AIGC content generation method and system based on multi-source feature fusion

Abstract

The invention discloses an AIGC content generation method and system based on multi-source feature fusion, and relates to the technical field of artificial intelligence, and the method comprises the following steps: importing a teaching outline of each learning section, determining a knowledge boundary of each learning section, setting a knowledge point progressive rule according to the teaching outline and the knowledge boundary, setting a generation rule according to a content type, and generating the content according to the generation rule; based on the knowledge point progressive rule and the set learning section standard rule, the knowledge point progressive rule, the generation rule and the learning section standard rule are fused, and the AIGC content is generated in combination with the user generation request content. According to the method, a generation rule covering courseware and analysis and a learning section standard rule are set, multi-modal materials are fused through a cross-modal semantic matching algorithm, and a multi-dimensional verification mechanism is established, so that the accuracy and normalization of multi-modal content fusion are effectively improved, and the high adaptability and teaching application value of final output content are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an AIGC content generation method and system based on multi-source feature fusion. Background Technology

[0002] In recent years, with the rapid development of AI-generated content technology, its application in the education field has become increasingly widespread, especially in the automated generation of teaching resources, where it has shown great potential. AIGC technology can now generate multimodal content such as text, images, and audio based on user input, providing new possibilities for personalized learning and teaching assistance. The generation of educational content has high professional requirements, needing to strictly follow the knowledge boundaries and progressive logic of the teaching syllabus, while taking into account the cognitive characteristics and learning needs of students at different grade levels. How to deeply integrate AIGC technology with professional knowledge in the field of education to achieve content generation that both conforms to teaching principles and meets personalized needs has become an important research direction.

[0003] Existing technologies and methods lack systematic processing of teaching outlines and knowledge boundaries, making it difficult to accurately capture the vertical hierarchy and horizontal connections between knowledge points. The generated content is poor in terms of logical coherence and knowledge progression. Existing technologies are relatively simplistic in setting content generation rules and grade level standards, and cannot flexibly adjust generation strategies according to the cognitive characteristics of students at different grade levels. The generated content is difficult to match actual teaching needs in terms of language style, presentation, and difficulty gradient. The multimodal content fusion generation and verification mechanism is not yet perfect, resulting in significant room for improvement in the accuracy, adaptability, and standardization of the final generated content. Summary of the Invention

[0004] The technical problem addressed by this invention is that existing methods lack systematic processing of teaching outlines and knowledge boundaries, making it difficult to accurately capture the vertical hierarchy and horizontal relationships between knowledge points. The generated content is poor in terms of logical coherence and knowledge progression. Existing technologies are relatively simplistic in setting content generation rules and grade-level standards, failing to flexibly adjust generation strategies according to the cognitive characteristics of students at different grade levels. The generated content is difficult to match actual teaching needs in terms of language style, presentation, and difficulty gradient. The multimodal content fusion generation and verification mechanism is still imperfect, resulting in significant room for improvement in the accuracy, adaptability, and standardization of the final generated content.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an AIGC content generation method based on multi-source feature fusion, comprising the following steps: Step S1: Import the teaching syllabus for each learning stage, determine the knowledge boundaries for each learning stage, and set the knowledge point progression rules according to the teaching syllabus and knowledge boundaries. Step S2: Based on the knowledge point progression rules, set generation rules according to content type. The generation rules include courseware template generation rules and parsing template generation rules. Step S3: Based on the knowledge point progression rules, set the learning stage standard rules, which include lower learning stage standard rules, middle learning stage standard rules and upper learning stage standard rules; Step S4: Integrate the knowledge point progression rules, generation rules, and learning stage standard rules, and generate AIGC content in conjunction with the user-generated request content.

[0006] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, step S1 specifically includes: Import the teaching syllabus for each grade level, including lower, middle and upper grades. Standardize the format of the teaching syllabus, verify and complete it, and determine the knowledge boundaries of each grade level based on the processed teaching syllabus. The rules for progressively advancing knowledge points are set according to the aforementioned syllabus and knowledge boundaries.

[0007] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, the determination of the knowledge boundaries for each learning stage specifically includes: Based on the processed teaching syllabus, the core attributes of each knowledge point are determined. The core attributes include the knowledge point name, the knowledge point connotation, the knowledge point extension, and the relationship between the knowledge point and other knowledge points, and are stored in a structured knowledge base. The terminology mentioned includes terms commonly used in the field of academic qualifications. The knowledge points mentioned include their core concepts, principles, formulas, and conditions for use. The scope of the knowledge points includes their application scenarios; The relationships mentioned include both vertical and horizontal hierarchical relationships between various knowledge points.

[0008] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, the step of setting knowledge point progression rules according to the teaching syllabus and knowledge boundaries specifically includes: The knowledge point progression rule includes marking the prerequisite and subsequent knowledge points for core and non-core knowledge points respectively. For core knowledge points, subject matter experts manually annotate the prerequisite and subsequent knowledge points of the core knowledge points in the structured knowledge base to obtain the manual annotation results; For non-core knowledge points, semantic encoding is performed on the text containing the non-core knowledge points, and feature vectors of the non-core knowledge points are extracted. The co-occurrence frequency and teaching sequence of non-core knowledge points in historical teaching data are analyzed according to the association rule mining algorithm, and the preceding and subsequent knowledge points of the non-core knowledge points are automatically labeled to obtain automatic labeling results.

