A method for identifying implicit structure of teaching segments

Abstract

The present application belongs to the technical field of educational informatization, and discloses a method for identifying implicit structure of teaching links, which comprises the following steps: first, obtaining disordered teaching texts formed after deleting teaching link names and adjusting teaching content order; then, determining teaching mode, course type and core concept, and matching standard teaching link ordered sequence from a macro template library; searching candidate content segments from a structured teaching link knowledge base to obtain micro teaching content information; calculating content coverage rate between macro teaching structure information and micro teaching content information, and selecting the standard teaching link ordered sequence as a mandatory order constraint or selecting the micro teaching link order as a reference order according to the content coverage rate; and inputting the fused structured information into a lightweight large language model to obtain an ordered teaching link sequence. The present application can improve the accuracy and robustness of teaching link division, name identification and order recovery.

Description

Technical Field

[0001] This invention relates to the field of educational informatization technology, specifically to a method for identifying implicit structures in teaching processes. Background Technology

[0002] With the continuous development of educational information technology, teaching materials such as lesson plans, instructional designs, and classroom records are increasingly stored and processed in digital text format. These teaching texts typically include multiple teaching segments such as introduction, explanation, interaction, and summary, with each segment interconnected according to a certain teaching logic. However, different teachers may use different segment names and division methods for the same teaching activity, and some teaching texts also suffer from missing segment names, unclear content boundaries, or disordered arrangement, resulting in structural information often existing implicitly within the teaching text.

[0003] Existing technologies typically employ text classification, semantic matching, vector retrieval, or large language models to analyze teaching texts. Text classification methods determine text categories based on predefined labels, semantic retrieval methods retrieve content related to the input text from a knowledge base, and retrieval-enhanced generation methods combine retrieval results with language models to assist in text generation or information extraction tasks. However, most of these methods focus on recognizing local semantic content and struggle to simultaneously consider the relationships between teaching stage boundaries, stage names, and stage sequences.

[0004] Especially when the names of teaching segments are deleted or the order of the corresponding content in each segment is scrambled, relying solely on local content matching or direct generation by a language model can easily lead to problems such as incomplete segment division, inaccurate segment name recognition, and incorrect order restoration. Meanwhile, lightweight large language models, limited by their semantic understanding and long text processing capabilities, struggle to independently restore complex teaching structures; while models with larger parameter scales increase computational resource consumption. Therefore, how to combine the overall teaching process information with local teaching content information to achieve accurate segmentation, name recognition, and order restoration of teaching segments in disordered teaching texts has become a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] To address the above technical problems, this invention provides a method for identifying implicit structures in teaching processes, the method comprising the following steps: S100. Obtain unordered teaching text, wherein the unordered teaching text is the text formed after deleting the names of teaching segments from the original teaching text and adjusting the order of the teaching content corresponding to each teaching segment. S200: Based on the unordered teaching text obtained in S100, determine the teaching mode, lesson type and core concept corresponding to the unordered teaching text, and match the corresponding ordered sequence of standard teaching links from the macro template library according to the teaching mode, lesson type and core concept to obtain macro teaching structure information; S300: Retrieve candidate content fragments related to the unordered teaching text from the structured teaching segment knowledge base, and obtain micro-level teaching content information containing the candidate teaching segment name and corresponding teaching content based on the candidate content fragments; S400: Based on the macro-level teaching structure information obtained in S200 and the micro-level teaching content information obtained in S300, determine the degree of consistency between the ordered sequence of the standard teaching steps and the names of the teaching steps in the micro-level teaching content information. S500: Based on the consistency level determined in S400, the macro-level teaching structure information and micro-level teaching content information are dynamically integrated; when the consistency level meets the preset conditions, the ordered sequence of the standard teaching steps is integrated as the sequence constraint information; when the consistency level does not meet the preset conditions, the order of the teaching steps in the micro-level teaching content information is integrated as the reference order to obtain structured integrated information. S600: Based on the structured fusion information obtained in S500, the disordered teaching text is divided into teaching segments, the names of teaching segments are identified, and the order of teaching segments is restored to obtain an ordered sequence of teaching segments.

[0006] As a further improvement of the present invention, S100 includes the following steps: S110. Obtain the original teaching text containing multiple teaching segment names and teaching content corresponding to each teaching segment name; S120. Delete the names of teaching segments in the original teaching text, and retain the teaching content corresponding to each teaching segment; S130. Adjust the order of teaching content corresponding to different teaching segments to form an unordered teaching text that does not contain explicit teaching segment names or the original order of teaching segments.

[0007] As a further improvement of the present invention, the macro template library in S200 is constructed according to the following steps: S210. Collect standard lesson plan texts and label each standard lesson plan text with teaching mode tags, lesson type tags, and core concept tags; S220. Based on the standard teaching process corresponding to each standard lesson plan text, extract and arrange the names of teaching links in each standard lesson plan text to obtain an ordered sequence of standard teaching links. S230. The teaching mode tags, lesson type tags, core concept tags and corresponding standard teaching steps are associated and stored in an ordered sequence to form the macro template library; Among them, the teaching mode label, lesson type label and core concept label are all single labels, and the core concept label is determined from a preset core concept vocabulary list, which can be expanded.

[0008] As a further improvement of the present invention, S200 further includes the following steps: S240. Based on multiple unordered teaching text samples and the teaching mode labels, lesson type labels, and core concept labels corresponding to each unordered teaching text sample, a classifier training dataset is constructed. S250. Extract the word frequency-inverse document frequency feature vector of each unordered teaching text sample, and train the teaching mode classifier, lesson type classifier and core concept classifier based on the word frequency-inverse document frequency feature vector respectively. S260. Input the unordered teaching text obtained in S100 into the teaching mode classifier, lesson type classifier and core concept classifier respectively to obtain the predicted teaching mode label, predicted lesson type label and predicted core concept label. S270. Combine the predicted teaching mode tag, predicted lesson type tag, and predicted core concept tag to form a search key, and perform matching in the macro template library according to the search key; S280. When a tag combination matching the search key exists in the macro template library, the ordered sequence of standard teaching steps corresponding to the tag combination is output; when no tag combination matching the search key exists, the macro teaching structure information is set to empty, and the injection of the ordered sequence of standard teaching steps is skipped.

