A micro course extraction method, a related method and a related device

Abstract

The application discloses a micro-lesson extraction method, related method and related device, and relates to the technical field of resource fragmentation, wherein the micro-lesson extraction method comprises the following steps: obtaining a target course video of a target subject; analyzing the content of the target course video to generate target text data with timestamps; performing knowledge point slicing on the target text data based on a slicing large model to obtain a segmentation result comprising at least the start and end times of a knowledge point segment, wherein the slicing large model is obtained by training a general large model base by using training text data annotated with real knowledge point segmentation results and supplemented with thinking process data for obtaining the real knowledge point segmentation results, and the training text data is text data corresponding to a course video sample of the target subject; and extracting a knowledge point video segment from the target course video according to the start and end times of the knowledge point segment to obtain a knowledge point video micro-lesson. The micro-lesson extraction method disclosed by the application can automatically extract an accurate knowledge point video micro-lesson from a course video.

Description

Technical Field

[0001] This application relates to the field of resource fragmentation technology, and in particular to a micro-lesson extraction method, related methods, and related devices. Background Technology

[0002] The smart education public service platform has gathered high-quality digital course resources, initially built a resource supply system that expands horizontally and extends vertically, and continues to promote the improvement of capabilities such as intelligent analysis, personalized push and intelligent search.

[0003] However, both the aforementioned platforms and digital education resource systems currently have limitations in terms of the flexibility of resource supply. The resources gathered by existing platforms are mostly systematic and structured complete courses, such as classroom recordings or full-length teaching videos. While these resources have advantages in the completeness of teaching content, their granularity is relatively large, and the data structure is relatively fixed. When teachers teach single knowledge points or students engage in fragmented learning, it is often difficult to directly extract the required knowledge segments from these complete videos.

[0004] Currently, converting complete course resources into micro-lessons that can be used in a fragmented and disjointed manner typically relies heavily on teachers' manual selection or secondary recording. This results in high costs and low efficiency in resource reconstruction, failing to fully utilize the vast amount of existing digital educational resources. Therefore, how to intelligently deconstruct existing complete course videos and automatically extract micro-lesson resources at the knowledge point level to support flexible resource reorganization and personalized supply has become a pressing technical problem to be solved in the field of digital educational resource construction. Summary of the Invention

[0005] In view of this, this application provides a micro-lesson extraction method, related methods, and related devices for intelligently deconstructing existing complete course videos and automatically extracting micro-lessons at the knowledge point level to support flexible resource reorganization and personalized supply. The technical solution is as follows:

[0006] A method for extracting micro-lessons, comprising:

[0007] Obtain target course videos for the target subject;

[0008] The content of the target course video is parsed to generate corresponding target text data with timestamps;

[0009] Based on the slice big model, the target text data is sliced ​​into knowledge points to obtain knowledge point segmentation results. The knowledge point segmentation results include at least the start and end times of the knowledge point segments. The slice big model is trained on a general big model base using training text data labeled with the actual knowledge point segmentation results, and supplemented by thinking process data to obtain the actual knowledge point segmentation results. The training text data is time-stamped text data corresponding to the course video samples of the target subject.

[0010] Based on the start and end times of the knowledge point segments, knowledge point video segments are extracted from the target course video to obtain knowledge point video micro-lessons.

[0011] In one possible implementation, parsing the content of the target course video and generating corresponding, timestamped target text data includes:

[0012] The target course video is subjected to audio-video separation to obtain audio data and silent video data;

[0013] By combining the video content of the silent video data, the audio data is processed into standardized, correct, time-stamped, and speaker-tagged structured text data to obtain the target text data corresponding to the target course video.

[0014] In one possible implementation, the step of combining the video content of the silent video data with the audio data to process the audio data into standardized, correct, timestamped, and speaker-tagged structured text data includes:

[0015] The audio data is subjected to speech recognition to obtain time-stamped recognized text. By distinguishing between teacher and student speech in the recognized text, text with time stamp and speaker label is obtained.

