Intelligent courseware production method and system based on AIGC

Abstract

The present application relates to AIGC and intelligent courseware making technical field, specifically for intelligent courseware making method and system based on AIGC, including: the acquisition of the preset static courseware assets;Build teaching knowledge graph;Generate global constraint vector representing the integrity of course structure;Trigger dynamic content generation process;The semantic feature extraction of personalized inquiry demand is obtained, and the query vector is obtained, and the inquiry deviation is determined by semantic deviation calculation;Through constraint probability distribution correction processing, generate new content module;Based on the new content module, the query vector and the global constraint vector, the fitness is determined;Judge whether the fitness is greater than the preset fitness threshold;In response to the fitness being greater than the preset fitness threshold, dynamically update the teaching knowledge graph and the global constraint vector;In response to the fitness not being greater than the preset fitness threshold, discard the new content module;The present application solves the technical pain point that content generation is uncontrollable and easy to deviate from the teaching syllabus.

Description

Technical Field

[0001] This invention relates to the field of AIGC and intelligent courseware production technology, specifically to an AIGC-based intelligent courseware production method and system. Background Technology

[0002] In the field of intelligent education technology, using AIGC to generate personalized and dynamic teaching content has become a key development direction. Its goal is to improve the real-time adaptability of the teaching process to the individual needs of students. Current technology is trying to shift from distributing pre-set, structured static courseware assets to a real-time content generation model driven by AIGC.

[0003] However, existing AIGC applications generally lack refined constraints and guidance mechanisms when generating educational content. While this unconstrained generation process offers high flexibility, it can also easily lead to content deviating from the pre-set syllabus and core knowledge points, resulting in uncontrollable teaching objectives and chaotic knowledge structures. In existing technologies, content generation systems struggle to effectively balance the goals of immediate adaptability and curriculum structure integrity. Furthermore, existing methods generally lack a closed-loop mechanism for quality assessment and feedback of new content, preventing the effective retention and reuse of high-quality content generated in response to student inquiry, thus hindering the self-evolution and iterative optimization of the teaching knowledge system. Therefore, how to provide an intelligent courseware production method that can dynamically respond to personalized inquiry, ensure that the generated content always conforms to teaching constraints, and achieve closed-loop self-reconstruction of the knowledge graph is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for creating intelligent courseware based on AIGC. Specifically, the technical solution of this invention is as follows:

[0005] A method for creating intelligent courseware based on AIGC, comprising:

[0006] Collect pre-set static courseware assets;

[0007] Based on static courseware assets, the logical relationships between core knowledge points are analyzed to construct a teaching knowledge graph;

[0008] Based on the teaching knowledge graph and combined with the preset teaching weights, the core knowledge points are vectorized and a global constraint vector representing the integrity of the course structure is initialized and generated.

[0009] In response to students' personalized inquiry needs captured in real time, a dynamic content generation process is triggered, including:

[0010] S1, extract semantic features from personalized exploration needs to obtain query vectors;

[0011] S2, combining the query vector and the global constraint vector, determines the exploration deviation through semantic deviation calculation;

[0012] S3 uses the query vector, exploration deviation, and global constraint vector as the generation context of AIGC, and generates new content modules through constraint probability distribution correction.

[0013] S4, based on the new content module, query vector and global constraint vector, evaluates and determines the survival suitability;

[0014] S5, determine whether the survival rate is greater than the preset survival rate threshold;

[0015] S6. In response to the survival rate being greater than the preset survival rate threshold, the teaching knowledge graph and global constraint vector are dynamically updated.

[0016] S7. If the survival rate is not greater than the preset survival rate threshold, the newly generated content module is discarded.

[0017] Optional, constrained probability distribution correction processing includes:

[0018] Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model.

[0019] The constrained probability distribution and the original output probability distribution of the AIGC model are dynamically interpolated based on the degree of deviation to obtain the corrected probability distribution.

[0020] The AIGC model uses a modified probability distribution to sample lexical terms and generate new content modules.

[0021] Optionally, assess and determine survivability, including:

[0022] Calculate the responsiveness of the new content module corresponding to the query vector and the consistency of the new content module corresponding to the global constraint vector, respectively.

[0023] Based on responsiveness and consistency, and combined with preset balancing weights, the fitness is calculated through linear weighted summation.

