An AI-based method and system for outputting teaching content

Abstract

This invention relates to an AI-based method for outputting teaching content, characterized by comprising: collecting and analyzing blackboard video and audio recordings from classroom teaching to obtain a sequence of blackboard elements appearing in chronological order, audio text, and knowledge point importance weights calculated based on audio features; constructing a dynamic knowledge graph based on the spatial layout and chronological order of the blackboard element sequence and logical connectives in the audio text; inputting the dynamic knowledge graph and preset courseware constraints into a constraint solving engine, wherein the constraints are at least based on visual saliency mapping constraints of knowledge point importance weights, animation sequence constraints based on chronological order, and structural mapping constraints based on logical relationships; the constraint solving engine outputs all courseware document structures and style data that meet the constraints; and generating an initial courseware file based on the courseware document structure and style data.

Description

Technical Field

[0001] This invention relates to teaching systems, and more particularly to an AI-based method and system for outputting teaching content. Background Technology

[0002] Current methods for generating teaching courseware largely rely on manual creation by teachers or simple blackboard / speech conversion tools, exhibiting significant technical shortcomings. On one hand, traditional courseware generation struggles to connect with dynamic classroom information, only statically extracting blackboard text or audio, failing to consider the spatial layout of blackboard elements, their temporal order, and logical connections within the audio. This results in blurred hierarchical relationships and logical derivations between knowledge points, making it difficult to recreate the dynamic process of classroom teaching. On the other hand, existing technologies lack quantification and visualization of the importance of knowledge points, failing to identify core knowledge points based on classroom speech characteristics (such as speech rate, pauses, and emphasis). This leads to a lack of visual distinction between key and secondary content in the courseware, hindering the delivery of key teaching points. Furthermore, the lack of a unified constraint mechanism in the courseware generation process makes logical inconsistencies or visual discrepancies in structural arrangement, animation sequences, and style design prone to occur. Manually compiling classroom content to generate courseware is also inefficient, failing to meet the needs for rapid reuse and iteration of teaching resources. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an AI-based teaching content output method and system to overcome the above-mentioned defects in the existing technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] An AI-based method for outputting teaching content includes:

[0006] The information collection steps involve collecting blackboard videos and audio recordings from classroom teaching, and then analyzing them to obtain the sequence of blackboard elements appearing in chronological order, the audio text, and the importance weights of knowledge points calculated based on speech features.

[0007] The knowledge graph construction steps involve constructing a dynamic knowledge graph based on the spatial layout and temporal order of the whiteboard element sequence and the logically related words in the spoken text.

[0008] The dynamic constraint step involves inputting the dynamic knowledge graph and preset courseware constraints into the constraint solving engine. The constraints are based at least on visual saliency mapping constraints based on the importance weight of knowledge points, animation sequence constraints based on time order, and structural mapping constraints based on logical relationships. The constraint solving engine outputs the courseware document structure and style data that meet all the constraints.

[0009] The courseware output step involves generating an initial courseware file based on the courseware document structure and courseware style data.

[0010] Preferably, the knowledge graph construction step includes an incremental strategy, which includes monitoring each teaching event and updating the dynamic knowledge graph according to the event type. The teaching events include knowledge point creation events, knowledge point modification events, and association establishment events. When a knowledge point creation event is detected, a new knowledge point node is created in the dynamic knowledge graph, and the content and its timestamp are recorded for the node. When a knowledge point modification event is detected, the existing knowledge point node targeted by the event is identified, and the attributes of the node are updated to reflect its emphasis. When an association establishment event is detected, at least two existing knowledge point nodes involved in the event are identified, and an edge representing a logical relationship is created between the at least two nodes.

[0011] Preferably, the visual saliency mapping constraint includes automatically mapping the weight attributes of the knowledge points to the style attributes of the corresponding visual elements in the courseware, and the mapping relationship is positively correlated. The animation sequence constraint includes stipulating that the order in which all knowledge points appear on the courseware page is consistent with the order of their timestamps in the dynamic knowledge graph.

[0012] Preferably, the method further includes an optimization step, which is used to render the initial courseware file into at least one page of courseware image, perform visual analysis on the courseware image to obtain the actual visual elements and their attributes, obtain the expected visual specification list of the dynamic knowledge graph, compare the actual visual elements and attributes with the expected visual specification list to generate a difference report, and automatically adjust the style or layout of the initial courseware file based on the difference report to generate an optimized courseware file.