[0009] As a further improvement of the present invention, the structured teaching knowledge base in S300 is constructed according to the following steps: S310. Obtain a lesson plan text containing the name of the teaching segment and the corresponding teaching content, and process the lesson plan text into multiple teaching segment data pairs. Each teaching segment data pair includes the name of the teaching segment and a content fragment corresponding to the name of the teaching segment. S320. Generate content embedding vectors for each content segment, and store the teaching segment name, the corresponding content segment, and the content embedding vector together. S330. Based on the content embedding vectors corresponding to each content segment, a vector index is established to form the knowledge base of the structured teaching process.

[0010] As a further improvement of the present invention, the candidate content fragment retrieval in S300 includes the following steps: S340. Generate a query embedding vector based on the unordered teaching text obtained in S100; S350. Calculate the cosine similarity between the query embedding vector and the embedding vectors of each content in the structured teaching knowledge base; S360. Sort each content segment according to the cosine similarity, and select the top-ranked content segments to form an initial candidate segment set.

[0011] As a further improvement of the present invention, S300 further includes the following steps: S370. The disordered teaching text obtained in S100 is concatenated with each candidate content segment in the initial candidate segment set, and the concatenation result is input into the cross encoder to obtain the correlation score between the disordered teaching text and each candidate content segment. S380. The initial candidate fragment set is reordered according to the relevance score, and candidate content fragments are selected from the reordered initial candidate fragment set to form the final candidate fragment set. S390. Obtain the number of query attention heads for the lightweight large language model used to execute S600, and compare the number of query attention heads with a preset format switching threshold. S391. When the number of query attention heads is not less than the preset format switching threshold, the teaching segment name in the final candidate segment set is combined with the corresponding complete content segment to obtain the micro-level teaching content information of the complete content format. S392. When the number of query attention heads is less than the preset format switching threshold, the teaching segment name in the final candidate segment set is combined with the first complete sentence of the corresponding content segment to obtain the micro-teaching content information of the first sentence content format.

[0012] As a further improvement of the present invention, S400 includes the following steps: S410. Extract candidate teaching segment names from the micro-teaching content information obtained in S300, and normalize the candidate teaching segment names according to the synonym normalization mapping to obtain a normalized ordered sequence of micro-teaching segment names. S420. For each macroscopic teaching segment name in the ordered sequence of standard teaching segments obtained in S200, determine the fuzzy matching degree between the macroscopic teaching segment name and each microscopic teaching segment name in the ordered sequence of normalized microscopic teaching segment names. S430. Based on the fuzzy matching degree, determine the number of macro-level teaching segment names that meet the preset matching conditions; S440. Based on the ratio between the number of macro-level teaching segment names that meet the preset matching conditions and the total number of macro-level teaching segment names contained in the ordered sequence of standard teaching segments, determine the content coverage rate between the macro-level teaching structure information and the micro-level teaching content information, and use the content coverage rate as the degree of consistency.

[0013] As a further improvement of the present invention, S500 includes the following steps: S510, Obtain the macro-level teaching structure information obtained in S200 and the content coverage rate determined in S400; S520. When the macro-level teaching structure information is not empty and the content coverage rate meets the preset coverage conditions, the ordered sequence of the standard teaching links is used as a mandatory sequence constraint and combined with the micro-level teaching content information obtained in S300 to form the first structured fusion information. S530. When the macro-teaching structure information is empty or the content coverage does not meet the preset coverage conditions, the normalized micro-teaching link name ordered sequence is used as a reference order and combined with the micro-teaching content information obtained in S300 to form the second structured fusion information. S540, take the first structured fusion information or the second structured fusion information as input to S600.

[0014] As a further improvement of the present invention, S600 includes the following steps: S610. Combine the structured fusion information obtained in S500 with the enhanced prompts, wherein the enhanced prompts include common process templates and sequence restoration rules for teaching steps; S620. Input the combined structured fusion information and enhanced prompts into the lightweight large language model; S630. Divide the unordered teaching text into teaching segments using the lightweight large language model, and determine the corresponding teaching segment name for each segmented teaching content. S640. Based on the sequence constraint information or reference sequence contained in the structured fusion information, arrange the names of each teaching segment and their corresponding teaching content to obtain an ordered sequence of teaching segments.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves synergistic utilization of macro-level teaching structure information and micro-level teaching content information by matching teaching modes, lesson types, and core concepts with standard teaching sequence and combining the micro-level content retrieval results from a structured teaching sequence knowledge base. Furthermore, it dynamically selects the order constraint method based on the content coverage between the two, avoiding erroneous constraints when the macro template does not match the current teaching content, thereby improving the accuracy and robustness of teaching sequence division, sequence name recognition, and sequence restoration. Simultaneously, this invention inputs the fused structured information into a lightweight large language model, improving the model's ability to process long, unordered teaching texts while reducing computational resource consumption, and enhancing the interpretability and applicability of the recognition results in educational scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the overall method provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the macro template library construction and 3D classifier training process provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the macro-knowledge matching process provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the knowledge base construction process for structured teaching activities provided in this embodiment of the invention; Figure 6 This is a schematic diagram of the microscopic knowledge retrieval process provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the dynamic route verification and fusion process provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the enhanced generation process provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] First, some technical terms used in this invention will be explained: Disordered teaching text: refers to the text formed after deleting the names of teaching segments from the original teaching text and rearranging the order of the corresponding teaching content for each teaching segment.

[0021] Implicit structure recognition refers to the process of dividing teaching texts into teaching segments, recognizing the names of teaching segments, and restoring the order of teaching segments based on the semantics and logic of the teaching content, when the names and original order of teaching segments are not explicitly preserved.

[0022] Macro template library: refers to a collection of teaching templates formed by associating and storing teaching mode tags, lesson type tags, core concept tags and corresponding standard teaching steps in an ordered sequence.

[0023] A structured teaching segment knowledge base is a knowledge base consisting of multiple teaching segment names and their corresponding content fragments, with embedded vectors associated and stored for each content fragment.

[0024] Micro-level teaching content information refers to the candidate teaching segment names and corresponding teaching content obtained by retrieving candidate content fragments from the structured teaching segment knowledge base based on unordered teaching text, and then reordering and dynamically formatting them.

[0025] Content coverage rate: refers to the proportion of the number of teaching segments in the ordered sequence of standard teaching segments that meet the preset matching conditions with the names of normalized micro-level teaching segments, to the total number of standard teaching segments.