[0016] Extract key video frames from the silent video data, and extract text information from the key video frames;

[0017] Based on a large language model, and combined with text information extracted from the key video frames, the text with timestamps and speaker tags is semantically, grammatically, and with key information corrected. The corrected text is then subjected to long sentence segmentation, formula standardization, and semantic-based paragraph division to obtain standardized, correct, time-stamped, and speaker-tagged structured text.

[0018] In one possible implementation, the step of slicing the target text data into knowledge points based on a large slice model to obtain knowledge point segmentation results includes:

[0019] Based on the large-scale slicing model, combined with the knowledge graph and curriculum standards of the target subject, the target text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results;

[0020] Alternatively, based on the large-scale slicing model, and combined with the knowledge point list of the chapter to which the target course video belongs, the target text data can be sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0021] In one possible implementation, the step of slicing the target text data into knowledge points based on a large slice model to obtain knowledge point segmentation results includes:

[0022] Based on the large-scale slicing model, the target text data is initially sliced ​​into knowledge points to obtain preliminary knowledge point segmentation results.

[0023] Based on the aforementioned large-scale model of knowledge points, the preliminary knowledge point segmentation results are optimized according to the knowledge graph and curriculum standards of the target subject, as well as the knowledge related to the knowledge point segments, to obtain the final knowledge point segmentation results.

[0024] In one possible implementation, the step of slicing the target text data into knowledge points based on a large slice model to obtain knowledge point segmentation results includes:

[0025] Obtain the curriculum syllabus for the target subject;

[0026] The course syllabus is segmented according to the teaching stages to obtain the syllabus segmentation results;

[0027] Based on the teaching syllabus segmentation results, text fragments that are irrelevant to the knowledge point segments in the target text data are filtered out;

[0028] Based on the aforementioned large-scale slicing model, the filtered text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0029] In one possible implementation, the knowledge point segmentation result further includes: the title of the knowledge point micro-lesson, the summary of the knowledge point micro-lesson, and the text content of the knowledge point micro-lesson;

[0030] The micro-lesson extraction method also includes:

[0031] Construct metadata that includes the title of the micro-lesson for the knowledge point, the summary of the micro-lesson for the knowledge point, and the tags for the knowledge point;

[0032] The metadata and the text content of the knowledge point micro-lessons are packaged and encapsulated with the knowledge point video micro-lessons to obtain complete knowledge point video micro-lesson resources;

[0033] The video micro-lesson resources for the knowledge points are associated with the corresponding knowledge points in the knowledge graph of the target subject.

[0034] In one possible implementation, the micro-lesson extraction method further includes:

[0035] Based on the aforementioned large-scale slicing model and combined with the typical example question slicing rules, the target text data is sliced ​​into typical example questions to obtain a segmentation result that includes at least the start and end times of the typical example question segments. Then, typical example question video segments are extracted from the target course video according to the start and end times of the typical example question segments to obtain typical example question video micro-lessons.

[0036] And / or, based on the slice big model and combined with the scenario material slicing rules, the target text data is sliced ​​into scenario materials to obtain a segmentation result that includes at least the start and end times of the scenario material segments, and scenario material video segments are extracted from the target course video according to the start and end times of the scenario material segments to obtain scenario material video micro-lessons;

[0037] And / or, based on the slice model and combined with the classroom activity slice rules, the target text data is sliced ​​into classroom activities to obtain a segmentation result that includes at least the start and end times of the classroom activity segments. Classroom activity video segments are then extracted from the target course video according to the start and end times of the classroom activity segments to obtain classroom activity video micro-lessons.

[0038] In one possible implementation, the step of slicing the target text data into typical example questions based on the slice model and in conjunction with the typical example question slicing rules includes:

[0039] Based on the target text data, keywords related to the typical example questions are obtained, and the presentation method of the typical example questions in the classroom is determined; based on the segmentation model, combined with the typical example question segmentation rules, the keywords, and the presentation method of the typical example questions in the classroom, the target text data is segmented into typical example questions.

