A large model-based online video training method

Abstract

The application provides an online video training method based on a large model, which comprises the following steps: constructing teaching data and displaying the teaching data to a user in a visual manner; performing vectorization processing on the teaching data, dividing the teaching data according to the similarity between vectors, and adding knowledge point positioning to the teaching data according to the division result; generating a multi-type question bank according to the characteristics of different types of questions and the teaching data carrying the knowledge point positioning, and selecting part of the questions from the multi-type question bank to interact with the user; performing semantic migration on the questions so that the questions for each interaction with the user are not completely the same, playing back part of the teaching data according to the interaction result of the user and the questions, and generating a personalized review plan and learning suggestions. The application has the beneficial effects that long videos are divided into short videos, and a quick search module and a self-question generation module are added, so that the interactivity is increased, the learning interest of students is improved, and the learning efficiency and learning effect are improved.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and in particular relates to an online video training method based on a large model. Background Technology

[0002] Online video education and training has developed rapidly in recent years, primarily driven by the widespread adoption of mobile devices, 5G communication, and the internet, as well as technological advancements. Distance learning has become mainstream, with many educational institutions and platforms transforming themselves to offer a wealth of online courses to meet the diverse needs of learners. This trend not only increases the flexibility of learning but also creates opportunities for the dissemination and globalization of educational resources.

[0003] The main drawbacks of online video training at present include the following:

[0004] 1. Training videos are too long: Some online videos are very long. With a huge amount of information, this can cause information overload for trainees, making it difficult to grasp the key points, resulting in learning difficulties and low efficiency.

[0005] 2. Insufficient interactivity: Traditional video training is usually a one-way communication method, where students can only passively receive information, leading to a decline in learning interest; in traditional question banks, students directly memorize practice answers in exams, which prevents them from applying what they have learned to other situations and achieves a true understanding.

[0006] 3. Limited depth of learning: Many traditional video training courses may only provide a superficial introduction, lacking in-depth discussion and detailed analysis. Often, information copied and pasted into search engines like Baidu is disorganized, requiring time-consuming summarization. Alternatively, generalized models may offer more knowledge but can create a knowledge illusion, potentially misleading learners. All of these factors increase the difficulty of acquiring knowledge, often leading learners to abandon further research. Summary of the Invention

[0007] In view of this, the present invention aims to propose an online video training method based on a large model, in order to solve at least one of the above-mentioned technical problems.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0009] An online video training method based on a large model, the method comprising:

[0010] Build teaching data and present it to users in a visual way;

[0011] The teaching data is vectorized, and the data is divided based on the similarity between vectors. Knowledge points are then added to the teaching data based on the division results.

[0012] Based on the characteristics of different question types and combined with teaching data containing knowledge point locations, a multi-question question bank is generated, and some questions are selected from the multi-question question bank for interaction with users;

[0013] By performing semantic transfer on the questions to ensure that the questions are not exactly the same each time the user interacts with them, partial teaching data is replayed based on the user's interaction with the questions, and personalized review plans and learning suggestions are generated.

[0014] Furthermore, the process of vectorizing the teaching data includes:

[0015] The system pre-sets teaching documents and videos, separates the teaching videos to obtain teaching audio, transcribes the teaching audio into target text, and performs vectorization processing; it also associates the time points of the teaching audio, teaching videos, and target text.

[0016] Extract the logical hierarchy from the teaching documents and output them in a structured format to obtain teaching text and teaching images. Then, vectorize the teaching text and teaching images respectively.

[0017] Furthermore, the process of segmenting teaching data based on the similarity between vectors and adding knowledge point locations to the teaching data according to the segmentation results includes:

[0018] Calculate the cosine similarity between vectors to obtain the segmentation boundary of knowledge points. Divide the teaching data into two segments based on the distribution of knowledge points. Then, use recursion to divide the results of the first segmentation into two segments to obtain multiple data segments that meet the teaching requirements.

[0019] Add corresponding labels to each data segment and obtain the timestamp between two adjacent data segments. Add knowledge point location to the teaching data based on the timestamp.