[0026] Dynamic routing refers to the process of selecting either the ordered sequence of standard teaching steps as a mandatory sequence constraint or the ordered sequence of normalized micro-level teaching step names as a reference sequence, based on whether the macro-level teaching structure information is empty and whether the content coverage meets the preset coverage conditions.

[0027] Lightweight large language model: refers to a language model with relatively low computational resource requirements that performs teaching segment division, teaching segment name recognition and teaching segment order restoration on unordered teaching text based on structured fusion information and enhanced prompts.

[0028] Example 1 As can be seen from the background technology, existing natural language processing methods are prone to problems such as inaccurate division of teaching segments, incorrect identification of segment names, and disordered restoration of order when processing long teaching texts with missing teaching segment names and disordered teaching content. In addition, they lack a processing mechanism for dynamically integrating macro-level teaching processes and micro-level teaching content.

[0029] Based on the above problems, this embodiment provides a method for identifying implicit structures in teaching segments. By acquiring macro-level teaching structure information and micro-level teaching content information, and selecting the corresponding order constraint method according to the degree of consistency between the two, the method can realize the division of teaching segments, name recognition, and order restoration of unordered teaching texts.

[0030] like Figures 1 to 8 This embodiment illustrates a method for recognizing implicit structures in teaching processes. Assume the acquired lesson plan text contains a total of... The first The lesson plan text is recorded as This results in a collection of lesson plan texts: Each lesson plan text Each includes the names of several teaching segments. and corresponding teaching content The names of the teaching segments in the original lesson plan text are deleted, and the order of the corresponding teaching content paragraphs for each teaching segment is shuffled, resulting in a disordered teaching text with the teaching segment names removed and the order shuffled. .

[0031] For disordered teaching texts First, the teaching content is divided into multiple segments corresponding to different teaching stages, and a name for each teaching stage is assigned to each segment. Then, the identified teaching stage names and their corresponding content are restored in sequence. This task is represented as: in, Indicates the use of disordered teaching texts The teaching process is divided into segments, resulting in unordered names for the teaching segments. and corresponding teaching content ; This indicates that the order of the divided teaching segment names and their corresponding teaching content is restored to obtain an ordered list of teaching segment names. and corresponding teaching content .

[0032] like Figures 1 to 8 As shown, this embodiment includes the following steps.

[0033] S100. Obtain the unordered teaching text. The unordered teaching text is the text formed after deleting the names of the teaching segments from the original teaching text and adjusting the order of the teaching content corresponding to each teaching segment.

[0034] S110. Obtain the original teaching text containing multiple teaching segment names and the teaching content corresponding to each teaching segment name.

[0035] Each original teaching text includes multiple teaching segments arranged according to the original teaching process, and each teaching segment includes the name of the teaching segment and its corresponding teaching content.

[0036] S120. Delete the names of teaching segments in the original teaching text, and retain the teaching content corresponding to each teaching segment.

[0037] After the names of the teaching segments are removed, each segment of teaching content no longer has an explicit label that directly indicates the teaching segment it belongs to.

[0038] S130. Adjust the order of teaching content corresponding to different teaching segments to form an unordered teaching text that does not contain explicit teaching segment names or the original order of teaching segments.

[0039] The rearranged teaching content, along with the other sections, together constitutes a disordered teaching text. Disordered teaching texts As input to S200 and S300.

[0040] S200: Based on the unordered teaching text obtained in S100, determine the teaching mode, lesson type and core concept corresponding to the unordered teaching text, and match the corresponding ordered sequence of standard teaching links from the macro template library according to the teaching mode, lesson type and core concept to obtain macro teaching structure information.

[0041] S210. Collect standard lesson plan texts and label each standard lesson plan text with teaching mode tags, lesson type tags, and core concept tags.

[0042] The teaching model tags include heuristic teaching, lecture-based teaching, inquiry-based teaching, and project-based teaching; the lesson type tags include new lesson, practice lesson, review lesson, and experimental lesson; the core concept tags include "area of ​​a circle" and "quadratic equation in one variable".

[0043] Teaching models, lesson types, and core concepts are all labeled with single tags. Core concepts use a closed vocabulary, but the macro template library supports dynamic expansion. When the same standard lesson plan text involves multiple concepts, the main concepts explicitly mentioned in the key teaching chapters are prioritized as core concept tags.

[0044] S220. Based on the standard teaching process corresponding to each standard lesson plan text, extract the names of the teaching segments from each standard lesson plan text, and arrange the names of the teaching segments according to the standard teaching process to obtain an ordered sequence of standard teaching segments.

[0045] For example, the standard teaching sequence for a new lesson is: introduction to the lesson - explanation of knowledge - consolidation exercises - summary of the lesson - assignment of homework.

[0046] S230. Link and store teaching mode tags, lesson type tags, core concept tags and corresponding standard teaching steps in an ordered sequence to form a macro template library.

[0047] Macro Template Library Represented as: in, Indicates the first Teaching mode tags for each template, Indicates the first A template for lesson type tags, Indicates the first The core concept of a template is tags. This indicates the total number of templates in the macro template library. Indicates the first The standard teaching steps corresponding to each template are in an ordered sequence: in, , , Indicates the first The names of the teaching segments arranged according to the standard teaching process in each template.

[0048] S240. Based on multiple unordered teaching text samples and the teaching mode labels, lesson type labels, and core concept labels corresponding to each unordered teaching text sample, a classifier training dataset is constructed.

[0049] The classifier training dataset is represented as follows: in, For the first A sample of unordered teaching text (the names of teaching segments have been removed, and the order of the corresponding paragraphs has been shuffled). The corresponding teaching mode tags, lesson type tags, and core concept tags are provided.

[0050] The classifier training dataset is divided into training, validation, and test sets, with the training set accounting for 70%, the validation set accounting for 15%, and the test set accounting for 15%. Stratified sampling is used to ensure that the distribution of each category in the training, validation, and test sets is consistent.

[0051] S250. Extract the word frequency-inverse document frequency feature vector of each unordered teaching text sample, and use the word frequency-inverse document frequency feature vector as input to train the teaching mode classifier, lesson type classifier and core concept classifier respectively.

[0052] The formula for calculating the term frequency-inverse document frequency feature is: .

[0053] The maximum number of features parameter in TF-IDF is: The TF-IDF phrase range parameter is: The teaching mode classifier, lesson type classifier, and core concept classifier all use the decision tree algorithm. The splitting criterion of the decision tree is information gain, the maximum depth is not limited, the minimum number of leaf node samples is 1, and the minimum number of split samples is 2.

[0054] The TF-IDF feature vectors of unordered teaching text samples are used as inputs, and the corresponding teaching mode labels, lesson type labels, and core concept labels are used as output targets to train the teaching mode classifier, lesson type classifier, and core concept classifier.

[0055] S260. Input the unordered teaching text obtained in S100 into the teaching mode classifier, lesson type classifier and core concept classifier respectively to obtain the predicted teaching mode label, predicted lesson type label and predicted core concept label.

[0056] For the input unordered teaching text The prediction process of the three classifiers is represented as follows: in, Indicates a predictive teaching model label. Indicates the predicted lesson type label, Indicates the core concept label for prediction. Represents a teaching mode classifier. This represents a lesson type classifier. This represents the core concept classifier.

[0057] S270. Combine the predicted teaching mode tags, predicted lesson type tags, and predicted core concept tags to form a search key, and perform matching in the macro template library based on the search key.

[0058] The search key is represented as: in, This indicates a string concatenation operation.

[0059] Search by key in macro template library In order to perform precise matching.

[0060] S280. When a tag combination that perfectly matches the search key exists in the macro template library, output the ordered sequence of standard teaching steps corresponding to the tag combination; when no perfectly matching tag combination exists, set the macro teaching structure information to empty.

[0061] The macro-level teaching structure information is represented as follows: If a perfect match of 3D labels exists, return the corresponding ordered sequence of standard teaching steps. If no perfectly matching combination of 3D labels exists, an empty set is returned.

[0062] S300: Based on the unordered teaching text obtained in S100, retrieve candidate content fragments related to the unordered teaching text from the structured teaching segment knowledge base, and obtain micro-level teaching content information containing the candidate teaching segment name and corresponding teaching content based on the candidate content fragments.

[0063] S310. Obtain the lesson plan text containing the names of teaching segments and the corresponding teaching content, and process the lesson plan text into multiple teaching segment data pairs. Each teaching segment data pair includes the name of the teaching segment and the content segment corresponding to the name of the teaching segment.

[0064] High-quality lesson plan data was selected as the data source for the structured teaching process knowledge base. The selected lesson plans were converted into TXT format, the content related to the teaching process was extracted, and the content was batch processed into fragment pairs of "teaching process name - corresponding content".

[0065] Structured teaching process knowledge base Represented as: in, This constitutes a segment of the knowledge base for structured teaching activities. Indicates the first Name of each teaching segment Indicates the first Detailed text content corresponding to each teaching segment. This represents the total number of segments in the knowledge base of structured teaching sessions.

[0066] S320. Generate the content embedding vector for each content segment, and store the teaching segment name, corresponding content segment and content embedding vector together.

[0067] The BGE embedding model is used for each content fragment in the knowledge base of the structured teaching process. Generate content embedding vectors: in, Indicates content fragments The corresponding content embedding vector.

[0068] S330. Based on the content embedding vectors corresponding to each content segment, establish a vector index to form a knowledge base for structured teaching links.

[0069] Collect the content embedding vectors corresponding to all content fragments and construct a content fragment vector index: S340: Generate a query embedding vector based on the unordered teaching text obtained in S100.

[0070] For the input unordered teaching text Generate query embedding vectors using the BGE embedding model: in, Indicating unordered teaching texts The corresponding query embedding vector.

[0071] S350. Calculate the cosine similarity between the query embedding vector and the embedding vectors of each content in the structured teaching knowledge base.

[0072] Query embedding vector With content embedding vector The cosine similarity between them is expressed as: S360. Sort each content segment according to cosine similarity, and select the top-ranked content segments to form an initial candidate segment set.

[0073] The content fragments in the knowledge base of the structured teaching process are sorted in descending order based on the cosine similarity score. (Return to previous...) Several relevant content fragments are used to form an initial set of candidate fragments: in, In this embodiment, the initial number of candidate segments to be selected is: S370. The unordered teaching text obtained in S100 is concatenated with each candidate content fragment in the initial candidate fragment set, and the concatenation result is input into the cross encoder to obtain the correlation score between the unordered teaching text and each candidate content fragment.

[0074] For the initial set of candidate fragments: Disordered teaching texts With each candidate content fragment After concatenation, the data is input into the cross-encoder model to obtain the corresponding relevance scores: in, This indicates a text concatenation operation.

[0075] The cross encoder model adopts: The cross encoder is used to reorder the initial retrieval results without additional training on data specific to this domain; the maximum input length of the cross encoder uses the model's default setting.

[0076] When the combined length of the unordered teaching text and candidate content fragments exceeds the maximum input length of the cross encoder, a right-side truncation method is used to truncate the content at the end of the candidate content fragments, while retaining the unordered teaching text. The full content.

[0077] The score output by the cross encoder is used as a relevance score. The higher the relevance score, the higher the semantic matching degree between the unordered teaching text and the candidate content fragments.

[0078] S380. Reorder the initial candidate fragment set according to the relevance score, and select candidate content fragments from the reordered initial candidate fragment set to form the final candidate fragment set.

[0079] The initial candidate fragment set is reordered in descending order based on the relevance score: From the reordered collection of fragments Before being selected The most relevant segments are selected to form the final candidate segment set.

[0080] A typical lesson plan contains 3 to 7 teaching segments; therefore, in this embodiment: or: S390: Obtain the number of query attention heads for the lightweight large language model used to execute S600, and compare the number of query attention heads with a preset format switching threshold.

[0081] Target: Lightweight Large Language Model The query attention head count is represented as: in, This represents the number of query attention heads for the target lightweight large language model. The number of attention heads refers to the number of query attention heads, not the number of key-value attention heads.

[0082] The target lightweight large language model and its query attention head count used in this embodiment are as follows: The preset format switching threshold is: The three lightweight large language models are each adapted to dynamic formatting rules; for Internlm2_5-7b-chat-1m, the results obtained by using the format "teaching segment name + first sentence content" and the format "teaching segment name + complete content" are similar.

[0083] S391. When the number of query attention heads is not less than the preset format switching threshold, the teaching segment name in the final candidate segment set is combined with the corresponding complete content segment to obtain the micro-level teaching content information of the complete content format.

[0084] The complete content format is as follows: in, Indicates and content fragments The corresponding teaching segment name, This indicates the operation of combining the name of a teaching segment with a content fragment.

[0085] S392. When the number of query attention heads is less than the preset format switching threshold, the teaching segment name in the final candidate segment set is combined with the first complete sentence of the corresponding content segment to obtain the micro-teaching content information of the first sentence content format.

[0086] The format of the first sentence is as follows: in, Indicates the return of a content fragment. The first complete sentence in the text.

[0087] The dynamic formatting strategy is represented as: like The output after dynamic formatting for That is, use the format of "Teaching segment name + complete content"; if The output after dynamic formatting for That is, adopt the format of "teaching segment name + first sentence content".

[0088] in, This represents the micro-level teaching content information obtained after dynamic formatting.

[0089] S400: Based on the macro-level teaching structure information obtained in S200 and the micro-level teaching content information obtained in S300, determine the degree of consistency between the ordered sequence of standard teaching steps and the names of teaching steps in the micro-level teaching content information.

[0090] S410. Extract candidate teaching segment names from the micro-level teaching content information obtained in S300, and normalize the candidate teaching segment names according to the synonym normalization mapping to obtain a normalized ordered sequence of micro-level teaching segment names.

[0091] From the perspective of micro-level teaching content information The names of each candidate teaching segment are extracted according to the current arrangement, and a normalized ordered sequence of micro-level teaching segment names is obtained through synonym normalization mapping. .

[0092] Synonym normalization only processes semantically equivalent teaching segment names according to semantic equivalence rules. For example, "review and revisit" is normalized to "review", and "class summary" is normalized to "summary". Teaching segment names with different meanings are not merged. For example, "independent exploration" is not normalized to "explanation".

[0093] To preserve the ordered sequence of the names of the teaching segments, it is not an unordered set.

[0094] S420. For each macroscopic teaching segment name in the ordered sequence of standard teaching segments obtained in S200, determine the fuzzy matching degree between the macroscopic teaching segment name and each microscopic teaching segment name in the ordered sequence of normalized microscopic teaching segment names.

[0095] Macro-level teaching segment name Ordered sequence of normalized microscopic teaching segment names The matching score between them is expressed as: in, This represents the name of a micro-teaching segment in an ordered sequence of normalized micro-teaching segment names. This represents a fuzzy string matching function, and its value range is: S430. Determine the number of macro-level teaching segment names that meet the preset matching conditions based on the fuzzy matching degree.

[0096] The binary matching indicator function is expressed as: in, In this embodiment, the matching similarity threshold is represented as follows: When the name of the macro-teaching segment When the matching score between the micro-teaching segment name and at least one micro-teaching segment name in the ordered sequence of normalized micro-teaching segment names is not less than 60. ;otherwise, .

[0097] S440. Based on the ratio between the number of macro-level teaching segment names that meet the preset matching conditions and the total number of macro-level teaching segment names contained in the ordered sequence of standard teaching segments, determine the content coverage rate between macro-level teaching structure information and micro-level teaching content information, and use the content coverage rate as the degree of consistency.

[0098] Content coverage Represented as: in, Indicates content coverage. This represents the total number of teaching segment names in an ordered sequence of standard teaching segments.

[0099] When the macro-level teaching structure information output by S280 is empty, the content coverage calculation and macro-level prior injection are skipped. The subsequent retrieval process relies solely on the micro-level teaching content information, and a normalized ordered sequence of micro-level teaching segment names is used as a weak reference order. In this embodiment, the macro-level teaching structure information is used as auxiliary information, and the micro-level teaching content information is used as the primary information.

[0100] S500: Based on the degree of consistency determined in S400, the macro-level teaching structure information and the micro-level teaching content information are dynamically integrated.

[0101] S510: Obtain the macro-level teaching structure information obtained in S200 and the content coverage rate determined in S400.

[0102] Set coverage threshold: Based on the relationship between content coverage and coverage threshold, determine whether to inject macro-level teaching structure information: S520. When the macro-level teaching structure information is not empty and the content coverage meets the preset coverage conditions, the ordered sequence of standard teaching links is used as a mandatory sequence constraint and combined with the micro-level teaching content information obtained in S300 to form the first structured fusion information.

[0103] when At that time, the standard teaching steps will be arranged in an orderly sequence. As a mandatory sequential prompt, it is related to micro-level teaching content information. The first structured fusion information is represented as follows: in, This indicates that the ordered sequence of standard teaching steps serves as a mandatory sequential constraint.

[0104] S530. When the macro-level teaching structure information is empty or the content coverage does not meet the preset coverage conditions, the normalized ordered sequence of micro-level teaching link names is used as the reference order and combined with the micro-level teaching content information obtained in S300 to form the second structured fusion information.

[0105] When the macro-level teaching structure information is empty, or when At the same time, instead of injecting macro-level teaching structure information into the lightweight large language model, only the normalized micro-level teaching segment names are ordered. As a weak reference order, it is related to micro-level teaching content information. The fusion process is performed, and the second structured fusion information is represented as follows: in, This indicates that the ordered sequence of normalized microscopic teaching segment names is used as a weak reference order.

[0106] The complete representation of structured fused information is as follows: When the macro-level teaching structure information is empty, the second structured fusion information is used directly.

[0107] S540, take the first structured fusion information or the second structured fusion information as input to S600.

[0108] Based on the structured fusion information obtained in S500, S600 performs teaching segment division, teaching segment name recognition, and teaching segment order restoration on disordered teaching texts to obtain an ordered teaching segment sequence.

[0109] S610, Combine the structured fusion information obtained in S500 with the enhanced prompts.

[0110] Enhanced prompts It includes common process templates for teaching procedures and clear rules for sequential restoration.

[0111] Common teaching process templates include: introduction - explanation - practice - summary - homework; the sequence restoration rule requires a lightweight large language model to identify the name of the teaching process corresponding to each teaching content segment, and to sort the teaching processes according to the teaching logic.

[0112] S620: Input the combined structured fusion information and enhanced prompts into the lightweight large language model.

[0113] The augmentation generation process of a lightweight large language model can be represented as follows: in, This represents the structured fusion information output by the S500. This indicates enhanced prompts. This represents a lightweight, large language model.

[0114] S630. Divide the unordered teaching text into teaching segments using a lightweight large language model, and determine the corresponding teaching segment name for each segment of the teaching content.

[0115] The lightweight large language model divides the unordered teaching text based on the candidate teaching segment names and corresponding content in the micro-level teaching content information, obtains the content corresponding to each teaching segment, and determines the corresponding teaching segment name for each segment.

[0116] S640. Based on the sequence constraint information or reference sequence contained in the structured fusion information, arrange the names of each teaching segment and their corresponding teaching content to obtain an ordered sequence of teaching segments.

[0117] When structured fusion information includes and At that time, the names of each teaching segment and their corresponding teaching content are arranged according to the orderly sequence of standard teaching segments; when the structured integrated information includes and At that time, the normalized ordered sequence of micro-level teaching segment names was used as a weak reference order to arrange the names of each teaching segment and their corresponding teaching content.

[0118] The final output of the ordered sequence of teaching steps is represented as follows: in, Indicates the order after restoration. Name of each teaching segment Indicates the relationship with the first The teaching content corresponding to each teaching segment name.

[0119] After obtaining the ordered sequence of teaching segments, the results of segment division and sequence restoration are quantitatively evaluated in multiple dimensions, including content completeness, sequence consistency, accuracy of segment boundaries, and standardization of names.

[0120] Content completeness is represented as: in, This indicates the total number of segments in the standard answer. Indicates the first The content of a standard answer excerpt Indicates the first The content of each predicted segment. For each standard answer segment, calculate its relationship with each predicted segment. The scores are calculated, and the maximum value is selected. Then, the average of the maximum values ​​corresponding to all standard answer segments is calculated.

[0121] Sequential consistency is represented as: in, Indicates the number of predicted segments. Indicates the first The location of each predicted segment in the standard answer Indicates an indicator function; when When the value is 1, the indicator function takes the value 1; otherwise, it takes the value 0.

[0122] The accuracy of the process boundary is calculated using character positions, with a tolerance window of 50 characters. The accuracy of the process boundary is expressed as: in, This represents the total number of predicted boundary points. Indicates the first The character positions of the predicted boundary points Represents the set of boundary points of the standard answer. Indicates the boundary points of the standard answer. This represents the character distance between two boundary points.

[0123] For each predicted boundary point, calculate its character distance from each standard answer boundary point and select the minimum character distance. When the minimum character distance is not greater than 50, score it progressively using (1-d / 50); when the minimum character distance is greater than 50, the score for that predicted boundary point is 0. The average score of all predicted boundary points is taken as the link boundary accuracy.

[0124] The predicted boundary and the standard answer boundary are extracted according to the same boundary detection rules, which include: recognizing the format of “teaching session name”, recognizing the format of “teaching session name”, recognizing the teaching session name marked with double quotes, recognizing the teaching session name ending with a Chinese colon, and recognizing independent lines consisting of at least two Chinese characters.

[0125] The standardization of names is calculated based on the degree of fuzzy matching between the predicted teaching segment names and the standard teaching segment names, and is expressed as follows: in, This indicates the number of predicted teaching segment names. This indicates the number of teaching segment names in the standard answer. Indicates the first The name of each predictive teaching segment. This represents a function for fuzzy string matching.

[0126] This invention achieves synergistic utilization of macro-level teaching structure information and micro-level teaching content information by matching teaching modes, lesson types, and core concepts with standard teaching sequence and combining the micro-level content retrieval results from a structured teaching sequence knowledge base. Furthermore, it dynamically selects the order constraint method based on the content coverage between the two, avoiding erroneous constraints when the macro template does not match the current teaching content, thereby improving the accuracy and robustness of teaching sequence division, sequence name recognition, and sequence restoration. Simultaneously, this invention inputs the fused structured information into a lightweight large language model, improving the model's ability to process long, unordered teaching texts while reducing computational resource consumption, and enhancing the interpretability and applicability of the recognition results in educational scenarios.

[0127] For each predicted teaching segment name, calculate its fuzzy matching score with each standard teaching segment name in the standard teaching segment name set, and select the maximum fuzzy matching score; average the maximum fuzzy matching scores corresponding to all predicted teaching segment names, and multiply by the quantity balance factor, which is used to prevent over-segmentation of disordered teaching text.

[0128] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, it should be understood that the sequence number of each step in the above embodiments does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The actions or steps recorded in the claims can be performed in a different order than that in the above embodiments and can still achieve the desired result. In addition, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0129] Example 2 Based on the same inventive concept, corresponding to any of the above embodiments, this disclosure also provides a system for recognizing implicit structures in teaching processes, including an unordered teaching text acquisition module, a macro-level teaching structure acquisition module, a micro-level teaching content acquisition module, a consistency degree determination module, a dynamic fusion module, and an enhanced generation module.

[0130] The module for obtaining disordered teaching text is used to obtain the original teaching text, delete the names of teaching segments from the original teaching text, and adjust the order of the teaching content corresponding to each teaching segment to form disordered teaching text.

[0131] The macro-level teaching structure acquisition module is used to determine the teaching mode, lesson type, and core concepts corresponding to unordered teaching texts. Based on the teaching mode, lesson type, and core concepts, it matches the corresponding ordered sequence of standard teaching steps from the macro-level template library to obtain macro-level teaching structure information.

[0132] The macro-level teaching structure acquisition module includes a macro-level template library construction unit, a classifier training unit, and a macro-level template matching unit.

[0133] The macro template library construction unit is used to collect standard lesson plan texts and label each standard lesson plan text with teaching mode tags, lesson type tags, and core concept tags; according to the standard teaching process, it generates a corresponding ordered sequence of standard teaching steps for each standard lesson plan text; and it stores the teaching mode tags, lesson type tags, core concept tags, and the corresponding ordered sequence of standard teaching steps in association to form the macro template library.

[0134] The classifier training unit is used to construct a training dataset containing unordered teaching text samples and corresponding teaching mode labels, lesson type labels and core concept labels. It extracts the word frequency-inverse document frequency features of each unordered teaching text sample and trains the teaching mode classifier, lesson type classifier and core concept classifier respectively.

[0135] The macro-template matching unit is used to input unordered teaching text into the teaching mode classifier, lesson type classifier, and core concept classifier respectively, and obtain predicted teaching mode labels, predicted lesson type labels, and predicted core concept labels. Based on these predicted labels, it performs precise matching in the macro-template library. When a perfectly matching label combination exists, it outputs the corresponding ordered sequence of standard teaching steps; when no perfectly matching label combination exists, the macro-teaching structure information is set to empty.

[0136] The micro-level teaching content acquisition module is used to retrieve candidate content fragments related to unordered teaching texts from the structured teaching segment knowledge base, and obtain micro-level teaching content information containing the candidate teaching segment name and corresponding teaching content based on the candidate content fragments.

[0137] The micro-level teaching content acquisition module includes a knowledge base construction unit, a vector coarse retrieval unit, a cross encoder reordering unit, and a dynamic formatting unit.

[0138] The knowledge base construction unit is used to process lesson plan texts into fragment pairs of teaching segment names and corresponding content segments, generate content embedding vectors for each content segment, and establish vector indexes based on the content embedding vectors to form a structured teaching segment knowledge base.

[0139] The vector coarse retrieval unit is used to generate query embedding vectors for unordered teaching texts, calculate the cosine similarity between the query embedding vectors and the content embedding vectors in the structured teaching knowledge base, and select an initial candidate fragment set based on the cosine similarity.

[0140] The cross-encoder reordering unit is used to concatenate the unordered teaching text with each candidate content fragment in the initial candidate fragment set, input the concatenation result into the cross-encoder model, obtain the relevance score between the unordered teaching text and each candidate content fragment, reorder the initial candidate fragment set according to the relevance score, and select candidate content fragments from the reordered initial candidate fragment set to form the final candidate fragment set.

[0141] The dynamic formatting unit is used to obtain the number of query attention heads of the target lightweight large language model and compare the number of query attention heads with a preset format switching threshold. When the number of query attention heads is not less than the preset format switching threshold, the candidate teaching segment name is combined with the corresponding complete content segment. When the number of query attention heads is less than the preset format switching threshold, the candidate teaching segment name is combined with the first complete sentence of the corresponding content segment to obtain micro-level teaching content information.

[0142] The consistency determination module is used to extract candidate teaching segment names from micro-level teaching content information and perform synonym normalization on the candidate teaching segment names to obtain a normalized ordered sequence of micro-level teaching segment names. It determines the fuzzy matching degree between each teaching segment name in the ordered sequence of standard teaching segments and the normalized micro-level teaching segment names. Based on the ratio between the number of teaching segment names that meet the preset matching conditions and the total number of teaching segment names in the ordered sequence of standard teaching segments, it determines the content coverage rate and uses the content coverage rate as the consistency degree between macro-level teaching structure information and micro-level teaching content information.

[0143] The dynamic fusion module is used to dynamically fuse macro-level teaching structure information and micro-level teaching content information based on whether the macro-level teaching structure information is empty and the relationship between content coverage and coverage threshold.

[0144] When the macro-level teaching structure information is not empty and the content coverage rate is not less than the coverage rate threshold, the dynamic fusion module will use the ordered sequence of standard teaching links as a mandatory sequential constraint and merge it with the micro-level teaching content information.

[0145] When the macro-level teaching structure information is empty or the content coverage rate is less than the coverage rate threshold, the dynamic fusion module does not inject the macro-level teaching structure information. Instead, it uses the normalized ordered sequence of micro-level teaching segment names as a weak reference order and merges it with the micro-level teaching content information to obtain structured fusion information.

[0146] The enhancement generation module is used to input structured fusion information and enhanced prompts into a lightweight large language model. The lightweight large language model divides the unordered teaching text into teaching segments, determines the corresponding teaching segment name for each segmented teaching content, and arranges the teaching segment names and corresponding teaching content according to the mandatory order constraints or weak reference order in the structured fusion information to obtain an ordered sequence of teaching segments.

[0147] In this embodiment, the construction method of the macro template library, the training parameters of the classifier, the vector retrieval method, the cross-encoder model, the dynamic formatting parameters, the content coverage calculation method, the threshold setting, the dynamic fusion method, and the enhancement generation process are all the same as in Embodiment 1, and will not be repeated here.

[0148] The system described in the above embodiments is used to implement a corresponding method for identifying implicit structures in teaching processes in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0149] It should be noted that the aforementioned system for recognizing implicit structures in teaching processes is embodied in the form of functional units. The term "module" here can be implemented in software and / or hardware, without specific limitations.

[0150] For example, a "module" can be a software program, hardware circuit, or a combination of both that implements the above functions. Hardware circuits may include application-specific integrated circuits (ASICs), electronic circuits, processors (e.g., shared processors, proprietary processors, or group processors) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functions.

[0151] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for identifying implicit structures in teaching processes, characterized in that, The method includes the following steps: S100. Obtain unordered teaching text, wherein the unordered teaching text is the text formed after deleting the names of teaching segments from the original teaching text and adjusting the order of the teaching content corresponding to each teaching segment. S200: Based on the unordered teaching text obtained in S100, determine the teaching mode, lesson type and core concept corresponding to the unordered teaching text, and match the corresponding ordered sequence of standard teaching links from the macro template library according to the teaching mode, lesson type and core concept to obtain macro teaching structure information; S300: Retrieve candidate content fragments related to the unordered teaching text from the structured teaching segment knowledge base, and obtain micro-level teaching content information containing the candidate teaching segment name and corresponding teaching content based on the candidate content fragments; S400: Based on the macro-level teaching structure information obtained in S200 and the micro-level teaching content information obtained in S300, determine the degree of consistency between the ordered sequence of the standard teaching steps and the names of the teaching steps in the micro-level teaching content information. S500: Based on the consistency level determined in S400, the macro-level teaching structure information and micro-level teaching content information are dynamically integrated; when the consistency level meets the preset conditions, the ordered sequence of the standard teaching steps is integrated as the sequence constraint information; when the consistency level does not meet the preset conditions, the order of the teaching steps in the micro-level teaching content information is integrated as the reference order to obtain structured integrated information. S600: Based on the structured fusion information obtained in S500, the disordered teaching text is divided into teaching segments, the names of teaching segments are identified, and the order of teaching segments is restored to obtain an ordered sequence of teaching segments.

2. The method according to claim 1, characterized in that, S100 includes the following steps: S110. Obtain the original teaching text containing multiple teaching segment names and teaching content corresponding to each teaching segment name; S120. Delete the names of teaching segments in the original teaching text, and retain the teaching content corresponding to each teaching segment; S130. Adjust the order of teaching content corresponding to different teaching segments to form an unordered teaching text that does not contain explicit teaching segment names or the original order of teaching segments.

3. The method according to claim 1, characterized in that, The macro template library in S200 is constructed according to the following steps: S210. Collect standard lesson plan texts and label each standard lesson plan text with teaching mode tags, lesson type tags, and core concept tags; S220. Based on the standard teaching process corresponding to each standard lesson plan text, extract and arrange the names of teaching links in each standard lesson plan text to obtain an ordered sequence of standard teaching links. S230. The teaching mode tags, lesson type tags, core concept tags and corresponding standard teaching steps are associated and stored in an ordered sequence to form the macro template library; Among them, the teaching mode label, lesson type label and core concept label are all single labels, and the core concept label is determined from a preset core concept vocabulary list, which can be expanded.

4. The method according to claim 3, characterized in that, S200 further includes the following steps: S240. Based on multiple unordered teaching text samples and the teaching mode labels, lesson type labels, and core concept labels corresponding to each unordered teaching text sample, a classifier training dataset is constructed. S250. Extract the word frequency-inverse document frequency feature vector of each unordered teaching text sample, and train the teaching mode classifier, lesson type classifier and core concept classifier based on the word frequency-inverse document frequency feature vector respectively. S260. Input the unordered teaching text obtained in S100 into the teaching mode classifier, lesson type classifier and core concept classifier respectively to obtain the predicted teaching mode label, predicted lesson type label and predicted core concept label. S270. Combine the predicted teaching mode tag, predicted lesson type tag, and predicted core concept tag to form a search key, and perform matching in the macro template library according to the search key; S280. When a tag combination matching the search key exists in the macro template library, the ordered sequence of standard teaching steps corresponding to the tag combination is output; when no tag combination matching the search key exists, the macro teaching structure information is set to empty, and the injection of the ordered sequence of standard teaching steps is skipped.

5. The method according to claim 1, characterized in that, The structured teaching knowledge base in S300 is constructed according to the following steps: S310. Obtain a lesson plan text containing the name of the teaching segment and the corresponding teaching content, and process the lesson plan text into multiple teaching segment data pairs. Each teaching segment data pair includes the name of the teaching segment and a content fragment corresponding to the name of the teaching segment. S320. Generate content embedding vectors for each content segment, and store the teaching segment name, the corresponding content segment, and the content embedding vector together. S330. Based on the content embedding vectors corresponding to each content segment, a vector index is established to form the knowledge base of the structured teaching process.

6. The method according to claim 5, characterized in that, The candidate content fragment retrieval in S300 includes the following steps: S340. Generate a query embedding vector based on the unordered teaching text obtained in S100; S350. Calculate the cosine similarity between the query embedding vector and the embedding vectors of each content in the structured teaching knowledge base; S360. Sort each content segment according to the cosine similarity, and select the top-ranked content segments to form an initial candidate segment set.

7. The method according to claim 6, characterized in that, The S300 further includes the following steps: S370. The disordered teaching text obtained in S100 is concatenated with each candidate content segment in the initial candidate segment set, and the concatenation result is input into the cross encoder to obtain the correlation score between the disordered teaching text and each candidate content segment. S380. The initial candidate fragment set is reordered according to the relevance score, and candidate content fragments are selected from the reordered initial candidate fragment set to form the final candidate fragment set. S390. Obtain the number of query attention heads for the lightweight large language model used to execute S600, and compare the number of query attention heads with a preset format switching threshold. S391. When the number of query attention heads is not less than the preset format switching threshold, the teaching segment name in the final candidate segment set is combined with the corresponding complete content segment to obtain the micro-level teaching content information of the complete content format. S392. When the number of query attention heads is less than the preset format switching threshold, the teaching segment name in the final candidate segment set is combined with the first complete sentence of the corresponding content segment to obtain the micro-teaching content information of the first sentence content format.

8. The method according to claim 1, characterized in that, The S400 includes the following steps: S410. Extract candidate teaching segment names from the micro-teaching content information obtained in S300, and normalize the candidate teaching segment names according to the synonym normalization mapping to obtain a normalized ordered sequence of micro-teaching segment names. S420. For each macroscopic teaching segment name in the ordered sequence of standard teaching segments obtained in S200, determine the fuzzy matching degree between the macroscopic teaching segment name and each microscopic teaching segment name in the ordered sequence of normalized microscopic teaching segment names. S430. Determine the number of macro-level teaching segment names that meet the preset matching conditions based on the fuzzy matching degree. S440. Based on the ratio between the number of macro-level teaching segment names that meet the preset matching conditions and the total number of macro-level teaching segment names contained in the ordered sequence of standard teaching segments, determine the content coverage rate between the macro-level teaching structure information and the micro-level teaching content information, and use the content coverage rate as the degree of consistency.

9. The method according to claim 8, characterized in that, The S500 includes the following steps: S510, Obtain the macro-level teaching structure information obtained in S200 and the content coverage rate determined in S400; S520. When the macro-level teaching structure information is not empty and the content coverage rate meets the preset coverage conditions, the ordered sequence of the standard teaching links is used as a mandatory sequence constraint and combined with the micro-level teaching content information obtained in S300 to form the first structured fusion information. S530. When the macro-teaching structure information is empty or the content coverage does not meet the preset coverage conditions, the normalized micro-teaching link name ordered sequence is used as a reference order and combined with the micro-teaching content information obtained in S300 to form the second structured fusion information. S540, take the first structured fusion information or the second structured fusion information as input to S600.

10. The method according to claim 9, characterized in that, The S600 includes the following steps: S610. Combine the structured fusion information obtained in S500 with the enhanced prompts, wherein the enhanced prompts include common process templates and sequence restoration rules for teaching steps; S620. Input the combined structured fusion information and enhanced prompts into the lightweight large language model; S630. Divide the unordered teaching text into teaching segments using the lightweight large language model, and determine the corresponding teaching segment name for each segmented teaching content. S640. Based on the sequence constraint information or reference sequence contained in the structured fusion information, arrange the names of each teaching segment and their corresponding teaching content to obtain an ordered sequence of teaching segments.

Images