A course video-based knowledge point learning recommendation method and system

Abstract

The application discloses a kind of based on course video's knowledge point learning recommendation method and system, belong to wisdom education technical field, can solve the problem that existing course video learning mode is difficult to meet the individualized learning needs of student.Said method includes: S1, the association between multiple knowledge points is determined, and based on multiple knowledge points, using optical character recognition method and large language model, course video is processed, and the learning video of each knowledge point is generated;S2, using large language model determines the knowledge point in each test question in question bank, and determines the learning corresponding relationship between knowledge point, learning video and test question three;S3, according to association and learning corresponding relationship, construct knowledge graph, and according to the specific demand of knowledge demander and knowledge graph, learning video and test question matched with specific demand are pushed to knowledge demander.The application is used to carry out knowledge point learning recommendation.

Description

Technical Field

[0001] The present invention relates to a method and system for recommending knowledge point learning based on course videos, belonging to the technical field of smart education. Background Art

[0002] In the medical and nursing fields, teachers typically record multiple course videos through live broadcasts, recorded broadcasts, and face-to-face lectures, which students then watch sequentially to complete their studies. Because these videos typically correspond to chapters in the textbook, each video is relatively long and covers a wide range of key points, making it suitable for students with no prior knowledge or ample time for systematic learning.

[0003] However, for students with a certain foundation and limited study time, they don't need to learn every knowledge point one by one; they only need to strengthen their study of the knowledge points they are not proficient in. However, existing course videos cover a wide range of knowledge points, take a long time to learn, and use a single learning method. They cannot support students in quickly and accurately identifying the knowledge points they are not proficient in and conducting targeted learning, thus failing to meet their personalized learning needs. This not only increases students' learning burden and reduces their learning enthusiasm, but also seriously affects their learning efficiency and effectiveness. Summary of the Invention

[0004] The present invention provides a method and system for recommending knowledge points based on course videos, which can solve the problem that existing course video learning methods are difficult to meet the personalized learning needs of students.

[0005] In one aspect, the present invention provides a method for recommending knowledge points based on course videos, the method comprising:

[0006] S1. Determine the association between multiple knowledge points, and based on the multiple knowledge points, use optical character recognition methods and large language models to process the course video to generate a learning video for each knowledge point;

[0007] S2. Use the large language model to determine the knowledge points in each test question in the question bank and determine the learning correspondence between the knowledge points, learning videos, and test questions;

[0008] S3. Construct a knowledge graph based on the association relationship and the learning correspondence relationship, and push learning videos and test questions that match the specific needs to the knowledge demander based on the specific needs of the knowledge demander and the knowledge graph.

[0009] Optionally, in S1, based on multiple knowledge points, the course video is processed using an optical character recognition method and a large language model to generate a learning video for each knowledge point, specifically:

[0010] Based on multiple knowledge points, the optical character recognition method is used to perform text recognition on the image information in the course video, and the course video is cut according to the text recognition results to generate video clips for each knowledge point;

[0011] The large language model is used to perform text processing on the audio information of each video clip, and text information is added to each video clip based on the text processing results to generate a learning video for each knowledge point.

[0012] Optionally, the text recognition of the image information in the course video is performed using an optical character recognition method based on multiple knowledge points, specifically:

[0013] Decompose the course video into multiple image frames;

[0014] Based on multiple knowledge points, optical character recognition method is used to perform text recognition on the image information in each image frame.

[0015] Optionally, after generating the video clip for each knowledge point, the method further includes:

[0016] Based on multiple knowledge points, the course video is segmented using a large language model to generate validation video clips for each knowledge point;

[0017] Determine the similarity between the video clip of each knowledge point and the verification video clip, and adjust the cutting position of the corresponding video clip in the course video when the similarity is lower than a preset threshold, and generate the video clip of the corresponding knowledge point again.

[0018] Optionally, the text information includes one or more of video subtitles, video summaries, and video tags.

[0019] Optionally, before generating a learning video for each knowledge point, the method further includes:

[0020] A portrait recognition method is used to identify an image frame that meets preset conditions from each video clip, and the image frame is set as the cover of the corresponding video clip.