[0020] Furthermore, during the process of selecting some questions and interacting with the user:

[0021] Based on the knowledge points corresponding to the questions, each question is assigned a priority, and the questions are sorted according to their priority. Questions with higher priority are selected for user interaction first. The priority of selection from high to low is as follows: teaching text, target text, and teaching images.

[0022] Furthermore, after setting the priority:

[0023] Record user interaction traces and dynamically adjust questions based on these traces. For the same knowledge point, use different question types to interact with the user, and the question types may differ between two consecutive interactions.

[0024] Furthermore, the generation process of the multi-question bank includes:

[0025] The retrieval-enhanced generation technology is used to set different answer generation strategies for different question types. The answer generation strategies and teaching data are then input into the large language model to generate corresponding questions and answers for different knowledge points in the teaching data.

[0026] Furthermore, the answer generation strategy includes:

[0027] If the question is a true / false question, the generation strategy is to determine the truth or falsehood of the statement, replace the knowledge points with synonyms or opposites, and generate a binary option.

[0028] If the question is a multiple choice, the generation strategy is to provide multiple answers based on the statement, and at least one of the answers is the correct option that corresponds to the knowledge point.

[0029] If the question is a fill-in-the-blank question, the generation strategy is to fill in the missing part based on the context or prompts, partially delete the content of the knowledge point, and record the deleted content as the answer;

[0030] If the question type is a short answer question, the user's answer will be recorded and awaited manual processing.

[0031] Furthermore, the process of semantic transfer of the question includes:

[0032] Based on the vectorization processing results, select the content with high repetition in the same knowledge point as keywords, conduct online searches on the keywords, and perform synonym replacements on the keywords based on the search results;

[0033] By filtering out highly repetitive content from different knowledge points as general terms, these general terms are then removed from the keyword list.

[0034] Furthermore, the process of replaying part of the teaching data based on the user's interaction with the questions includes:

[0035] If the user fails to answer or answers incorrectly within the specified time, the teaching data will be rolled back to the location of the previous knowledge point and then re-visualized to the user.

[0036] Furthermore, the process of generating personalized review plans and learning suggestions includes:

[0037] Select questions that users answered incorrectly, identify the corresponding knowledge points as the user's weaknesses, compile statistics on these weaknesses, and provide the user with a review plan and learning suggestions based on these weaknesses.

[0038] Compared with existing technologies, the online video training method based on a large model described in this invention has the following beneficial effects:

[0039] Based on documents and training models, online long videos are intelligently segmented into short videos according to knowledge points. Based on courseware and video understanding, corresponding question banks are generated. After users watch each key segment, the system automatically pops up the corresponding questions to help reinforce learning.

[0040] When encountering unfamiliar knowledge points, students can directly select them in the system for intelligent retrieval and summarization, saving them time.

[0041] Dividing long videos into short videos, along with a quick search module and a self-generated question module, increases interactivity and helps improve students' learning interest, efficiency, and effectiveness. Attached Figure Description

[0042] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0043] Figure 1 This is a schematic diagram of the online video training method described in an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the target text vectorization process described in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram illustrating the vectorization process of teaching texts and teaching images as described in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram illustrating the process of storing vectorized teaching data into a vector database according to an embodiment of the present invention;

[0047] Figure 5 This is a flowchart illustrating the training question bank model M1 as described in an embodiment of the present invention;

[0048] Figure 6 This is a schematic diagram of the question bank generation process according to an embodiment of the present invention;

[0049] Figure 7 This is a schematic diagram illustrating the optimized process of the question bank according to an embodiment of the present invention;

[0050] Figure 8 This is a schematic diagram of the visual interactive interface described in an embodiment of the present invention;

[0051] Figure 9 This is a schematic diagram of the visual interactive interface described in an embodiment of the present invention;

[0052] Figure 10 This is a schematic diagram of the visual interactive interface described in an embodiment of the present invention. Detailed Implementation

[0053] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0054] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0055] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0056] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0057] An online video training method based on a large model includes:

[0058] S1. Construct teaching data and present it to users in a visual way. Screenshots of the interactive visualization interface are shown below. Figure 8-10 As shown.