[0024] Optional, dynamic updates for S6 include:

[0025] Find the best connection points in the teaching knowledge graph;

[0026] New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the best connection points.

[0027] Obtain the original number of knowledge points in the teaching knowledge graph;

[0028] Set the survival rate as the teaching weight for new knowledge points;

[0029] By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

[0030] An AIGC-based intelligent courseware production system includes:

[0031] The teaching intent analysis module is used to collect preset static courseware assets, construct a teaching knowledge graph, and initialize and generate a global constraint vector representing the integrity of the course structure.

[0032] The demand capture module is used to capture students' personalized inquiry needs in real time.

[0033] The intelligent courseware generation module is designed to respond to personalized inquiry needs by dynamically generating content and reconstructing graphs. The intelligent courseware generation module includes:

[0034] The demand vectorization unit is used to extract semantic features from personalized exploration demands to obtain query vectors.

[0035] The deviation calculation unit is used to combine the query vector and the global constraint vector to determine the exploration deviation.

[0036] The content generation unit is used to generate new content modules by using query vectors, exploration deviations, and global constraint vectors as the generation context for AIGC.

[0037] The survival assessment unit is used to assess and determine survival based on the new content module, query vector, and global constraint vector.

[0038] The graph update unit is used to dynamically update the teaching knowledge graph and global constraint vector in response to the survival degree being greater than the preset survival degree threshold.

[0039] The content discarding unit is used to discard newly generated content modules in response to a survival rate not exceeding a preset survival rate threshold.

[0040] Optional, content generation unit, used for:

[0041] Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model.

[0042] The constrained probability distribution and the original output probability distribution of the AIGC model are dynamically interpolated based on the degree of deviation to obtain the corrected probability distribution.

[0043] The AIGC model uses a modified probability distribution to sample lexical terms and generate new content modules.

[0044] Optional, a survival assessment unit, used for:

[0045] Calculate the responsiveness of the new content module corresponding to the query vector and the consistency of the new content module corresponding to the global constraint vector, respectively.

[0046] Based on responsiveness and consistency, and combined with preset balancing weights, the fitness is calculated through linear weighted summation.

[0047] Optional, a map update unit, used for:

[0048] Find the best connection points in the teaching knowledge graph;

[0049] New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the best connection points.

[0050] Obtain the original number of knowledge points in the teaching knowledge graph;

[0051] Set the survival rate as the teaching weight for new knowledge points;

[0052] By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. This invention solves the technical pain points of uncontrollable content generation and easy deviation from the teaching syllabus by constructing a global constraint vector representing the integrity of the course structure and performing probability correction during AIGC generation;

[0055] 2. This invention introduces the exploration deviation as a dynamic weight to dynamically interpolate the original output probability and constraint probability of AIGC, thereby achieving a precise balance between responding to students' immediate exploration and maintaining the structural integrity of the teaching main line.

[0056] 3. This invention establishes an objective survival evaluation mechanism. New content must simultaneously meet the requirements of responsiveness to student needs and consistency with the teaching syllabus in order to pass the quality threshold judgment, effectively preventing low-quality or irrelevant content from polluting the knowledge graph.

[0057] 4. This invention realizes the closed-loop self-reconstruction of the teaching knowledge graph. High-quality new content that has passed the fitness assessment is added back to the knowledge graph, and the global constraint vector is dynamically updated, enabling the courseware system to self-evolve and iteratively optimize during the teaching process. Attached Figure Description

[0058] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0059] Figure 1 This is a flowchart of the method of the present invention;

[0060] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0062] Example 1:

[0063] Please see Figure 1 A method for creating intelligent courseware based on AIGC, comprising:

[0064] Collect pre-set static courseware assets;

[0065] Based on static courseware assets, the logical relationships between core knowledge points are analyzed to construct a teaching knowledge graph;

[0066] Based on the teaching knowledge graph and combined with the preset teaching weights, the core knowledge points are vectorized and a global constraint vector representing the integrity of the course structure is initialized and generated.

[0067] In response to students' personalized inquiry needs captured in real time, a dynamic content generation process is triggered, including:

[0068] S1, extract semantic features from personalized exploration needs to obtain query vectors;

[0069] S2, combining the query vector and the global constraint vector, determines the exploration deviation through semantic deviation calculation;

[0070] S3 uses the query vector, exploration deviation, and global constraint vector as the generation context of AIGC, and generates new content modules through constraint probability distribution correction.