[0021] Optionally, the large language model is used in S2 to determine the knowledge points in each question in the question bank, specifically:

[0022] Use a large language model to analyze each test question in the question bank, and determine the knowledge points in each test question based on the analysis results.

[0023] Optionally, after constructing the knowledge graph according to the association relationship and the learning relationship, the method further includes:

[0024] Pushing multiple test questions to the knowledge demander based on the knowledge demander's characteristic information, and obtaining the knowledge demander's answers to the multiple test questions;

[0025] The specific needs of the knowledge demander are determined based on the answer situation.

[0026] Optionally, determining the association relationship between multiple knowledge points in S1 is specifically:

[0027] The association relationship between the multiple knowledge points is determined based on the hierarchical relationship, causal relationship or parallel relationship between each two knowledge points in the multiple knowledge points.

[0028] On the other hand, the present invention further provides a system based on any of the above-mentioned knowledge point learning recommendation methods based on course videos, the system comprising:

[0029] The video processing module is used to determine the association between multiple knowledge points and process the course video based on the multiple knowledge points using optical character recognition methods and large language models to generate a learning video for each knowledge point;

[0030] The learning correspondence module is used to use the large language model to determine the knowledge points in each test question in the question bank and determine the learning correspondence between the knowledge points, learning videos, and test questions;

[0031] The learning push module is used to construct a knowledge graph based on the association relationship and the learning correspondence relationship, and push learning videos and test questions that match the specific needs to the knowledge demander based on the specific needs of the knowledge demander and the knowledge graph.

[0032] The beneficial effects that the present invention can produce include:

[0033] The present invention is based on multiple knowledge points, uses optical character recognition methods and large language models to process course videos, and generates learning videos for each knowledge point; then, based on the learning correspondence between knowledge points, learning videos, and test questions, as well as the correlation between multiple knowledge points, a knowledge graph is constructed; finally, based on the knowledge graph and the specific needs of the knowledge demander, learning videos and test questions that match their specific needs are pushed to the knowledge demander. This can help the knowledge demander quickly and accurately obtain the required learning content, allowing the knowledge demander to learn in a fragmented, precise, and flexible manner, thereby meeting the knowledge demander's personalized learning needs, helping to reduce the students' learning burden, increase their learning enthusiasm, and achieve the purpose of improving learning efficiency and learning results. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flowchart of a method for recommending knowledge points based on course videos provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The present invention is described in detail below with reference to the embodiments, but the present invention is not limited to these embodiments.

[0036] The embodiment of the present invention provides a method for recommending knowledge points based on course videos. Figure 1 As shown, the method includes:

[0037] S1. Determine the relationships between multiple knowledge points and, based on these knowledge points, use optical character recognition (OCR) and a large language model to process the course video and generate a learning video for each knowledge point.

[0038] S2. Use the large language model to determine the knowledge points in each test question in the question bank and determine the learning correspondence between the knowledge points, learning videos, and test questions;

[0039] S3. Build a knowledge graph based on association relationships and learning correspondences, and push learning videos and test questions that match the specific needs of knowledge demanders based on their specific needs and knowledge graph.

[0040] Specifically, the learning correspondence means that for each knowledge point, the knowledge point should be associated with its corresponding learning video and corresponding test questions to form a correspondence between knowledge point - learning video - test questions, thereby obtaining segmented learning resources for each knowledge point.

[0041] In this embodiment, the association relationship between multiple knowledge points is determined in S1, which may be:

[0042] The association relationship between the multiple knowledge points is determined based on the hierarchical relationship, causal relationship or parallel relationship between each two knowledge points in the multiple knowledge points.

[0043] In this embodiment, S1 processes the course video based on multiple knowledge points using the OCR method and a large language model to generate a learning video for each knowledge point, which may be:

[0044] Based on multiple knowledge points, the OCR method is used to perform text recognition on the image information in the course video, and the course video is cut according to the text recognition results to generate video clips for each knowledge point.

[0045] Furthermore, based on multiple knowledge points, the OCR method is used to perform text recognition on the image information in the course video, which can be specifically:

[0046] Decompose the course video into multiple image frames;

[0047] Based on multiple knowledge points, the OCR method is used to perform text recognition on the image information in each image frame.

[0048] Specifically, this embodiment uses tools such as FFmpeg to decompose the course video into a series of image frames.

[0049] Specifically, the image information may include PPT text and other text information in the course video. This embodiment uses the OCR method to extract and recognize text information from the image information.

[0050] Specifically, this embodiment cuts the course video according to the text recognition results to generate video clips for each knowledge point.

[0051] The large language model is used to perform text processing on the audio information of each video clip, and text information is added to each video clip based on the text processing results to generate a learning video for each knowledge point.

[0052] Specifically, the above text information may include one or more of video subtitles, video summary, and video tags.

[0053] Specifically, text processing includes converting audio information into text and performing in-depth analysis of the text.

[0054] Specifically, this embodiment uses Automatic Speech Recognition (ASR) to convert the audio information of each video clip into text. Natural Language Processing (NLP) methods within a large language model are then used to perform in-depth analysis of this text, including word segmentation, named entity recognition (NER), and syntactic analysis. Based on the results of this in-depth analysis, this embodiment then adds video captions and a video summary to each video clip. Furthermore, video tags are added to each video clip, such as chapter name, key concepts covered, and applicable exam scope, to enhance the video clip's searchability.

[0055] In this embodiment, before generating a learning video for each knowledge point, the method may further include:

[0056] A portrait recognition method is used to identify an image frame that meets preset conditions from each video clip, and the image frame is set as the cover of the corresponding video clip.

[0057] Specifically, the preset conditions may include requirements for the teacher's movements and expressions, such as requiring that the teacher's image in the image frame should keep the mouth open, eyes open, and the face clear.

[0058] After the video clips are processed as described above, a learning video for each knowledge point can be obtained.

[0059] Furthermore, after generating the video clip of each knowledge point, the method may further include:

[0060] Based on multiple knowledge points, the course video is segmented using a large language model to generate validation video clips for each knowledge point.

[0061] Determine the similarity between the video clip of each knowledge point and the validation video clip, and adjust the cutting position of the corresponding video clip in the course video when the similarity is lower than a preset threshold, and generate the video clip of the corresponding knowledge point again.

[0062] Specifically, the large language model can be a Hunyuan large model, a Tingwu large model, etc.

[0063] Specifically, in the course video of this embodiment, the teacher teaches the course in combination with PPT, and each page of the PPT is marked with the knowledge points of that page. This embodiment can use the Hunyuan large model to identify the knowledge points of each page of the PPT, as well as the two image frames corresponding to the start and end of each page of the PPT, output the start and end time of each page of the PPT, and then cut the course video according to the start and end time of each page of the PPT to obtain the verification video clips of each knowledge point.

[0064] This embodiment uses two methods (a method combining OCR with a large language model, and a method using only a large model) to cut the course video, and then compares the cutting effects of the two methods. The cutting position is adjusted according to the comparison results, which can achieve mutual verification of the two methods and effectively avoid the problem of inaccurate cutting caused by large semantic understanding deviations in the field of traditional Chinese medicine when using a single method. Therefore, each video segment focuses on one knowledge point, ensuring the integrity and accuracy of the video segment.

[0065] In this embodiment, the large language model is used in S2 to determine the knowledge points in each question in the question bank, which can be specifically:

[0066] Use a large language model to analyze each test question in the question bank, and determine the knowledge points in each test question based on the analysis results.

[0067] Specifically, the test questions include real exam questions, AI-generated practice questions, etc.

[0068] Specifically, the large language model can perform semantic analysis based on the question stem, options, and answers, thereby identifying the key knowledge points within the question. For example, if the question involves "classification of mineral-based traditional Chinese medicines," the large language model can identify this as an important key knowledge point within the classification of mineral-based traditional Chinese medicines.

[0069] Specifically, after the large language model determines the knowledge points in each test question, this embodiment can verify the knowledge points of each test question to ensure accurate correspondence between the knowledge points and the test questions.

[0070] Furthermore, this embodiment can determine the learning correspondence between knowledge points, learning videos, and test questions based on the learning videos and test questions corresponding to the knowledge points. Simultaneously, by combining the associations between multiple knowledge points, the relationships between knowledge points, learning videos, and test questions can be linked to construct a lightweight knowledge graph. The nodes in the knowledge graph represent different knowledge points, and the lines between the nodes represent the relationships between the knowledge points (such as hierarchical relationships, causal relationships, parallel relationships, etc.). Furthermore, this embodiment can also accurately match knowledge points with test points in the syllabus.

[0071] In this embodiment, after constructing the knowledge graph based on the association relationship and the learning relationship, the method may further include:

[0072] Obtain characteristic information of knowledge seekers;

[0073] Push multiple test questions to the knowledge demander based on the knowledge demander's characteristic information, and obtain the knowledge demander's answer status to the multiple test questions;

[0074] Determine the specific needs of knowledge seekers based on their responses.

[0075] Then, this embodiment pushes learning videos and test questions that match the specific needs to the knowledge demander based on the specific needs of the knowledge demander and the knowledge graph.

[0076] Specifically, this embodiment can design a questionnaire on the product side to obtain characteristic information such as the education level, age, whether or not the knowledge seeker, and whether the job is related to pharmacy of the knowledge seeker, so as to construct the student's initial learning profile, including knowledge level, learning goals, time arrangements, etc., so as to tailor differentiated learning plans for the students.

[0077] Specifically, this embodiment can utilize the Tingwu model and the Hunyuan model to develop personalized learning plans for students. For example, based on the student's educational background, age, medical examination experience, work background, and other characteristic information, this embodiment can reduce the review of basic knowledge points for students with a pharmaceutical background and directly move on to learning more difficult knowledge points; while for students without a pharmaceutical background, this embodiment can provide more learning videos and test questions on basic knowledge points.

[0078] Specifically, this embodiment can also use the knowledge graph to randomly generate 10-20 knowledge points and push them to students. The large language model can mark the knowledge points that need to be focused on based on the students' answers and feedback. This not only allows the students' learning plans to be dynamically adjusted, but also allows for a cold start for new students. For example, if a student's answer to a certain knowledge point is not ideal, this embodiment can increase the push of learning videos and test questions for that knowledge point, extending the student's learning time for that knowledge point; if a student has already mastered a certain knowledge point, this embodiment can push learning videos and test questions for that knowledge point, shortening the student's learning time for that knowledge point, and push more learning videos and test questions for other knowledge points that have not been mastered to the student.

[0079] Specifically, this embodiment can also collect students' negative feedback on the learning correspondence between knowledge points, learning videos, and test questions, and then adjust the cutting position of the video clips and regenerate the learning video according to the students' negative feedback to adjust the learning correspondence between the three.

[0080] Another embodiment of the present invention provides a system based on any of the above-mentioned knowledge point learning recommendation methods based on course videos, and the system may include:

[0081] The video processing module is used to determine the association between multiple knowledge points and process the course video based on multiple knowledge points using OCR methods and large language models to generate learning videos for each knowledge point;

[0082] The learning correspondence module is used to use the large language model to determine the knowledge points in each test question in the question bank and determine the learning correspondence between the knowledge points, learning videos, and test questions;

[0083] The learning push module is used to build a knowledge graph based on association relationships and learning correspondence relationships, and push learning videos and test questions that match the specific needs of knowledge demanders based on their specific needs and knowledge graphs.

[0084] The present invention is based on multiple knowledge points and uses OCR methods and large language models to process course videos to generate learning videos for each knowledge point; then, a knowledge graph is constructed based on the learning correspondence between knowledge points, learning videos, and test questions, as well as the correlation between multiple knowledge points; finally, based on the knowledge graph and the specific needs of the knowledge demander, learning videos and test questions that match their specific needs are pushed to the knowledge demander. This can help the knowledge demander quickly and accurately obtain the required learning content, allowing the knowledge demander to learn in a fragmented, precise, and flexible manner, thereby meeting the knowledge demander's personalized learning needs, helping to reduce the students' learning burden, increase their learning enthusiasm, and achieve the purpose of improving learning efficiency and learning results.

[0085] The present invention changes the traditional online course sequential learning and all-in-one learning methods, as well as the situation in which traditional online subjects are questions, classes are lessons, and test questions and course videos are unrelated, which is not conducive to efficient learning for students. The present invention uses knowledge graphs to conduct closed-loop learning recommendations for students in the form of test-learning-test-mastery, thereby achieving more accurate knowledge point learning recommendations.

[0086] The present invention matches the learning videos of knowledge points with the test questions in the learning path. It can push a small number of test questions to students to evaluate their mastery of multiple knowledge points in the field of medicine and nursing, thereby providing students with more systematic and accurate learning recommendations, solving the cold start problem of students.

[0087] The present invention adopts a method that combines the OCR method and the large language model to cut the course video. At the same time, the large language model is also used alone to cut the course video. The two video cutting results are then verified. This can effectively avoid the problem of low accuracy of a single cutting method and improve the matching degree between video clips and knowledge points.

[0088] The above descriptions are merely a few embodiments of the present application and do not constitute any form of limitation to the present application. Although the present application discloses the preferred embodiments as above, they are not intended to limit the present application. Any technical personnel familiar with the present profession, without departing from the scope of the technical solution of the present application, using the technical content disclosed above to make slight changes or modifications are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A knowledge point learning recommendation method based on course videos, characterized in that: The method comprises: S1. Determine the association relationship between multiple knowledge points, decompose the course video into multiple image frames, perform text recognition on the image information in each image frame using an optical character recognition method based on the multiple knowledge points, and cut the course video according to the text recognition results to generate a video segment for each knowledge point; use a large language model to perform text processing on the audio information of each video segment, and add text information to each video segment based on the text processing results to generate a learning video for each knowledge point; the text information includes one or more of video subtitles, video summaries, and video tags; S2. Use the large language model to determine the knowledge points in each test question in the question bank and determine the learning correspondence between the knowledge points, learning videos, and test questions; S3. Constructing a knowledge graph based on the association relationship and the learning correspondence relationship, pushing multiple test questions to the knowledge demander based on the knowledge demander's characteristic information, and obtaining the knowledge demander's answer status to the multiple test questions; determining the specific needs of the knowledge demander based on the answer status, and pushing learning videos and test questions that match the specific needs to the knowledge demander based on the specific needs and the knowledge graph; Use a large language model to perform text processing on the audio information of each video clip, including: The audio information of each video clip is converted into text using an automatic speech recognition method, and then the text is deeply analyzed using a natural language processing method in a large language model to generate the text processing result; the deep analysis includes word segmentation recognition, named entity recognition, and syntactic analysis; After generating the video clip for each knowledge point, the method further includes: Mark the knowledge points on each PPT page of the course video, use the large language model to identify the knowledge points on each PPT page and the two image frames corresponding to the start and end of each PPT page, and output the start and end time of each PPT page. Cut the course video according to the start and end time of the PPT corresponding to each knowledge point to obtain the validation video clip for each knowledge point; Determine the similarity between the video clip of each knowledge point and the verification video clip, and adjust the cutting position of the corresponding video clip in the course video when the similarity is lower than a preset threshold, and generate the video clip of the corresponding knowledge point again.

2. The method according to claim 1, characterized in that Before generating a learning video for each knowledge point, the method further includes: A portrait recognition method is used to identify an image frame that meets preset conditions from each video clip, and the image frame is set as the cover of the corresponding video clip.

3. The method according to claim 1, characterized in that In S2, the large language model is used to determine the knowledge points in each question in the question bank, specifically: Use a large language model to analyze each test question in the question bank, and determine the knowledge points in each test question based on the analysis results.

4. The method according to claim 1, wherein The association relationship between multiple knowledge points is determined in S1, specifically: The association relationship between the multiple knowledge points is determined based on the hierarchical relationship, causal relationship or parallel relationship between each two knowledge points in the multiple knowledge points.

5. A system based on the knowledge point learning recommendation method based on course videos according to any one of claims 1 to 4, characterized in that: The system comprises: The video processing module is used to determine the association between multiple knowledge points and process the course video based on the multiple knowledge points using optical character recognition methods and large language models to generate a learning video for each knowledge point; The learning correspondence module is used to use the large language model to determine the knowledge points in each test question in the question bank and determine the learning correspondence between the knowledge points, learning videos, and test questions; The learning push module is used to construct a knowledge graph based on the association relationship and the learning correspondence relationship, and push learning videos and test questions that match the specific needs to the knowledge demander based on the specific needs of the knowledge demander and the knowledge graph.

Images