[0009] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, step S2 specifically includes: The courseware template generation rules include a sequence of instructions for generating courseware content, which includes an introduction instruction, an explanation instruction, a derivation instruction, an analysis instruction, and a summary instruction. The introduction instructions are used to introduce target knowledge points by combining real-life examples, interesting questions, or stories from the history of the subject. The explanation instructions are used to present the target knowledge points in the form of definitions and concrete explanations; The derivation instruction is used to generate a complete derivation process of the formula and principle of the target knowledge point; The analysis command is used to generate typical examples of different difficulties based on the target knowledge points. The typical examples include the question stem, solution steps, and a summary of the thought process. The summary instruction is used to generate the core points and related relationships of the target knowledge point, and the related relationships include the relationship between the knowledge point and the preceding knowledge point and the relationship between the knowledge point and the subsequent knowledge point; The parsing template generation rules include a sequence of instructions for generating parsing content, which includes instructions for reviewing the question, brainstorming, calculating, answering, and prompting. The question review instructions are used to analyze the question stem of the target question and extract key information and known conditions from the question stem. The key information includes the question's focus, the question stem's limiting conditions, the core research object, data information, and clues to related knowledge points. The conceptualization instruction is used to determine the type of the target question based on the key information and known conditions, and to generate a textual description including the problem-solving approach and the reasons for the selection. The question types include basic concept application questions, logical reasoning questions, comprehensive cross-knowledge point questions, practical scenario application questions, calculation and derivation questions, and proof and derivation questions. The calculation instructions are used to generate the solution steps for the target problem and the reasoning process for the solution steps; The answering instruction is used to generate the final answer to the target question; The prompt instructions are used to generate common mistakes and avoidance methods for the question type.

[0010] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, step S3 specifically includes: The aforementioned standards and rules for lower grades include: For text-based sentences, the vocabulary is selected from a list of commonly used words in primary school, and the sentence length is within the first preset range; For image-based content, a cartoon style will be used; For audio and video categories, the audio uses child-friendly voiceovers and simple background music, the video duration is within a second preset range, and the calculation process is demonstrated with animation; The standards and rules for the secondary school level include: For text-based sentences, the vocabulary uses subject-specific terminology, the sentence length is within the third preset range, and simple compound sentences are used. For image-based content, a style combining cartoon and realistic elements is adopted; For audio and video categories, the audio uses standard Mandarin dubbing without background music, the video length is within the fourth preset range, and experimental demonstrations and formula derivations are added. The aforementioned standards and rules for higher education include: For text-based texts, subject-specific terminology is used, along with complex sentence structures; For image-based content, use a realistic style or professional charts; For audio and video categories, the audio uses academic lecture-style narration without background music, and the video length is within the fifth preset range, with the addition of academic discussions, literature citations, and experimental demonstrations.

[0011] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, step S4 specifically includes: The system receives user-generated content requests, which include a target learning stage, content type, and target knowledge points. Based on the request content, it determines the knowledge point progression rules, content type generation rules, and learning stage standard rules for the target knowledge points. The system then integrates these rules with the user-generated content to generate AIGC content.

[0012] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, the specific steps of fusing the knowledge point progression rules, generation rules, and academic stage standard rules, and combining them with user-generated request content to generate AIGC content, include: Based on the knowledge point progression rule, the prerequisite and subsequent knowledge points of the target knowledge point are determined, and the prerequisite and subsequent knowledge points are added before and after the content of the target knowledge point to generate an initial content framework. The generation rule is invoked, and an instruction sequence is generated using AIGC. Based on the instruction sequence, fill text matching the initial content frame is generated in the corresponding part of the initial content frame. Simultaneously with the generation of the fill text, according to the standard rules of the learning stage, the content of relevant elements is matched from the preset material library through a cross-modal semantic matching algorithm. The relevant elements include text, images, audio and video. Based on the content logic of the fill text, the content of the relevant elements is added to the corresponding position of the fill text to generate the final text. The final text is subjected to multi-dimensional verification. If the multi-dimensional verification passes, AIGC content is generated. If the multi-dimensional verification fails, the corresponding verification step is returned and AIGC content is regenerated.

[0013] As a preferred embodiment of the AIGC content generation method based on multi-source feature fusion described in this invention, the multi-dimensional verification of the final text specifically includes: The multi-dimensional verification includes logical verification, format verification and modal verification. If any dimension verification fails, the multi-dimensional verification is deemed to have failed. The logical verification compares the matching degree between the final text and the knowledge point progression rule using a semantic similarity algorithm. If the matching degree is greater than a preset first threshold, the logical verification is deemed to have passed; if the matching degree is less than or equal to the preset first threshold, the logical verification is deemed to have failed. The format validation is used to check whether the final text's layout, font, font size, paragraph spacing, heading level, and citations conform to the generation rules and general specifications. If they conform, the format validation is considered to have passed; otherwise, the format validation is considered to have failed. The modal validation calculates the correlation between the content of the relevant elements and the corresponding final text fragment using a cross-modal semantic matching algorithm. If the correlation is greater than a preset second threshold, the format validation is deemed to have passed; if the correlation is less than or equal to the preset second threshold, the format validation is deemed to have failed.

[0014] The AIGC content generation system based on multi-source feature fusion is applied to the AIGC content generation method based on multi-source feature fusion, and includes a delimitation module, a scaling block, a scaling module, and a generation module. The boundary definition module is used to import the teaching syllabus of each learning stage, determine the knowledge boundaries of each learning stage, and set the knowledge point progression rules according to the teaching syllabus and knowledge boundaries. The fixed-size block is used to set generation rules based on the knowledge point progression rules and according to the content type. The generation rules include courseware template generation rules and parsing template generation rules. The calibration module is used to set grade level standard rules based on the knowledge point progression rules. The grade level standard rules include lower grade level standard rules, middle grade level standard rules and upper grade level standard rules. The generation module is used to integrate the knowledge point progression rules, generation rules, and academic stage standard rules to generate AIGC content.