[0040] The step of slicing the target text data into contextual elements based on the slice model and in conjunction with contextual element slicing rules includes:

[0041] Determine the scenario type of the scenario material involved in the target text data; based on the slice model, and combining the scenario type with the slice rules corresponding to the scenario type in the scenario material slicing rules, slice the target text data into scenario material slices;

[0042] The step of slicing the target text data into classroom activity segments based on the slice model and combined with classroom activity slicing rules includes:

[0043] Determine the activity type of the classroom activities involved in the target text data; based on the slice model, and combining the activity type with the slice rules corresponding to the activity type in the classroom activity slice rules, slice the target text data into classroom activities.

[0044] In one possible implementation, the micro-lesson extraction method further includes:

[0045] The solution process data of the example problems is obtained from the example problem fragments, and the solution process data of the example problems is associated with the example problem video micro-lessons;

[0046] Based on the scenario material fragments, obtain the key elements and / or teaching intentions of the scenario material, and associate the key elements and / or teaching intentions with the scenario material video micro-lessons;

[0047] Based on the classroom activity segments, obtain the activity process analysis results, and / or teaching objectives, and / or teacher-student interaction information, and associate the activity process analysis results, and / or the teaching objectives, and / or the teacher-student interaction information with the classroom activity video micro-lesson. The activity process analysis results include some or all of the following information: activity organization form information, activity participant role information, activity process information, and time allocation information for each stage of the activity.

[0048] In one possible implementation, the micro-lesson extraction method further includes:

[0049] Obtain the lesson plan document corresponding to the target course video;

[0050] The lesson plan document is parsed to obtain the document parsing results;

[0051] Based on the document parsing results, identify the document structure of the lesson plan document;

[0052] Based on the document parsing results and the document structure, structured lesson plan text data is obtained, and the structured lesson plan text data is sliced ​​into knowledge points to obtain knowledge point documents.

[0053] Associate the knowledge point document with the corresponding knowledge point video micro-lesson, and / or associate the knowledge point document with the corresponding knowledge point in the knowledge graph of the target subject.

[0054] In one possible implementation, the micro-lesson extraction method further includes:

[0055] Generate a teaching script for the target knowledge point, where the target knowledge point is the knowledge point corresponding to the video micro-lesson of the knowledge point;

[0056] Based on the teaching script for the target knowledge point, generate the teaching audio for the target knowledge point to obtain the audio micro-lesson for the knowledge point;

[0057] The audio micro-lessons on the knowledge points are associated with the corresponding knowledge points in the knowledge graph of the target subject.

[0058] In one possible implementation, the teaching script for generating the target knowledge points includes:

[0059] Based on the target knowledge points and the instructional design information of the target knowledge points, a teaching script is constructed to generate prompts. The instructional design information includes some or all of the following: dialogue design logic, plot arrangement information, content organization information, and character design and configuration information.

[0060] The teaching script generation prompts are input into a large language model to generate the teaching script, thus obtaining the teaching script for the target knowledge point.

[0061] A second aspect of this application provides a resource retrieval method, comprising:

[0062] Obtain query information related to knowledge points input by the user;

[0063] Retrieve resources that match the query information from a pre-built resource library, wherein the resource library includes resources associated with knowledge points in a subject knowledge graph, and the resources in the resource library include micro-lesson resources extracted using the micro-lesson extraction method described above.

[0064] A third aspect of this application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0065] The memory is used to store computer programs;

[0066] The processor is used to execute the computer program so that the electronic device can implement any of the above-described micro-lesson extraction methods or the above-described resource retrieval methods.

[0067] A fourth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement any of the micro-lesson extraction methods or the resource retrieval methods described above.

[0068] The fifth aspect of this application discloses a computer program product comprising computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement any of the micro-lesson extraction methods described above, or to implement the resource retrieval methods described above.

[0069] The sixth aspect of this application provides a resource processing management system, including: a terminal and a server;

[0070] The terminal is used to receive target course videos uploaded by users and upload the target course videos to the server;

[0071] The server is used to extract micro-lessons from the target course video using the micro-lesson extraction method described above.