[0013] Preferably, the step of obtaining the expected visual specification list includes obtaining a dynamic knowledge graph, which contains multiple knowledge point nodes with attributes and edges representing logical relationships.

[0014] Access a predefined library of instructional demonstration visualization patterns, which stores multiple visualization patterns, each associated with one or more graph semantic structures, and defines a corresponding visual layout template;

[0015] Semantic structure analysis is performed on the dynamic knowledge graph to identify one or more semantic subgraphs contained therein;

[0016] For each of the identified semantic subgraphs, at least one applicable target visualization pattern is retrieved and matched from the teaching demonstration visualization pattern library;

[0017] The attribute parameters of the knowledge points in the semantic subgraph are bound to the visual attribute parameters corresponding to the target visualization mode, thereby generating the instantiated visual specification of the semantic subgraph.

[0018] Based on the instantiated visual specifications of all semantic subgraphs, an overall layout plan is performed to generate a list of structured expected visual specifications to guide the generation of courseware.

[0019] Preferably, the automatic adjustment includes acquiring a difference report, generating a numerical state vector representing the current visual state of the courseware, inputting the state vector into an automatic learning model, selecting a style adjustment action from a preset action space, the action space containing multiple atomic operations that can modify the style attributes of the courseware, executing the selected style adjustment action, modifying the initial courseware to generate an intermediate courseware file, performing detailed rendering and visual analysis on the intermediate courseware file, comparing the results with the expected visual specification list, calculating a new difference report, and acquiring a reward signal based on the changes in this difference report compared to the previous round. The training data consisting of the state vector, the selected action, the immediate reward signal, and the new state vector is used to update the automatic learning model.

[0020] Preferably, the calculation of the importance weight of the knowledge point includes obtaining the font size and position of the corresponding blackboard writing, the volume and speech rate changes when the knowledge point is mentioned in speech, and whether it is emphasized by annotation symbols, and obtaining the importance weight of the knowledge point based on the above content.

[0021] An AI-based teaching content output system includes:

[0022] The multimodal teaching behavior analysis module is used to extract dynamic sequences of blackboard writing and speech features from teaching video streams and to construct dynamic knowledge graphs.

[0023] A multi-constraint courseware generation engine is used to receive the dynamic knowledge graph and courseware design constraints, and generate initial courseware files;

[0024] The visibility closed-loop optimization module is used to render, parse, compare, and automatically adjust the initial courseware file, and output the optimized final courseware file.

[0025] Preferably, the multi-constraint courseware generation engine is equipped with an engine solver to solve the constraint satisfaction problem consisting of visual saliency mapping constraints, animation sequence constraints, and structural mapping constraints, so as to determine the optimal visual presentation style of the courseware.

[0026] The beneficial effects of this invention are as follows: By collecting whiteboard videos and audio recordings, and combining them with temporal sequence, spatial layout, and logical connectives to construct a dynamic knowledge graph, this invention fully restores the rhythm and logical connections of knowledge points in classroom teaching. This solves the problems of static and fragmented knowledge points in traditional courseware, making the courseware more aligned with the actual teaching process. Based on speech features, the importance weights of knowledge points are calculated, and visual saliency mapping constraints are used to achieve differentiated display of key content, helping learners quickly focus on core knowledge points and improving the efficiency of information delivery. Through the combination of multi-dimensional constraints (visual saliency, animation sequences, structural mapping) and a constraint solving engine, the invention ensures that the courseware document structure is logically coherent and the style is consistent, while also adapting to the structured requirements of knowledge presentation in teaching scenarios, avoiding the logical confusion and cluttered styles of traditional courseware. Attached Figure Description

[0027] Figure 1 This is an overall flowchart of the present invention;

[0028] Figure 2 This is a flowchart of the expected visual specification list acquisition process of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0032] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings:

[0033] like Figure 1-2 As shown, this invention provides an AI-based method for outputting teaching content, including:

[0034] The information collection process involves acquiring and analyzing blackboard writing videos and audio recordings from classroom teaching. This yields a chronological sequence of blackboard elements, audio text, and knowledge point importance weights calculated based on speech features. Through synchronous audio and video acquisition and multi-dimensional feature extraction in a classroom setting, the objective calculation of knowledge point importance weights is achieved, providing data support for subsequent dynamic knowledge graph construction and focused presentation of courseware.

[0035] The calculation of the importance weight of knowledge points includes obtaining the font size and position of the corresponding blackboard writing, the volume and speech rate changes when the knowledge point is mentioned in speech, and whether it is emphasized by annotation symbols, and obtaining the importance weight of knowledge points based on the above content.

[0036] The knowledge graph construction steps involve building a dynamic knowledge graph based on the spatial layout and temporal order of the whiteboard element sequence and the logical connectives in the spoken text.

[0037] The knowledge graph construction steps include an incremental strategy, which involves monitoring each teaching event and updating the dynamic knowledge graph according to the event type. Teaching events include knowledge point creation events, knowledge point modification events, and association establishment events. When a knowledge point creation event is detected, a new knowledge point node is created in the dynamic knowledge graph, and the content and its timestamp are recorded for the node. When a knowledge point modification event is detected, the existing knowledge point node targeted by the event is identified, and the node's attributes are updated to reflect its emphasis. When an association establishment event is detected, at least two existing knowledge point nodes involved in the event are identified, and an edge representing the logical relationship is created between the at least two nodes. The importance weighting of knowledge points employs a three-dimensional fusion calculation model based on blackboard features, speech features, and annotation features. Each dimension is quantitatively scored and then weighted and summed. The blackboard feature dimension uses quantitative scoring based on font size and position. Image recognition algorithms are used to extract the pixel height of the blackboard text corresponding to each knowledge point, mapping the pixel height to a font level. The blackboard area is divided into three levels: core area, secondary core area, and edge area. If a blackboard element of the same knowledge point spans multiple areas, its position level is determined by the area where its center of gravity is located. If a knowledge point is presented multiple times on the blackboard, the score corresponding to the highest position level is taken. The speech feature dimension uses quantitative scoring based on changes in volume and speech rate, extracted from the speech text. The system extracts audio segments corresponding to the target knowledge points, performs audio frame segmentation on these segments, calculates the short-time energy of each frame, and takes the average short-time energy within a segment as the baseline volume. A volume threshold is set for judgment. Simultaneously, it analyzes speech rate changes, calculates the speech rate of the audio segments corresponding to the knowledge points, and compares it with the teacher's average speech rate throughout the entire lesson. If a knowledge point is mentioned multiple times in the audio, additional repetition emphasis points are added. Annotation symbol dimension: Based on the quantitative scoring of emphasis marks, an image recognition algorithm scans the area around the blackboard elements to identify whether annotation symbols exist. The total scores of the blackboard feature dimension, the speech feature dimension, and the annotation feature dimension are weighted and summed to obtain the importance weight of the knowledge points.

[0038] The dynamic constraint step inputs the dynamic knowledge graph and pre-defined courseware constraints into the constraint solving engine. These constraints are based at least on visual saliency mapping constraints (based on knowledge point importance weights), animation sequence constraints (based on time sequence), and structural mapping constraints (based on logical relationships). The constraint solving engine outputs the courseware document structure and style data that meet all constraints. This achieves the core aspects of ensuring logical coherence, standardized style, and highlighting key points in the courseware. By constructing a multi-dimensional constraint system and combining it with the automated reasoning of the constraint solving engine, the dynamic knowledge graph is transformed into a courseware structure and style that conforms to teaching principles. The pre-defined courseware constraints are designed based on the core needs of presenting teaching content, covering three dimensions: visual saliency, animation sequence, and structural logic. Each dimension achieves precise constraints through quantified rules and mapping relationships. The constraint solving engine, as the core execution module of the dynamic constraint step, outputs the courseware document structure and style data that meet all constraints through a modeling-reasoning-resolution process.

[0039] Visual saliency mapping constraints include automatically mapping the weight attributes of knowledge points to the style attributes of corresponding visual elements in the courseware, with a positive correlation. Animation sequence constraints stipulate that the order in which all knowledge points appear on the courseware page must match their timestamp order in the dynamic knowledge graph. The positive correlation mapping between knowledge point importance weights and courseware visual element styles transforms abstract weights into perceptible visual differences through preset quantification rules, ensuring that important content is more easily noticed in the courseware. All visual style attribute mappings must strictly adhere to the positive correlation principle of increasing visual saliency as weight increases. When multiple high-weight knowledge points compete for the visual focus area, the knowledge point with the higher weight is prioritized, and the rest are distributed to adjacent areas in descending order of weight, ensuring that saliency ranking is still satisfied even in case of positional conflicts. Animation sequence constraints aim to recreate the natural rhythm of knowledge point appearance in classroom teaching through courseware animation, ensuring that the dynamic presentation of the courseware is consistent with the temporal logic of the teaching process.

[0040] The courseware output steps involve generating an initial courseware file based on the courseware document structure and style data.

[0041] It also includes an optimization step, which renders the initial courseware file into at least one page of courseware image, performs visual analysis on the courseware image to obtain the actual visual elements and their attributes, obtains the expected visual specification list of the dynamic knowledge graph, compares the actual visual elements and attributes with the expected visual specification list to generate a difference report, and automatically adjusts the style or layout of the initial courseware file based on the difference report to generate an optimized courseware file. The optimization process involves closed-loop verification and precise correction of the initial courseware files. Through an automated process of rendering, analysis, comparison, and adjustment, it resolves potential visual style deviations and layout logic conflicts in the initial courseware, ensuring that the final output fully conforms to the preset specifications. To accurately capture the actual display effect of the initial courseware, the courseware document structure and style data are rendered into standardized images to ensure the authenticity and accuracy of the visual analysis. Computer vision algorithms are used to perform pixel-level analysis of the rendered images, accurately extracting the actual attributes of all visual elements in the courseware and forming an actual attribute list as a comparison benchmark. The expected visual specification list is a compliance benchmark generated based on the constraints of the dynamic constraint step, containing the expected visual, structural, and animation requirements for all knowledge points. Algorithms are used to automatically compare the actual attribute list and the expected visual specification list element by element and attribute by attribute, identifying all deviations and generating a structured difference report. Based on the difference report, automated adjustment algorithms correct the deviations in the initial courseware, generating optimized courseware that conforms to all specifications and ensuring no new conflicts arise after adjustment.

[0042] The expected steps for obtaining the visual specification list include: acquiring a dynamic knowledge graph, which contains multiple knowledge point nodes with attributes and edges representing logical relationships; firstly, acquiring the constructed dynamic knowledge graph, performing structured parsing of the semantic relationships between its node attributes and edges, traversing all knowledge point nodes in the graph, extracting key attribute parameters to form a list of node attributes, identifying the semantic relationships represented by the edges connecting nodes in the graph, and labeling them according to commonly used logic categories in teaching scenarios. Common semantic relationship types and representation features include subordinate relationships, parallel relationships, causal relationships, process relationships, and comparative relationships.

[0043] Access the predefined teaching demonstration visualization pattern library, which stores multiple visualization patterns. Each pattern is associated with one or more graph semantic structures and defines a corresponding visual layout template. The predefined teaching demonstration visualization pattern library is the core resource for standardized visual presentation. Its design revolves around the one-to-one correspondence between semantic structure and visual layout to ensure pattern adaptability and teaching practicality.

[0044] Semantic structure analysis is performed on dynamic knowledge graphs to identify one or more semantic subgraphs contained therein. A semantic subgraph is the smallest unit with independent semantic logic in a dynamic knowledge graph. Through structured analysis, the graph is decomposed into sub-modules that can match visualization patterns.

[0045] For each identified semantic subgraph, at least one applicable target visualization pattern is retrieved and matched from the teaching demonstration visualization pattern library; for each semantic subgraph, a suitable visualization pattern is selected from the pattern library to ensure that the visual layout is highly matched with the semantic logic.

[0046] By binding the attribute parameters of knowledge points in the semantic subgraph to the visual attribute parameters corresponding to the target visualization mode, an instantiated visual specification for the semantic subgraph is generated. Furthermore, by precisely binding the attribute parameters of knowledge points in the semantic subgraph to the visual attribute parameters of the target visualization mode, the instantiation of the visualization mode is achieved.

[0047] Based on the instantiated visual specifications of all semantic subgraphs, an overall layout plan is performed, generating a structured list of expected visual specifications to guide courseware generation. Based on the instantiated visual specifications of all semantic subgraphs, overall layout optimization across subgraphs is performed to ensure the continuity and visual consistency of the courseware pages, ultimately generating a structured list of expected visual specifications.

[0048] Automatic adjustment involves acquiring a difference report, generating a numerical state vector representing the current visual state of the courseware, inputting the state vector into the automatic learning model, selecting a style adjustment action from a preset action space (containing multiple atomic operations that modify the courseware's style attributes), executing the selected style adjustment action, modifying the initial courseware to generate an intermediate courseware file, performing detailed rendering and visual analysis on the intermediate courseware file, comparing the result with the expected visual specification list, calculating a new difference report, and obtaining a reward signal based on the changes in this difference report compared to the previous round. The training data, consisting of the state vector, the selected action, the immediate reward signal, and the new state vector, updates the automatic learning model. The core of automatic adjustment is a closed-loop optimization mechanism based on a reinforcement learning framework, encompassing state perception, action selection, reward feedback, and model iteration. By numerically representing the courseware's visual state, atomically adjusting actions, and quantifying reward signals, the automatic learning model is driven to continuously optimize the adjustment strategy, achieving precise alignment between the courseware and the expected visual specifications. The state vector is the core input for the automatic learning model to perceive the visual deviations in the current courseware. By extracting key indicators from the difference reports and quantifying them, the visual deviation state is transformed into vector data that the model can process. The action space is the model's selectable adjustment toolbox, containing a series of atomic operations on the style attributes of the courseware, covering the entire dimensions of visual, structural, and animation, ensuring the accuracy and comprehensiveness of the adjustments. The automatic learning model adopts a deep Q-network (DQN) to adapt to sequential decision-making scenarios. By learning the mapping relationship between state, action, and reward, it outputs the optimal adjustment strategy. The reward signal is the guiding force for the model's learning, quantifying the adjustment effect based on the changes in the difference reports between the previous and subsequent rounds, guiding the model to optimize in the direction of reducing deviations and improving compliance. The model parameters are continuously optimized through the training data generated in each round of adjustment, improving the accuracy of the adjustment strategy.

[0049] Initialization: Input the initial courseware's difference report, construct the initial state vector, and load the untrained / converged automatic learning model; Action Selection: The model outputs the optimal atomic operation based on the current state vector; Execution Adjustment: Execute the atomic operation to generate intermediate courseware, re-render and parse to generate a new difference report; Reward Calculation: Calculate the immediate reward based on the new and old difference reports; Data Storage and Model Update: Store the quadruples in the experience replay pool, and batch sample to update the model parameters; Termination Judgment: If the termination condition is met, output the optimized courseware; otherwise, return to action selection and repeat the adjustment.

[0050] An AI-based teaching content output system includes:

[0051] The multimodal teaching behavior analysis module is used to extract dynamic sequences of blackboard writing and speech features from teaching video streams and to construct dynamic knowledge graphs.

[0052] A multi-constraint courseware generation engine is used to receive dynamic knowledge graphs and courseware design constraints, and generate initial courseware files;

[0053] The visibility closed-loop optimization module is used to render, parse, compare, and automatically adjust the initial courseware file, and output the optimized final courseware file.

[0054] The multi-constraint courseware generation engine has an engine solver that solves the constraint satisfaction problem consisting of visual saliency mapping constraints, animation sequence constraints, and structural mapping constraints, in order to determine the optimal visual presentation style of the courseware.

[0055] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for outputting teaching content based on AI, characterized in that, include: The information collection steps involve collecting blackboard video and audio recordings from classroom teaching, and analyzing them to obtain the sequence of blackboard elements appearing in chronological order, the audio text, and the importance weight of knowledge points calculated based on speech features. The calculation of the importance weight of knowledge points includes obtaining the font size and position of the corresponding blackboard writing, the volume and speech rate changes when the knowledge point is mentioned in the audio, and whether it is emphasized by annotation symbols, and obtaining the importance weight of knowledge points based on the above content. The knowledge graph construction steps involve constructing a dynamic knowledge graph based on the spatial layout and temporal order of the whiteboard element sequence and the logically related words in the spoken text. The dynamic constraint step involves inputting the dynamic knowledge graph and preset courseware constraints into the constraint solving engine. The constraints are based at least on visual saliency mapping constraints based on the importance weight of knowledge points, animation sequence constraints based on time order, and structural mapping constraints based on logical relationships. The constraint solving engine outputs the courseware document structure and style data that meet all the constraints. The visual saliency mapping constraint includes automatically mapping the weight attributes of the knowledge points to the style attributes of the corresponding visual elements in the courseware, and the mapping relationship is positively correlated. The animation sequence constraint includes stipulating that the order in which all knowledge points appear on the courseware page should be consistent with the order of their timestamps in the dynamic knowledge graph. The courseware output step involves generating an initial courseware file based on the courseware document structure and courseware style data.

2. The method for outputting teaching content based on AI according to claim 1, characterized in that, The knowledge graph construction steps include an incremental strategy, which involves monitoring each teaching event and updating the dynamic knowledge graph according to the event type. The teaching events include knowledge point creation events, knowledge point modification events, and association establishment events. When a knowledge point creation event is detected, a new knowledge point node is created in the dynamic knowledge graph, and the content and its timestamp are recorded for the node. When a knowledge point modification event is detected, the existing knowledge point node targeted by the event is identified, and the node's attributes are updated to reflect its emphasis. When an association establishment event is detected, at least two existing knowledge point nodes involved in the event are identified, and an edge representing a logical relationship is created between the at least two nodes.

3. The method for outputting teaching content based on AI according to claim 1, characterized in that, It also includes an optimization step, which is used to render the initial courseware file into at least one page of courseware image, perform visual analysis on the courseware image to obtain the actual visual elements and their attributes, obtain the expected visual specification list of the dynamic knowledge graph, compare the actual visual elements and attributes with the expected visual specification list to generate a difference report, and automatically adjust the style or layout of the initial courseware file based on the difference report to generate an optimized courseware file.

4. The AI-based teaching content output method according to claim 3, characterized in that, The expected steps for obtaining the visual specification list include obtaining a dynamic knowledge graph, which contains multiple knowledge point nodes with attributes and edges representing logical relationships. Access a predefined library of instructional demonstration visualization patterns, which stores multiple visualization patterns, each associated with one or more graph semantic structures, and defines a corresponding visual layout template; Semantic structure analysis is performed on the dynamic knowledge graph to identify one or more semantic subgraphs contained therein; For each of the identified semantic subgraphs, at least one applicable target visualization pattern is retrieved and matched from the teaching demonstration visualization pattern library; The attribute parameters of the knowledge points in the semantic subgraph are bound to the visual attribute parameters corresponding to the target visualization mode, thereby generating the instantiated visual specification of the semantic subgraph. Based on the instantiated visual specifications of all semantic subgraphs, an overall layout plan is performed to generate a list of structured expected visual specifications to guide the generation of courseware.

5. The AI-based teaching content output method according to claim 3, characterized in that, The automatic adjustment includes acquiring a difference report, generating a numerical state vector representing the current visual state of the courseware, inputting the state vector into an automatic learning model, selecting a style adjustment action from a preset action space, the action space containing multiple atomic operations that can modify the style attributes of the courseware, executing the selected style adjustment action, modifying the initial courseware to generate an intermediate courseware file, performing detailed rendering and visual analysis on the intermediate courseware file, comparing the result with the expected visual specification list, calculating a new difference report, and acquiring a reward signal based on the changes in this difference report compared to the previous round. The training data consisting of the state vector, the selected action, the reward signal, and the new state vector is used to update the automatic learning model.

6. An AI-based teaching content output system, applicable to the AI-based teaching content output method described in any one of claims 1 to 5, characterized in that... include: The multimodal teaching behavior analysis module is used to extract dynamic sequences of blackboard writing and speech features from teaching video streams and to construct dynamic knowledge graphs. A multi-constraint courseware generation engine is used to receive the dynamic knowledge graph and courseware design constraints, and generate initial courseware files; The visibility closed-loop optimization module is used to render, parse, compare, and automatically adjust the initial courseware file, and output the optimized final courseware file.

7. The AI-based teaching content output system according to claim 6, characterized in that, The multi-constraint courseware generation engine is equipped with an engine solver, which is used to solve the constraint satisfaction problem consisting of visual saliency mapping constraints, animation sequence constraints, and structural mapping constraints, so as to determine the optimal visual presentation style of the courseware.

Images