[0015] The beneficial effects of this invention are as follows: By standardizing and verifying the format of the teaching syllabus and constructing a structured knowledge base containing core attributes and relationships, this invention can systematically process the teaching syllabus and knowledge boundaries, accurately capture the vertical hierarchy and horizontal relationships between knowledge points, thereby ensuring that the generated content has rigorous logical coherence and scientific knowledge progression. It sets generation rules covering courseware and analysis, as well as grade-level standard rules, integrating knowledge point progression rules, content types, and grade-level standards. This allows the generation strategy to be flexibly adjusted according to the cognitive characteristics of students at different grade levels, ensuring that the generated content accurately matches actual teaching needs in terms of language style, presentation, and difficulty gradient. Furthermore, by integrating multimodal materials through a cross-modal semantic matching algorithm and establishing a multi-dimensional verification mechanism including logic, format, and modality, it effectively improves the accuracy and standardization of multimodal content integration, ensuring the high adaptability and teaching application value of the final output content. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the basic process of an AIGC content generation method based on multi-source feature fusion, provided as an embodiment of the present invention.

[0017] Figure 2 This is a basic flowchart of an AIGC content generation system based on multi-source feature fusion, provided as an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1 As an embodiment of the present invention, an AIGC content generation method based on multi-source feature fusion is provided, comprising the following steps: Step S1: Import the teaching syllabus for each learning stage, determine the knowledge boundaries for each learning stage, and set the rules for the progression of knowledge points based on the teaching syllabus and knowledge boundaries.

[0020] Step S2: Based on the knowledge point progression rule, set generation rules according to the content type. The generation rules include courseware template generation rules and parsing template generation rules.

[0021] Step S3: Based on the knowledge point progression rule, set the standard rules for each learning stage. The standard rules for each learning stage include the standard rules for lower learning stages, the standard rules for middle learning stages, and the standard rules for higher learning stages.

[0022] Step S4: Integrate the knowledge point progression rules, generation rules, and learning stage standard rules, and generate AIGC content based on the user-generated request content.

[0023] It can effectively improve the adaptability and reliability of AIGC teaching content. Teachers no longer need to spend a lot of time screening and adjusting the content generated by AIGC to adapt to the grade level or supplement the knowledge logic. They can directly use the generated content for teaching practice, which greatly improves the efficiency of lesson preparation. The content that students come into contact with is always in line with their own grade level and cognitive ability, and they can absorb knowledge step by step, avoiding learning confusion or loss of interest caused by inappropriate content difficulty. At the same time, standardized courseware and analysis templates can also ensure the accuracy and coherence of knowledge transmission.

[0024] Step S1 specifically includes: Import the teaching syllabus for each grade level, including lower, middle and upper grades. Standardize the format of the teaching syllabus, verify and complete it, and determine the knowledge boundaries of each grade level based on the processed teaching syllabus.

[0025] The rules for progressively advancing knowledge points are set according to the teaching syllabus and knowledge boundaries.

[0026] By standardizing the format, the differences in expression and structural logic between teaching syllabi of different educational stages are eliminated, and the originally fragmented and inconsistent syllabus content is transformed into a standardized dataset, providing a unified benchmark for subsequent cross-educational knowledge comparison and connection analysis.

[0027] Verification and completion processes ensure the completeness and accuracy of the teaching syllabus, preventing deviations in the subsequent division of knowledge boundaries due to issues such as missing knowledge points, vague descriptions, or gaps in the connection between learning stages in the original syllabus. At the same time, they make the teaching requirements for each learning stage clearer and more practical.

[0028] Based on the processed syllabus, knowledge boundaries are determined and knowledge point progression rules are set. This clearly defines the knowledge scope and depth for lower, middle, and higher learning stages, avoiding confusion or gaps in knowledge between stages. At the same time, it constructs a knowledge point progression path that conforms to students' cognitive patterns, ensuring that the subsequently generated teaching content can be matched to the learning abilities of students at different stages in a step-by-step manner, without problems such as being too difficult or too superficial. This provides a scientific and rigorous knowledge framework to support the entire AIGC teaching content generation process.

[0029] Determining the knowledge boundaries for each learning stage specifically includes: Based on the processed syllabus, the core attributes of each knowledge point are determined. The core attributes include the knowledge point name, the knowledge point connotation, the knowledge point extension, and the relationship between the knowledge point and other knowledge points, and are stored in a structured knowledge base.

[0030] The knowledge points include terms that are generally applicable to the academic field.

[0031] The names of knowledge points should uniformly adopt terms that are generally applicable to the academic field, avoiding the use of regional or personalized terms, otherwise it will affect the universality of the knowledge base and the accuracy of subsequent cross-grade knowledge comparisons.

[0032] The content of a knowledge point includes its core concepts, principles, formulas, and conditions for use.

[0033] The scope of knowledge points includes their application scenarios.

[0034] The relationships include both vertical and horizontal hierarchical relationships between various knowledge points.

[0035] The vertical and horizontal hierarchical relationships between knowledge points are the core logic for building a knowledge system and clarifying the boundaries of knowledge at different grade levels. The vertical hierarchical relationship reflects the progressive difficulty and depth of knowledge points and is a "hierarchical" connection of subject knowledge. For example, in mathematics, the understanding of numbers extends downward to integers and fractions, and fractions are further refined into the meaning of fractions and fraction operations. This relationship determines the boundary division of lower grades focusing on basic knowledge points and higher grades delving into more detailed content. The horizontal hierarchical relationship is the "parallel" collaboration of knowledge points at the same level of difficulty. For example, in narrative writing, character description, environmental description, and plot development, although there is no progression in difficulty, they all support the writing goal and may also exist across disciplines (such as volume calculation in mathematics and volume measurement in physics). Clarifying the knowledge combination that needs to be mastered simultaneously at the same grade level, sorting out these two relationships can accurately define the depth of mastery and scope of collaboration of knowledge points, providing logical support for subsequent rule setting and content generation.