[0072] Using the above-mentioned technical solution, the micro-lesson extraction method provided in this application first acquires the target course video, then parses the content of the target course video to generate corresponding target text data with timestamps, then segments the target text data into knowledge points based on a large-scale slicing model adapted to the teaching scenario, obtaining a segmentation result that includes at least the start and end times of the knowledge point segments, and finally extracts knowledge point video segments from the target course video according to the start and end times of the knowledge point segments to obtain knowledge point video micro-lessons. Since the micro-lesson extraction method provided in this application automatically extracts knowledge point video micro-lessons from existing course videos based on a pre-trained large-scale slicing model, no manual intervention is required. Therefore, the micro-lesson extraction cost is low, the micro-lesson extraction efficiency is high, and existing digital education resources can be fully utilized. In addition, since the large-scale slicing model is trained on a general large-scale model base using training text data (text data corresponding to course video samples) labeled with real knowledge point segmentation results, and supplemented by thinking process data to obtain real knowledge point segmentation results, the large-scale slicing model can accurately segment the target text data into knowledge points. Therefore, accurate knowledge point video micro-lessons can be extracted based on the knowledge point segmentation results of the large-scale slicing model. Attached Figure Description

[0073] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0074] Figure 1 A flowchart illustrating the micro-lesson extraction method provided in this application embodiment;

[0075] Figure 2 This is a schematic diagram illustrating the process of acquiring knowledge point video micro-lessons as provided in an embodiment of this application;

[0076] Figure 3 A flowchart illustrating the process of parsing the content of a target course video and generating corresponding, timestamped target text data for the purposes of this application embodiment;

[0077] Figure 4This is a schematic diagram illustrating the process of acquiring typical example video micro-lessons provided in the embodiments of this application;

[0078] Figure 5 This is a schematic diagram illustrating the process of acquiring scenario material video micro-lessons provided in the embodiments of this application;

[0079] Figure 6 This is a schematic diagram illustrating the process of acquiring classroom activity video micro-lessons provided in an embodiment of this application.

[0080] Figure 7 A flowchart illustrating the process of processing target classroom videos to obtain knowledge point video micro-lessons, typical example question video micro-lessons, scenario material video micro-lessons, and classroom activity video micro-lessons, provided for embodiments of this application;

[0081] Figure 8 A schematic diagram illustrating the parallel execution of the resource slicing process, resource annotation process, and PPT frame extraction and key information extraction process provided in the embodiments of this application;

[0082] Figure 9 A flowchart illustrating the process of slicing knowledge points from a PDF teaching plan document, as provided in this embodiment of the application.

[0083] Figure 10 This is a schematic diagram illustrating the process of generating audio micro-lessons for knowledge points, provided in an embodiment of this application.

[0084] Figure 11 This is a schematic diagram of the micro-lesson extraction device provided in the embodiments of this application;

[0085] Figure 12 This is a schematic diagram of the structure of the resource processing management system provided in the embodiments of this application. Detailed Implementation

[0086] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0087] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0088] 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 terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0089] In order to transform complete course resources into micro-lessons of knowledge points that can be used in a fragmented and disorganized manner, so as to support flexible resource reorganization and personalized supply, this application provides a micro-lesson extraction method. The following embodiments will introduce the micro-lesson extraction method provided by this application.

[0090] Please see Figure 1 The diagram illustrates a flowchart of a micro-lesson extraction method provided in an embodiment of this application. This micro-lesson extraction method may include:

[0091] Step S101: Obtain the target course video.

[0092] In this embodiment, the target course video is a long video of the target subject from which micro-lessons are to be extracted.

[0093] The target course video can be a past teaching video from a school or educational institution (such as a complete classroom teaching recording). Of course, this embodiment is not limited to this. For example, the target course video can also be a lecture video recorded by the teacher or lecturer (a long lecture video recorded by the teacher or lecturer using a mobile phone, camera or screen recording software), a replay video recorded during live teaching (from various live streaming platforms), online course videos (such as course videos on MOOC platforms and knowledge payment platforms), and course videos on video sharing platforms (such as open courses, lecture recordings, etc.).

[0094] Step S102: Parse the content of the target course video and generate corresponding target text data with timestamps.

[0095] In this embodiment, the target course video is an audio video, which can be parsed into analyzable text data in a multi-dimensional and multi-modal manner to obtain the target text data corresponding to the target course video.

