters, each topic or chapter contains relevant knowledge points and concepts; divide the text by sentence, each sentence represents a basic semantic unit; divide the text into several parts based on keywords or concepts, each part revolves around a core concept; according to the results of semantic analysis, identify the parts that are related to each other and process them as a chunk.

[0046] Using the above text chunking rules, the standard teaching input text is processed into chunks. During the chunking process, combined with the semantic information of the teaching input text, it can be identified which parts should belong to the same chunk. The chunked text becomes multiple smaller semantic units, forming a teaching text chunking set, which serves as input data for subsequent processing.

[0047] NER is a natural language processing task that aims to identify entities with practical significance from text, such as names, place names, organization names, time, date, subject names, formulas, etc. Using NER to extract entities from a set of teaching text blocks, after entity extraction, relationship classification is performed, i.e. identifying the relationship between entities. Relationship classification can be based on predefined rules or using machine learning models, for example, using a BERT-based relationship classification model to identify the relationship between the two and classify it. After completing entity extraction and relationship classification, structured processing is performed to obtain the target teaching text structure, which includes clear entities and their relationships, which provides a basis for subsequent teaching content generation, knowledge query and recommendation system, etc.

[0048] Further, the obtained basic teaching text content includes:

[0049] The target teaching knowledge graph is used to map the content of the target teaching text structure to obtain matching teaching text content. According to the teaching content rule library, content specification rules, teaching logic rules and ability adaptation rules are determined. Based on the content specification rules, teaching logic rules and ability adaptation rules, the matching teaching text content is filtered and integrated to obtain the basic teaching text content.

[0050] The target teaching knowledge graph can provide information such as the relationship between knowledge points, concept structure, and context association. The target teaching text structure includes annotated entities and their grammatical structure. Using the target teaching knowledge graph, each teaching module in the target teaching text structure is matched with the corresponding node in the graph. The goal of this process is to associate the specific content in the teaching text with the abstract knowledge points, theorems, formulas, definitions, etc. in the knowledge graph. Specifically, the entities in the target teaching text structure are directly matched with the entities in the target teaching knowledge graph. If the text does not directly mention the entities in the knowledge graph, semantic similarity matching or synonym matching can be used for mapping. Through the above mapping process, each part of the target teaching text structure is associated with the information in the target teaching knowledge graph, and finally a matching teaching text content consistent with the target teaching knowledge graph is obtained.

[0051] The teaching content rule library is a database containing various teaching specifications and rules, covering how to organize teaching content, how to design teaching structure, and how to adjust teaching materials to meet the needs of different student groups. According to the teaching content rule library, content specification rules, teaching logic rules, and ability adaptation rules are determined. The content specification rules are used to determine the standardization and consistency of teaching content, such as the writing specification of formulas, the unified definition of terms, the standardization of paragraph structure, etc. The teaching logic rules are used to ensure that the teaching content is presented in a reasonable logical order, such as introducing basic knowledge first and then gradually introducing complex concepts to ensure that students can understand the content step by step. The ability adaptation rules are used to adjust the difficulty and presentation of teaching content according to the learning ability, interest and knowledge background of students, such as more concise and clear explanations for beginners and more in-depth and challenging content for advanced learners.

[0052] According to the aforementioned content specification rules, teaching logic rules and ability adaptation rules, content filtering is performed to remove parts that do not meet the specifications or are unnecessary, making the final teaching text more concise and efficient. The filtered content includes terms or symbols that do not meet the content specifications, paragraphs with unclear logic or unreasonable order, and overly complex or overly simple content that is not suitable for specific student groups. After content filtering, the remaining content is integrated in accordance with the specifications and logical order to form a structured and easy-to-understand basic teaching text content that meets all the content specification rules, teaching logic rules and ability adaptation rules.

[0053] Further, the target student portrait is constructed based on the learning behavior history data, including:

[0054] A set of portrait dimensions is defined, including knowledge mastery, learning style and interest preference. The learning behavior history data is analyzed in multiple dimensions according to the set of portrait dimensions to obtain a set of multi-dimensional portrait parameters. Cluster analysis is performed on the set of multi-dimensional portrait parameters to obtain student group clustering results, and the target student portrait is constructed based on the student group clustering results.

[0055] The image dimension set is a set of key factors to be considered when constructing a student image, which is used to describe the multi-dimensional characteristics of students and provide detailed information required for personalized teaching. The knowledge mastery indicates the mastery of students on course content and subject knowledge points, which can be measured by students' homework scores, test scores, examination results, etc. The learning style indicates the preferred way and method of students in the learning process, for example, visual learners (obtaining information through images, charts, etc.), auditory learners (learning through lectures, discussions, etc.) or hands-on learners (learning through experiments and practical operations). The interest preference indicates the interest of students in a specific subject or knowledge field, which can be evaluated by the courses selected by students, activities participated in, reading materials, etc.

[0056] The learning behavior history data refers to various records and data generated by students in the learning process. Through multi-dimensional analysis of the learning behavior history data based on the image dimension set defined above, for example, based on the students' score data, the students' mastery of different subjects and knowledge points is analyzed; the students' learning style is inferred from their learning activity records and participation methods, for example, some students often watch teaching videos and mark notes, indicating their visual learning style; the students' interest preference is inferred from the courses they choose, reading materials, and learning activities they participate in. After multi-dimensional analysis, a multi-dimensional image parameter set containing various dimensions is obtained, reflecting the characteristics and performance of each student in each dimension.

[0057] Clustering analysis is an unsupervised learning method, which aims to divide a data set into different clusters so that the members in each cluster are as similar as possible in some characteristics, while the clusters are as different as possible. In this step, clustering analysis methods such as K-means, hierarchical clustering, etc. are used to group students based on the multi-dimensional image parameter set, to identify similar learner groups. The clustering analysis result shows the characteristics of different student groups, for example, most students in one group have mastered basic knowledge and tend to learn through videos; another group has a high knowledge mastery level but is very interested in physics and prefers experimental learning. According to the results of clustering analysis, a target student image is constructed for each student group, reflecting the detailed characteristics of students in different dimensions.

[0058] Further, the determination of the adaptive teaching text content comprises:

[0059] Adjusting strategy analysis is performed on the dimension information in the image dimension set to obtain a multi-dimensional teaching adjustment strategy set; strategy matching is performed on the target student image based on the multi-dimensional teaching adjustment strategy set to determine a target teaching adjustment strategy; and the target teaching adjustment strategy is used to dynamically adjust the basic teaching text content to determine the adaptive teaching text content.

[0060] The adjustment strategy is to adjust the teaching content and teaching method according to the different learning characteristics and needs of students, so that it is more suitable for the personalized learning needs of specific students. First, each dimension in the portrait dimension set is analyzed in detail to find out how each dimension affects the adjustment of the teaching strategy. For example, if a student has poor mastery of a certain knowledge point, more detailed explanations, simpler examples, or additional practice and guidance are needed. If the mastery is good, more challenging content or direct entry into deeper learning can be provided. If a student tends to visual learning, the adjustment strategy can include more illustrations, video explanations, animations, etc. If it is an auditory learner, more use of explanations or audio resources is needed. If a student has strong interest in a certain field, the teaching content can be personalized according to the student's interest, adding content and examples related to the student's interest to stimulate their learning motivation. By integrating the information of each dimension, a multi-dimensional teaching adjustment strategy set suitable for different student groups is analyzed, aiming to maximize the adaptation of the teaching process to the individual needs of students through precise adjustment.

[0061] Based on the multi-dimensional teaching adjustment strategy set, the target student portrait is matched with the strategy. The goal is to find the most suitable teaching adjustment strategy for each student, so that the teaching content and method can meet their learning needs. Through this process, a specific and personalized target teaching adjustment strategy can be determined for each student.

[0062] Using the determined target teaching adjustment strategy, the basic teaching text content is dynamically adjusted to better meet the individual learning needs of each student. Ultimately, an adapted teaching text content that adapts to each student is generated. The adapted teaching text content includes personalized learning resources, suitable teaching methods, and reasonable knowledge depth, ensuring that the teaching content maximizes the needs of students.

[0063] Further, the generation of the target teaching text content includes:

[0064] Defining a template matching rule, the template matching rule includes teaching goal matching, subject matching and difficulty matching; using the template matching rule to perform template matching and text injection on the adapted teaching text content based on the teaching template library to obtain the target teaching text content.

[0065] The template matching rule is defined to ensure that the teaching content can be adjusted and optimized according to the matching of different teaching objectives, subject requirements and student ability, so as to generate personalized and standard teaching text. The template matching rule includes teaching objective matching, subject matching and difficulty matching. The teaching objective is the core achievement or ability to be achieved in the teaching activity. The teaching objective matching rule needs to ensure that the teaching content is consistent with the teaching objective. For example, the teaching objective is "mastering Newton's second law in physics". The template matching rule needs to ensure that the selected template can effectively convey this goal, and ensure that the required theoretical explanation, examples and exercises, etc. are around Newton's second law. The subject matching rule ensures that the selected teaching template is consistent with the subject being taught. For example, the teaching template of physics is different from the template of mathematics or chemistry in structure and content. According to the subject category that the student is learning, the template related to the subject is selected. The difficulty matching rule selects the teaching template with appropriate difficulty according to the student's learning level. For example, beginners need more basic content and more examples, while advanced students need more challenging tasks or in-depth analysis. The three matching rules of teaching objective, subject and difficulty work together to help accurately select the most suitable template and provide personalized learning resources for students.

[0066] The adapted teaching text content is input, and template matching is performed from the teaching template library according to the template matching rule. For example, if the teaching objective is "mastering Newton's second law", and the student is a beginner, a template containing basic definitions, illustrations and simple application examples is selected. After template matching is completed, the adapted teaching text content is inserted into the template, and the personalized teaching content meeting the student's needs, i.e. the target teaching text content, is finally generated. This process ensures that the teaching content not only meets the requirements in structure, but also meets the student's learning needs in depth, difficulty, subject relevance and teaching objective.

[0067] Further, the target teaching text content is obtained, including:

[0068] The template matching rule is used to perform similarity calculation and template optimization on the adapted teaching text content based on the teaching template library, to obtain a target teaching template. The adapted teaching text content is injected into the target teaching template for field replacement and dynamic generation, to obtain the target teaching text content.

[0069] The text similarity calculation method in natural language processing, such as cosine similarity, can be used to evaluate the matching degree of the adapted teaching text content and the template. For example, the text similarity calculation can evaluate the similarity between the template and the adapted teaching content by analyzing keywords, syntactic structure, semantic information, etc. After similarity calculation between all available templates and the adapted teaching text content, the templates are sorted according to the similarity scores. The template with the highest similarity is the most suitable template for the teaching content, which is the target teaching template.

[0070] The target teaching template usually contains multiple placeholders or dynamic fields, which can be dynamic parts of the template, to be filled with specific content. The adapted teaching text content is inserted into the corresponding fields in the template. In addition to simple field replacement, some fields in the template are dynamically generated according to the specific teaching content. For example, the "example" part in the template can automatically select relevant examples for display according to the teaching content, or generate exercises related to the knowledge points. The content of all field replacement and dynamic generation is integrated to form a complete target teaching text content. At this time, the target teaching text content already contains all the necessary information, and the structure is clear, the content is complete, which meets the teaching goal, the subject requirement and the learning needs of students.

[0071] In summary, the teaching text generation method based on natural language processing provided by the embodiments of the present application has the following technical effects:

[0072] By performing entity recognition and relation extraction on the target teaching knowledge base, the key concepts in the teaching content and their relationships are automatically extracted, which can efficiently convert the scattered teaching data into a structured knowledge graph, facilitating subsequent knowledge reasoning and content generation; through the structured processing of the target teaching input text by natural language processing algorithms, the key information in the text can be identified and organized into a standardized format, not only making the original text content more standardized, but also providing a clear framework for subsequent content mapping and generation, improving the understandability and operability of the teaching content; by pre-setting the teaching content rule library, the content in the target teaching text structure is automatically mapped to the corresponding concepts and terms in the teaching knowledge graph, ensuring that the generated teaching text content is consistent with the known teaching knowledge base and subject knowledge, combined with content filtering, which helps to generate more refined and teaching demand-oriented basic teaching text content; based on the learning behavior history data, the target student portrait can be constructed, which can reflect the student's knowledge mastery, learning style and interest preference, etc. Personalized information provides data support for subsequent teaching text adjustment, combined with the target student portrait, automatically adjusts the basic teaching text content to better meet the needs of specific students, significantly improving the individualization of teaching, and helping to improve the learning effect of students; by building a teaching template library, the appropriate template is automatically selected according to the adaptive teaching text content to fill and inject, ensuring that the generation of teaching text not only meets the teaching outline, but also efficiently generates structured and standardized teaching materials. After matching the appropriate template, the teaching content is automatically injected into the template to complete the field replacement and dynamic generation. This automatic generation process not only reduces the workload of manual editing, but also ensures that the generated teaching text meets the expected teaching goals and subject requirements, greatly improving the efficiency and accuracy of teaching design.

[0073] Embodiment 2

[0074] Based on the same inventive concept as the natural language processing-based teaching text generation method in the foregoing embodiments, as shown in Figure 2 The embodiments of the present application provide a natural language processing-based teaching text generation system, which comprises:

[0075] The relationship extraction module 10 is configured to collect a target teaching knowledge base, perform entity recognition and relationship extraction on the target teaching knowledge base, and construct a target teaching knowledge graph; the structured processing module 20 is configured to obtain a target teaching input text, perform text structured processing on the target teaching input text by using a natural language processing algorithm, and determine a target teaching text structure; the content filtering module 30 is configured to preset a teaching content rule library, perform content mapping and content filtering on the target teaching text structure based on the teaching content rule library by using the target teaching knowledge graph, and obtain basic teaching text content; the dynamic adjustment module 40 is configured to construct a target student portrait based on learning behavior historical data, dynamically adjust the basic teaching text content in combination with the target student portrait, and determine adaptive teaching text content; and the text injection module 50 is configured to build a teaching template library, perform template matching and text injection on the adaptive teaching text content based on the teaching template library, and generate target teaching text content.

[0076] Further, the relationship extraction module 10 is configured to perform the following operation steps:

[0077] According to a text data application standard, a data multi-level cleaning program is determined; the target teaching knowledge base is subjected to data cleaning processing according to the data multi-level cleaning program, and a standard teaching knowledge base is obtained; entity labeling and relationship extraction are performed on the standard teaching knowledge base, and a teaching knowledge entity set and a knowledge entity relationship set are obtained; graph construction and verification optimization are performed based on the teaching knowledge entity set and the knowledge entity relationship set, and the target teaching knowledge graph is constructed.

[0078] Further, the relationship extraction module 10 is configured to perform the following operation steps:

[0079] A standard entity system is built, entity recognition labeling is performed on the standard teaching knowledge base according to the standard entity system, and an entity labeling sample set is obtained; the entity labeling sample set is divided in proportion and labeled by using a BERT network structure, and an entity labeling model is obtained; entity type labeling is performed on the standard teaching knowledge base based on the entity labeling model, and a teaching knowledge entity set is obtained; an entity relationship rule template is obtained, relationship extraction and template matching are performed on the teaching knowledge entity set by using the entity relationship rule template, and a knowledge entity relationship set is obtained.

[0080] Further, the structured processing module 20 is configured to perform the following operation steps:

[0081] The natural language processing algorithm is used to denoise and format convert the target teaching input text to obtain a standard teaching input text; the PCFG is used to perform syntax analysis and semantic recognition on the standard teaching input text to obtain teaching input text semantic information; a text blocking rule is preset, and the text blocking rule is used to block the standard teaching input text based on the teaching input text semantic information to obtain a teaching text blocking set; the NER is used to extract entities and classify relationships based on the teaching text blocking set, and the target teaching text structure is output.

[0082] Further, the content filtering module 30 is configured to perform the following operation steps:

[0083] The target teaching knowledge graph is used to map the target teaching text structure to obtain matching teaching text content; the content specification rule, the teaching logic rule, and the ability adaptation rule are determined according to the teaching content rule library; the matching teaching text content is integrated based on the content specification rule, the teaching logic rule, and the ability adaptation rule to obtain the basic teaching text content.

[0084] Further, the dynamic adjustment module 40 is configured to perform the following operation steps:

[0085] A portrait dimension set is defined, the portrait dimension set includes knowledge mastery, learning style, and interest preference; the learning behavior history data is analyzed in multiple dimensions according to the portrait dimension set to obtain a multi-dimensional portrait parameter set; the multi-dimensional portrait parameter set is clustered to obtain a student group clustering result, and the target student portrait is constructed according to the student group clustering result.

[0086] Further, the dynamic adjustment module 40 is configured to perform the following operation steps:

[0087] Each dimension information in the portrait dimension set is analyzed to obtain a multi-dimensional teaching adjustment strategy set; the target student portrait is matched based on the multi-dimensional teaching adjustment strategy set to determine a target teaching adjustment strategy; the basic teaching text content is dynamically adjusted based on the target teaching adjustment strategy to determine the adaptive teaching text content.

[0088] Further, the text injection module 50 is configured to perform the following operation steps:

[0089] A template matching rule is defined, the template matching rule includes teaching target matching, subject matching, and difficulty matching; the template matching rule is used to match the adaptive teaching text content based on the teaching template library to obtain target teaching text content.

[0090] Further, the text injection module 50 is configured to perform the following operation steps:

[0091] The template matching rule is used to perform similarity calculation and template optimization on the adaptive teaching text content based on the teaching template library, so as to obtain a target teaching template; and the adaptive teaching text content is injected into the target teaching template for field replacement and dynamic generation, so as to obtain the target teaching text content.

[0092] Through the foregoing detailed description of the teaching text generation method based on natural language processing, those skilled in the art can clearly understand the teaching text generation system based on natural language processing in the embodiments. Since the system corresponds to the method disclosed in the embodiments, the system is described relatively simply, and the relevant part can be referred to the method part.

[0093] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating teaching text based on natural language processing, characterized in that, The method comprises: Collecting a target teaching knowledge base, performing entity recognition and relation extraction on the target teaching knowledge base, and constructing a target teaching knowledge graph; Obtaining a target teaching input text, performing text structuring processing on the target teaching input text by using a natural language processing algorithm, and determining a target teaching text structure; Pre-setting a teaching content rule library, performing content mapping and content filtering on the target teaching text structure based on the teaching content rule library by using the target teaching knowledge graph, and obtaining basic teaching text content; Based on learning behavior historical data, a target student portrait is constructed, and the basic teaching text content is dynamically adjusted combined with the target student portrait to determine adaptive teaching text content; A teaching template library is built, and the adaptive teaching text content is matched with a template and injected with text based on the teaching template library to generate target teaching text content; Wherein, the basic teaching text content comprises: Using the target teaching knowledge graph, each teaching module in the target teaching text structure is matched with the corresponding node in the graph, each part of the target teaching text structure is associated with the information in the target teaching knowledge graph, and matching teaching text content consistent with the target teaching knowledge graph is obtained; According to the teaching content rule library, content specification rules, teaching logic rules and ability adaptation rules are determined; Based on the content specification rules, teaching logic rules and ability adaptation rules, the matching teaching text content is filtered and integrated to obtain the basic teaching text content. 2.The natural language processing based teaching text generating method according to claim 1, characterized in that, The construction of the target teaching knowledge graph comprises: According to the text data application standard, a multi-level data cleaning program is determined; According to the data multi-level cleaning program, the target teaching knowledge base is cleaned to obtain a standard teaching knowledge base; The standard teaching knowledge base is annotated and relation extracted to obtain a teaching knowledge entity set and a knowledge entity relation set; Based on the teaching knowledge entity set and the knowledge entity relation set, the graph is constructed and verified and optimized to construct the target teaching knowledge graph. 3.The natural language processing based teaching text generation method according to claim 2, characterized in that, The obtaining of the teaching knowledge entity set and the knowledge entity relation set comprises: Building an entity standardization system, performing entity recognition annotation on the standard teaching knowledge base according to the entity standardization system to obtain an entity annotation sample set; Using the BERT network structure to proportionally divide and annotate the entity annotation sample set to obtain an entity annotation model; Based on the entity annotation model, the standard teaching knowledge base is annotated by entity type to obtain a teaching knowledge entity set; Obtaining an entity relation rule template, using the entity relation rule template to extract relations and match templates for the teaching knowledge entity set to obtain a knowledge entity relation set. 4.The natural language processing based teaching text generating method according to claim 1, wherein, The determination of the target teaching text structure comprises: Using a natural language processing algorithm to perform denoising processing and format conversion on the target teaching input text to obtain a standard teaching input text; Using PCFG to perform syntax analysis and semantic recognition on the standard teaching input text to obtain teaching input text semantic information; A preset text chunking rule is used to chunk the standard teaching input text based on the teaching input text semantic information, to obtain a teaching text chunking set; The teaching text chunking set is subjected to entity extraction and relationship classification based on NER, and the target teaching text structure is output in a structured manner. 5.The natural language processing based teaching text generating method according to claim 1, wherein, The target student portrait is constructed based on the learning behavior history data, including: A portrait dimension set is defined, including knowledge mastery, learning style, and interest preference; The learning behavior history data is subjected to multi-dimensional analysis according to the portrait dimension set, to obtain a multi-dimensional portrait parameter set; The multi-dimensional portrait parameter set is subjected to cluster analysis, to obtain a student group clustering result, and the target student portrait is constructed based on the student group clustering result. 6.The natural language processing based teaching text generating method according to claim 5, characterized in that, The adaptive teaching text content is determined, including: Each dimension information in the portrait dimension set is subjected to adjustment strategy analysis, to obtain a multi-dimensional teaching adjustment strategy set; The target student portrait is subjected to strategy matching based on the multi-dimensional teaching adjustment strategy set, to determine a target teaching adjustment strategy; The target teaching adjustment strategy is used to dynamically adjust the basic teaching text content, to determine the adaptive teaching text content. 7.The natural language processing based teaching text generating method according to claim 1, wherein, The target teaching text content is generated, including: A template matching rule is defined, including teaching target matching, subject matching, and difficulty matching; The template matching rule is used to perform template matching and text injection on the adaptive teaching text content based on the teaching template library, to obtain the target teaching text content. 8.The natural language processing based teaching text generating method according to claim 7, characterized in that, The target teaching text content is obtained, including: The template matching rule is used to perform similarity calculation and template optimization on the adaptive teaching text content based on the teaching template library, to obtain a target teaching template; The adaptive teaching text content is injected into the target teaching template for field replacement and dynamic generation, to obtain the target teaching text content.

9. A teaching text generation system based on natural language processing, characterized by, A system for implementing the natural language processing-based teaching text generation method of any one of claims 1-8, the system comprising: A relationship extraction module configured to collect a target teaching knowledge base, perform entity recognition and relationship extraction on the target teaching knowledge base, and construct a target teaching knowledge graph; A structured processing module configured to obtain a target teaching input text, perform text structured processing on the target teaching input text using a natural language processing algorithm, and determine a target teaching text structure; A content filtering module configured to preset a teaching content rule library, perform content mapping and content filtering on the target teaching text structure based on the teaching content rule library using the target teaching knowledge graph, and obtain a basic teaching text content; A dynamic adjustment module configured to construct a target student portrait based on learning behavior history data, dynamically adjust the basic teaching text content in combination with the target student portrait, and determine an adaptive teaching text content; A text injection module configured to build a teaching template library, perform template matching and text injection on the adaptive teaching text content based on the teaching template library, and generate a target teaching text content.

Images