[0071] S4, based on the new content module, query vector and global constraint vector, evaluates and determines the survival suitability;

[0072] S5, determine whether the survival rate is greater than the preset survival rate threshold;

[0073] S6. In response to the survival rate being greater than the preset survival rate threshold, the teaching knowledge graph and global constraint vector are dynamically updated.

[0074] S7. If the survival rate is not greater than the preset survival rate threshold, the newly generated content module is discarded.

[0075] This embodiment provides a method for creating intelligent courseware based on AIGC;

[0076] Collect pre-set static courseware assets:

[0077] Static courseware assets refer to teaching materials that are pre-set by teachers and have a fixed structure; their purpose is to provide the system with the original input for building the teaching syllabus; in this embodiment, it may specifically include a Word or PDF document of the teaching syllabus, static PowerPoint courseware, teacher handouts, reference textbook chapters, etc.

[0078] Based on static courseware assets, the logical relationships between core knowledge points are analyzed to construct a teaching knowledge graph:

[0079] The purpose of this step is to transform unstructured or semi-structured teaching materials into machine-readable structured knowledge networks. In this embodiment, the system utilizes natural language processing techniques, such as the named entity recognition and relation extraction model based on the G06F40 / 00 classification, to automatically extract core knowledge points from the collected assets. and the logical relationships between knowledge points Examples include prerequisites, inclusion in, and derived from, thereby constructing an initial pedagogical knowledge graph, denoted as . This atlas It represents the teacher's pre-set teaching intentions and the structural integrity of the course;

[0080] Based on the teaching knowledge graph and combined with the preset teaching weights, the core knowledge points are vectorized to initialize and generate a global constraint vector representing the integrity of the course structure.

[0081] The purpose of this step is to transform abstract teaching intentions. Quantized into a computable global constraint vector that represents the structural integrity of the courseware, denoted as ; In this invention, it serves as the semantic anchor or centroid of the teaching syllabus; in this embodiment, its calculation method is as follows:

[0082]

[0083] in, The global constraint vector is a high-dimensional semantic vector.

[0084] The total number of knowledge points is a scalar; its source is the pedagogical knowledge graph. Core knowledge points The total number;

[0085] : No. The teaching weight of each knowledge point is a scalar; its source can be pre-set by the teacher based on teaching experience, for example... The larger the value, the more important it is. Alternatively, it can be calculated using graph topology algorithms, such as PageRank, to reflect the importance of knowledge points in the graph.

[0086] : No. The semantic feature vectors of each knowledge point are high-dimensional vectors; they are derived from pre-trained language models, such as BERT or GPT, based on the knowledge points. The text description is encoded to obtain it;

[0087] This formula uses the semantic vectors of all knowledge points in the graph. By performing a weighted average, a centroid vector representing the core intent of the entire course was created. ;Should It will serve as the core constraint benchmark and be passed on to subsequent deviation calculations and AIGC content generation steps;

[0088] Responding to students' personalized inquiry needs captured in real time, such as when a student inputs a question through the courseware's interactive interface: "Why is the Laplace transform used here?", the system triggers a dynamic content generation process, including:

[0089] S1, extract semantic features from personalized exploration needs to obtain the query vector:

[0090] Query vector, denoted as It is a vectorized representation of students' real-time exploration needs; its purpose is to transform students' natural language questions into interactive representations. Semantic vectors for mathematical comparisons; in this embodiment, their source is the use and generation. The same pre-trained language model is used for real-time queries from students. That is, personalized exploration needs are vectorized and encoded.

[0091] S2, combining the query vector and the global constraint vector, determines the exploration deviation through semantic deviation calculation:

[0092] Explore the degree of deviation, denoted as It is used to determine student needs. Whether it belongs to the broader scope or cross-disciplinary research is a key indicator; its purpose is to quantify immediate adaptability. With structural integrity The semantic distance between them; in this embodiment, it is calculated as follows:

[0093]

[0094] in, This refers to the degree of deviation, which is a scalar between 0 and 2.

[0095] : Query vector, derived from step S1;

[0096] : Global constraint vector, derived from the initialization steps;

[0097] The initial source of this formula is a variant of standard cosine similarity. ; A higher value indicates a greater need for student inquiry. Compared with the preset outline The greater the semantic deviation; This will be passed as a key parameter to step S3 to dynamically adjust the constraint strength generated by AIGC.