[0096] Step S103: Based on the large-scale slicing model, the target text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0097] In this embodiment, after obtaining the target text data corresponding to the target course video, the target text data is sliced ​​into knowledge points based on the pre-trained large-scale slicing model to obtain the knowledge point segmentation result.

[0098] It should be noted that knowledge point slicing of target text data refers to dividing the target text data into micro-units around a single knowledge point, with the knowledge point as the core unit. Due to the different types of knowledge points, their content composition also varies. For example, for knowledge points of concepts, processes, and methods, it usually includes reasoning, argumentation, analysis, and exploration. For knowledge points of graphics, it usually focuses on recognition, drawing, measurement, feature description, and application. For knowledge points of function graphs, it often covers functional relationships, graph drawing, property analysis, key point research, and the combination of numbers and shapes. For knowledge points of mathematical thinking, it focuses on the penetration of ideas and the formation of literacy.

[0099] The knowledge point segmentation results obtained by slicing the target text data into knowledge points should at least include the start and end times of the knowledge point segments (i.e., the start and end times of the knowledge point micro-lessons). It should be noted that a knowledge point segment is a segment surrounding a single knowledge point, and the start and end times of the knowledge point segment are used to indicate where the knowledge point begins and ends.

[0100] In this embodiment, the large-scale model is obtained by training text data labeled with the actual knowledge point segmentation results, supplemented by thinking process data that obtains the actual knowledge point segmentation results, and training the general large-scale model base. The training text data is time-stamped text data corresponding to the course video samples of the target subject.

[0101] To train a high-performance slice model, a large number of course video samples of the target subject can be collected in advance (e.g., 6000+ subject course video samples) to ensure comprehensive coverage of knowledge points and clear boundaries. Then, training data can be constructed based on the large number of course video samples collected (each training data is constructed based on a course video sample, which includes training text data labeled with the actual knowledge point segmentation results, as well as thinking process data for obtaining the actual knowledge point segmentation results).

[0102] It should be noted that the purpose of introducing thinking process data (such as thinking paths, derivation steps, typical misconceptions and error correction processes in the teaching process) during training is to help the model possess "teaching and research level" thinking and explanatory abilities.

[0103] like Figure 2 As shown, when performing knowledge point slicing on target text data based on the large slice model, knowledge point slicing prompts can be constructed based on the target text data, and then the knowledge point slicing prompts can be input into the large slice model to perform knowledge point slicing (logical slicing).

[0104] Step S104: Based on the start and end times of the knowledge point segments, extract the knowledge point video segments from the target course video to obtain the knowledge point video micro-lessons.

[0105] After obtaining the start and end times of the knowledge point segments (i.e., the start and end times of the micro-lessons on the knowledge points), such as Figure 2 As shown, knowledge point video segments (i.e., physics slices) can be extracted from the target course video based on the start and end times of the knowledge point segments, thus obtaining knowledge point video micro-lessons.

[0106] Preferably, the knowledge point segmentation results obtained based on the large-scale slicing model can include, in addition to the start and end times of the knowledge point segments, also the titles of the knowledge point segments (i.e., the titles of the micro-lessons), the summaries of the knowledge point segments (i.e., the summaries of the micro-lessons), and the text content of the knowledge point segments (i.e., the text content of the micro-lessons). The micro-lesson titles can include both a main title and a subtitle; alternatively, they can include only the main title. The text content of the micro-lessons is a text segment located from the target text data based on the start and end times of the knowledge point segments. The micro-lesson titles and summaries are generated from this text segment.

[0107] Preferably, the micro-lesson extraction method may further include: constructing metadata that includes at least the title of the knowledge point micro-lesson, the summary of the knowledge point micro-lesson, and the knowledge point tags; packaging the metadata and the text content of the knowledge point micro-lesson together with the knowledge point video micro-lesson (or only packaging the metadata and the knowledge point video micro-lesson together) to obtain a complete knowledge point video micro-lesson resource.

[0108] Preferably, the metadata of the knowledge point video micro-lessons may include not only the title, summary, and tags of the knowledge point micro-lessons, but also the applicable grade level and keywords.

[0109] Preferably, after obtaining complete knowledge point video micro-lesson resources, these resources can be associated with corresponding knowledge points in the knowledge graph of the target subject to support knowledge point-based retrieval and personalized recommendations. For example, when a user makes a mistake on a certain knowledge point, a knowledge point video micro-lesson associated with that knowledge point can be directly recommended to the user for repeated study. Similarly, when a user searches for a concept, they can directly locate the knowledge point video micro-lesson explaining that concept, thereby improving learning efficiency.

[0110] Furthermore, knowledge graphs define the sequential relationships between knowledge points. Linking knowledge point video micro-lessons to corresponding knowledge points means that each knowledge point in the learning path has specific teaching content to support it. In addition, after linking knowledge point video micro-lessons to knowledge graphs, these resources are no longer isolated and discrete, but become organized "knowledge assets," facilitating teachers' course design and management. Essentially, linking knowledge point video micro-lessons to knowledge graphs deeply integrates discrete teaching resources with the systematic knowledge logic, laying a solid foundation for future intelligent learning and precision teaching.

[0111] The micro-lesson extraction method provided in this application first acquires the target course video, then parses the content of the target course video to generate corresponding target text data with timestamps. Next, based on a pre-trained large-scale slicing model, the target text data is sliced ​​into knowledge points to obtain a segmentation result that includes at least the start and end times of the knowledge point segments. Finally, knowledge point video segments are extracted from the target course video according to the start and end times of the knowledge point segments to obtain knowledge point video micro-lessons. Since the micro-lesson extraction method provided in this application automatically extracts knowledge point micro-lessons from existing course videos based on the large-scale slicing model without manual intervention, the extraction cost of knowledge point micro-lessons is low, the extraction efficiency is high, and existing digital educational resources can be fully utilized. In addition, since the large-scale slicing model is trained on a general large-scale model base using training text data (text data corresponding to course video samples) labeled with real knowledge point segmentation results, and supplemented by thinking process data to obtain real knowledge point segmentation results, the large-scale slicing model can accurately slice the target text data into knowledge points. Therefore, accurate knowledge point video micro-lessons can be extracted based on the knowledge point segmentation results of the large-scale slicing model.

[0112] As can be seen from the micro-lesson extraction method provided in the above embodiments, the basis of micro-lesson extraction is target text data. In some embodiments of this application, the implementation process of "analyzing the content of the target course video and generating corresponding target text data with timestamps" in the above embodiments is introduced.

[0113] Please see Figure 3 This diagram illustrates the process of parsing the content of a target course video and generating corresponding, timestamped target text data, which may include:

[0114] Step S301: Perform audio-video separation on the target course video to obtain audio data and silent video data.

[0115] Optionally, ffmpeg can be used to separate the audio and video of the target course video. It should be noted that ffmpeg is a powerful open-source multimedia processing tool, suitable for batch processing or scenarios requiring high precision. Of course, this embodiment is not limited to this; other methods for separating audio and video of the target course video are also applicable to this application.

[0116] Preferably, when separating audio data from the target course video, background noise suppression and endpoint detection can be performed to ensure that subsequent speech processing has a higher signal-to-noise ratio.

[0117] Step S302: Combine the video content of the silent video data with the audio data to process the audio data into standardized, correct, time-stamped, and speaker-tagged structured text data to obtain the target text data corresponding to the target course video.

[0118] In one possible implementation, the process of processing audio data into standardized, correct, timestamped, and speaker-tagged structured text data, in conjunction with the video content of silent video data, may include:

[0119] Step S3021a: Perform speech recognition on the audio data to obtain the recognized text with timestamps.

[0120] Speech recognition can be performed on audio data based on a target speech recognition model to obtain timestamped recognized text. In one possible implementation, the target speech recognition model can be a general-purpose speech recognition model. To improve the model's recognition rate for subject-specific vocabulary, terminology, and proper nouns in teaching scenarios (such as classroom settings), the general-purpose speech recognition model can be fine-tuned using teaching audio and corresponding text from these scenarios. This fine-tuned model then serves as the target speech recognition model. Furthermore, the target speech recognition model can incorporate punctuation prediction and sentence boundary segmentation capabilities, allowing the recognized text to be directly used as input for natural language processing.

[0121] Step S3022a: By distinguishing between teacher and student speech in the identified text with timestamps, a conversational text with timestamps and speaker labels is obtained.

[0122] Speaker recognition can be performed on audio data to obtain a series of time periods with speaker tags. Then, the speaker tags are aligned with the timestamps of the recognized text to obtain text with timestamps and speaker tags (i.e., conversational text with timestamps).

[0123] Step S3021b: Extract key video frames from silent video data.

[0124] Representative key video frames can be extracted from silent video data through methods such as scene change detection, camera switch recognition, and visual consistency analysis.

[0125] Step S3022b: Extract text information from key video frames.

[0126] Text information in key video frames can be extracted by performing OCR recognition on the key video frames.

[0127] Step S3023: Based on the large language model, combined with the text information extracted from key video frames, the text with timestamps and speaker tags is corrected in terms of semantics, syntax and key information. The corrected text is then processed by long sentence segmentation, formula standardization and semantic-based paragraph division to obtain standardized, correct, time-stamped and speaker-tagged structured text.

[0128] After obtaining the text with timestamps and speaker tags, preliminary corrections can be made, including but not limited to some or all of the following processes: spell correction, punctuation correction, synonym unification, terminology standardization, removal of duplicate paragraphs, timestamp alignment verification, and formatted output.

[0129] After initial corrections, the text can be further revised based on a large language model. Specifically, based on the large language model and combined with textual information extracted from key video frames, the text with timestamps and speaker tags is corrected semantically, grammatically, and in terms of key information. That is, grammatical errors and semantic deviations in the text are corrected, and key information in the text is corrected based on textual information extracted from key video frames (including supplementing missing content and correcting erroneous content).

[0130] After obtaining the corrected text, long sentence segmentation, formula standardization, and semantic paragraph segmentation can be performed on the corrected text based on a large language model to obtain a standardized, correct, timestamped, and speaker-labeled structured text.

[0131] Segmentation based on semantics means dividing paragraphs according to the semantic logic of the context. When dividing paragraphs, the size of the paragraph can be controlled, for example, a single paragraph can not exceed 10 complete semantic units (such as sentences).

[0132] When performing long sentence segmentation, long sentence segmentation rules can be introduced to enable the large language model to segment long sentences according to these rules. For example, long sentence segmentation rules may include: ① prohibiting segmentation in the following locations: inside mathematical formulas, combinations of technical terms, and fixed collocations; ② speaker sentence segmentation check: performing sentence segmentation based on speaker characteristic differences, with each speaker's conversational unit forming an independent sentence.

[0133] When performing formula standardization, a unified formula standard specification can be introduced so that large language models can perform formula standardization according to the unified formula standard specification. For example, standardized formula specifications may include: ① Establishing a standard mathematical symbol lookup table and implementing unified unit conversion; ② Standardizing formula terms according to LaTeX, including: "square root" → \sqrt{}, "pi" → \pi, "cube" → 3^3; ③ Standardizing fractional expressions: original expression: a / b, converted expression: $\frac{a}{b}$; original expression: 3 1 / 2, converted expression: 3$\frac{1}{2}$; ④ Chinese mathematical description conversion rules: original expression: square root of two plus three squares, converted expression: $\sqrt{2}+3^2$; ⑤ Standardizing quantifiers: original expression: 5 kilohms, converted expression: 5 kΩ; ⑥ Formula encapsulation format: inline formula: $e^{i\pi}+1=0$, independent formula: $$\int_{a}^{b}f(x)dx$$; ⑦ Prohibiting the embedding of Chinese characters in formulas, and prohibiting the appearance of Chinese punctuation and characters.

[0134] After obtaining the target text data (i.e., standardized, correct, time-stamped, and speaker-tagged structured text) corresponding to the target course video, the target text data can be sliced ​​into knowledge points based on the slice big model to obtain the knowledge point segmentation results. