Interactive teaching courseware and video generation method based on artificial intelligence

Abstract

The invention provides an interactive teaching courseware and video generation method based on artificial intelligence. The method comprises the following steps: inputting a course core target, and carrying out semantic analysis to extract key subject terms; intelligently analyzing the attachment, and updating the knowledge base; relevant knowledge points of the course target are retrieved and positioned; searching materials, supplementing and writing into a knowledge base, and updating a material reference list of the course outline; configuring a personalized style, and storing the personalized style as a personalized template; generating a course outline and PPT details; editing and modifying each page of PPT, and rendering and previewing after adjustment is completed; exporting a course package; according to the method, content credibility is ensured through the knowledge base, courseware editing and dynamic demonstration generation are supported, the recognition accuracy of complex content is improved, the copyright risk is avoided, and efficient and personalized generation of teaching resources is achieved.

Description

Technical Field

[0001] This invention relates to the field of educational technology and artificial intelligence, and in particular to an artificial intelligence-based method for generating interactive teaching courseware and videos. Background Technology

[0002] With the advancement of the digital education strategy, smart education platforms have been explicitly required to meet the core requirements of "interactive, diagnostic, and iterative" teaching resources. These resources must support interactive formats such as in-class quizzes and virtual experiments, generate knowledge gap reports based on student learning data, and enable dynamic updates. However, current teaching resource generation technologies still have many shortcomings and struggle to meet the actual needs of teachers and students.

[0003] However, current teaching resource generation technologies mainly fall into two categories: one is rule-based or template-based semi-automatic generation technology, which extracts metadata through OCR and speech recognition and matches pre-made templates to complete surface content such as intros and transition pages. This type of technology relies on limited templates, cannot perform semantic-level customization according to subjects and grade levels, and interactive elements need to be pre-fixed, making it impossible to dynamically adjust according to students' learning progress. Creating a 45-minute lesson still takes 2-3 hours, resulting in low efficiency. The other type is AIGC-driven multimodal resource generation technology. Although it can achieve one-click production, the generation process is "black box," meaning teachers cannot modify it frame-by-frame or word-by-word in real time. Fine-tuning requires re-running the template, which is more time-consuming than manual operation. Furthermore, the generated content is disconnected from real-time learning progress, and multi-dimensional feedback is only used for post-lesson statistics, failing to enable real-time error correction and content rearrangement.

[0004] Therefore, the existing technology has the following drawbacks:

[0005] First, the data source parsing capability is limited, only supporting PPT or short text input, and the accuracy rate for recognizing complex content such as formulas and tables is less than 75%.

[0006] Secondly, it has poor operability, as the generated results are packaged into an unremovable package, making it difficult to make real-time adjustments in class.

[0007] Third, the visualization effect is insufficient. The final product is still a "high-definition PPT". The main body of the screen is still in the "head picture + screen recording" mode. The visualization effect of abstract concepts is poor and it is difficult to stimulate students' interest.

[0008] Fourth, existing technologies directly capture publicly available materials from the internet, posing risks to copyright and data security. The lack of copyright verification for these materials exposes schools to the dual risks of copyright lawsuits and administrative penalties. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies by proposing an artificial intelligence-based method for generating interactive teaching courseware and videos.

[0010] To achieve the above object, the present invention adopts the following technical solutions:

[0011] An interactive teaching courseware and video generation method based on artificial intelligence, comprising the following steps:

[0012] S1: Input the core objectives of the course, and semantically analyze and extract the key topic words;

[0013] The user interacts and inputs in the text box the core objectives of this course in natural language. The core objectives can be long or short. Immediately semantically analyze the core objectives and extract the key topic words;

[0014] The key topic words include the teaching object, subject, knowledge points and teaching objectives;

[0015] Specifically, it includes:

[0016] S11: The user interacts and inputs the core objectives of the course;

[0017] Input natural language through the text box set on the front-end page to describe the core objectives of the course;

[0018] The core objectives are the core demands of the course clearly stated by the teacher in natural language, including the content in three dimensions of knowledge, ability and scenario;

[0019] The knowledge dimension includes the core concepts, formulas, theorems or skills to be mastered;

[0020] The ability dimension includes the practical, analytical or application abilities that students need to achieve;

[0021] For example, the core objective input is: A 45-minute course for eighth-grade junior high school physics, enabling students to understand the principle of electromagnetic induction, master the right-hand rule to judge the direction of induced current, including 1 group interaction experiment and 1 in-class test;

[0022] S12: Semantic analysis;

[0023] Preprocess, segment words, and perform word-pos tagging and screening on the input text of the core objectives to obtain the key topic words;

[0024] S121: Preprocessing of the input text of the core objectives;

[0025] Receive the input text of the core objectives from the user, remove line breaks, filter special symbols and modal particles;分句处理 is not clear in the original text, assuming it means "split sentences by punctuation marks" here. Split the long text by punctuation marks to obtain the preprocessed text.

[0026] The special symbols include non-teaching-related symbols such as @, #, $, &, *;

[0027] The modal particles include "ah", "ya", "ne", etc.;

[0028] S122: Perform word segmentation and词性 tagging on the preprocessed text;

[0029] Perform word segmentation on the text through the Jieba word segmentation tool, and load a subject-specific dictionary in the Jieba word segmentation tool to achieve accurate word segmentation; then perform词性 tagging through the HanLP tool, screen out core词性 such as nouns, verbs, and adjectives, and filter out词性 without real meaning such as auxiliary words ("的", "地") and conjunctions ("和", "与"). Then obtain the logical relationship between words through dependency syntactic analysis.

[0030] The Jieba word segmentation tool is a Chinese word segmentation tool used to split continuous Chinese text into a sequence of meaningful words.

[0031] The HanLP tool is a natural language processing toolkit with rich functions, including词性 tagging, named entity recognition, syntactic analysis, etc.

[0032] S123: Semantic unit splitting;

[0033] According to the logical relationship of词性 tagging and syntactic analysis, screen out the key topic words;

[0034] The key topic words include knowledge keywords, ability verbs, and scenario qualifiers.

[0035] Knowledge keywords are nouns or noun phrases containing knowledge dimensions;

[0036] Ability verbs are verbs or verb phrases containing ability dimensions;

[0037] Scenario qualifiers are restrictive phrases containing scenario dimensions.

[0038] S13: Obtain the key words for semantic parsing, write them into the confirmation window, and form a summary card. The permission of the confirmation window is set to be editable for teachers to confirm again and edit and modify to avoid missing key topic words. It truly achieves "starting with one sentence and having a zero-threshold entry".

[0039] S2: Intelligent parsing of attachments and updating the knowledge base;

[0040] Set an "attachment upload" button on the front-end page to respond to the user's drag or selection of a local file to upload an attachment; start a multi-modal large model to jointly parse the text, formulas, charts, voices, and images in the attachment, extract "concept-relationship-example" triples, and write them into the knowledge base to achieve incremental update of the attachment to the knowledge base and copyright compliance warehousing.

[0041] S21: Attachment upload;

[0042] Define the attachment upload trigger event as drag or selection, respond to the user's operation to trigger upload, and upload the local file to the server.

[0043] Local files include various formats such as PPT, PDF, Word, video, and images. The upload process uses breakpoint resume and format pre-check to ensure stable transmission of large files into the database.

[0044] S22: Annex preprocessing;

[0045] Receive uploaded attachments and use a format conversion tool to convert various attachments into a standard parsing format;

[0046] Convert PPT, PDF, Word and other document types to PDF format while preserving the layout and structure of text, formulas and charts;

[0047] The video is then transcoded into H.264 encoded MP4, sliced ​​into 10-second segments, and the audio stream (WAV format) and frame images are extracted.

[0048] Images are uniformly converted to JPG in RGB color mode.

[0049] S22: Parallel parsing of multimodal data;

[0050] Launch the multimodal large model and perform joint analysis on the text, formulas, charts, audio, and images in the attachment to obtain multimodal data.

[0051] S221: Text parsing;

[0052] For PDF files, PDFBox is used to extract copyable text, preserving paragraph structure and page number information. For text in scanned PDFs or images, MathpixOCR and Baidu OCR are used in conjunction to obtain the text. Mathpix is ​​used for formula recognition to obtain the corresponding LaTeX source code. HanLP is then used to perform part-of-speech tagging and syntactic analysis on the text. Part-of-speech tagging includes nouns, verbs, and adjectives; syntactic analysis examines the logical relationships between words, including subordination, causation, and derivation.

[0053] S222: Formula Explanation;

[0054] Obtain the LaTeX source code of the corresponding formula in the formula area, and use a MathML to LaTeX conversion tool to extract the formula.

[0055] S223: Chart Analysis;

[0056] 1) Identify chart types: The YOLOv8 object detection model is used to identify chart types, including line charts, bar charts, pie charts, tables, etc.

[0057] 2) Extract chart data;

[0058] For tables, the system identifies table borders, extracts cell coordinates, and uses OCR to recognize cell text, generating a structured table in a two-dimensional array format while preserving the relationship between the table header and the data.

[0059] For graph types (line charts, bar charts, pie charts), the Python matplotlib library is used to parse the axes (meaning, units, and scale values ​​of the horizontal and vertical axes), legends, and data points, extracting numerical information and generating key-value pair formatted data. The numerical information includes the height values ​​of the bar charts and the coordinate points of the line charts. The generated key-value pair formatted data includes the horizontal axis, vertical axis, and data points.

[0060] S224: Speech analysis;

[0061] The extracted audio stream is transcribed using the speech recognition API, converting speech into text. The text is then parsed and processed.

[0062] S225: Image analysis;

[0063] For frame images and pictures, image content is identified: OCR is used to recognize text within the image, and the CLIP model is called to identify scenes, objects, and actions in the image to obtain recognized image labels. The text and image labels are then associated.

[0064] S23: Extract the "concept-relation-example" triple;

[0065] S231: Extracting concepts;

[0066] For the parsed multimodal data, nouns with part-of-speech tagging are extracted as candidate concepts, and a BERT model trained on educational concepts is used. The candidate concepts are identified, and the identified concept names are output, along with annotations on their sources.

[0067] S232: Extraction relation;

[0068] The process involves acquiring multimodal data. For text parsing, syntactic analysis is used to identify relation types. For formula parsing, the relationships between variables and constants are determined. For chart parsing, coordinate axis correspondences and data trends are used to identify conceptual relationships. For speech parsing, syntactic analysis of the text is used to determine if conceptual relationships exist. The resulting concepts and relation types are then extracted. Relationship types include subordination, causation, and derivation.

[0069] S233: Extraction example;

[0070] For text parsing, extract the content following keywords such as "for example," "like," and "example," and perform syntactic analysis to identify subordinate relationships within the content following these keywords. For image parsing, extract images with image labels and, based on the associated text, determine the subordinate relationships between the image labels. For chart parsing, extract the charts and, based on the horizontal and vertical axes or table headers, determine the subordinate relationships between the charts.

[0071] S24: Triple encapsulation is written to the knowledge base;

[0072] The extracted "concept-relationship-example" triples are encapsulated in a unified JSON format, with source identifiers added, and incrementally written into the knowledge base in the form of traceable IDs.

[0073] Furthermore, it also includes: S25: Visualizing the analysis results, which are then written to the knowledge base after user adjustment and confirmation;

[0074] S251: Generate visualization results;

[0075] ECharts is used to draw a visual tree diagram, with concepts as core nodes, relationships as edges, and examples as node-specific information. The node-specific information is displayed through a floating window, triggered by mouse movement or clicking.

[0076] S252: User interaction confirmation;

[0077] Users can add or delete concept nodes and set concept weights in the generated visual tree diagram. After adjustment, they can confirm the final result.

[0078] S253: Final results are entered into the database;

[0079] Based on the content of the final result's visualized tree diagram, the modified "concept-relationship-example" triples are obtained, encapsulated in a unified JSON format, and incrementally written into the knowledge base.

[0080] A knowledge base is a database used to store trusted, structured knowledge in the field of education; the knowledge base is deployed on a server. The knowledge base stores knowledge content including textbooks, school-based question banks, authoritative journals, and teachers' past courseware.

[0081] S3: Retrieve relevant knowledge points for locating course objectives;

[0082] Based on the key keywords of the core course objectives, a search is conducted in the knowledge base to obtain the content contained in the key keywords. Users select and confirm the relevant knowledge points, obtain the triples of the confirmed knowledge points, build an index list, and generate a list of material references.

[0083] S4: Search for materials to supplement and add to the knowledge base and update the material reference list in the course outline;

[0084] The front-end page also features an AI search button that responds to user clicks to trigger search commands. Based on course objectives and confirmed knowledge points, it generates search terms and performs a reliable search in the registered educational resource library, open journals, and shared materials within the school to obtain materials. Users select the required materials, and the selected materials are transcoded and sliced ​​and written into the knowledge base. The selected materials are added to the material citation list, and the material citation list is updated.

[0085] S41: Generate search terms;

[0086] Obtain the key keywords of the core course objectives and the triples of the confirmed knowledge points. Calculate the weight of the words in the data using the TF-IDF algorithm, select the top 8 high-frequency core words, extract basic search terms, and combine the extracted basic search terms to generate 5-8 search terms.

[0087] S42: Reliable search based on search terms;

[0088] Based on the obtained search terms, a reliable search is performed in the registered educational resource database, open journals, and shared materials within the school to obtain the materials.

[0089] S43: Visual presentation of materials;

[0090] The retrieved materials are divided into four resource types, including concept animations, case study videos, interactive 3D, and extended reading; each type of resource is presented in card format.

[0091] Concept animations are obtained by filtering animation formats (WebM / MP4), knowledge point demonstrations, and other material resources.

[0092] Case study videos are obtained by selecting real-life videos, case explanations, and other material resources.

[0093] Interactive 3D is obtained by filtering 3D models or interactive resources in GLB / FBX format.

[0094] Further reading involves selecting and filtering document-type (PDF / EPUB) or literature resources.

[0095] S5: Configure a personalized style and save it as a personalized template;

[0096] Personalized style includes visual theme, color scheme, font hierarchy, sound, and delivery method;

[0097] Users can independently choose visual themes, color schemes, font levels, and sound or narration methods for the rendering interface and audio. The narration methods include three preset modes: story-based, deductive, and case-based. All configurations will be saved as a "personalized template," and all subsequently generated courseware, videos, and scripts will use this template uniformly, avoiding repeated settings and truly achieving "set once, reuse many times."

[0098] S6: Generate course outline and PPT details;

[0099] Based on the latest knowledge base and material reference list, a two-level course outline is generated, and two deliverables are generated in parallel for each PPT page: one is editable HTML code and a WYSIWYG visual editing canvas that supports dragging, adding and deleting elements; the other is the presentation script for the corresponding page, which includes narration text, animation sequence, and interactive node descriptions. The two are bound together by a unique page number.

[0100] S61: Generate a two-level curriculum syllabus;

[0101] Based on the material usage list, the confirmed knowledge point triples and AI-supplemented materials are obtained from the latest knowledge base. A knowledge topology graph is constructed based on the relationship between the triple concepts, and the hierarchical relationship between the concepts is extracted, dividing it into a two-level course outline of chapter → section → knowledge point.

[0102] S62: Generate PPT details;

[0103] S621: Page allocation and layout;

[0104] Obtain the course outline, generate at least one PPT slide for each knowledge point in the course outline, and define a basic structure template for each slide, including a title area, content area, interactive area, and material area; the basic structure template is adapted to HTML5 responsive layout;

[0105] S622: Generate HTML code;

[0106] First, a standardized HTML5 skeleton is generated based on the Vue / React component library;

[0107] A standardized HTML5 skeleton includes a header, body, and footer. The header includes dependencies for CSS, JavaScript, SVG / WebGL, and audio / video tags. The body is divided into different sections based on the page template, including the heading area. <h1>Content area <section>material section <video> / <canvas>Interactive Area <button> / <input> Add page number indicators at the end, etc.

[0108] Next, populate the content and embed interactive components;

[0109] Based on the material reference list, for static content, extract knowledge point texts, formulas, and charts from the knowledge base and populate them into the corresponding sections;

[0110] For dynamic assets, bind the video / animation asset path to the partition; for 3D models, call Three.js to load the GLTF model.

[0111] For interactive components, standardized interactive components are embedded and bound to basic interactive logic; standardized interactive components include voice answering microphones, function graph sliders, etc.

[0112] Iterate through each slide of the PPT and generate HTML code;

[0113] S623: Generate a visual editing canvas

[0114] This visual editor, developed based on Quill / CKEditor, supports drag-and-drop, adding and deleting elements, editing content, loading personalized templates, rendering objects based on generated HTML code, and monitoring changes to objects on the canvas to update the HTML code. Object changes include changes in object position and content.

[0115] S624: Generate the demonstration script;

[0116] Based on the HTML code, extract the text from the title and content areas. According to the narration style in the personalized template, use a Large Language Model (LLM) to generate the narration text. Based on the voice selected in the personalized template, use voices from the sound library to generate the narration voice-over. The sound library stores several male, female, or neutral synthesized voices with different speaking speeds and emotional intensities. Arrange the animation sequence according to the logical hierarchy and teaching order of the elements in the HTML code. Analyze the appearance, emphasis, and exit logic of each visual element on the page, assign precise timing points to each animation action, and ensure the animation sequence strictly corresponds to the narration segments.

[0117] Obtain the control elements in the interactive area, define the trigger conditions for control interaction, feedback logic, and generate interactive node descriptions;

[0118] Triggering conditions include clicking, dragging, and timed pop-ups;

[0119] Feedback logic includes playback, page redirection, pop-up windows, etc.

[0120] Bind the demo script to the page number tags of the HTML code, listen for editing and modification events in the HTML code, and update the demo script synchronously.

[0121] Furthermore, it also includes identifying and generating dynamic presentation content;

[0122] First, the NLP model scans the knowledge point text to identify the keywords that need to be dynamically demonstrated. Then, the subject animation template library is called to generate vector animations or 3D scenes in real time based on the parameter range, time series and causal relationship in the knowledge base.

[0123] Keywords that need to be dynamically demonstrated include abstract concepts such as electromagnetic induction, the limit of calculus, and DNA replication. A preset keyword list stores the keywords that need to be dynamically demonstrated.

[0124] The subject animation template library is a collection of standardized animation resources designed for teaching scenarios in various subjects, used to quickly obtain dynamic demonstration materials that are suitable for knowledge points.

[0125] S7: Edit and modify each slide of the PPT, and render a preview after adjustment;

[0126] In the visual editing canvas, you can modify and adjust each slide of the PPT. Double-click text boxes to change wording, drag and drop to replace images and animations, or adjust the trigger conditions and timing of interactive buttons. Employing differential rendering technology, only the modified layers are redrawn, and the latest effect is displayed within seconds. A snapshot is generated for each slide after each rendering, and these snapshots are stored according to different versions. You can roll back the snapshot using the undo button. A history snapshot function is provided, allowing you to revert to any version at any time.

[0127] After fine-tuning, the generated animation script is obtained. It is then played on a student's device to verify the smoothness of the animation and the correctness of the interaction logic, and to confirm whether there are any problems.

[0128] S8: Export the course package.

[0129] After confirmation, teachers can export the course package with a single click. The course package includes high-definition videos (with voiceover, animation, and subtitles, 1080P universal encoding, and videos rendered by the presentation script) for online teaching; and PDF courseware (statically formatted PPTs, printable, PPTs rendered from HTML code and converted to PDF format) for offline classes. If the source files are needed, the HTML code and presentation script can be downloaded as a package. The final version is simultaneously written back to the knowledge base and a version number is generated for use in the next round of course iterations, forming a closed loop of continuous evolution.

[0130] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Before generating the data, the present invention first converts the textbooks, school-based question banks, and authoritative materials into structured knowledge items and performs retrieval enhancement with the large model. Each step of the model output must correspond to a credible source in the knowledge base to prevent fabrication out of thin air; if the data is missing, the system will remind the teacher to supplement it instead of guessing, thereby ensuring that the content is rigorous and based on evidence, and that the generated course is based on evidence, avoiding the illusion of simply generating the large model.

[0131] (2) The courseware generated by this invention is broken down into independently draggable atomic components such as text, graphics, animations, and interactive buttons. Teachers can add, delete, change colors, and rearrange orders at any time in the WYSIWYG interface. Modifications only require partial redrawing, without having to re-run the entire package, achieving instant refinement "like editing a PPT", significantly improving classroom adaptability. The courseware content can be edited page by page, improving editing flexibility.

[0132] (3) This invention uses a multimodal large model to jointly parse multiple formats of attachments such as PPT, PDF, video, and pictures. It optimizes the recognition algorithm for complex content such as formulas, tables, and handwritten notes, improving the recognition accuracy to over 92%. The parsing results are generated into triples and written into the knowledge base to fully capture knowledge points, ensuring that the generated courseware content is accurately matched and avoiding omissions of knowledge points, thus providing solid data support for personalized teaching.

[0133] (4) All materials are taken from authorized school-based resources and open-access documents, and source identifiers are automatically added. Data is encrypted using national cryptographic algorithms at different levels during transmission and storage, and sensitive information is desensitized in advance to prevent copyright disputes and privacy leaks from the source. Data is regenerated based on the knowledge base to avoid intellectual property infringement. Attached Figure Description

[0134] Figure 1 This is a flowchart illustrating the steps of the AI-based interactive teaching courseware and video generation method of the present invention. Detailed Implementation

[0135] To provide a further understanding of the purpose, structure, features, and functions of the present invention, detailed descriptions are provided below with reference to specific embodiments.

[0136] An AI-based method for generating interactive teaching materials and videos includes the following steps:

[0137] S1: Input the core course objectives and extract key keywords through semantic parsing;

[0138] Users can use natural language to interactively input descriptions of the core objectives of this course in the text box. The core objectives can be long or short. The core objectives are analyzed in real time to extract key keywords.

[0139] The key topic words include the teaching object, subject, knowledge points, and teaching objectives;

[0140] Specifically include:

[0141] S11: The user interacts to input the core objective of the course;

[0142] Enter natural language through the text box set on the front-end page to describe the core objective of the course;

[0143] The core objective is the core requirement of the course clearly stated by the teacher in natural language, including the content in three dimensions of knowledge, ability, and scenario;

[0144] The knowledge dimension includes the core concepts, formulas, theorems, or skills that need to be mastered;

[0145] The ability dimension includes the practical, analytical, or application abilities that students need to achieve;

[0146] For example, the core objective input is: A 45-minute course for eighth-grade junior high school physics, enabling students to understand the principle of electromagnetic induction, master the right-hand rule to judge the direction of the induced current, including 1 group interaction experiment and 1 in-class test;

[0147] S12: Semantic parsing;

[0148] Preprocess, segment, and perform word-pos tagging and screening on the input text of the core objective to obtain the key topic words;

[0149] S121: Preprocessing of the text input of the core objective;

[0150] Receive the text of the core objective input by the user, remove line breaks, filter special symbols and modal particles;分句 process the long text according to punctuation marks to obtain the preprocessed text.

[0151] The special symbols include non-teaching-related symbols such as @, #, $, &, *, etc.;

[0152] Modal particles include "ah", "ya", "ne", etc.;

[0153] S122: Segment and perform word-pos tagging on the preprocessed text;

[0154] Perform word segmentation on the text through the Jieba word segmentation tool, and load a subject-specific dictionary in the Jieba word segmentation tool to achieve accurate word segmentation; then perform word-pos tagging through the HanLP tool, screen out core word classes such as nouns, verbs, and adjectives, and filter out word classes without substantial meaning such as auxiliary words ("de", "di") and conjunctions ("he", "yu"). Then obtain the logical relationship between words through dependency syntactic analysis.

[0155] Jieba is a Chinese word segmentation tool used to divide continuous Chinese text into meaningful word sequences.

[0156] HanLP is a feature-rich natural language processing toolkit that includes functions such as part-of-speech tagging, named entity recognition, and syntactic analysis.

[0157] S123: Semantic unit decomposition;

[0158] Based on the logical relationship between part-of-speech tagging and syntactic analysis, key topic words are selected;

[0159] Key keywords include knowledge keywords, ability verbs, and scenario-specific words.

[0160] Knowledge keywords are nouns or noun phrases that contain knowledge dimensions;

[0161] Ability verbs are verbs or verb phrases that include a capability dimension;

[0162] Scene qualifiers are restrictive phrases that include the scene dimension.

[0163] S13: Obtain the key terms for semantic analysis, write them into the confirmation window, and form a summary card. The confirmation window is set to editable, allowing teachers to confirm and edit it to avoid missing key terms. It truly achieves "one-sentence start, zero-threshold entry".

[0164] S2: Intelligently analyzes attachments and updates the knowledge base;

[0165] The front-end page features an "Attachment Upload" button, responding to users dragging and dropping or selecting local files to upload attachments; a multimodal large model is launched to jointly parse the text, formulas, charts, audio, and images in the attachments, extract the "concept-relationship-example" triples, and write them into the knowledge base, realizing incremental updates of attachments to the knowledge base and copyright-compliant entry into the database.

[0166] S21: Upload attachments;

[0167] Define the attachment upload trigger event as drag-and-drop or selection, responding to user actions to trigger the upload and upload local files to the server.

[0168] Local files include various formats such as PPT, PDF, Word, video, and images. The upload process uses breakpoint resume and format pre-check to ensure stable transmission of large files into the database.

[0169] S22: Annex preprocessing;

[0170] Receive uploaded attachments and use a format conversion tool to convert various attachments into a standard parsing format;

[0171] Convert PPT, PDF, Word and other document types to PDF format while preserving the layout and structure of text, formulas and charts;

[0172] The video is then transcoded into H.264 encoded MP4, sliced ​​into 10-second segments, and the audio stream (WAV format) and frame images are extracted.

[0173] Images are uniformly converted to JPG in RGB color mode.

[0174] S22: Parallel parsing of multimodal data;

[0175] Launch the multimodal large model and perform joint analysis on the text, formulas, charts, audio, and images in the attachment to obtain multimodal data.

[0176] S221: Text parsing;

[0177] For PDF files, PDFBox is used to extract copyable text, preserving paragraph structure and page number information. For text in scanned PDFs or images, MathpixOCR and Baidu OCR are used in conjunction to obtain the text. Mathpix is ​​used for formula recognition to obtain the corresponding LaTeX source code. HanLP is then used to perform part-of-speech tagging and syntactic analysis on the text. Part-of-speech tagging includes nouns, verbs, and adjectives; syntactic analysis examines the logical relationships between words, including subordination, causation, and derivation.

[0178] S222: Formula Explanation;

[0179] Obtain the LaTeX source code of the corresponding formula in the formula area, and use a MathML to LaTeX conversion tool to extract the formula.

[0180] S223: Chart Analysis;

[0181] 1) Identify chart types: The YOLOv8 object detection model is used to identify chart types, including line charts, bar charts, pie charts, tables, etc.

[0182] 2) Extract chart data;

[0183] For tables, the system identifies table borders, extracts cell coordinates, and uses OCR to recognize cell text, generating a structured table in a two-dimensional array format while preserving the relationship between the table header and the data.

[0184] For graph types (line charts, bar charts, pie charts), the Python matplotlib library is used to parse the axes (meaning, units, and scale values ​​of the horizontal and vertical axes), legends, and data points, extracting numerical information and generating key-value pair formatted data. The numerical information includes the height values ​​of the bar charts and the coordinate points of the line charts. The generated key-value pair formatted data includes the horizontal axis, vertical axis, and data points.

[0185] S224: Speech analysis;

[0186] The extracted audio stream is transcribed using the speech recognition API, converting speech into text. The text is then parsed and processed.

[0187] S225: Image analysis;

[0188] For frame images and pictures, image content is identified: OCR is used to recognize text within the image, and the CLIP model is called to identify scenes, objects, and actions in the image to obtain recognized image labels. The text and image labels are then associated.

[0189] S23: Extract the "concept-relation-example" triples;

[0190] S231: Extracting concepts;

[0191] For the parsed multimodal data, nouns with part-of-speech tagging are extracted as candidate concepts, and a BERT model trained on educational concepts is used. The candidate concepts are identified, and the identified concept names are output, along with annotations on their sources.

[0192] S232: Extraction relation;

[0193] The process involves acquiring multimodal data. For text parsing, syntactic analysis is used to identify relation types. For formula parsing, the relationships between variables and constants are determined. For chart parsing, coordinate axis correspondences and data trends are used to identify conceptual relationships. For speech parsing, syntactic analysis of the text is used to determine if conceptual relationships exist. The resulting concepts and relation types are then extracted. Relationship types include subordination, causation, and derivation.

[0194] S233: Extraction example;

[0195] For text parsing, extract the content following keywords such as "for example," "like," and "example," and perform syntactic analysis to determine the subordinate relationships within the content following these keywords. For image parsing, extract images with image labels and determine the subordinate relationships of the image labels based on the associated text. For chart parsing, extract the charts and determine the subordinate relationships between the charts based on their horizontal and vertical axes or table headers.

[0196] S24: Triple encapsulation is written to the knowledge base;

[0197] The extracted "concept-relationship-example" triples are encapsulated in a unified JSON format, with source identifiers added, and incrementally written into the knowledge base in the form of traceable IDs.

[0198] Furthermore, it also includes: S25: Visualizing the analysis results, which are then written to the knowledge base after user adjustment and confirmation;

[0199] S251: Generate visualization results;

[0200] ECharts is used to draw a visual tree diagram, with concepts as core nodes, relationships as edges, and examples as node-specific information. The node-specific information is displayed through a floating window, triggered by mouse movement or clicking.

[0201] S252: User interaction confirmation;

[0202] Users can add or delete concept nodes and set concept weights in the generated visual tree diagram. After adjustment, they can confirm the final result.

[0203] S253: Final results are entered into the database;

[0204] Based on the content of the final visualized tree diagram, the modified "concept-relationship-example" triples are obtained, encapsulated in a unified JSON format, and incrementally written into the knowledge base.

[0205] A knowledge base is a database used to store trusted, structured knowledge in the field of education; the knowledge base is deployed on a server. The knowledge base stores knowledge content including textbooks, school-based question banks, authoritative journals, and teachers' past courseware.

[0206] S3: Retrieve relevant knowledge points for locating course objectives;

[0207] Based on the key keywords of the core course objectives, a search is conducted in the knowledge base to obtain the content contained in the key keywords. Users select and confirm the relevant knowledge points, obtain the triples of the confirmed knowledge points, build an index list, and generate a list of material references.

[0208] S4: Search for materials to supplement and add to the knowledge base and update the material reference list in the course outline;

[0209] The front-end page also features an AI search button that responds to user clicks to trigger search commands. Based on course objectives and confirmed knowledge points, it generates search terms and performs a reliable search in the registered educational resource library, open journals, and shared materials within the school to obtain materials. Users select the required materials, and the selected materials are transcoded and sliced ​​and written into the knowledge base. The selected materials are added to the material citation list, and the material citation list is updated.

[0210] S41: Generate search terms;

[0211] Obtain the key keywords of the core course objectives and the triples of the confirmed knowledge points. Calculate the weight of the words in the data using the TF-IDF algorithm, select the top 8 high-frequency core words, extract basic search terms, and combine the extracted basic search terms to generate 5-8 search terms.

[0212] S42: Reliable search based on search terms;

[0213] Based on the obtained search terms, a reliable search is performed in the registered educational resource database, open journals, and shared materials within the school to obtain the materials.

[0214] S43: Visual presentation of materials;

[0215] The retrieved materials are divided into four resource types, including concept animations, case study videos, interactive 3D, and extended reading; each type of resource is presented in card format.

[0216] Concept animations are obtained by filtering animation formats (WebM / MP4), knowledge point demonstrations, and other material resources.

[0217] Case study videos are obtained by selecting real-life videos, case explanations, and other material resources.

[0218] Interactive 3D is obtained by filtering 3D models or interactive resources in GLB / FBX format.

[0219] Further reading involves selecting and filtering document-type (PDF / EPUB) or literature resources.

[0220] S5: Configure a personalized style and save it as a personalized template;

[0221] Personalized style includes visual theme, color scheme, font hierarchy, sound, and delivery method;

[0222] Users can independently choose visual themes, color schemes, font levels, sound, or narration methods for the rendering interface and audio. The narration methods include three preset modes: storytelling, deduction, and case studies. All configurations will be saved as a "personalized template," and all subsequently generated courseware, videos, and scripts will use this template uniformly, avoiding repeated settings and truly achieving "set once, reuse many times."

[0223] S6: Generate course outline and PPT details;

[0224] Based on the latest knowledge base and material reference list, a two-level course outline is generated, and two deliverables are generated in parallel for each PPT page: one is editable HTML code and a WYSIWYG visual editing canvas that supports dragging, adding and deleting elements; the other is the presentation script for the corresponding page, which includes narration text, animation sequence, and interactive node descriptions. The two are bound together by a unique page number.

[0225] S61: Generate a two-level curriculum syllabus;

[0226] Based on the material usage list, the confirmed knowledge point triples and AI-supplemented materials are obtained from the latest knowledge base. A knowledge topology graph is constructed based on the relationship between the triple concepts, and the hierarchical relationship between the concepts is extracted, dividing it into a two-level course outline of chapter → section → knowledge point.

[0227] S62: Generate PPT details;

[0228] S621: Page allocation and layout;

[0229] Obtain the course outline, generate at least one PPT slide for each knowledge point in the course outline, and define a basic structure template for each slide, including a title area, content area, interactive area, and material area; the basic structure template is adapted to HTML5 responsive layout;

[0230] S622: Generate HTML code;

[0231] First, a standardized HTML5 skeleton is generated based on the Vue / React component library;

[0232] A standardized HTML5 skeleton includes a header, body, and footer. The header includes dependencies for CSS, JavaScript, SVG / WebGL, and audio / video tags. The body is divided into different sections based on the page template, including the heading area. <h1>Content area <section>material section <video> / <canvas> Interactive Area< / canvas> < / video> < / section> < / h1> < / button> <button> / <input> Add page number indicators at the end, etc.

[0233] Next, populate the content and embed interactive components;

[0234] Based on the material reference list, for static content, extract knowledge point texts, formulas, and charts from the knowledge base and populate them into the corresponding sections;

[0235] For dynamic assets, bind the video / animation asset path to the partition; for 3D models, call Three.js to load the GLTF model.

[0236] For interactive components, standardized interactive components are embedded and bound to basic interactive logic; standardized interactive components include voice answering microphones, function graph sliders, etc.

[0237] Iterate through each slide of the PPT and generate HTML code;

[0238] S623: Generate a visual editing canvas

[0239] This visual editor, developed based on Quill / CKEditor, supports drag-and-drop, adding and deleting elements, editing content, loading personalized templates, rendering objects based on generated HTML code, and monitoring changes to objects on the canvas to update the HTML code. Object changes include changes in object position and content.

[0240] S624: Generate the demonstration script;

[0241] Based on the HTML code, extract the text from the title and content areas. According to the narration style in the personalized template, use a Large Language Model (LLM) to generate the narration text. Based on the voice selected in the personalized template, use voices from the sound library to generate the narration voice-over. The sound library stores several male, female, or neutral synthesized voices with different speaking speeds and emotional intensities. Arrange the animation sequence according to the logical hierarchy and teaching order of the elements in the HTML code. Analyze the appearance, emphasis, and exit logic of each visual element on the page, assign precise timing points to each animation action, and ensure the animation sequence strictly corresponds to the narration segments.

[0242] Obtain the control elements in the interactive area, define the trigger conditions for control interaction, feedback logic, and generate interactive node descriptions;

[0243] Triggering conditions include clicking, dragging, and timed pop-ups;

[0244] Feedback logic includes playback, page redirection, pop-up windows, etc.

[0245] Bind the demo script to the page number tags of the HTML code, listen for editing and modification events in the HTML code, and update the demo script synchronously.

[0246] Furthermore, it also includes identifying and generating dynamic presentation content;

[0247] First, the NLP model scans the knowledge point text to identify the keywords that need to be dynamically demonstrated. Then, the subject animation template library is called to generate vector animations or 3D scenes in real time based on the parameter range, time series and causal relationship in the knowledge base.

[0248] Keywords that need to be dynamically demonstrated include abstract concepts such as electromagnetic induction, the limit of calculus, and DNA replication. A preset keyword list stores the keywords that need to be dynamically demonstrated.

[0249] The subject animation template library is a collection of standardized animation resources designed for teaching scenarios in various subjects, used to quickly obtain dynamic demonstration materials that are suitable for knowledge points.

[0250] S7: Edit and modify each slide of the PPT, and render a preview after adjustment;

[0251] In the visual editing canvas, you can modify and adjust each slide of the PPT. Double-click text boxes to change wording, drag and drop to replace images and animations, or adjust the trigger conditions and timing of interactive buttons. Employing differential rendering technology, only the modified layers are redrawn, and the latest effect is displayed within seconds. A snapshot is generated for each slide after each rendering, and these snapshots are stored according to different versions. You can roll back the snapshot using the undo button. A history snapshot function is provided, allowing you to revert to any version at any time.

[0252] After fine-tuning, the generated animation script is obtained. It is then played on a student's device to verify the smoothness of the animation and the correctness of the interaction logic, and to confirm whether there are any problems.

[0253] S8: Export the course package.

[0254] After confirmation, teachers can export the course package with a single click. The course package includes high-definition videos (with voiceover, animation, and subtitles, 1080P universal encoding, and videos rendered by the presentation script) for online teaching; and PDF courseware (statically formatted PPTs, printable, PPTs rendered from HTML code and converted to PDF format) for offline classes. If the source files are needed, the HTML code and presentation script can be downloaded as a package. The final version is simultaneously written back to the knowledge base and a version number is generated for use in the next round of course iterations, forming a closed loop of continuous evolution.

[0255] This invention adds a "trusted knowledge base" as the sole source of facts in the generation process. After teachers input course objectives or upload attachments, the multimodal parsing engine first unifies heterogeneous materials such as textbooks, school-based question banks, authoritative journals, and teachers' previous courseware into "concept-relationship-example" triples, which are then injected into the knowledge base with traceable IDs. Subsequently, when generating each PPT slide or video narration, the large model must execute a "retrieve first, then generate" retrieval enhancement (RAG) process: the model first requests relevant entries from the knowledge base, then writes content based on the returned fragments, while simultaneously marking the source link in the sidebar of the courseware. If the knowledge base lacks corresponding information, the system will automatically pop up a "to be supplemented" prompt, rather than fabricating it itself. Teachers can review each item before publication or modify it in real time during class: each page's content is broken down into independent HTML components, supporting fine-tuning word by word, image by image, and animation by animation, and local changes only require differential rendering, taking effect in seconds. Therefore, the courseware and videos are always "based on evidence," and teachers have absolute control over the entire process, making it visible, modifiable, and traceable. This ensures that the generated courseware and videos are based on evidence, making the content controllable and adjustable.

[0256] This invention automatically identifies keywords indicating "required dynamic demonstration" when generating presentation scripts and calls upon the subject-specific animation template library. Based on parameter ranges, time series, and causal relationships in the knowledge base, the system generates vector animations or 3D scenes in real time. The animation script and courseware HTML code are output synchronously, allowing for additions, deletions, color changes, and reordering at any time within a WYSIWYG interface. Modifications only require partial redrawing, eliminating the need to rewrite the entire package, enabling instant refinement "like editing a PowerPoint presentation." This significantly improves classroom responsiveness, and courseware content can be edited page-by-page, enhancing editing flexibility.

[0257] This invention outputs each page of courseware as standardized HTML5 code, naturally supporting SVG, Canvas, WebGL, audio / video tags, and JavaScript interaction. Teachers can download and open it directly in a browser, or import it into any online editor for secondary development: for example, inserting draggable sliders to observe function graph changes, embedding WebXR for solid geometry experiences, and using a microphone for voice-based question answering. The HTML structure is clear, readable, and maintainable, allowing teachers or research groups to iterate and reuse versions like managing web pages.

[0258] The present invention has been described in the above-described embodiments; however, these embodiments are merely examples for implementing the present invention. It must be noted that the disclosed embodiments do not limit the scope of the present invention. Conversely, any modifications and refinements made without departing from the spirit and scope of the present invention are within the scope of patent protection of the present invention.< / button> < / canvas> < / video> < / section> < / h1>

Claims

1. A method for generating interactive teaching courseware and videos based on artificial intelligence, characterized in that: Includes the following steps: S1: Input the core course objectives and extract key keywords through semantic parsing; S2: Intelligently analyzes attachments and updates the knowledge base; Respond to user drag-and-drop or selection of local file upload attachments; The multimodal large model is launched to jointly parse the text, formulas, charts, audio, and images in the attachments, extract the "concept-relationship-example" triples, and write them into the knowledge base; S3: Retrieve relevant knowledge points for locating course objectives; Based on key keywords, a search is performed in the knowledge base to obtain the content contained in the key keywords in the knowledge base. Users select and confirm relevant knowledge points, obtain the triples of the confirmed knowledge points, build an index list, and generate a list of material references. S4: Search for materials to supplement and add to the knowledge base and update the material reference list in the course outline; In response to search commands, the system generates search terms based on course objectives and confirmed knowledge points, performs a reliable search in the registered educational resource database, open journals, and shared materials within the school to obtain materials, allows users to select the required materials, transcodes and slices the selected materials and writes them into the knowledge base, adds entries for the selected materials to the material reference list, and updates the material reference list. S5: Configure a personalized style and save it as a personalized template; S6: Generate course outline and PPT details; Based on the latest knowledge base and material reference list, a two-level course outline is produced, and two deliverable PPT details are generated in parallel for each PPT page: one is editable HTML code and visual editing canvas; the other is the presentation script for the corresponding page. S7: Edit and modify each slide of the PPT, and render a preview after adjustment; S8: Export course package; After confirming that everything is correct, generate a course package based on the PPT details and export the course package.

2. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 1, characterized in that: Step S1 includes: S11: Core objectives of the user interaction input course; The core objective is for teachers to clearly define the core course requirements through natural language, including content in three dimensions: knowledge, skills, and scenarios. S12: Semantic parsing; The core target text is preprocessed, segmented, and part-of-speech tagged to obtain key keywords; S13: Obtain the key words for semantic parsing, write them into the confirmation window, and form a summary card; The confirmation window is set to editable, allowing teachers to confirm and modify it.

3. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 2, characterized in that: S12 includes: S121: Text preprocessing of the core target input; The system receives the core target text input by the user, removes line breaks, filters special characters and modifiers; and segments long texts according to punctuation marks to obtain preprocessed text. S122: Perform word segmentation and part-of-speech tagging on the preprocessed text; The text was segmented using the Jieba word segmentation tool, and part-of-speech tagging was performed using the HanLP tool to select the parts of speech of nouns, verbs, and adjectives, while filtering out auxiliary words and conjunctions. Finally, the logical relationships between words were obtained through dependency parsing. S123: Semantic unit decomposition; Based on the logical relationship between part-of-speech tagging and syntactic analysis, key topic words are selected; Key keywords include knowledge keywords, ability verbs, and scenario-limiting words; Knowledge keywords are nouns or noun phrases that contain knowledge dimensions; Ability verbs are verbs or verb phrases that include a capability dimension; Scene qualifiers are restrictive phrases that include the scene dimension.

4. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 1, characterized in that: Step S2 includes: S21: Upload attachments; S22: Annex preprocessing; Receive uploaded attachments and use a format conversion tool to convert various attachments into a standard parsing format; S22: Parallel parsing of multimodal data; Launch the multimodal large model to jointly parse the text, formulas, charts, audio, and images in the attachment to obtain multimodal data; S23: Extract the "concept-relation-example" triples; Concepts, relations, and examples are extracted from multimodal data to form "concept-relation-example" triples; S24: Triple encapsulation is written to the knowledge base; The extracted "concept-relationship-example" triples are encapsulated in a unified JSON format, with source identifiers added, and incrementally written into the knowledge base in the form of traceable IDs.

5. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 4, characterized in that: The standard parsing format is as follows: PPT, PDF, and Word documents are converted to PDF format; videos are transcoded to H.264 encoded MP4, sliced ​​into 10-second segments, and the audio stream and frame images are extracted; images are uniformly converted to JPG in RGB color mode. S22 includes: S221: Text parsing; For PDF files, PDFBox is used to extract copyable text while preserving paragraph structure and page number information. For text in scanned PDFs or images, MathpixOCR and Baidu OCR are used in combination to parse the text. Mathpix is ​​used for formula recognition to obtain the LaTeX source code of the corresponding formulas. HanLP is used to perform part-of-speech tagging and syntactic analysis on the text. S222: Formula Explanation; Obtain the LaTeX source code of the corresponding formula in the formula area, and use a MathML to LaTeX conversion tool to extract the formula; S223: Chart Analysis; 1) Identify chart types: The YOLOv8 object detection model is used to identify chart types, including graphs and tables; 2) Extract chart data; For tables, the system identifies table borders, extracts cell coordinates, and uses OCR to recognize cell text, generating a structured table in a two-dimensional array format while preserving the relationship between the table header and the data. For graphs, the Python matplotlib library is used to parse coordinate axes, legends, and data points, extract numerical information, and generate key-value pair formatted data. S224: Speech analysis; The extracted audio stream is transcribed using the speech recognition API, converting speech into text. S225: Image analysis; For frame images and pictures, identify image content: use OCR to recognize text in the image, call the CLIP model to recognize scenes, objects and actions in the image, obtain image labels, and associate text with image labels.

6. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 4, characterized in that: Step S23 includes: S231: Extracting concepts; For the multimodal data obtained after parsing, extract the nouns in the text as candidate concepts, use the BERT model trained on educational concepts to identify the candidate concepts, output the name of the identified concept, and label the source of the concept name; S232: Extraction relation; For multimodal data, syntactic analysis is used to obtain relation types for text parsing data, the relationships between variables and constants in formulas are obtained for formula parsing results, and conceptual relationships are determined by coordinate axis correspondence and data change trends for graph parsing results. For speech parsing results, syntactic analysis of the text is used to determine whether there are relationships between concepts, and the corresponding concepts and relation types are extracted. Relationship types include subordinate relationships, causal relationships, and inferential relationships. S233: Extraction example; For the text obtained from text parsing, extract the content after keywords such as "for example," "like," and "example," and perform syntactic analysis to obtain the concept that the content after the keywords has a subordinate relationship. For image parsing data, extract images with image labels and obtain the concept of image label belonging based on the associated text; for chart parsing results, extract charts and associate the concept of chart belonging based on the horizontal and vertical axes or table headers.

7. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 4, characterized in that: It also includes S25: Visualized analysis results, which are then written to the knowledge base after user adjustment and confirmation; S251: Generate visualization results; Draw a visual tree diagram, with concepts as core nodes, relationships as edges, and examples as node-specific information for concepts; S252: User interaction confirmation; Users can add or delete concept nodes and set concept weights in the generated visual tree diagram. After adjustment, they can confirm the final result. S253: Final results are entered into the database; Based on the content of the final visualized tree diagram, the modified "concept-relationship-example" triples are obtained, encapsulated in a unified JSON format, and incrementally written into the knowledge base.

8. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 1, characterized in that: S4 includes: S41: Generate search terms; Obtain the key keywords of the core course objectives and the triplet of the confirmed knowledge points. Calculate the weight of the words in the data using the TF-IDF algorithm, select the Top 8 core words, extract basic search terms, and combine the extracted basic search terms to generate 5-8 search terms. S42: Reliable search based on search terms; Based on the obtained search terms, a reliable search is performed in the registered educational resource database, open journals, and shared materials within the school to obtain the materials; S43: Visual presentation of materials; The retrieved materials are divided into four resource types, including concept animations, case study videos, interactive 3D, and extended reading; each type of resource is presented in card format.

9. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 1, characterized in that: S6 includes: S61: Generate a two-level curriculum syllabus; Based on the material usage list, the confirmed knowledge point triples and AI-supplemented materials are obtained from the latest knowledge base. The knowledge topology graph is constructed based on the relationship between the triple concepts, the hierarchical relationship between the concepts is extracted, and the course outline is divided into two levels: chapters, sections, and knowledge points. S62: Generate PPT details; S621: Page allocation and layout; Obtain the course outline, and based on the knowledge points in the course outline, generate at least one PPT slide for each knowledge point, and define a basic structure template for each slide; S622: Generate HTML code; First, a standardized HTML5 skeleton is generated based on the Vue / React component library; A standardized HTML5 skeleton includes a header, body, and footer. The header includes dependencies for CSS, JavaScript, SVG / WebGL / audio / video tags; the body generates different sections according to the page template, including a title section, content section, material section, and interactive section; and the footer adds page number markers. Next, populate the content and embed interactive components; Based on the material reference list, for static content, extract knowledge point texts, formulas, and charts from the knowledge base and populate them into the corresponding sections; For dynamic assets, bind the video / animation asset path to the partition; for 3D models, call Three.js to load the GLTF model. For interactive components, embed standardized interactive components and bind basic interactive logic; Iterate through each slide of the PPT and generate HTML code; S623: Generate a visual editing canvas Based on a visual editor, load personalized templates, render objects according to the generated HTML code, listen for changes in objects on the canvas, and update the HTML code accordingly. S624: Generate the demonstration script; Based on the HTML code, extract the text from the title and content areas. Based on the explanation method in the personalized template, call the Large Language Model (LLM) to generate the narration text. Based on the sound selected in the personalized template, call the sound from the sound library to generate the narration voice-over. Based on the logical hierarchy of elements in the HTML code and the teaching order, the animation timing is arranged to assign precise time points to each animation action; Obtain the control elements in the interactive area, define the trigger conditions for control interaction, feedback logic, and generate interactive node descriptions; Bind the demo script to the page number tags of the HTML code, listen for editing and modification events in the HTML code, and update the demo script synchronously.

10. The method for generating interactive teaching courseware and videos based on artificial intelligence as described in claim 9, characterized in that: Step S624 also includes identifying and generating dynamic presentation content; First, the NLP model scans the knowledge point text to identify the keywords that need to be dynamically demonstrated. Then, the subject animation template library is called to generate vector animations or 3D scenes based on the parameter range, time series and causal relationship in the knowledge base. A pre-defined keyword list stores the keywords that need to be dynamically demonstrated.

Images