[0098] S3 uses the query vector, exploration deviation, and global constraint vector as the generation context for AIGC, and generates new content modules through constraint probability distribution correction.

[0099] This step is the content generation phase of AIGC; , and They are used together as the AIGC generation context; AIGC in the module that generates new content is denoted as At that time, its generation process is affected and The common constraints; the specific implementation of this constraint probability distribution correction process will be detailed in Example 2;

[0100] S4, based on the new content module, query vector, and global constraint vector, evaluates and determines the survival suitability:

[0101] The purpose of this step is to generate S3. Conduct a quality assessment to determine its eligibility for integration back into a permanent instructional knowledge graph. Medium; survivability, denoted as This is a comprehensive evaluation score; the specific implementation method of this evaluation will be detailed in Example 3.

[0102] S5, Determine if the sustainability is greater than the preset sustainability threshold:

[0103] The preset survival threshold is denoted as: This is a pre-set quality threshold by the system; its source is based on the experience of instructional designers, or preferably, it is determined after evaluating historical generated data, for example... This threshold is used to ensure that only high-quality and relevant content is adopted.

[0104] S6, in response to a survival rate greater than a preset survival rate threshold, i.e. Then, the teaching knowledge graph and global constraint vector are dynamically updated:

[0105] This step signifies the creation of new content modules. Content deemed high-quality and relevant will undergo self-evolution of the graph; the specific implementation of this dynamic update will be detailed in Example 4.

[0106] S7, in response to the survival rate not exceeding a preset survival rate threshold, i.e. Then discard the newly generated content module:

[0107] This step means If a data entry fails the quality assessment, it is deemed invalid and generated to prevent it from contaminating the knowledge graph.

[0108] Example 2:

[0109] Constrained probability distribution correction processing includes:

[0110] Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model.

[0111] The constrained probability distribution and the original output probability distribution of the AIGC model are dynamically interpolated based on the degree of deviation to obtain the corrected probability distribution.

[0112] The AIGC model uses a modified probability distribution to sample lexical terms and generate new content modules.

[0113] This embodiment is a concretization and optimization of the constrained probability distribution correction process in Embodiment 1; the purpose of this process is to, in real time, modify the original output probability of each word generated by AIGC with a constrained probability distribution. The guided constraint probabilities are dynamically fused, and this process specifically includes:

[0114] Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model:

[0115] The purpose of this step is to calculate a probability distribution that reflects each lexical unit in the vocabulary. With the teaching syllabus The semantic relevance; the constrained probability distribution is denoted as The calculation method is as follows:

[0116]

[0117] in, Constrained probability distribution, scalar, representative lexical unit exist Probability under constraints;

[0118] : lexical element, These are the lexical units currently being evaluated. Traverse the entire vocabulary of the AIGC model;

[0119] : Lexical vector, a high-dimensional vector; its source is the vector representation of the corresponding lexical in the word embedding layer inside the AIGC model;

[0120] : Global constraint vector, derived from the steps in Example 1;

[0121] Temperature parameter, a scalar; its source is either empirically preset or determined through experimental optimization, for example... Used for regulation The sharpness of the distribution; The smaller the value, the more concentrated the distribution is. Lexical units with high semantic similarity;

[0122] This formula is calculated. and The dot product, i.e., the semantic similarity score, is transformed into an efficient probability distribution that sums to 1 over the entire vocabulary using the Softmax function. ;

[0123] By dynamically interpolating the constrained probability distribution with the original output probability distribution of the AIGC model based on the exploration deviation, the corrected probability distribution is obtained:

[0124] This step utilizes S2 calculations. As dynamic interpolation weights; the corrected probability distribution is denoted as... The calculation method is as follows:

[0125]

[0126] in, : Corrected probability distribution, scalar, the probability that AIGC will ultimately use for sampling;

[0127] The original output probability distribution of the AIGC model is a scalar; its source is the original output of the AIGC model, such as the last Softmax layer of the Transformer.

[0128] : Constraint probability distribution, scalar, derived from step 1 of this embodiment;

[0129] : Investigating the deviation, a scalar, derived from step S2 of Example 1;

[0130] The global constraint strength hyperparameter is a scalar; it is set experimentally and finely on a set of calibration datasets, for example, 0.1-0.5, and its range is constrained to ensure... The number of items is no greater than 1, used for control. The overall strength of the constraint;

[0131] This formula utilizes That is, to explore the degree of deviation, it acts as a dynamic weight adjuster;

[0132] Derivation 1, Inquiry within the syllabus: When students inquire... With the outline When highly consistent, At this point, the weighted term The formula degenerates into AIGC will primarily use its raw probabilities. This process ensures both the smoothness and creativity of the generated content.

[0133] Derivation 2, Inquiry Beyond the Syllabus: When Students Inquire With the outline When the deviation is large, ;at this time, Constrained distribution Activated and weighted; AIGC generates lexical units At that time, they tend to choose those High-probability morphemes, i.e., those related to the teaching syllabus Semantic related lexical units, thereby bringing the deviated exploration back to the main teaching line;

[0134] The AIGC model uses a modified probability distribution to sample terms and generate new content modules.

[0135] AIGC no longer uses the original Instead, use Lexical sampling, such as Top-k, Top-p, or kernel sampling, is performed to generate new content modules word by word, which are then aggregated into new content modules. .

[0136] Example 3:

[0137] The assessment determines suitability, including:

[0138] Calculate the responsiveness of the new content module corresponding to the query vector and the consistency of the new content module corresponding to the global constraint vector, respectively.

[0139] Based on responsiveness and consistency, and combined with preset balancing weights, the fitness is calculated through linear weighted summation.

[0140] This embodiment is a concretization and optimization of the assessment and determination of survivability in Embodiment 1; the purpose of this assessment is to use a quantifiable indicator. Objectively judge the new content modules The overall quality, i.e., whether it should be adopted;

[0141] The assessment specifically includes:

[0142] Calculate the responsiveness of the newly generated content module corresponding to the query vector, and the consistency of the newly generated content module corresponding to the global constraint vector, respectively:

[0143] responsiveness The purpose is to evaluate To what extent did it solve the students' inquiry problem? ;

[0144] Consistency The purpose is to evaluate To what extent does it conform to the teaching syllabus? ;

[0145] Based on responsiveness and consistency, and combined with preset balanced weights, the survivability is calculated through a linear weighted sum:

[0146] The purpose of this step is to combine the assessments from the two dimensions mentioned above into a single fitness score. Fractions; in this embodiment, they are calculated as follows:

[0147]

[0148] in, This refers to scalar fitness, a quantity such as 0 or 1.

[0149] Preset balance weights, scalars, for example Its origin lies in the priority set by instructional designers for the two objectives of responding to student needs and maintaining syllabus consistency.

[0150] Similarity functions, such as the standard cosine similarity function;

[0151] : The semantic vector of the new content module, a high-dimensional vector; it is obtained by using and , Same model pair The text content is encoded to obtain it;

[0152] : Query vector, derived from step S1;

[0153] : Global constraint vector, derived from the initialization steps;

[0154] The formula is passed The weighting achieves a comprehensive consideration of responsiveness and consistency, where responsiveness is... Consistency level is Only when At the same time, when it meets the needs of students well and does not deviate excessively from the syllabus, that is... and At the same time, when it is relatively high, its overall survivability Only then will it be high.

[0155] Example 4:

[0156] S6's dynamic updates include:

[0157] Find the best connection points in the teaching knowledge graph;

[0158] New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the best connection points.

[0159] Obtain the original number of knowledge points in the teaching knowledge graph;

[0160] Set the survival rate as the teaching weight for new knowledge points;

[0161] By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

[0162] This embodiment is a concretization and optimization of the dynamic update of S6 described in Embodiment 1; the purpose of this update is to update high-quality new content that has passed the adaptability assessment. Permanently integrate it back into the teaching system to realize knowledge graph and constraint vector Closed-loop iteration;

[0163] This dynamic update specifically includes:

[0164] Finding the optimal connection points in the instructional knowledge graph:

[0165] The purpose of this step is to... In the existing map The system calculates the most suitable mounting location; in this embodiment, the system calculates... Vectors and graphs All existing knowledge points Based on semantic similarity, the knowledge points with the highest similarity are identified and determined as the best connection points. ;

[0166] New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the optimal connection points:

[0167] This step is the map. Structural evolution; in this embodiment, the system will As a new knowledge point Add to graph In the middle, and establish and New logical relationships between them, such as those related to or derived from, form an updated knowledge graph. ;

[0168] The number of original knowledge points obtained from the teaching knowledge graph:

[0169] That is, get the update before Total number of knowledge points This parameter originates from the initialization steps of Example 1;

[0170] Set the survival rate as the teaching weight for new knowledge points:

[0171] New knowledge points ,Right now The teaching weight is denoted as And it is dynamically set; its source is set to the fitness level calculated in Example 3. ,Right now ;

[0172] The value itself represents The overall quality and relevance are used as the weight for new knowledge points. This means that high-quality, highly relevant new content will be abundant in the future. It occupies a higher weight in the calculation, which is a positive feedback incentive for the successful generation of AIGC;

[0173] By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

[0174] This step is The purpose of iterative evolution is to make Able to reflect The changes; the updated constraint vector is denoted as The calculation method is as follows:

[0175]

[0176] in, : The updated constraint vector, a high-dimensional vector;

[0177] The original number of knowledge points is a scalar, derived from step 3 of this embodiment.

[0178] : The original constraint vector, a high-dimensional vector, is derived from the initialization steps in Example 1;

[0179] The new point weight, a scalar, originates from step 4 of this embodiment, i.e. ;

[0180] New module vector, high-dimensional vector, i.e. semantic vector;

[0181] According to the definition in Example 1, Equivalent to That is, the sum of the centroids of the original vectors; this formula Its physical meaning is to assign weights of New knowledge points It was efficiently incorporated into the centroid calculation, resulting in a new, representative... global constraint vector ;

[0182] Will replace This serves as the input for the system to execute step S2 of Example 1 next time, thereby achieving... A complete technological closed loop.

[0183] Example 5:

[0184] Please see Figure 2 An AIGC-based intelligent courseware production system includes:

[0185] The teaching intent analysis module is used to collect preset static courseware assets, construct a teaching knowledge graph, and initialize and generate a global constraint vector representing the integrity of the course structure.

[0186] The demand capture module is used to capture students' personalized inquiry needs in real time.

[0187] The intelligent courseware generation module is designed to respond to personalized inquiry needs by dynamically generating content and reconstructing graphs. The intelligent courseware generation module includes:

[0188] The demand vectorization unit is used to extract semantic features from personalized exploration demands to obtain query vectors.

[0189] The deviation calculation unit is used to combine the query vector and the global constraint vector to determine the exploration deviation.

[0190] The content generation unit is used to generate new content modules by using query vectors, exploration deviations, and global constraint vectors as the generation context for AIGC.

[0191] The survival assessment unit is used to assess and determine survival based on the new content module, query vector, and global constraint vector.

[0192] The graph update unit is used to dynamically update the teaching knowledge graph and global constraint vector in response to the survival degree being greater than the preset survival degree threshold.

[0193] The content discarding unit is used to discard newly generated content modules in response to a survival rate not exceeding a preset survival rate threshold.

[0194] This embodiment correspondingly provides an AIGC-based intelligent courseware production system, which is configured to execute the method described in Embodiment 1 above; in this embodiment, the system, for example deployed on a cloud server or local computer, includes:

[0195] The instructional intent parsing module aims to perform initialization; it is configured to collect preset static courseware assets, construct an instructional knowledge graph, and initialize and generate a global constraint vector representing the integrity of the course structure. ;

[0196] The requirement capture module aims to capture students' real-time input; it is configured to capture students' personalized inquiry needs in real time. ;

[0197] The intelligent courseware generation module aims to perform a dynamic generation and reconstruction cycle. It is configured to respond to personalized inquiry needs, performing dynamic content generation and self-reconstruction of courseware maps. This module specifically includes:

[0198] Demand vectorization unit: used to execute S1, Convert to query vector ;

[0199] Deviation calculation unit: used to execute S2, combined with and Determine the degree of deviation in the investigation ;

[0200] Content generation unit: used to execute S3, which will... , and As context, generate new content modules ;

[0201] Survivability Assessment Unit: Used to execute S4 and assess and determine survivability. ;

[0202] Map update unit: used to execute S6, in The teaching knowledge graph is updated dynamically in time. and global constraint vector ;

[0203] Content discarding unit: used to execute S7, in At that time, discard the new content module. ;

[0204] The method and system described in this embodiment construct a global constraint vector representing the teaching intention. Furthermore, by dynamically constraining the AIGC generation process, it solves the technical pain points of uncontrollable AIGC generated content and easy deviation from the teaching syllabus in existing technologies. At the same time, through adaptability assessment and closed-loop self-reconstruction of the knowledge graph, the courseware system can not only respond to students' personalized exploration, i.e., immediate adaptability, but also ensure that the generated content always conforms to the structural integrity of the course, and realize the self-evolution and iterative optimization of the knowledge graph in the teaching process.

[0205] Example 6:

[0206] Content generation unit, used for:

[0207] Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model.

[0208] The constrained probability distribution and the original output probability distribution of the AIGC model are dynamically interpolated based on the degree of deviation to obtain the corrected probability distribution.

[0209] The AIGC model uses a modified probability distribution to sample lexical terms and generate new content modules.

[0210] In the system embodiment corresponding to Example 5, the content generation unit is specifically configured as follows:

[0211] Perform step 1 above, based on , and Calculate the constrained probability distribution ;

[0212] Perform step 2 above, based on , , and The corrected probability distribution is calculated through dynamic interpolation. ;

[0213] Perform step 3 above, and use Perform lexical sampling to generate new content modules ;

[0214] This embodiment introduces... and The calculation provides a refined AIGC output constraint mechanism; this mechanism does not impose hard constraints on AIGC, such as content filtering, but rather performs soft-guided probability correction; especially through... As dynamic weights, the AIGC generation process is... Dynamic anchoring; deviation The larger the value, the stronger the anchoring pull, thus ensuring that when AIGC responds to inquiries beyond the syllabus, its generated results are both innovative and do not deviate from the main teaching line, achieving a precise balance between the degree of freedom of generation and the constraints of teaching.

[0215] Example 7:

[0216] The survival assessment unit is used for:

[0217] Calculate the responsiveness of the new content module corresponding to the query vector and the consistency of the new content module corresponding to the global constraint vector, respectively.

[0218] Based on responsiveness and consistency, and combined with preset balancing weights, the fitness is calculated through linear weighted summation.

[0219] In the system embodiment corresponding to Example 5, the survivability assessment unit is specifically configured as follows:

[0220] Perform step 1 above to calculate the response degree. and consistency ;

[0221] Perform step 2 above, based on Response and consistency are calculated using a linear weighted sum to obtain the fitness score. ;

[0222] This embodiment defines... The calculation method provides a quantifiable, objective, and multi-objective quality assessment standard; it avoids subjective judgment of the generated content and ensures that only high-quality content that simultaneously meets the two core objectives of responding to students' individual needs and conforming to the integrity of the syllabus structure is produced. Only when the content is selected can it be adopted by the system and proceed to the next stage of the knowledge graph update process, effectively preventing low-quality or irrelevant content from polluting the teaching knowledge graph.

[0223] Example 8:

[0224] The map update unit is used for:

[0225] Find the best connection points in the teaching knowledge graph;

[0226] New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the best connection points.

[0227] Obtain the original number of knowledge points in the teaching knowledge graph;

[0228] Set the survival rate as the teaching weight for new knowledge points;

[0229] By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

[0230] In the system embodiment corresponding to Example 5, the map update unit is specifically configured as follows:

[0231] Perform steps 1 and 2 above to find the optimal connection point. And in Add ;

[0232] Perform steps 3 and 4 above to obtain the original number of knowledge points. And set the teaching weight for new knowledge points. ;

[0233] Perform step 5 above, combined with , , and The updated constraint vector is generated by calculating the incremental weighted average. ;

[0234] This embodiment defines a spectrum. and constraint vector Its dynamic update mechanism provides a complete closed-loop feedback system, enabling the teaching system to self-evolve and dynamically reconstruct itself. New knowledge generated by AIGC in response to student needs and proven to be of high quality will, in turn, enhance and expand the original curriculum, thus updating it. and through The iterative updates dynamically and progressively adjust the constraints on subsequent content generation, enabling the courseware system to be continuously optimized during use, becoming increasingly aligned with actual teaching needs and students' research interests.

[0235] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for creating intelligent courseware based on AIGC, characterized in that, include: Collect pre-set static courseware assets; Based on static courseware assets, the logical relationships between core knowledge points are analyzed to construct a teaching knowledge graph; Based on the teaching knowledge graph and combined with the preset teaching weights, the core knowledge points are vectorized and a global constraint vector representing the integrity of the course structure is initialized and generated. In response to students' personalized inquiry needs captured in real time, a dynamic content generation process is triggered, including: S1, extract semantic features from personalized exploration needs to obtain query vectors; S2, combining the query vector and the global constraint vector, determines the exploration deviation through semantic deviation calculation; S3 uses the query vector, exploration deviation, and global constraint vector as the generation context of AIGC, and generates new content modules through constraint probability distribution correction. S4, based on the new content module, query vector and global constraint vector, evaluates and determines the survival suitability; S5, determine whether the survival rate is greater than the preset survival rate threshold; S6. In response to the survival rate being greater than the preset survival rate threshold, the teaching knowledge graph and global constraint vector are dynamically updated. S7. If the survivability is not greater than the preset survivability threshold, then discard the newly generated content module. Constrained probability distribution correction processing includes: Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model. The constrained probability distribution and the original output probability distribution of the AIGC model are dynamically interpolated based on the degree of deviation to obtain the corrected probability distribution. The AIGC model uses a modified probability distribution to sample words and generate new content modules. The assessment determines suitability, including: Calculate the responsiveness of the new content module corresponding to the query vector and the consistency of the new content module corresponding to the global constraint vector, respectively. Based on responsiveness and consistency, and combined with preset balancing weights, the fitness is calculated through linear weighted summation.

2. The method for creating intelligent courseware based on AIGC according to claim 1, characterized in that, S6's dynamic updates include: Find the best connection points in the teaching knowledge graph; New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the best connection points. Obtain the original number of knowledge points in the teaching knowledge graph; Set the survival rate as the teaching weight for new knowledge points; By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

3. An AIGC-based intelligent courseware production system, based on the AIGC-based intelligent courseware production method according to any one of claims 1-2, characterized in that, include: The teaching intent analysis module is used to collect preset static courseware assets, construct a teaching knowledge graph, and initialize and generate a global constraint vector representing the integrity of the course structure. The demand capture module is used to capture students' personalized inquiry needs in real time. The intelligent courseware generation module is designed to respond to personalized inquiry needs by dynamically generating content and reconstructing graphs. The intelligent courseware generation module includes: The demand vectorization unit is used to extract semantic features from personalized exploration demands to obtain query vectors. The deviation calculation unit is used to combine the query vector and the global constraint vector to determine the exploration deviation. The content generation unit is used to generate new content modules by using query vectors, exploration deviations, and global constraint vectors as the generation context for AIGC. The survival assessment unit is used to assess and determine survival based on the new content module, query vector, and global constraint vector. The graph update unit is used to dynamically update the teaching knowledge graph and global constraint vector in response to the survival degree being greater than the preset survival degree threshold. The content discarding unit is used to discard newly generated content modules in response to a survival rate not exceeding a preset survival rate threshold.

4. The AIGC-based intelligent courseware production system according to claim 3, characterized in that, Content generation unit, used for: Based on the global constraint vector, the constraint probability distribution is calculated on the vocabulary of the AIGC model. The constrained probability distribution and the original output probability distribution of the AIGC model are dynamically interpolated based on the degree of deviation to obtain the corrected probability distribution. The AIGC model uses a modified probability distribution to sample lexical terms and generate new content modules.

5. The AIGC-based intelligent courseware production system according to claim 3, characterized in that, The survival assessment unit is used for: Calculate the responsiveness of the new content module corresponding to the query vector and the consistency of the new content module corresponding to the global constraint vector, respectively. Based on responsiveness and consistency, and combined with preset balancing weights, the fitness is calculated through linear weighted summation.

6. The AIGC-based intelligent courseware production system according to claim 3, characterized in that, The map update unit is used for: Find the best connection points in the teaching knowledge graph; New content modules are added as new knowledge points to the teaching knowledge graph, and new logical relationships are established with the best connection points. Obtain the original number of knowledge points in the teaching knowledge graph; Set the survival rate as the teaching weight for new knowledge points; By combining the global constraint vector, new knowledge points, teaching weights, and the original number of knowledge points, an updated constraint vector is generated through incremental weighted averaging.

Images