Learning content generation method and related device

Abstract

The embodiment of the invention discloses a learning content generation method and a related device. The method comprises the following steps: acquiring target knowledge content; determining a knowledge outline based on the target knowledge content, wherein the knowledge outline comprises a plurality of target knowledge points; retrieving a pre-constructed learning resource library to obtain a target resource fragment corresponding to each target knowledge point, wherein the learning resource library comprises a plurality of resource fragments; and inputting the knowledge outline and the target resource fragment into a pre-trained large language model to obtain learning content output by the large language model based on the knowledge outline, the learning content including the target resource fragment. By combining the ability of the large language model, the learner can learn the learning resource fragments related to the interested knowledge points in the learning content at one time, so that the learner can conveniently learn in a targeted manner, and the learning efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and related apparatus for generating learning content. Background Technology

[0002] With the development of computer and internet technologies, learners can easily access a wealth of online learning resources. As an online platform that provides learning resources and services, a learning platform can utilize internet technology and multimedia tools to integrate educational content, learning tools, and interactive functions, providing learners with a flexible and convenient learning space.

[0003] Traditional learning platforms typically present learning content at the course level, with each course designed and produced independently to cover a specific field or topic of knowledge. Therefore, some overlap in knowledge and learning content is inevitable between different courses. When learners want to learn a particular topic, they may need to switch between multiple courses and select chapters of interest, resulting in low learning efficiency. Summary of the Invention

[0004] This application provides a learning content generation method and related apparatus, enabling learners to learn learning resource fragments related to knowledge points of interest all at once, facilitating targeted learning and improving learning efficiency.

[0005] The first aspect of this application provides a method for generating learning content, which can be applied in a learning platform, including:

[0006] Acquire target knowledge content; target knowledge content refers to the knowledge areas that learners choose based on their personal interests or learning needs and hope to understand and master in depth.

[0007] A knowledge outline is determined based on the target knowledge content. The knowledge outline includes multiple target knowledge points and typically consists of one or more levels, progressively detailed. The knowledge outline can be pre-stored in the learning platform or generated based on the target knowledge content.

[0008] The learning resource repository retrieves target resource fragments corresponding to each target knowledge point from a pre-built learning resource library. This library contains multiple resource fragments; the learning resources can be various types of media data fragments, including course videos, teaching documents, presentations, exercises, mock exams, case studies, etc., while the resource fragments can be part or all of the content of these learning resources. Based on the determined knowledge points in the knowledge outline, searches can be performed in the learning resource library to obtain the target resource fragments corresponding to each target knowledge point.

[0009] A knowledge outline and target resource fragments are input into a pre-trained large language model to obtain learning content output by the large language model based on the knowledge outline. This learning content includes the target resource fragments. The large language model can handle various natural language tasks. After inputting the knowledge outline and target resource fragments, the model's understanding capabilities allow it to integrate the target resource fragments based on the knowledge outline, resulting in learning content. This learning content displays various target knowledge points through the knowledge outline, and each target knowledge point includes a corresponding target resource fragment.

[0010] The learning content generation method provided in this application first obtains the target knowledge content specified by the user, then determines a knowledge outline based on the target knowledge content, and further determines multiple target knowledge points corresponding to the target knowledge content based on the knowledge outline. Based on the target knowledge points, corresponding target resource fragments are retrieved from a pre-built learning resource library, and the knowledge outline and target resource fragments are combined using the natural language processing capabilities of a large language model to output the learning content. This learning content integrates multiple target resource fragments related to the target knowledge content. These target resource fragments may come from different courses. Users or learners can learn or understand these knowledge-related fragments at once through this learning content, without having to switch between multiple courses, thus improving learning efficiency.

[0011] In one possible implementation, before retrieving the target resource fragment corresponding to each target knowledge point from a pre-built learning resource library, the following steps are also included:

[0012] Access learning resources;

[0013] The learning resource content is split into multiple resource fragments based on preset rules;

[0014] A learning resource library is built based on multiple resource fragments.

[0015] Understandably, learning resources can be obtained directly from the learning platform, or they can be other learning resources uploaded by users and approved. Depending on the type of learning resource, it can be split according to pre-defined rules to obtain multiple resource fragments. These fragments are then imported into a database or cloud server to complete the construction of the learning resource library.

[0016] In one possible implementation, a learning resource library is constructed based on multiple resource fragments, including:

[0017] The resource fragments are processed to obtain retrieval information, which includes at least one of feature vectors, descriptive text, and keywords.

[0018] A learning resource library is constructed based on resource fragments carrying retrieval information;

[0019] The target resource fragments corresponding to each target knowledge point are retrieved from the pre-built learning resource library, including:

[0020] Based on the retrieved information, the target resource fragments corresponding to each target knowledge point are retrieved from the learning resource library.

[0021] Understandably, after obtaining multiple resource fragments, to facilitate accurate retrieval of target resource fragments related to the target knowledge point, these fragmented fragments can be processed to obtain retrieval information. This retrieval information is then incorporated as part of the resource fragments and used to construct a learning resource library. Specifically, when retrieving based on the target knowledge point, the target resource fragment can be obtained by matching the retrieval information of each resource fragment in the learning resource library.

[0022] In one possible implementation, the learning resource content includes document data, and the resource fragment includes a first text fragment and a first image;

[0023] The learning resource content is split into multiple resource fragments based on preset rules, including:

[0024] The text content in the document data is split based on the first rule to obtain the first text fragment. The first rule includes at least one of the following: directory level, specified word count, and semantic understanding.

[0025] Extract the image information from the document data to obtain the first image.

[0026] Understandably, the document data in learning resources includes document-based data such as textbooks, courseware, academic papers, and research reports. Its main characteristic is that it consists primarily of text, and may include illustrations. For the text portion of the document data, it can be split into multiple text fragments based on rules such as directory hierarchy, specified word count, and semantic understanding, including the first text fragment. Simultaneously or subsequently, traversing the entire document allows for the extraction of image information, including the first image.

[0027] In a further implementation of this possible method, after extracting the image information from the document data to obtain the first image, the method further includes:

[0028] Based on the position of the first image in the document data, the first associated text of the first image is extracted from the text content. The first associated text is used to generate retrieval information for the first image.

[0029] Understandably, after extracting image information, the image's position within the document data can be recorded, and surrounding text can be obtained based on that position to generate potential titles or descriptions. This text data can then be used as associated text for the image information, generating retrieval information for that image.

[0030] In one possible implementation, the learning resource content includes presentation data, and the resource fragments include a second text fragment and a second presentation medium;

[0031] The learning resource content is split into multiple resource fragments based on preset rules, including:

[0032] Layout element identification is performed on the presentation data to obtain title text and description text;

[0033] The description text is split based on the title text to obtain the second text fragment;

[0034] Extract media information from the presentation data to obtain the second presentation media.

[0035] Understandably, presentation data integrates text with various media formats such as images, audio, and video. For presentation data, layout element identification can be performed first to obtain the title and description text, typically achieved by analyzing the page's visual hierarchy, font size, and color contrast. After obtaining the title and description text, the description text can be broken down into multiple secondary text segments related to the title, based on the content or structure of the title text. Media information in presentation data includes multimedia elements such as images, audio, and video. The entire presentation can be traversed, using image processing, file parsing, or layout element identification techniques to identify the image, audio, and video files on each page. For media files embedded in the presentation, specific tools or libraries may be required for extraction.

[0036] In a further implementation of this possible method, after extracting the image information from the presentation data to obtain the second presentation media, the method further includes:

[0037] Based on the results of layout element identification, the second associated text of the second presentation media is extracted, and the second associated text is used to generate retrieval information for the second presentation media.

[0038] Understandably, after extracting media information, it's possible to extract related text based on the previous layout element identification of the presentation. This related text can be text data on the same page as the media information, or it can include text data from other pages belonging to the same title text. This text data can be used as related text for the media information to generate retrieval information for that media information.

[0039] In one possible implementation, the learning resource content includes video data, and the resource segments are video clips;

[0040] The learning resource content is split into multiple resource fragments based on preset rules, including:

[0041] Perform text recognition on video data frames;

[0042] Perform speech recognition on the audio of video data;

[0043] Based on the results of text recognition and speech recognition, and combined with semantic understanding technology, the video data is split into multiple video segments.

[0044] Understandably, this video data can be educational videos, science videos, live stream recordings, etc. Typically, video data containing learning content will describe key information points with text, and the knowledge points presented in the video will be displayed in a logical order. Therefore, the video data can be broken down into video frames and audio. For the video frames, text recognition technology can be used to extract the text data, and for the audio, speech recognition technology can be used to convert the audio into text. The resulting text content can then be combined with semantic understanding technology to further segment the video data into multiple video clips.

[0045] A further implementation of this possible method, after segmenting the video data into multiple video segments based on the results of character recognition and speech recognition, combined with semantic understanding technology, also includes:

[0046] Based on the results of text recognition and speech recognition, the third associated text of the video clip is extracted, and the third associated text is used to generate retrieval information for the video clip.

[0047] Understandably, after obtaining multiple video clips, it is possible to extract related text based on the previous text recognition and speech recognition results. This related text can be the text content obtained from the text recognition and speech recognition results, and this text content can be used to generate retrieval information for the corresponding video clips.

[0048] In one possible implementation, a knowledge outline is determined based on the target knowledge content, including:

[0049] The target knowledge content is input into the large language model to obtain the knowledge outline generated by the large language model.

[0050] Understandably, with a pre-built learning resource library, both the target knowledge content and the learning resource library can be input into the large language model. Since the resource fragments are text content or have corresponding descriptive text, the understanding and summarizing capabilities of the large language model are applied to match the knowledge points in the learning resource library with the target knowledge content, thereby constructing a knowledge outline associated with the resource fragments in the learning resource library. This avoids the problem that the generated knowledge outline may contain knowledge points not covered in the learning resource library, which could lead to the inability to output learning content or the output of learning content without a basis.

[0051] In one possible implementation, after inputting the knowledge outline and target resource fragments into a pre-trained large language model to obtain the learned content output by the large language model based on the knowledge outline, the method further includes:

[0052] The learning content is input into the large language model to obtain test questions and / or code experiments for each target knowledge point based on the learning content.

[0053] Understandably, after receiving the learning content, it is input into a large language model. Upon receiving the input, the model utilizes its internal deep learning algorithms and extensive training data to analyze the content, identifying key knowledge points, concepts, and terminology. Based on the user's instructions, it generates corresponding test questions or code experiments. The generated test questions may include multiple-choice, fill-in-the-blank, and short-answer questions, designed to assess the learner's understanding and mastery of the knowledge points. The generated code experiments may include example programs, function implementations, and data processing scripts, used to help learners deepen their understanding through practice.

[0054] A second aspect of this application provides a learning content generation apparatus, comprising:

[0055] The knowledge acquisition module is used to acquire target knowledge content;

[0056] The outline acquisition module is also used to determine a knowledge outline based on the target knowledge content, which includes multiple target knowledge points;

[0057] The retrieval module is used to retrieve the target resource fragments corresponding to each target knowledge point from the pre-built learning resource library;

[0058] The output module is used to input the knowledge outline and target resource fragments into the pre-trained large language model to obtain the learning content output by the large language model based on the knowledge outline, which includes the target resource fragments.

[0059] One possible implementation also includes:

[0060] The module is used to acquire learning resource content; split the learning resource content according to preset rules to obtain multiple resource fragments; and build a learning resource library based on multiple resource fragments.

[0061] In one possible implementation,

[0062] The module is specifically used to process resource fragments to obtain retrieval information, which includes at least one of feature vectors, descriptive text, and keywords; and to build a learning resource library based on the resource fragments carrying the retrieval information.

[0063] The retrieval module is specifically used to retrieve the target resource fragments corresponding to each target knowledge point from the learning resource library based on the retrieval information.

[0064] In one possible implementation, the learning resource content includes document data, and the resource fragment includes a first text fragment and a first image;

[0065] The module is specifically used to split the text content in the document data according to a first rule to obtain a first text fragment. The first rule includes at least one of the following: directory level, specified word count, and semantic understanding; and to extract image information from the document data to obtain a first image.

[0066] Furthermore, the building module is also used to extract the first associated text of the first image from the text content based on the position of the first image in the document data. The first associated text is used to generate retrieval information for the first image.

[0067] In one possible implementation, the learning resource content includes presentation data, and the resource fragments include a second text fragment and a second presentation medium;

[0068] The module is specifically used to identify layout elements in presentation data to obtain title text and description text.

[0069] The description text is split based on the title text to obtain the second text fragment;

[0070] Extract media information from the presentation data to obtain the second presentation media.

[0071] Furthermore, the construction module is also used to extract the second associated text of the second presentation media based on the results of layout element recognition. The second associated text is used to generate retrieval information for the second presentation media.

[0072] In one possible implementation, the learning resource content includes video data, and the resource segments are video clips;

[0073] The module is specifically used for text recognition of video data frames.

[0074] Perform speech recognition on the audio of video data;

[0075] Based on the results of text recognition and speech recognition, and combined with semantic understanding technology, the video data is split into multiple video segments.

[0076] Furthermore, the module is also used to extract third-party related text from video clips based on the results of text recognition and speech recognition. This third-party related text is used to generate retrieval information for the video clips.

[0077] In one possible implementation, with the learning resource library constructed, the outline acquisition module is specifically used to input the target knowledge content into the large language model to obtain the knowledge outline generated by the large language model.

[0078] In one possible implementation, the output module is also used to input the learning content into the large language model to obtain test questions and / or code experiments for each target knowledge point based on the learning content.

[0079] A third aspect of this application provides a computing device, including a processor and a memory;

[0080] The processor is used to execute computer programs or computer instructions in memory to perform the methods described in the first aspect above.

[0081] The fourth aspect of this application provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the implementation as described in the first aspect.

[0082] The fifth aspect of this application provides a computer-readable storage medium including computer instructions that, when executed on a computer, cause the computer to perform the implementation as described in the first aspect.

[0083] The beneficial effects of the technical solutions provided in the second to fifth aspects above can be referred to the beneficial effects of the technical solutions in the first aspect, and will not be repeated here. Attached Figure Description

[0084] Figure 1This is a flowchart of the learning content generation method provided in the embodiments of this application;

[0085] Figure 2 Example diagram of a learning outline provided for an embodiment of this application;

[0086] Figure 3 A schematic diagram of a presentation page provided for an embodiment of this application;

[0087] Figure 4 This is a diagram illustrating the overall technical architecture of the learning content generation method provided in the embodiments of this application.

[0088] Figures 5a to 5c A schematic diagram showing the breakdown of various learning resource contents provided in the embodiments of this application;

[0089] Figure 6 This is a flowchart illustrating the construction process of the learning resource library provided in the embodiments of this application;

[0090] Figure 7 A schematic diagram of the product form interaction interface of the learning content generation method provided in the embodiments of this application;

[0091] Figure 8 This is a schematic diagram of the learning content generation device provided in the embodiments of this application;

[0092] Figure 9 This is another schematic diagram of the learning content generation device provided in the embodiments of this application;

[0093] Figure 10 This is a schematic diagram of the structure of the computing device in the embodiments of this application. Detailed Implementation

[0094] This application provides a learning content generation method and related apparatus. First, a knowledge outline is determined based on the knowledge content that learners are interested in. Then, each knowledge point in the knowledge outline is retrieved from a learning resource library to obtain learning resource fragments corresponding to each knowledge point. Finally, based on the capabilities of a large language model, learning content is output according to the knowledge outline, and the learning resource fragments corresponding to each knowledge point are embedded in the learning content. This allows learners to learn learning resource fragments related to the knowledge points of interest all at once, facilitating targeted learning and improving learning efficiency.

[0095] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application are described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Those skilled in the art will understand that with the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0096] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in this application does not imply that the steps in the method flow must be performed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical purpose, as long as the same or similar technical effect is achieved. The division of units in this application is a logical division. In practical applications, there may be other division methods. For example, multiple units may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the shown or discussed mutual coupling, direct coupling, or communication connection may be through some interface, and the indirect coupling or communication connection between units may be electrical or other similar forms, none of which are limited in this application. Furthermore, the units or sub-units described as separate components may or may not be physically separated, may or may not be physical units, or may be distributed among multiple circuit units. Some or all of the units can be selected to achieve the purpose of the solution in this application according to actual needs.

[0097] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, pre-trained model technology, operating / interactive systems, and mechatronics. Pre-trained models (PTMs), also known as large models or foundational models, refer to deep neural networks (DNNs) with a large number of parameters. These DNNs are trained on massive amounts of unlabeled data. Leveraging the function approximation capabilities of large-parameter DNNs, PTMs extract common features from the data. Through fine-tuning, parameter efficient fine-tuning (PEFT), and prompt-tuning techniques, they are suitable for downstream tasks. Therefore, pre-trained models can achieve ideal results in small-shot or zero-shot scenarios. PTMs can be categorized according to the data modality they process, such as language models (ELMO, BERT, GPT), visual models (swin-transformer, ViT, V-MOE), speech models (VALL-E), and multimodal models (ViBERT, CLIP, Flamingo, Gato). Multimodal models refer to models that establish feature representations for two or more data modalities. Pre-trained models are important tools for outputting AI-generated content (AIGC) and can also serve as a general interface connecting multiple specific task models. Artificial intelligence software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning. Natural Language Processing (NLP) is a crucial area within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP deals with natural language, the language people use in daily life, and is closely related to linguistics; it also involves computer science and mathematics. Pre-trained models, a key technology for model training in artificial intelligence, evolved from large language models in NLP.After fine-tuning, large language models can be widely applied to downstream tasks. Natural language processing (NLP) technologies typically include text processing, semantic understanding, machine translation, robot question answering, and knowledge graphs. With the research and advancement of artificial intelligence (AI) technology, AI is being researched and applied in multiple fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, digital twins, virtual humans, robots, AI-generated content (AIGC), conversational interaction, smart healthcare, smart customer service, and game AI. It is believed that with technological development, AI will be applied in more fields and play an increasingly important role. The solutions provided in this application's embodiments involve AI technologies such as language processing. The following is a brief introduction to the technical background involved in this application:

[0098] With the development of computer and internet technologies, learners can easily access a wealth of online learning resources. As an online platform that provides learning resources and services, a learning platform can utilize internet technology and multimedia tools to integrate educational content, learning tools, and interactive functions, providing learners with a flexible and convenient learning space.

[0099] Traditional learning platforms typically present learning content at the course level, with each course designed and produced independently to cover a specific field or topic of knowledge. Therefore, different courses inevitably share some common knowledge points and related learning content. When learners want to study a particular knowledge point, they may need to switch between multiple courses, independently selecting chapters of interest, leading to low learning efficiency.

[0100] With the continuous advancement of artificial intelligence and natural language processing technologies, computers are able to better understand and produce human language, simulate human dialogue when interacting with users, and summarize and organize the information provided based on user needs through large language models, and output a set of logically related content.

[0101] Based on this, this application provides a learning content generation method and related apparatus. By retrieving learning resource fragments related to knowledge points of interest to learners from a learning resource library, and then using the semantic understanding capabilities of a large language model, these learning resource fragments are summarized and generalized to generate learning content corresponding to the knowledge point. This allows learners to learn all the content related to the knowledge point at once without having to switch between multiple courses, thereby improving learning efficiency.

[0102] The learning content generation method provided in the embodiments of this application will be described below. Please refer to [link / reference]. Figure 1 , Figure 1This is a flowchart illustrating the learning content generation method provided in this application embodiment. This method can be specifically applied to a learning platform, which typically includes various learning resources. These resources may include diverse learning materials such as course videos, teaching documents, presentations, exercises, mock exams, and case studies. These resources are created by professional teachers or educational institutions to meet the learning needs of different learners.

[0103] like Figure 1 As shown, the method provided in this application embodiment includes:

[0104] 101. Obtain the target knowledge content.

[0105] Target knowledge content refers to the knowledge areas that learners choose based on their personal interests or learning needs and wish to delve deeper into and master. The learning platform can provide a text input interface, guiding learners to input this knowledge content through text boxes in the interactive interface. The platform can also recommend relevant learning content based on learners' learning behaviors (such as clicks, browsing, course completion, etc.) using intelligent recommendation algorithms, and respond to learners' click actions to retrieve learning content.

[0106] 102. Determine the knowledge outline based on the target knowledge content. The knowledge outline includes multiple target knowledge points.

[0107] Understandably, after acquiring the target knowledge content, a related knowledge outline can be determined based on that target knowledge content. This outline aims to refine the broad knowledge areas that learners wish to master or understand into a series of specific, orderly, and interrelated learning objectives or knowledge points.

[0108] A knowledge outline typically comprises one or more levels, progressively refining each level. For example, at the top level, it might encompass an overview or topic covering the entire knowledge domain; these topics might then be broken down into more specific sub-topics or chapters; finally, each sub-topic or chapter is further refined into a series of target knowledge points—the specific information or skills learners need to master. The number of levels in a knowledge outline can be determined by the target knowledge content. For instance, if the knowledge content is relatively basic, the outline might only have one level, including just a series of knowledge points; when the knowledge content is more comprehensive, the outline might be divided into more levels.

[0109] In one possible implementation, the knowledge outline can be pre-stored in the learning platform. That is, the learning platform may have pre-built knowledge outlines for various courses and learning content based on different knowledge topics. After identifying the target knowledge content, a search is performed based on that target knowledge content to obtain knowledge outlines corresponding to the same or similar knowledge content. If multiple knowledge outlines are retrieved, they can all be displayed to the learner, allowing them to choose the one that best suits their needs.

[0110] In another possible implementation, the knowledge outline can be generated based on the target knowledge content. Specifically, step 102 can be:

[0111] 1021. Input the target knowledge content into the large language model to obtain the knowledge outline generated by the large language model.

[0112] It is understood that the method provided in this application embodiment can generate the knowledge outline based on the capabilities of a large language model. Taking the learner inputting "deep learning perceptron" as an example, the following prompt word template for generating the knowledge outline is input into the large language model:

[0113] "Please generate a well-structured knowledge outline based on the following knowledge points."

[0114] Key concept: Deep learning perceptron

[0115] Understandably, the output of a large language model is usually text. To facilitate display, it can be processed into an expandable table of contents using appropriate formatting templates. The specific conversion process is a common technique and will not be elaborated here. The resulting knowledge outline is as follows: Figure 2 As shown.

[0116] 103. Retrieve the target resource fragment corresponding to each target knowledge point from the pre-built learning resource library, which includes multiple resource fragments.

[0117] It is understandable that a pre-built learning resource library refers to a database that includes multiple learning resource fragments. Learning resources can be various types of media data fragments, including course videos, teaching documents, presentations, exercises, mock exams, case studies, etc., while resource fragments can be part or all of the content of the above learning resources.

[0118] Based on the defined knowledge outline and the target knowledge points, searches can be performed in the learning resource repository to obtain the corresponding target resource fragments. To facilitate retrieval, resource fragments in the learning resource repository can be categorized by subject, theme, or knowledge point, and the corresponding target resource fragments can be retrieved based on the categorization. Alternatively, metadata such as title, description, and keywords can be added to each resource fragment when constructing the learning resource repository, allowing retrieval of the corresponding target resource fragments based on keywords. Furthermore, database indexing techniques (such as inverted indexes) can be used to improve retrieval efficiency; inverted indexes map words in the content to resource fragments containing those words.

[0119] In one possible implementation, before step 103, the following is also included:

[0120] 1001. Obtain learning resources;

[0121] 1002. The learning resource content is split into multiple resource fragments based on preset rules;

[0122] 1003. Construct a learning resource library based on multiple resource fragments.

[0123] It is understandable that steps 1001 to 1003 constitute the process of pre-building a learning resource library, which can be executed in the offline phase of the learning platform.

[0124] First, obtain learning resources. These resources can be existing resources on the learning platform that can be obtained directly from it, or other learning resources uploaded by users and approved. These resources can come from various channels, such as textbooks, online courses, teaching videos, academic papers, research reports, and experimental guides.

[0125] Then, based on different types of learning resource content, it can be split according to corresponding preset rules to obtain multiple resource fragments. Each resource fragment can include one or more knowledge points. For example, a learning platform may include a Python programming course, which consists of multiple chapter-based course videos. Each course may cover one or more chapters, each chapter may be further subdivided into multiple knowledge points, and each knowledge point may exist in multiple chapters. In a simple and feasible method, the course video can be split into multiple shorter video fragments based on its length. This method ensures that each video fragment covers as few knowledge points as possible, making it easier to match more accurate videos when searching based on knowledge points. In another possible method, the video can be split according to its original thematic structure, treating each knowledge point in the learning resource as an independent resource fragment.

[0126] After breaking down the content of each learning resource, multiple resource fragments are obtained. These fragments are then imported into a database or cloud server to complete the construction of the learning resource library.

[0127] In one possible implementation, step 1003 specifically includes:

[0128] 10031. Process the resource fragments to obtain retrieval information, which includes at least one of feature vectors, descriptive text, and keywords;

[0129] 10032. Construct a learning resource library based on resource fragments carrying retrieval information;

[0130] Step 103 specifically includes:

[0131] Based on the retrieved information, the target resource fragments corresponding to each target knowledge point are retrieved from the learning resource library.

[0132] Understandably, after obtaining multiple resource fragments, to facilitate accurate retrieval of target resource fragments related to the target knowledge point, these fragmented fragments can be processed to obtain retrieval information usable for searching. For example, appropriate metadata annotations, such as titles, descriptions, and keywords, can be added. These metadata annotations can be used to quickly retrieve the required target resource fragments. Furthermore, since the above annotation information is textual, it can be used to extract features from these annotations using natural language processing techniques, converting the text into numerical representations to obtain resource feature vectors. After determining the target knowledge point, a query feature vector for the target knowledge point is constructed, and the pre-similarity between the query feature vector and the resource feature vector is calculated. The resource fragments with the highest similarity are then used as the retrieval results.

[0133] After obtaining the retrieval information for each resource fragment, this retrieval information can be incorporated as part of the resource fragment and used to construct a learning resource library. When performing a retrieval based on a target knowledge point, the target resource fragment can be obtained by matching the retrieval information of each resource fragment in the learning resource library.

[0134] It is understood that different methods of splitting and retrieving information are used for different types of learning resource content. For ease of understanding, the following embodiments of this application describe the methods of splitting document data, presentation data, and video data in learning resource content and the methods of retrieving information.

[0135] In one possible implementation, the learning resource content includes document data, and the resource fragment includes a first text fragment and a first image;

[0136] At this point, step 1002 specifically includes:

[0137] The text content in the document data is split based on the first rule to obtain the first text fragment. The first rule includes at least one of the following: directory level, specified word count, and semantic understanding.

[0138] Extract the image information from the document data to obtain the first image.

[0139] In this embodiment of the application, the document data in the learning resource content includes document-based data such as textbooks, courseware, academic papers, and research reports. Its characteristic is that it is mainly composed of text and may include illustrations.

[0140] For the text portion of the document data, it can be split into multiple text fragments based on rules such as directory hierarchy, specified word count, and semantic understanding, including the first text fragment.

[0141] For example, a learning resource is a textbook on "Machine Learning Fundamentals". The textbook contains multiple chapters, sections and illustrations. The first rule is a combination of table of contents hierarchy, specified word count and semantic understanding.

[0142] First, the textbook is divided into chapters (e.g., "Chapter 1 Overview of Machine Learning") and sections (e.g., "1.1 What is Machine Learning"), resulting in multiple parts, each corresponding to a chapter or section. Further, a word limit is set (e.g., no more than 500 words per paragraph), and each chapter or section is further divided into multiple text segments. The first segment comes from the beginning of "Chapter 1 Overview of Machine Learning," the second segment comes from the first 500 words of "1.1 What is Machine Learning" (after specifying 500 words, relevant sentences are retained or deleted based on semantic understanding), the third segment comes from the remaining content of "1.1 What is Machine Learning," and so on, dividing the textbook into multiple text segments.

[0143] By traversing the entire document during or after text splitting, image information, including the first image, can be extracted. For example: Image 1: filename "fig1.png"; Image 2: filename "fig2.png", and so on.

[0144] Furthermore, after obtaining the first image, it also includes:

[0145] Based on the position of the first image in the document data, the first associated text of the first image is extracted from the text content. The first associated text is used to generate retrieval information for the first image.

[0146] Understandably, after extracting image information, the image's position within the document data can be recorded, and surrounding text can be obtained based on that position to generate potential titles or descriptions. For example: Image 1: filename "fig1.png", located in "Chapter 1 Overview of Machine Learning", described as "Machine Learning Overview Diagram"; Image 2: filename "fig2.png", located in "1.1 What is Machine Learning", with no additional description. And so on. This text data can serve as associated text for the image information, used to generate retrieval information for that image.

[0147] In one possible implementation, the learning resource content includes presentation data, and the resource fragments include a second text fragment and a second presentation medium;

[0148] At this point, step 1002 specifically includes:

[0149] Layout element identification is performed on the presentation data to obtain title text and description text;

[0150] The description text is split based on the title text to obtain the second text fragment;

[0151] Extract media information from the presentation data to obtain the second presentation media.

[0152] In this embodiment of the application, the learning content includes presentation data. It is understood that the presentation data integrates text as well as various media formats such as images, audio, and video.

[0153] For presentation data, layout element recognition can be performed first to obtain title and description text. This is typically done by analyzing the page's visual hierarchy, font size, color contrast, etc. Specifically, if the presentation is in image format, each page can be converted into editable text format using optical character recognition (OCR) technology. Then, based on attributes such as font size, position, and bolding, the title text (usually larger, centered, or bold) and description text (usually located below the title, with a smaller font) can be identified.

[0154] Please see Figure 3 , Figure 3 This is a schematic diagram of a presentation page provided for an embodiment of this application.

[0155] like Figure 3As shown, after identifying the layout elements of the page, the title text "Clinical and Scientific Big Data Applications" in larger font at the top of the page is identified; the four text boxes in the middle of the page are descriptive texts about "Clinical Decision Support", "Gene Sequencing and Precision Medicine", "Medical Data Demonstration" and "New Drug Development"; and the video in the lower half of the page is image information within the media information.

[0156] After obtaining the title text and description text, the description text can be split into multiple second text segments related to the title, based on the content or structure of the title text. This process can be aided by natural language processing (NLP) techniques to understand the text content. Specifically, NLP techniques are used to understand the theme or keywords of the title text. Based on these themes or keywords, algorithms using semantic similarity or keyword matching are used to find and extract relevant paragraphs or sentences from the description text, thus obtaining the second text segments.

[0157] The media information in the presentation data includes multimedia elements such as images, audio, and video. The entire presentation can be traversed, and image processing, file parsing, or layout element recognition technologies can be used to identify the image, audio, and video files on each page. For media files embedded in the presentation, specific tools or libraries may be needed for extraction. The identified media files are then extracted from the presentation, including secondary presentation media. For example: Media 1: filename "fig3.png"; Media 2: filename "video1.mp4"; Media 3: filename "audio.mp3", etc.

[0158] Furthermore, after obtaining the second demonstration medium, it also includes:

[0159] Based on the results of layout element identification, the second associated text of the second presentation media is extracted, and the second associated text is used to generate retrieval information for the second presentation media.

[0160] Understandably, after extracting media information, it's possible to extract related text based on the previous layout element identification of the presentation. This related text can be text data on the same page as the media information, or it can include text data from other pages under the same title text. For example: Media 1: filename "fig3.png", title "Big Data Application", description "Decision Support, Effect Prediction, Teaching Demonstration, Technology Transfer"; Media 2: filename "video1.mp4", title "Industrial Robot", description "Logistics Robots Handling Goods in Logistics Warehouses"; Media 3: filename "audio.mp3", title "Speech Recognition", description "Calling a Smart Assistant via Command". This text data can serve as related text for the media information, used to generate retrieval information for that media information.

[0161] In one possible implementation, the learning resource content includes video data, and the resource segments are video clips;

[0162] At this point, step 1002 specifically includes:

[0163] Perform text recognition on video data frames;

[0164] Perform speech recognition on the audio of video data;

[0165] Based on the results of text recognition and speech recognition, and combined with semantic understanding technology, the video data is split into multiple video segments.

[0166] In this embodiment of the application, the learning content includes video data, which may be teaching videos, popular science videos, live screen recordings, etc.

[0167] Typically, video data related to learning content will describe some key information points through text, and the knowledge points involved in the video will also be presented in a certain logical order. Therefore, the video data can be split into video frames and audio. For video frames, text recognition technology can be used to obtain the text data, and for audio, speech recognition technology can be used to convert the audio into text.

[0168] In one possible implementation, the text content obtained from character recognition and speech recognition is combined with semantic understanding technology to segment the text content according to certain rules, specifically the first rule in the aforementioned document data. Based on the segmented text fragments and their positions in the video data, the video data can then be segmented into multiple video segments.

[0169] In another possible implementation, the content displayed by the video data may be a presentation. In this case, the layout elements of the video frames can be identified based on the content described in the layout element recognition technology for presentations above, the title text and description text of each frame can be determined, and the video data can be split based on the content of the title text and description text.

[0170] Furthermore, after obtaining multiple video clips, it also includes:

[0171] Based on the results of text recognition and speech recognition, the third associated text of the video clip is extracted, and the third associated text is used to generate retrieval information for the video clip.

[0172] Understandably, after obtaining multiple video clips, it is possible to extract related text based on the previous text recognition and speech recognition results. This related text can be the text content obtained from the text recognition and speech recognition results, and this text content can be used to generate retrieval information for the corresponding video clips.

[0173] In one possible implementation, given the construction of a learning resource library, the knowledge outline determined in step 1021 above can also be generated based on resource fragments in the learning resource library. That is, step 1021 specifically includes:

[0174] The target knowledge content and learning resource library are input into the large language model to obtain a knowledge outline generated by the large language model based on the learning resource library.

[0175] Understandably, with a pre-built learning resource library, both the target knowledge content and the learning resource library can be input into the large language model. Since the resource fragments are text content or have corresponding descriptive text, the understanding and summarizing capabilities of the large language model are applied to match the knowledge points in the learning resource library with the target knowledge content, thereby constructing a knowledge outline associated with the resource fragments in the learning resource library. This avoids the problem that the generated knowledge outline may contain knowledge points not covered in the learning resource library, which could lead to the inability to output learning content or the output of learning content without a basis.

[0176] It is understandable that among the above implementation methods for different types of learning resources, the retrieval information can be used to retrieve the corresponding resource fragments. When the resource fragment is text content, the retrieval information can be the text content itself, or it can be used to extract features and keywords from the text content to obtain the corresponding feature vectors and keywords.

[0177] Next, we return to step 103 above. After step 103, that is, after determining the knowledge outline and target resource fragments, the following steps are included:

[0178] 104. Input the knowledge outline and target resource fragments into the pre-trained large language model to obtain the learning content output by the large language model based on the knowledge outline, which includes the target resource fragments.

[0179] Understandably, a pre-trained large language model (LLM) refers to a deep learning model trained on a large amount of text data, capable of generating natural language text or understanding the meaning of language text. Large language models can handle various natural language tasks, such as text classification, question answering, and dialogue. After inputting a knowledge outline and target resource fragments into the large language model, its understanding capabilities allow it to integrate the target resource fragments based on the knowledge outline, resulting in learning content. This learning content displays various target knowledge points through the knowledge outline, and each target knowledge point includes a corresponding target resource fragment.

[0180] The learning content generation method provided in this application first obtains the target knowledge content specified by the user, then determines a knowledge outline based on the target knowledge content, and further determines multiple target knowledge points corresponding to the target knowledge content based on the knowledge outline. Based on the target knowledge points, corresponding target resource fragments are retrieved from a pre-built learning resource library, and the knowledge outline and target resource fragments are combined using the natural language processing capabilities of a large language model to output the learning content. This learning content integrates multiple target resource fragments related to the target knowledge content. These target resource fragments may come from different courses. Users or learners can learn or understand these knowledge-related fragments at once through this learning content, without having to switch between multiple courses, thus improving learning efficiency.

[0181] In one possible implementation, after step 104, the method further includes:

[0182] 105. Input the learning content into the large language model to obtain test questions and / or code experiments for each target knowledge point based on the learning content.

[0183] Understandably, after receiving the learning content, it is input into a large language model. Upon receiving the input, the model utilizes its internal deep learning algorithms and extensive training data to analyze the content, identifying key knowledge points, concepts, and terminology. Based on the user's instructions, it generates corresponding test questions or code experiments. The generated test questions may include multiple-choice, fill-in-the-blank, and short-answer questions, designed to assess the learner's understanding and mastery of the knowledge points. The generated code experiments may include example programs, function implementations, and data processing scripts, used to help learners deepen their understanding through practice.

[0184] To facilitate understanding, the application of the learning content generation method provided in this application will be described in detail below. Please refer to... Figure 4 , Figure 4 This is a diagram illustrating the overall technical architecture of the learning content generation method provided in this application embodiment.

[0185] like Figure 4 As shown, the entire process can be divided into two parts: offline and online. The offline part involves building the learning resource library: first, various types of learning resource content can be obtained from the learning platform, such as document data, presentation data, and video data; then, according to the splitting rules corresponding to each type, the learning resource content is split into resource fragments.

[0186] Please see Figure 5a The text content in the document data can be segmented (split) according to rules such as directory hierarchy, specified length, and semantic understanding; for images in the document, the corresponding title and surrounding text can be extracted based on the image's position in the document, and corresponding search information can be generated.

[0187] Please see Figure 5b Presentation data can be identified using layout element recognition technology to identify elements such as titles, description text, and media information. For description text, it can be segmented based on its title; for media information, it can generate corresponding search information based on the text content in the presentation page corresponding to the media information.

[0188] Please see Figure 5c For video data, it can be split into video frames and audio. Text recognition or layout element recognition is performed on the video frames, and speech recognition is performed on the audio data. Based on the recognition results of the video frames and audio, and combined with semantic understanding, the video data is split into video segments. The text content obtained based on the above recognition technology can be used as retrieval information for the corresponding video segments.

[0189] Among the various implementation methods for different types of learning resources, retrieval information can be used to retrieve corresponding resource fragments. When the resource fragment is text content, the retrieval information can be the text content itself, or it can be obtained by extracting features and keywords from the text content to obtain corresponding feature vectors and keywords. When the resource fragment is an image, video, or other media content, the retrieval information can be the text itself extracted using the above methods (descriptive text), or it can be obtained by extracting features and keywords from the descriptive text to obtain feature vectors and keywords.

[0190] In addition to obtaining existing learning resources from the learning platform, users can also upload their own learning resources. Please refer to [link / reference]. Figure 6 , Figure 6This is a flowchart illustrating the construction process of the learning resource library provided in this embodiment of the application.

[0191] like Figure 6 As shown, users can upload various types of learning resource files through the relevant interfaces, including learning resource content in the form of document data, video data, and presentation document data, and... Figure 4 The corresponding offline component works similarly, splitting learning resources based on file type, such as document splitting, video splitting, and presentation splitting. Then, retrieval information is generated for the split resource fragments, and a learning resource library is built based on these fragments carrying the retrieval information.

[0192] The learning resources on the learning platform are broken down into smaller parts, such as videos, text in documents, relevant text extracted from images in documents, descriptive text in presentations, and relevant text extracted from media information in presentations. Finally, a corresponding learning resource library is constructed. In the learning resource library, text fragments, video fragments, and other media content all have corresponding feature vectors. Below are examples illustrating data samples of text, video, and other media content in the knowledge base:

[0193] For text, it is segmented into multiple text fragments (segmented by knowledge point granularity or specific rules), and each text fragment has its corresponding feature vector extracted through an embedding model. For example, the following text fragments and feature vectors:

[0194] "In deep learning, the concept of the perceptron is extended to the multi-layer perceptron (MLP), which consists of multiple layers, including an input layer, one or more hidden layers, and an output layer. Each layer consists of multiple neurons connected by weights. Each neuron performs a weighted summation of the input data and then transforms it through an activation function (such as ReLU, Sigmoid, or Tanh) to produce the output."

[0195] [0.1,0.2,0.3,…,0.9,0.8,0.7]

[0196] "Deep learning perceptrons can be applied to various tasks, including image recognition, speech recognition, and natural language processing. With the development of deep learning, many more complex network structures have emerged, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs), which perform better on specific tasks."

[0197] For videos, after the course videos are segmented according to knowledge point granularity, each video segment corresponds to a piece of content or topic description text. Similarly, an embedding model is used to extract feature vectors from the description text of the video segments. For example, the following are the text descriptions and feature vectors of video segments:

[0198] [0.2,0.1,0.3,…,0.5,0.9,0.7]

[0199] This video introduces the different types of loss functions in deep learning.

[0200] For other media content, such as images in a document, the descriptive text of the image can be extracted based on the title or surrounding text. Similarly, an embedding model is used to extract feature vectors from the descriptive text. For example, the following image descriptive text and feature vectors...

[0201] [0.2,0.9,0.6,…,0.5,0.8,0.7]

[0202] "The Focal Loss function curve in deep learning".

[0203] The aforementioned feature vectors and descriptive text can be used as retrieval information to retrieve corresponding resource fragments, and can also allow the large language model to insert relevant resource fragments at appropriate positions in the generated content based on the descriptive text.

[0204] Next Figure 4 The online portion of the program will be introduced. This portion is based on a pre-built learning resource library and combines the capabilities of a large language model to generate learning content, test questions, and code experiments based on the knowledge content input by the user.

[0205] After a user or learner inputs the knowledge content they want to learn, a related knowledge outline is generated based on that content. This knowledge outline can be generated using a large language model. Taking the user inputting "deep learning perceptron" as an example, a hierarchical and systematic knowledge outline can be generated using a large language model. The prompt word template for generating the knowledge outline can be:

[0206] "Please generate a well-structured knowledge outline based on the following knowledge points."

[0207] Key concept: Deep learning perceptron

[0208] Generate a good knowledge outline, such as Figure 2 As shown.

[0209] Furthermore, with a pre-built learning resource library, both the target knowledge content and the learning resource library can be input into the large language model. Since the resource fragments are text content or have corresponding descriptive text, the understanding and summarizing capabilities of the large language model are applied to enable the large language model to match the knowledge points in the learning resource library based on the target knowledge content, and construct a knowledge outline associated with the resource fragments in the learning resource library. This avoids the problem that the generated knowledge outline includes knowledge points not covered in the learning resource library, which would prevent the subsequent output of learning content or the output of learning content from having no basis.

[0210] For each knowledge point in the knowledge outline, relevant resource fragments, including target video clips, target text clips, and other target media content, can be retrieved from the learning resource library using vector similarity or keyword retrieval. For example, for the knowledge point "Perceptron Algorithm - Loss Function," a similarity search in the learning resource library retrieves text clips, video clips, and images. These relevant contents (descriptive text or search information for video clips and images) are then combined and placed into a prompt word template for generating course content. This allows the large language model to generate learning content about the knowledge point "Perceptron Algorithm - Loss Function" based on this relevant content. An example of a prompt word template for generating learning content is shown below.

[0211] "Based on the following background information, please generate learning content for the relevant knowledge points. If the background information includes images or videos, please insert relevant image or video tags at appropriate locations in the generated content, based on the text descriptions of the images or videos, to create learning content that combines text and images."

[0212] Background information:

[0213] {

[0214] "Text": ["The perceptron algorithm is a linear binary classification model whose goal is to find a linear hyperplane that can classify data into two categories. The perceptron algorithm uses a loss function to measure the difference between the model's predictions and the actual values, and attempts to improve the model by minimizing this loss.", "For the perceptron, commonly used loss functions are misclassification loss or 0-1 loss. This loss function is very simple: if a sample is correctly classified, the loss is 0; if it is misclassified, the loss is 1. However, the 0-1 loss function is discontinuous and non-differentiable, making it difficult to use in actual optimization processes."]

[0215] "Image":[{"id":"image_1","description":"Cross-entropy loss function formula"},{"id":"image_2","description":"Focal Loss function curve"}],

[0216] “video”:[{“id”:”video_1”,“description”:“Introduction to the loss function of deep learning perceptron”}]

[0217] }

[0218] Key concept: Perceptron algorithm - loss function

[0219] The final learning content generated by the large language model is as follows:

[0220] Introduction to the loss function of deep learning perceptrons:

[0221] <video_1>

[0222] The cross-entropy loss function characterizes the distance between two probability distributions and is more suitable for classification problems.

[0223] <image_1>

[0224] The Focal Loss function is designed to address the problem of extreme imbalance between positive and negative samples in one-stage object detection.

[0225] <image_2>

[0226] When training a perceptron, the loss is typically summed over all samples, and then the total loss is minimized. This process can be implemented using stochastic gradient descent (SGD) or other optimization algorithms.

[0227] After obtaining the specific learning content, you can also use a large language model to generate relevant test questions and code experiment content based on test question prompt templates or code experiment prompt templates.

[0228] Example of a test prompt template:

[0229] "Based on the above learning content, please generate relevant multiple-choice questions according to the following knowledge points."

[0230] Key concept: Loss function of deep learning perceptron

[0231] Understandably, different prompt templates can be set for different test questions, including templates for true / false questions, single-choice questions, and multiple-choice questions. These templates are stored in the corresponding storage media and can be retrieved as needed.

[0232] Example of code experiment prompt template:

[0233] "Based on the above learning content, please generate relevant experimental code according to the following knowledge points."

[0234] Key concept: Cross-entropy loss function of deep learning perceptron.

[0235] Understandably, different prompt word templates can be set for different experimental codes, including AI experiments, big data experiments, operating system experiments, etc. These templates are stored in the corresponding storage media and can be retrieved as needed.

[0236] Based on the above architecture, learning content that interests learners can be generated, and the content is presented in a systematic and hierarchical manner. In addition, the learning content is accompanied by corresponding test questions and experiments, covering learning, practice, and testing.

[0237] This application also provides a possible product form illustration, such as... Figure 7 As shown, Figure 7 This is a schematic diagram of the product form interaction interface of the learning content generation method provided in the embodiments of this application.

[0238] The interactive interface diagram shown in the embodiments of this application includes a knowledge outline, a knowledge area, an experimental area, and an exercise area.

[0239] The knowledge outline can be located in the left sidebar, displaying the course outline using a tree structure. Each chapter or section can be a collapsible panel, with the panel title displaying the name of the knowledge points for that chapter and section. Clicking the chapter title expands or collapses all sections under that chapter. Clicking the section title loads and displays the learning content corresponding to the knowledge points in that section in the knowledge area.

[0240] Based on the selected section in the knowledge outline, relevant learning content is dynamically loaded and displayed. The content is presented in rich text format, including text paragraphs, images, and video links (which can be embedded for playback). Further features include text highlighting, note-taking, and content sharing.

[0241] The experiment area can be used to showcase experimental code examples closely related to the current learning content. Code blocks should be highly readable and editable, supporting syntax highlighting. A code input box should also be provided, allowing learners to directly modify or write code. A run button should be located below or next to the input box, allowing execution of the code in a local or cloud environment. The results or output of code execution should be displayed in real time, supporting error messages and debugging information.

[0242] The exercise area displays relevant practice questions or quizzes based on the current learning section. Question types can include multiple choice, fill-in-the-blank, and programming problems. Learners answer on the interface, and the correctness of their answers and explanations are displayed immediately after submission. For programming problems, code submission and execution may be required to verify the results. Further features include recording learner performance data such as accuracy and time taken, which can be used for learning progress assessment and personalized recommendations.

[0243] Based on the above design, an online learning platform that is both clearly structured and highly interactive can be built, effectively improving learners' learning experience and outcomes.

[0244] This application also provides a learning content generation apparatus; please refer to [link / reference]. Figure 8 , Figure 8 A schematic diagram of the learning content generation device provided in this application includes:

[0245] Knowledge acquisition module 801 is used to acquire target knowledge content;

[0246] The outline acquisition module 802 is used to determine a knowledge outline based on the target knowledge content. The knowledge outline includes multiple target knowledge points.

[0247] The retrieval module 803 is used to retrieve the target resource fragments corresponding to each target knowledge point from the pre-built learning resource library;

[0248] The output module 804 is used to input the knowledge outline and target resource fragments into the pre-trained large language model to obtain the learning content output by the large language model based on the knowledge outline, which includes the target resource fragments.

[0249] The learning content generation apparatus provided in this application first determines a knowledge outline based on the knowledge content that learners are interested in. Then, it retrieves each knowledge point in the knowledge outline from the learning resource library to obtain learning resource fragments corresponding to each knowledge point. Finally, based on the capabilities of a large language model, it outputs learning content according to the knowledge outline and embeds the learning resource fragments corresponding to each knowledge point into the learning content. This allows learners to learn learning resource fragments related to the knowledge points they are interested in at once, facilitating targeted learning and improving learning efficiency.

[0250] Please see Figure 9 , Figure 9 This is another schematic diagram of the learning content generation apparatus provided in an embodiment of this application. (See diagram below.) Figure 9 As shown, the knowledge acquisition module 901, outline acquisition module 902, retrieval module 903, and output module 904 are respectively connected to... Figure 8The knowledge acquisition module 801, outline acquisition module 802, retrieval module 803, and output module 804 correspond to each other, and will not be elaborated further here. Additionally:

[0251] One possible implementation also includes:

[0252] Module 905 is used to acquire learning resource content; split the learning resource content according to preset rules to obtain multiple resource fragments; and build a learning resource library based on multiple resource fragments.

[0253] Understandably, learning resources can be obtained directly from the learning platform, or they can be other learning resources uploaded by users and approved. Depending on the type of learning resource, it can be split according to pre-defined rules to obtain multiple resource fragments. These fragments are then imported into a database or cloud server to complete the construction of the learning resource library.

[0254] In one possible implementation,

[0255] Module 905 is specifically used to process resource fragments to obtain retrieval information, which includes at least one of feature vectors, descriptive text, and keywords; and to build a learning resource library based on resource fragments carrying retrieval information.

[0256] The retrieval module 903 is specifically used to retrieve the target resource fragments corresponding to each target knowledge point from the learning resource library based on the retrieval information.

[0257] Understandably, after obtaining multiple resource fragments, to facilitate accurate retrieval of target resource fragments related to the target knowledge point, these fragmented fragments can be processed to obtain retrieval information. This retrieval information is then incorporated as part of the resource fragments and used to construct a learning resource library. Specifically, when retrieving based on the target knowledge point, the target resource fragment can be obtained by matching the retrieval information of each resource fragment in the learning resource library.

[0258] In one possible implementation, the learning resource content includes document data, and the resource fragment includes a first text fragment and a first image;

[0259] Module 905 is specifically used to split the text content in the document data based on a first rule to obtain a first text fragment. The first rule includes at least one of the following: directory level, specified number of characters, and semantic understanding; and to extract image information from the document data to obtain a first image.

[0260] Understandably, the document data in learning resources includes document-based data such as textbooks, courseware, academic papers, and research reports. Its main characteristic is that it consists primarily of text, and may include illustrations. For the text portion of the document data, it can be split into multiple text fragments based on rules such as directory hierarchy, specified word count, and semantic understanding, including the first text fragment. Simultaneously or subsequently, traversing the entire document allows for the extraction of image information, including the first image.

[0261] Furthermore, the construction module 905 is also used to extract the first associated text of the first image from the text content based on the position of the first image in the document data. The first associated text is used to generate retrieval information for the first image.

[0262] Understandably, after extracting image information, the image's position within the document data can be recorded, and surrounding text can be obtained based on that position to generate potential titles or descriptions. This text data can then be used as associated text for the image information, generating retrieval information for that image.

[0263] In one possible implementation, the learning resource content includes presentation data, and the resource fragments include a second text fragment and a second presentation medium;

[0264] Module 905 is specifically used to identify layout elements in presentation data to obtain title text and description text.

[0265] The description text is split based on the title text to obtain the second text fragment;

[0266] Extract media information from the presentation data to obtain the second presentation media.

[0267] Understandably, presentation data integrates text with various media formats such as images, audio, and video. For presentation data, layout element identification can be performed first to obtain the title and description text, typically achieved by analyzing the page's visual hierarchy, font size, and color contrast. After obtaining the title and description text, the description text can be broken down into multiple secondary text segments related to the title, based on the content or structure of the title text. Media information in presentation data includes multimedia elements such as images, audio, and video. The entire presentation can be traversed, using image processing, file parsing, or layout element identification techniques to identify the image, audio, and video files on each page. For media files embedded in the presentation, specific tools or libraries may be required for extraction.

[0268] Furthermore, the construction module 905 is also used to extract the second associated text of the second presentation media based on the results of layout element recognition, and the second associated text is used to generate retrieval information of the second presentation media.

[0269] Understandably, after extracting media information, it's possible to extract related text based on the previous layout element identification of the presentation. This related text can be text data on the same page as the media information, or it can include text data from other pages belonging to the same title text. This text data can be used as related text for the media information to generate retrieval information for that media information.

[0270] In one possible implementation, the learning resource content includes video data, and the resource segments are video clips;

[0271] Module 905 is specifically used for text recognition of video data frames.

[0272] Perform speech recognition on the audio of video data;

[0273] Based on the results of text recognition and speech recognition, and combined with semantic understanding technology, the video data is split into multiple video segments.

[0274] Understandably, this video data can be educational videos, science videos, live stream recordings, etc. Typically, video data containing learning content will describe key information points with text, and the knowledge points presented in the video will be displayed in a logical order. Therefore, the video data can be broken down into video frames and audio. For the video frames, text recognition technology can be used to extract the text data, and for the audio, speech recognition technology can be used to convert the audio into text. The resulting text content can then be combined with semantic understanding technology to further segment the video data into multiple video clips.

[0275] Furthermore, the construction module 905 is also used to extract the third associated text of the video segment based on the results of text recognition and speech recognition. The third associated text is used to generate retrieval information for the video segment.

[0276] Understandably, after obtaining multiple video clips, it is possible to extract related text based on the previous text recognition and speech recognition results. This related text can be the text content obtained from the text recognition and speech recognition results, and this text content can be used to generate retrieval information for the corresponding video clips.

[0277] In one possible implementation, with the learning resource library constructed, the outline acquisition module 902 is specifically used to input the target knowledge content into the large language model to obtain the knowledge outline generated by the large language model.

[0278] Understandably, with a pre-built learning resource library, both the target knowledge content and the learning resource library can be input into the large language model. Since the resource fragments are text content or have corresponding descriptive text, the understanding and summarizing capabilities of the large language model are applied to match the knowledge points in the learning resource library with the target knowledge content, thereby constructing a knowledge outline associated with the resource fragments in the learning resource library. This avoids the problem that the generated knowledge outline may contain knowledge points not covered in the learning resource library, which could lead to the inability to output learning content or the output of learning content without a basis.

[0279] In one possible implementation, the output module 904 is also used to input the learning content into the large language model to obtain test questions and / or code experiments for each target knowledge point based on the learning content.

[0280] Understandably, after receiving the learning content, it is input into a large language model. Upon receiving the input, the model utilizes its internal deep learning algorithms and extensive training data to analyze the content, identifying key knowledge points, concepts, and terminology. Based on the user's instructions, it generates corresponding test questions or code experiments. The generated test questions may include multiple-choice, fill-in-the-blank, and short-answer questions, designed to assess the learner's understanding and mastery of the knowledge points. The generated code experiments may include example programs, function implementations, and data processing scripts, used to help learners deepen their understanding through practice.

[0281] This application also provides a computing device; please refer to [link / reference]. Figure 10 This application provides a schematic diagram of the structure of a computing device according to an embodiment. The computing device includes a processor 1101 and a memory 1102. Optionally, the communication device further includes a transceiver 1103.

[0282] A computing device can be used to execute computer programs or computer instructions stored in memory to perform... Figure 1 The methods in the illustrated embodiments can be referred to the relevant descriptions in the above method embodiments.

[0283] This application also provides a computer program product including instructions that, when run on a computer, cause the computer to perform the above-described actions. Figure 1 The learning content generation method shown in the embodiment.

[0284] This application also provides a computer-readable storage medium, including computer instructions that, when executed on a computer, cause the computer to perform the actions described above. Figure 1 The method of the embodiment shown.

[0285] This application also provides a chip device, including a processor for connecting to a memory and calling a program stored in the memory, so that the processor executes the above-described... Figure 1 The method of the embodiment shown.

[0286] The processor mentioned above can be a general-purpose central processing unit, a microprocessor, an application-specific integrated circuit (ASIC), or one or more devices used to control the above. Figure 1 The illustrated embodiment is an integrated circuit for program execution of the method. The memory mentioned above may be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, such as random access memory (RAM).

[0287] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0288] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0289] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0290] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0291] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for generating learning content, characterized in that, include: Acquire the target knowledge content; A knowledge outline is determined based on the target knowledge content, and the knowledge outline includes multiple target knowledge points; The target resource fragments corresponding to each target knowledge point are retrieved from the pre-built learning resource library, which includes multiple resource fragments; The knowledge outline and the target resource fragment are input into a pre-trained large language model to obtain the learning content output by the large language model based on the knowledge outline, wherein the learning content includes the target resource fragment.

2. The method according to claim 1, characterized in that, Before retrieving the target resource fragment corresponding to each target knowledge point from the pre-built learning resource library, the method further includes: Access learning resources; The learning resource content is split according to preset rules to obtain the multiple resource fragments; The learning resource library is constructed based on the multiple resource fragments.

3. The method according to claim 2, characterized in that, The construction of the learning resource library based on the multiple resource fragments includes: The resource fragments are processed to obtain retrieval information, which includes at least one of feature vectors, descriptive text, and keywords; The learning resource library is constructed based on the resource fragments carrying the retrieval information; The step of retrieving the target resource fragment corresponding to each target knowledge point from the pre-built learning resource library includes: Based on the retrieval information, the target resource fragment corresponding to each target knowledge point is retrieved from the learning resource library.

4. The method according to claim 3, characterized in that, The learning resource content includes document data, and the resource fragment includes a first text fragment and a first image; The learning resource content is split according to preset rules to obtain multiple resource fragments, including: The text content in the document data is split based on a first rule to obtain a first text fragment. The first rule includes at least one of the following: directory level, specified number of characters, and semantic understanding. The image information is extracted from the document data to obtain the first image.

5. The method according to claim 4, characterized in that, After extracting the image information from the document data to obtain the first image, the process further includes: Based on the position of the first image in the document data, a first associated text of the first image is extracted from the text content, and the first associated text is used to generate the retrieval information of the first image.

6. The method according to claim 3, characterized in that, The learning resource content includes presentation data, and the resource fragments include second text fragments and second presentation media; The learning resource content is split according to preset rules to obtain multiple resource fragments, including: The presentation data is analyzed to identify layout elements, resulting in title text and description text. The description text is split based on the title text to obtain the second text fragment; Extract the media information from the presentation data to obtain the second presentation media.

7. The method according to claim 6, characterized in that, After extracting the image information from the presentation data to obtain the second presentation media, the process further includes: Based on the results of the layout element identification, the second associated text of the second presentation media is extracted, and the second associated text is used to generate the retrieval information of the second presentation media.

8. The method according to claim 3, characterized in that, The learning resources include video data, and the resource segments are video segments. The learning resource content is split according to preset rules to obtain multiple resource fragments, including: Perform text recognition on the video data frames; Perform speech recognition on the audio of the video data; Based on the results of the text recognition and speech recognition, the video data is split using semantic understanding technology to obtain multiple video segments.

9. The method according to claim 8, characterized in that, After the video data is segmented into multiple video segments based on the results of text recognition and speech recognition, combined with semantic understanding technology, the method further includes: Based on the results of the text recognition and the speech recognition, a third associated text is extracted from the video segment, and the third associated text is used to generate the retrieval information of the video segment.

10. The method according to any one of claims 1 to 9, characterized in that, The process of determining the knowledge outline based on the target knowledge content includes: The target knowledge content is input into the large language model to obtain the knowledge outline generated by the large language model.

11. The method according to any one of claims 1 to 10, characterized in that, After inputting the knowledge outline and the target resource fragment into a pre-trained large language model to obtain the learning content output by the large language model based on the knowledge outline, the method further includes: The learning content is input into the large language model to obtain test questions and / or code experiments for each target knowledge point output by the large language model based on the learning content.

12. A learning content generation device, characterized in that, include: The knowledge acquisition module is used to acquire target knowledge content; The outline acquisition module is also used to determine a knowledge outline based on the target knowledge content, wherein the knowledge outline includes multiple target knowledge points; The retrieval module is used to retrieve the target resource fragments corresponding to each target knowledge point from the pre-built learning resource library; The output module is used to input the knowledge outline and the target resource fragment into a pre-trained large language model to obtain the learning content output by the large language model based on the knowledge outline, wherein the learning content includes the target resource fragment.

13. A computing device, characterized in that, Including processor and memory; The processor is configured to execute computer programs or computer instructions in the memory to perform the method as described in any one of claims 1 to 11.

14. A computer program product containing instructions, characterized in that, When the instructions are executed by the computing device, the computing device performs the method as described in any one of claims 1 to 11.

15. A computer-readable storage medium, characterized in that, It includes computer program instructions, which, when executed by a computing device, cause the computing device to perform the method as described in any one of claims 1 to 11.

Images