[0059] S2. Vectorize the teaching data, divide the teaching data by the similarity between vectors, and add knowledge point locations to the teaching data based on the division results.

[0060] In this embodiment, the execution flow of step S2 includes:

[0061] The system pre-sets teaching documents and videos, separates the teaching videos to obtain teaching audio, transcribes the teaching audio into target text, and performs vectorization processing; it also associates the time points of the teaching audio, teaching videos, and target text.

[0062] Extract the logical hierarchy from the teaching documents and output them in a structured format to obtain teaching text and teaching images. Then, vectorize the teaching text and teaching images respectively.

[0063] Calculate the cosine similarity between vectors to obtain the segmentation boundary of knowledge points. Divide the teaching data into two segments based on the distribution of knowledge points. Then, use recursion to divide the results of the first segmentation into two segments to obtain multiple data segments that meet the teaching requirements.

[0064] Add corresponding labels to each data segment and obtain the timestamp between two adjacent data segments. Add knowledge point location to the teaching data based on the timestamp.

[0065] During the process of selecting some questions and interacting with the user:

[0066] Based on the knowledge points corresponding to each question, a priority is assigned to each question. Questions are then sorted according to priority, and higher-priority questions are selected for user interaction first. The selection priority, from highest to lowest, is: teaching text, target text, and teaching images. After setting the priority:

[0067] Record user interaction traces and dynamically adjust questions based on these traces. For the same knowledge point, use different question types to interact with the user, and the question types may differ between two consecutive interactions.

[0068] In some other embodiments, one execution flow for step S2 is as follows:

[0069] The course instructor decides on the course topic K1, writes the teaching document W1, and records the course video V1; the course instructor transfers the materials for course K1: video V1 and teaching document W1 (supporting rich formats of unstructured data such as markdown, pdf, doc, html, ppt, etc.) into the course topic K1;

[0070] like Figure 2 As shown, after the upload is completed, the audio and video separation task is executed to separate the video file into audio and video, resulting in audio A1. The audio-to-text task is then executed to convert audio A1 into target text VT1. The text vectorization task is then executed to vectorize the target text VT1 into Vec1. Finally, the time points of the target text VT1, Vec1, audio A1, and video V1 are associated.

[0071] like Figure 3As shown, it provides a general document parsing service, supporting the extraction of logical hierarchical structures such as titles and paragraphs from unstructured documents (text, tables, images, etc.) and outputting them in a structured format. The content of the teaching document W1 is parsed and obtained as text WT1 and image WP1.

[0072] The text vectorization task is performed, vectorizing the target text WT1 into Vec2 and the target image WP1 into Vec3. The image vectorization task is executed in two ways: first, an image content understanding service (imagetovec1) uses a multimodal large model to parse and understand the image content and perform text recognition, storing the parsed text vectors in a vector database; second, an image text recognition service (imagetovec2) uses OCR capabilities to recognize text in the image, storing the parsed text vectors in a vector database.

[0073] like Figure 4 As shown, Vec1, Vec2, and Vec3 are sliced ​​and stored in a vector database. The long video is then segmented into short videos by semantic segmentation of knowledge points. The segmentation boundary of knowledge points is determined by calculating the cosine similarity between vectors. When the cosine similarity decreases significantly, it means that a new knowledge point may be starting. During similarity matching, the threshold can be dynamically adjusted to ensure the rationality of the segmentation granularity and to ensure that overly similar content is not segmented into different knowledge points.

[0074] For certain knowledge points, recursive processing can be used, that is, further retrieval and segmentation, to further refine the knowledge points. For example, knowledge points in electricity can be further divided into more detailed sub-knowledge points such as current, voltage, and resistance.

[0075] The knowledge point vectors obtained through the above method are used to regenerate short video information. Here, only the short video title SV needs to be generated. SV corresponds to the corresponding video duration range. Assuming that the long video course is 2 hours long and there are 20 knowledge points, then the data of the short video collection is: K1 = {SV1, SV2, SV3, ..., SV19}. Since the start time of each short video is recorded, it is not necessary to physically cut the long video to locate the knowledge point video in the player device of this invention.

[0076] S3. Based on the characteristics of different question types and combined with teaching data containing knowledge point locations, generate a multi-question question bank, and select some questions from the multi-question question bank for interaction with users.

[0077] In this embodiment, the execution flow of step S3 includes:

[0078] Retrieval-enhanced generation techniques are used to set different answer generation strategies for different question types. These strategies, along with teaching data, are then input into a large language model to generate corresponding questions and answers for different knowledge points within the teaching data. The specific answer generation strategies are as follows:

[0079] If the question is a true / false question, the generation strategy is to determine the truth or falsehood of the statement, replace the knowledge points with synonyms or opposites, and generate a binary option.

[0080] If the question is a multiple choice, the generation strategy is to provide multiple answers based on the statement, and at least one of the answers is the correct option that corresponds to the knowledge point.

[0081] If the question is a fill-in-the-blank question, the generation strategy is to fill in the missing part based on the context or prompts, partially delete the content of the knowledge point, and record the deleted content as the answer;

[0082] If the question type is a short answer question, the user's answer will be recorded and awaited manual processing.

[0083] The process of semantic transfer of the question includes:

[0084] Based on the vectorization processing results, select the content with high repetition in the same knowledge point as keywords, conduct online searches on the keywords, and perform synonym replacements on the keywords based on the search results;

[0085] By filtering out highly repetitive content from different knowledge points as general terms, these general terms are then removed from the keyword list.

[0086] Retrieval-Augmented Generation (RAG) is an AI approach that combines retrieval and generative techniques to improve the relevance, accuracy, and diversity of content generated by the model. When processing generation tasks, RAG first retrieves the most relevant fragments to the input from a large amount of external data or knowledge base. Then, the retrieved information, along with the original input, is fed into a Large Language Model (LLM) as prompts or context to guide the model in generating more accurate and richer answers.

[0087] In some other embodiments, one execution flow of step S3 is as follows:

[0088] like Figure 5 As shown, the training question bank model M1 is configured with question types QT, including true / false questions QT1, multiple choice questions QT2, fill-in-the-blank questions QT3, and short answer questions QT4. The detailed training process is as follows:

[0089] True / False Questions QT1: In the retrieval phase, the model retrieves the definition of true / false questions: it requires judging whether a statement is true or false based on its content; in the generation phase, the model summarizes the characteristics of true / false questions based on the retrieved information: they usually have only two options (True / False) and test the direct judgment of a certain fact or knowledge point.

[0090] Multiple Choice Questions QT2: In the retrieval phase, the characteristics of the retrieved multiple choice questions are: multiple options for a single question, with only one or more correct answers; in the generation phase, the model generates a description of the multiple choice questions: the multiple choice questions are highly structured and test the ability to identify and select knowledge points.

[0091] Fill-in-the-blank questions in QT3: In the retrieval stage, the format and usage of fill-in-the-blank questions are retrieved; in the generation stage, a summary is generated. Fill-in-the-blank questions require candidates to fill in the missing parts based on the context or prompts, testing the application and accuracy of knowledge.

[0092] QT4 Question and Answer Questions: In the retrieval phase, the model retrieves feature information about the question and answer questions: generally, candidates are required to provide expanded answers based on the questions; in the generation phase, it is concluded that question and answer questions are usually used to test in-depth understanding and analytical ability, and the answers are flexible with no standard answer.

[0093] like Figure 6 As shown, a large amount of question data of different types is collected, including multiple choice questions, short answer questions, fill-in-the-blank questions, and true / false questions, so that the RAG model can learn the corresponding questions and answers. The preprocessed knowledge points are input into the LLM model to generate a preliminary question bank model framework. Through multiple iterations and feedback, the generation of question bank exercises is improved.

[0094] By batch inputting knowledge point content (SV) and processing it through the question bank model M1, a question bank (QD) for several question types can be obtained, which is then stored in a vector database. The large model is then used for further optimization, and the continuity is corrected using human language.

[0095] like Figure 7 As shown, the model is continuously optimized through user feedback (such as answer accuracy, answer time, and user ratings). By automatically evaluating questions and receiving feedback data, the RAG model can continuously fine-tune the generation algorithm to improve the quality and difficulty matching of questions; based on feedback, the model can enhance its control over certain question types (such as the quality of distractors generated in multiple-choice questions and the conciseness of answers in short-answer questions).

[0096] After obtaining the question bank (QD), measures are taken to prevent students from memorizing questions during real-time question generation. The question generation model M2 is used to generate dynamic questions based on the existing QD. The capabilities of model M2 are as follows: In each assessment, the numerical values ​​or contexts of the questions differ; the answer options are shuffled, and the question types are dynamically switched. Each corresponding answer requires a separate question generation model. This enriches the question variety, saves time in question generation, and enhances the user experience.

[0097] S4. By performing semantic transfer on the questions to ensure that the questions are not exactly the same each time the user interacts with them, some teaching data is replayed based on the user's interaction with the questions, and personalized review plans and learning suggestions are generated.

[0098] In this embodiment, the specific execution flow of step S4 includes:

[0099] If the user fails to answer or answers incorrectly within the specified time, the teaching data will be rolled back to the location of the previous knowledge point and then re-visualized to the user.

[0100] Select questions that users answered incorrectly, identify the corresponding knowledge points as the user's weaknesses, compile statistics on these weaknesses, and provide the user with a review plan and learning suggestions based on these weaknesses.

[0101] In some other embodiments, a specific execution flow of step S4 is as follows:

[0102] After the administrator completes the creation of the question bank, it must undergo strict review to form the final question bank set Q = {Q1, Q2, Q3, ..., Q10}. This question bank covers multiple key knowledge points to ensure the comprehensiveness and coverage of the exam.

[0103] The system performs limited semantic transfer processing on each question in the question bank based on the Retrieval Enhancement Generative Model (RAG); this process ensures that each learner will not encounter the exact same questions during the learning process, thereby improving the relevance and personalization of the questions and avoiding repetitive learning.

[0104] The playback device of this invention supports cross-platform operation and provides rich interfaces, enabling seamless integration with third-party systems. After authorized login, the user's learning performance and training results will be automatically reported to the third-party system through the interface, thereby realizing real-time sharing and feedback of learning outcomes.

[0105] After logging in, users can begin watching the course. The system provides an intuitive user interface. After each knowledge point is played, the system automatically pauses and displays the answer content over the player; for example, if it is a true / false question, the system will prompt the user to judge whether a knowledge point is true or false. If the user answers incorrectly or fails to provide a correct answer within the specified time, the system will automatically rewind to the previous video and replay it.

[0106] The questions are presented in a variety of ways, so even for the same knowledge point, different users will see different question formats, which increases the challenge of learning. If a user answers incorrectly multiple times in a row, the system will ask the user whether to skip the current knowledge point or conduct an internet search (such as through academic databases, Wikipedia, Zhihu and other knowledge-vertical websites) to obtain supplementary information on the relevant knowledge point from external resources, and then summarize and present it to the user again.

[0107] If a user successfully answers a question using online knowledge, the system reports the correct result to the question bank and updates the model in real time, thereby optimizing the generation of subsequent questions and the accuracy of answers.

[0108] After all video content has been played, the system will automatically summarize the user's learning performance and generate a personalized review plan for the user's weak knowledge points to help the user conduct targeted learning and consolidation.

[0109] The system generates a comprehensive exam based on a preset question ratio. Users need to complete the exam to assess their learning effectiveness. Based on the exam results, users can choose to review their weak areas or proceed directly to the graduation process. The system will record the final score and provide follow-up learning suggestions.

[0110] Based on documents and training models, the system intelligently segments online long videos into short videos according to knowledge points using a large model. It then generates summaries for each segment and, based on courseware and video comprehension, generates a corresponding question bank. After watching each key segment, the system automatically displays corresponding questions to reinforce learning. Users can directly select unfamiliar knowledge points for intelligent retrieval and summarization, saving time. The segmentation of long videos into short videos, combined with a quick search module and a self-generated question module, increases interactivity, helping to improve learning interest, efficiency, and effectiveness.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online video training based on a large model, characterized in that, The method includes: Build teaching data and present it to users in a visual way; The teaching data is vectorized, and the data is divided based on the similarity between vectors. Knowledge points are then added to the teaching data based on the division results. Based on the characteristics of different question types and combined with teaching data containing knowledge point locations, a multi-question question bank is generated, and some questions are selected from the multi-question question bank for interaction with users; By performing semantic transfer on the questions to ensure that the questions are not exactly the same each time the user interacts with them, partial teaching data is replayed based on the user's interaction with the questions, and personalized review plans and learning suggestions are generated. The process of vectorizing the teaching data includes: The system pre-sets teaching documents and videos, separates the teaching videos to obtain teaching audio, transcribes the teaching audio into target text, and performs vectorization processing; it also associates the time points of the teaching audio, teaching videos, and target text. Extract the logical hierarchy from the teaching documents and output them in a structured format to obtain teaching text and teaching images. Then, vectorize the teaching text and teaching images respectively. The process of segmenting teaching data based on the similarity between vectors and then adding knowledge point locations to the teaching data based on the segmentation results includes: Calculate the cosine similarity between vectors to obtain the segmentation boundary of knowledge points. Divide the teaching data into two segments based on the distribution of knowledge points. Then, use recursion to divide the results of the first segmentation into two segments to obtain multiple data segments that meet the teaching requirements. Add corresponding labels to each data segment and obtain the timestamp between two adjacent data segments. Add knowledge point location to the teaching data based on the timestamp. During the process of selecting questions to interact with the user, each question is assigned a priority based on the knowledge points it corresponds to. The questions are then sorted according to their priority, and questions with higher priority are selected for interaction with the user first. The selection priority, from highest to lowest, is as follows: teaching text, target text, and teaching images. After setting priorities, record user interaction traces and dynamically adjust questions based on these traces. For the same knowledge point, use different question types to interact with the user, and the question types will be different in two adjacent interactions. The generation process of the multi-question bank includes: The retrieval-enhanced generation technology is used to set different answer generation strategies for different question types. The answer generation strategies and teaching data are then input into the large language model to generate corresponding questions and answers for different knowledge points in the teaching data.

2. The online video training method based on a large model according to claim 1, characterized in that, The answer generation strategy includes: If the question is a true / false question, the generation strategy is to determine the truth or falsehood of the statement, replace the knowledge points with synonyms or opposites, and generate a binary option. If the question is a multiple choice, the generation strategy is to provide multiple answers based on the statement, and at least one of the answers is the correct option that corresponds to the knowledge point. If the question is a fill-in-the-blank question, the generation strategy is to fill in the missing part based on the context or prompts, partially delete the content of the knowledge point, and record the deleted content as the answer; If the question type is a short answer question, the user's answer will be recorded and awaited manual processing.

3. The online video training method based on a large model according to claim 1, characterized in that, The process of semantic transfer of the question includes: Based on the vectorization processing results, select the content with high repetition in the same knowledge point as keywords, conduct online searches on the keywords, and perform synonym replacements on the keywords based on the search results; By filtering out highly repetitive content from different knowledge points as general terms, these general terms are then removed from the keyword list.

4. The online video training method based on a large model according to claim 1, characterized in that, The process of replaying part of the teaching data based on the user's interaction with the questions includes: If the user fails to answer or answers incorrectly within the specified time, the teaching data will be rolled back to the location of the previous knowledge point and then re-visualized to the user.

5. The online video training method based on a large model according to claim 1, characterized in that, The process of generating personalized review plans and learning suggestions includes: Select questions that users answered incorrectly, identify the corresponding knowledge points as the user's weaknesses, compile statistics on these weaknesses, and provide the user with a review plan and learning suggestions based on these weaknesses.

Images