[0036] By clarifying the core attributes such as the name, connotation, extension, and relationships of knowledge points, the potentially abstract and fragmented content of the teaching syllabus is transformed into concrete and structured knowledge units. For example, the knowledge point of equations is broken down into its general name, the core concept of equations with unknowns, the application scenarios for solving practical problems, and its vertical hierarchical relationship with algebra and its horizontal relationship with word problems, making the boundaries and positioning of each knowledge point clear and identifiable.

[0037] Storing these structured attributes in a knowledge base provides a searchable and comparable basis for the subsequent division of knowledge boundaries for each learning stage, avoiding knowledge overlap or gaps between learning stages due to ambiguity in understanding knowledge points. For example, by using the vertical hierarchical relationship between linear equations and quadratic equations in the knowledge base, it can be clearly determined that the former belongs to the middle school stage and the latter belongs to the upper school stage, ensuring the scientific nature of the knowledge boundary division.

[0038] The clear understanding of the relationships between knowledge points, especially the vertical hierarchy (number recognition - integers - fractions) and the horizontal hierarchy (fractions - fraction addition and subtraction - fraction word problems), lays a logical foundation for setting subsequent rules for the progression of knowledge points. This ensures that the knowledge at each stage not only has clear boundaries but also forms a coherent knowledge network, avoiding knowledge fragmentation.

[0039] The rules for setting progressive knowledge points based on the teaching syllabus and knowledge boundaries specifically include: The knowledge point progression rule includes marking the prerequisite and subsequent knowledge points for core and non-core knowledge points respectively.

[0040] For core knowledge points, subject matter experts manually annotate the prerequisite and subsequent knowledge points in the structured knowledge base to obtain the manual annotation results.

[0041] For non-core knowledge points, semantic encoding is performed on the text containing the connotation of non-core knowledge points, and feature vectors of non-core knowledge points are extracted. Based on the association rule mining algorithm, the co-occurrence frequency and teaching sequence of non-core knowledge points in historical teaching data are analyzed, and the preceding and subsequent knowledge points of non-core knowledge points are automatically labeled to obtain automatic labeling results.

[0042] For core knowledge points, subject matter experts manually annotate preceding and subsequent knowledge points to ensure that their progression aligns with the nature of the subject and the principles of teaching. This avoids logical deviations caused by mechanical algorithmic judgments and builds a reliable framework for the knowledge system. For non-core knowledge points, semantic encoding and feature vector extraction combined with association rule mining algorithms analyze historical data, enabling efficient processing of the annotation of a large number of non-core knowledge points' progression relationships. By leveraging the teaching practice patterns contained in the data, the annotation process achieves scalability and automation while ensuring synergy with the progression logic of core knowledge points. This combination of methods not only guarantees the rigor of the progression of core knowledge but also improves the efficiency and coverage of the overall rule setting, providing a clear knowledge connection path for the subsequent generation of content for different learning stages.

[0043] Step S2 specifically includes: The courseware template generation rules include a sequence of instructions for generating courseware content. The instruction sequence includes an introduction instruction, an explanation instruction, a derivation instruction, an analysis instruction, and a summary instruction.

[0044] The "Introduction" instruction is used to introduce the target knowledge point by combining real-life examples, interesting questions, or stories from the history of the subject.

[0045] Explanation instructions are used to present target knowledge points in the form of definitions and concrete explanations.

[0046] The derivation command is used to generate the complete derivation process of the formulas and principles of the target knowledge point.

[0047] The analysis command is used to generate typical examples of varying difficulty based on the target knowledge points. Typical examples include the question stem, solution steps, and a summary of the thought process.

[0048] The summary command is used to generate the core points and related relationships of the target knowledge points. The related relationships include the relationship between the knowledge points and the prerequisite knowledge points and the relationship between the knowledge points and the subsequent knowledge points.

[0049] Real-life examples, engaging questions, and historical anecdotes are key introductory materials that connect students with new knowledge in a concrete way. Real-life examples are close to everyday experiences, such as using the fact that a person can float while swimming and that a stone can sink to introduce buoyancy, giving abstract concepts a real-world basis. Engaging questions stimulate thinking by creating suspense, such as asking whether the probability of getting tails on the 11th toss after 10 heads increases, thus breaking cognitive inertia. Historical anecdotes add interest by connecting knowledge with its development, such as telling the legend of Pythagoras discovering the Pythagorean theorem from a paving stone, making the knowledge points more vivid. All three can stimulate learning interest from different dimensions and lay a good foundation for subsequent teaching.

[0050] The implementation of each instruction needs to be adjusted in terms of weight and presentation according to the characteristics of the learning stage. For example, in the courseware for lower grades, the derivation instructions should simplify or even omit complex processes, focusing on the fun of introduction and explanation, while in higher grades, the rigor of derivation and the depth of analysis need to be strengthened. The real-life examples or stories selected for the introduction instructions should be highly relevant to the target knowledge points, avoiding deviation from the core of teaching due to excessive pursuit of fun. The difficulty gradient of the examples in the analysis instructions should match the rules of knowledge point progression and the rules of the learning stage standards, ensuring that the examples are both challenging and within the students' ability range. The presentation of the relationships in the summary instructions should be concise and clear. The emphasis of each instruction should be flexibly adjusted according to the characteristics of the subject. For example, the derivation instructions may be weakened in humanities subjects, while they need to be strengthened in science subjects.

[0051] The courseware content is designed to progress step-by-step according to cognitive principles. Introducing instructions uses real-life examples and engaging questions to reduce the unfamiliarity of knowledge points, quickly attracting students' attention and establishing a connection between knowledge and reality. Explaining instructions combines definitions with concrete descriptions, ensuring accuracy while making abstract concepts easy to understand. Derivation instructions present the complete derivation process of formulas and principles, helping students understand the origins and development of knowledge and cultivating logical thinking. Analysis instructions use typical examples of varying difficulty to bridge the gap between theory and application, enabling students to master the practical use of knowledge. Summarizing instructions outlines core points and their connections to previous and subsequent knowledge points, strengthening memory and building a systematic understanding. This set of rules ensures that the generated courseware covers the complete teaching chain of knowledge points while maintaining structural consistency and logic, avoiding fragmented content or blurred focus, and improving teaching efficiency.

[0052] The template generation rules include a sequence of instructions for generating the parsed content. The sequence of instructions includes instructions for reviewing the question, brainstorming, calculating, answering, and prompting.

[0053] The question review instruction is used to analyze the question stem of the target question and extract key information and known conditions from the question stem. Key information includes the question's focus, the limiting conditions of the question stem, the core research object, data information, and clues to related knowledge points.

[0054] The ideation command is used to determine the type of the target question based on key information and known conditions, and generate a textual explanation including the solution approach and the reasons for the choice. The question types include basic concept application questions, logical reasoning questions, comprehensive cross-knowledge point questions, real-world scenario application questions, calculation and derivation questions, and proof and derivation questions.

[0055] The calculation instructions are used to generate the solution steps for the target problem and the reasoning process for those steps.

[0056] The answer command is used to generate the final answer to the target question.

[0057] Hints are used to generate common mistakes and ways to avoid them for different question types.

[0058] The question-reading instructions help students accurately grasp the key information in the question stem, avoiding deviations in the solution direction due to information omission or misinterpretation. The problem-solving instructions clarify the question type and explain the solution approach, allowing students to understand "why we think this way" rather than just knowing "how to do it," thus cultivating problem-solving logic. The calculation instructions present the complete solution steps and reasoning process, making the solution path clear and traceable, facilitating student imitation and learning. The answering instructions provide standardized answers to ensure the accuracy of the results. The prompting instructions specifically point out common mistakes and avoidance methods, helping students avoid common errors in advance and improve their problem-solving accuracy. This set of instructions upgrades the analysis from simply providing steps and answers to a learning tool that teaches methods and avoids pitfalls, adapting to different types of questions and guiding students to form scientific problem-solving thinking.

[0059] Step S3 specifically includes: The standards and rules for lower grades include: For text-based sentences, the vocabulary is selected from a list of commonly used words in primary school, and the sentence length is within the first preset range.

[0060] For image-based content, a cartoon style will be used.

[0061] For audio and video categories, the audio uses child-friendly voiceovers and simple background music, the video duration is within a second preset range, and the calculation process is demonstrated with animation.

[0062] The standard rules for secondary school include: For text-based sentences, the vocabulary uses subject-specific terminology, the sentence length is within the third preset range, and simple compound sentences are used.

[0063] For image-based content, a style combining cartoon and realistic elements is adopted.

[0064] For both audio and video formats, the audio uses standard Mandarin dubbing without background music, the video length is within the fourth preset range, and experimental demonstrations and formula derivations are added.

[0065] The standards and rules for higher education include: For text-based content, the vocabulary uses subject-specific terminology and complex sentence structures.

[0066] For image-based content, use a realistic style or professional charts.

[0067] For audio and video categories, the audio uses academic lecture-style narration without background music, and the video length is within the fifth preset range, with the addition of academic discussions, literature citations, and experimental demonstrations.

[0068] The pre-defined ranges in the rules for each grade level combine the characteristics of the subject and the cognitive patterns of each grade level to clearly define quantitative standards. For lower grades, the sentence length for text-based activities is 5-15 characters for Chinese (e.g., "The kitten is eating fish in the yard") and 4-12 characters for mathematics (e.g., "Calculating the result of 5 minus 3"). For lower grades, the video duration for audio-visual activities is 2-4 minutes for Chinese (e.g., "Textbook animation") and 1.5-3 minutes for mathematics (e.g., "Demonstration of addition and subtraction"). For middle grades, the sentence length for text-based activities is 15-25 characters for Chinese (e.g., "Giving life to grass through personification") and 12-22 characters for mathematics (e.g., "Calculating angles based on the angle sum theorem"). For middle grades, the video duration for audio-visual activities is 5-8 minutes (e.g., "Derivation of geometric theorems," "Interpretation of classical Chinese texts"). For higher grades... The duration of audio-visual videos for different grade levels is set at 8-12 minutes for Chinese language (e.g., an analysis of "Ode to the Red Cliff") and 10-15 minutes for mathematics, physics, and chemistry (e.g., calculus formula derivation, chemical experiment analysis). These specific quantitative standards ensure consistency and adaptability in the presentation of teaching content across different grade levels and subjects. Regarding the transition in image style, an imbalance in the ratio of cartoonish to realistic elements should be avoided in middle school. Excessive emphasis on cartoons may not meet the cognitive development needs of middle school students, while excessive emphasis on realism may increase the difficulty of comprehension. Academic discussions and literature citations in higher grades ensure the authority and accuracy of the content, avoiding the use of incorrect or non-academic materials, which would affect the professionalism of the teaching content.

[0069] By setting targeted standard rules for lower, middle, and upper grades, the teaching content generated by AIGC can be precisely matched to the cognitive level and learning needs of students at different grades in terms of form, difficulty, and presentation. Rules for lower grades use common elementary school vocabulary, short sentences, cartoon images, and child-friendly audio and video to cater to the concrete and engaging cognitive characteristics of young students, lowering the learning threshold. Meanwhile, animated demonstrations of calculation processes make abstract knowledge more intuitive. Rules for middle grades transition to basic subject-specific terminology, simple compound sentences, and images combining cartoons and realistic depictions. Audio and video content removes background music and adds experimental demonstrations and formula derivations, thus balancing student cognitive development with practical application. The enhancement of abilities provides support for the deepening of subject knowledge, helping students gradually adapt to more logical learning content. The rules for higher grades focus on subject-specific terminology, complex sentence structures, realistic images, or professional charts. The audio and video use academic lecture-style dubbing, incorporating academic discussions and literature citations, which fully meet the needs of high school students and above for in-depth and academic knowledge. This helps them build a systematic and professional knowledge system, and creates a clear gradient difference in the teaching content of different grades in terms of text, images, and audio and video dimensions. This avoids the problem of the lower grades having content that is too difficult or the higher grades having content that is too shallow, thus ensuring the effectiveness of the teaching content.

[0070] Step S4 specifically includes: The system receives user-generated content requests, which include the target learning stage, content type, and target knowledge points. Based on the request content, it determines the knowledge point progression rules, content type generation rules, and learning stage standard rules for the target knowledge points. It then integrates these rules with the user-generated content requests to generate AIGC content.

[0071] Based on user requests, key rules are locked in to prevent generated content from deviating from user needs (e.g., if a user needs courseware on linear equations in one variable for middle school mathematics, the generated content should accurately match the progressive logic of the knowledge point, the instruction sequence of the courseware template, and the text and audio-visual standards for middle school). The integration of these three types of rules ensures that the generated content not only conforms to the progressive pattern of subject knowledge but also adapts to the cognitive characteristics of the target grade level and meets the presentation requirements of specific content types (such as courseware and explanations). For example, when generating audio for reciting ancient poems in elementary school Chinese, the generated content will simultaneously follow the progressive rules of pinyin assistance - poem explanation - emotional reading and the generation logic of audio content, as well as the lower grade standards of child-friendly dubbing + simple background music, ensuring that the content is both professional and suitable, and greatly improving the efficiency of users in obtaining teaching resources.

[0072] The AIGC content is generated by integrating knowledge point progression rules, generation rules, and grade level standard rules, and combining these with user-generated request content. Specifically, this includes: Based on the knowledge point progression rule, determine the prerequisite and subsequent knowledge points of the target knowledge point, and add the prerequisite and subsequent knowledge points before and after the content of the target knowledge point to generate the initial content framework.

[0073] The generation rules are invoked, and an instruction sequence is generated using AIGC. Based on the instruction sequence, fill text matching the initial content frame is generated in the corresponding part of the initial content frame.

[0074] While generating the fill text, according to the standard rules for each learning stage, the content of relevant elements is matched from the preset material library through a cross-modal semantic matching algorithm. The relevant elements include text, images, audio and video. Based on the content logic of the fill text, the content of the relevant elements is added to the corresponding position of the fill text to generate the final text.

[0075] Perform multi-dimensional validation on the final text. If the multi-dimensional validation passes, generate AIGC content. If the multi-dimensional validation fails, return to the corresponding validation step and regenerate AIGC content.

[0076] Setting a reasonable upper limit on the number of regeneration attempts is crucial to prevent the process from getting stuck in an infinite loop. When problems arise, such as logical conflicts between rules (e.g., a mismatch between the knowledge point progression requirements and the educational level standards in terms of content depth) or a lack of suitable elements in the preset material library (e.g., missing professional charts for a specific educational level), a reasonable limit on the number of attempts (e.g., 3-5 times) can provide the system with enough room for adjustment to optimize content quality. It can also trigger an interruption mechanism when multiple attempts still fail to meet the requirements, prompting manual intervention to investigate the root cause of the problem (e.g., contradictory rule settings or the need to supplement the material library). This ensures the reliability of the generated content while maintaining the efficient operation of the overall process and avoiding resource waste and time loss caused by technical bottlenecks.

[0077] The framework is built according to the progressive rules of knowledge points, and preceding and following knowledge points are added (such as a review of linear equations before a lesson on quadratic equations) to avoid fragmented content. The generation rules are called to generate fill-in text, and cross-modal algorithms are used to match and adapt materials (such as cartoon images and children's voiceovers for lower grade graphic recognition lessons) to make the content structured and vivid. Multi-dimensional verification and correction of deviations (such as correcting out-of-syllabus terms for middle school students) ensure professionalism. Overall, the content is upgraded from a single text to a structured, multi-modal, and compliant resource, improving its quality and value.

[0078] Multi-dimensional validation of the final text specifically includes: Multi-dimensional validation includes logical validation, format validation, and modal validation. If any one of the dimensions fails validation, the multi-dimensional validation is deemed to have failed.

[0079] The logical verification compares the final text with the knowledge point progression rules using a semantic similarity algorithm. If the matching degree is greater than a preset first threshold, the logical verification is deemed to have passed; if the matching degree is less than or equal to the preset first threshold, the logical verification is deemed to have failed.

[0080] Format validation is used to check whether the final text's layout, font, font size, paragraph spacing, heading levels, and citations conform to the generation rules and general specifications. If they conform, the format validation is considered passed; otherwise, it is considered failed.

[0081] Modal validation calculates the relevance between the content of related elements and the corresponding final text fragment using a cross-modal semantic matching algorithm. If the relevance is greater than a preset second threshold, the format validation is deemed to have passed; if the relevance is less than or equal to the preset second threshold, the format validation is deemed to have failed.

[0082] The first threshold is set according to the rigor of the subject. For science subjects such as mathematics and physics, it can be set at 80%-85% to ensure the logical progression of knowledge. For humanities subjects such as Chinese and history, it can be relaxed to 70%-75% to retain the flexibility of expression. The second threshold is adjusted according to the modality type. The correlation between images and text is set at 75%-80% to ensure the matching of core information. For audio and video, it can be reduced to 70%-75% to allow slight deviations in auxiliary information. Both types of thresholds need to be dynamically optimized in combination with teaching practice to balance quality control and content flexibility.

[0083] Logical verification uses semantic similarity algorithms to compare the matching degree between text and knowledge point progression rules, which can accurately identify knowledge logic deviations (such as missing the connection of previous knowledge points or incorrect connection of subsequent knowledge points), ensuring that the content conforms to the progression rules of subject knowledge. Format verification conducts standard checks on details such as layout, font, and heading level, which can unify the presentation standards of teaching content and avoid affecting the teacher's use and the student's reading experience due to chaotic format. Modal verification calculates the correlation between images, audio and video elements and text fragments through cross-modal algorithms, which can prevent multimodal element misalignment problems such as text explaining geometric theorems and illustrations being algebraic formulas, ensuring the consistency of content presentation. The three form a complementary verification system. If any dimension fails, the overall verification is judged as a failure, which can minimize the output of unqualified content and improve the reliability of AIGC teaching resources.

[0084] Example 2, refer to Figure 2 This invention provides an AIGC content generation system based on multi-source feature fusion, including a delimitation module, a scaling module, a scaling module, and a generation module.

[0085] The boundary definition module is used to import the teaching syllabus for each learning stage, determine the knowledge boundaries of each learning stage, and set the rules for the progression of knowledge points based on the teaching syllabus and knowledge boundaries.

[0086] Fixed-size blocks are used to set generation rules based on knowledge point progression rules and content type. The generation rules include courseware template generation rules and parsing template generation rules.

[0087] The calibration module is used to set standard rules for each learning stage based on the progressive rules of knowledge points. The standard rules for each learning stage include standard rules for lower learning stages, standard rules for middle learning stages, and standard rules for higher learning stages.

[0088] The generation module is used to integrate knowledge point progression rules, generation rules, and grade level standard rules to generate AIGC content.

[0089] This invention standardizes and verifies the teaching syllabus, and constructs a structured knowledge base containing core attributes and relationships. It systematically processes the teaching syllabus and knowledge boundaries, accurately capturing the vertical hierarchy and horizontal connections between knowledge points. This ensures the generated content possesses rigorous logical coherence and scientific knowledge progression. It sets generation rules covering courseware and analysis, as well as grade-level standards, integrating knowledge point progression rules, content types, and grade-level standards. This allows the generation strategy to be flexibly adjusted according to the cognitive characteristics of students at different grade levels, ensuring the generated content accurately matches actual teaching needs in terms of language style, presentation, and difficulty gradient. By integrating multimodal materials through a cross-modal semantic matching algorithm and establishing a multi-dimensional verification mechanism encompassing logic, format, and modality, it effectively improves the accuracy and standardization of multimodal content integration, guaranteeing the high adaptability and teaching application value of the final output content.

[0090] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0091] 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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An AIGC content generation method based on multi-source feature fusion, characterized in that, The method comprises the following steps: Step S1, importing the teaching syllabus of each stage, and determining the knowledge boundary of each stage, and setting the knowledge point progression rule according to the teaching syllabus and the knowledge boundary; Step S2, based on the knowledge point progression rule, setting the generation rule according to the content type, the generation rule including the courseware template generation rule and the analysis template generation rule; Step S3, based on the knowledge point progression rule, setting the stage standard rule, the stage standard rule including the low stage standard rule, the middle stage standard rule and the high stage standard rule; Step S4, fusing the knowledge point progression rule, the generation rule and the stage standard rule, and combining the user generated request content to generate AIGC content.

2. The AIGC content generation method based on multi-source feature fusion according to claim 1, characterized in that: The step S1 specifically comprises: Importing the teaching syllabus of each stage, the stages including low stage, middle stage and high stage, performing format standardization processing and checking and completing processing on the teaching syllabus, and determining the knowledge boundary of each stage according to the processed teaching syllabus; Setting the knowledge point progression rule according to the teaching syllabus and the knowledge boundary.

3. The AIGC content generation method based on multi-source feature fusion according to claim 2, characterized in that: The determination of the knowledge boundary of each stage specifically comprises: Based on the processed teaching syllabus, determining the core attribute of each knowledge point, the core attribute including the knowledge point name, the knowledge point connotation, the knowledge point extension and the association relationship between the knowledge point and other knowledge points, and storing in the structured knowledge base; The knowledge point name includes the general range of academic field; The knowledge point connotation includes the core idea, principle, formula and use condition of the knowledge point; The knowledge point extension includes the application scenario of the knowledge point; The association relationship includes the vertical hierarchical relationship and the horizontal hierarchical relationship between the knowledge points.

4. The AIGC content generation method based on multi-source feature fusion according to claim 3, characterized in that: The setting of the knowledge point progression rule according to the teaching syllabus and the knowledge boundary specifically comprises: The knowledge point progression rule includes the annotation of the prerequisite knowledge point and the subsequent knowledge point of the core knowledge point and the non-core knowledge point respectively; For the core knowledge point, the subject experts manually annotate the prerequisite knowledge point and the subsequent knowledge point of the core knowledge point in the structured knowledge base, and obtain the manual annotation result; For the non-core knowledge point, the connotation text of the non-core knowledge point is semantically coded, and the feature vector of the non-core knowledge point is extracted, the co-occurrence frequency and the teaching order of the non-core knowledge point in the historical teaching data are analyzed according to the association rule mining algorithm, and the prerequisite knowledge point and the subsequent knowledge point of the non-core knowledge point are automatically annotated, and the automatic annotation result is obtained.

5. The AIGC content generation method based on multi-source feature fusion according to claim 4, characterized in that: The step S2 specifically comprises: The courseware template generation rule includes the instruction sequence for generating courseware content, the instruction sequence including introduction instruction, explanation instruction, derivation instruction, analysis instruction and summary instruction; The introduction instruction is used to introduce the target knowledge point in combination with life examples, interesting problems or subject history stories; The explanation instruction is used to present the target knowledge point in the form of definition and objectification explanation; The derivation instruction is used to generate the complete derivation process of the formula and principle of the target knowledge point; The analysis instruction is used to generate different difficulty typical examples according to the target knowledge point, the typical examples including the question stem, the problem solving steps and the thought summary; The summary instruction is used to generate a core point of the target knowledge point and a related association relationship, and the related association relationship includes an association relationship between the knowledge point and a previous knowledge point and an association relationship between the knowledge point and a subsequent knowledge point; The analysis template generation rule includes an instruction sequence for generating analysis content, and the instruction sequence includes a question reviewing instruction, a thinking instruction, a calculation instruction, an answering instruction and a prompting instruction; The question reviewing instruction is used to analyze a title of a question and a stem of the question, and extract key information and known conditions of the stem, and the key information includes a question pointing direction, a stem limiting condition, a core research object, data information and an associated knowledge point clue; The thinking instruction is used to determine a question type of the question according to the key information and the known conditions, and generate a text description including a problem solving thought and a selection reason, and the question type includes a basic concept application type question, a logical reasoning type question, a comprehensive cross-knowledge point question, an actual scene application type question, a calculation type question and a proof derivation type question; The calculation instruction is used to generate a problem solving step of the question, and generate a reasoning process of the problem solving step; The answering instruction is used to generate a final answer of the question; The prompting instruction is used to generate a common error point and an avoiding method of the question type.

6. The AIGC content generation method based on multi-source feature fusion according to claim 5, characterized in that: The step S3 specifically includes: The low school stage standard rule includes: For a text type, a vocabulary selection is a commonly used vocabulary table for primary school, and a sentence length is within a first preset range; For an image type, a cartoon style is adopted; For an audio type and a video type, a child voice is adopted for audio, simple background music is adopted, a video time is within a second preset range, and an animation is adopted to demonstrate a calculation process; The middle school stage standard rule includes: For a text type, a vocabulary adopts a subject basic term, a sentence length is within a third preset range, and a simple complex sentence is used; For an image type, a cartoon and a realistic combination style is adopted; For an audio type and a video type, a standard Mandarin voice is adopted for audio, no background music is adopted, a video time is within a fourth preset range, and experimental demonstration and formula derivation contents are added; The high school stage standard rule includes: For a text type, a vocabulary adopts a subject professional term, and a complex sentence type is used; For an image type, a realistic style or a professional chart is adopted; For an audio type and a video type, an academic lecture type voice is adopted for audio, no background music is adopted, a video time is within a fifth preset range, and academic discussion, literature reference and experimental demonstration contents are added.

7. The AIGC content generation method based on multi-source feature fusion according to claim 6, characterized in that: The step S4 specifically includes: Receiving a user generated request content, the user generated request content including a target school stage, a content type and a target knowledge point, determining a knowledge point progression rule of the target knowledge point, a generation rule of the content type and a school stage standard rule of the target school stage according to the request content, fusing the knowledge point progression rule, the generation rule and the school stage standard rule, and generating an AIGC content in combination with the user generated request content.

8. The AIGC content generation method based on multi-source feature fusion according to claim 7, characterized in that: Fusing the knowledge point progression rule, the generation rule and the school stage standard rule, and generating an AIGC content in combination with the user generated request content specifically includes: According to the knowledge point progression rule, the prerequisite knowledge point and the subsequent knowledge point of the target knowledge point are determined, and the prerequisite knowledge point and the subsequent knowledge point are added before and after the content of the target knowledge point respectively to generate an initial content framework; The generation rule is called, and an instruction sequence is generated by using AIGC, and based on the instruction sequence, the filling text matching the initial content framework is generated in the corresponding part of the initial content framework; At the same time of generating the filling text, according to the learning stage standard rule, the content of the related elements is matched from the preset material library by a cross-modal semantic matching algorithm, the related elements include text, image, audio and video, and the content of the related elements is added to the corresponding position of the filling text according to the content logic of the filling text to generate a final text; The final text is subjected to multi-dimensional checking, if the multi-dimensional checking is passed, AIGC content is generated, if the multi-dimensional checking is not passed, the corresponding checking step is returned, and AIGC content is regenerated.

9. The AIGC content generation method based on multi-source feature fusion according to claim 8, characterized in that: The multi-dimensional checking of the final text specifically includes: The multi-dimensional checking includes logic checking, format checking and modal checking, if any dimension checking is not passed, it is determined that the multi-dimensional checking is not passed; The logic checking compares the matching degree of the final text and the knowledge point progression rule by a semantic similarity algorithm, if the matching degree is greater than a preset first threshold, it is determined that the logic checking is passed, if the matching degree is less than or equal to the preset first threshold, it is determined that the logic checking is not passed; The format checking is used to check whether the layout, font, font size, paragraph spacing, title level and reference annotation of the final text meet the generation rule and general specification, if they meet, it is determined that the format checking is passed, if they do not meet, it is determined that the format checking is not passed; The modal checking calculates the correlation degree of the content of the related elements and the corresponding final text segment by a cross-modal semantic matching algorithm, if the correlation degree is greater than a preset second threshold, it is determined that the format checking is passed, if the correlation degree is less than or equal to the preset second threshold, it is determined that the format checking is not passed.

10. The AIGC content generation system based on multi-source feature fusion, applied to the AIGC content generation method based on multi-source feature fusion as claimed in any one of claims 1-9, characterized in that, It includes a boundary determining module, a scale determining module, a label determining module and a generating module; The boundary determining module is used to import the teaching syllabus of each learning stage, determine the knowledge boundary of each learning stage, and set the knowledge point progression rule according to the teaching syllabus and the knowledge boundary; The scale determining module is used to set the generation rule according to the content type based on the knowledge point progression rule, the generation rule includes the courseware template generation rule and the analysis template generation rule; The label determining module is used to set the learning stage standard rule based on the knowledge point progression rule, the learning stage standard rule includes the low learning stage standard rule, the middle learning stage standard rule and the high learning stage standard rule; The generating module is used to fuse the knowledge point progression rule, the generation rule and the learning stage standard rule, and generate AIGC content.

Images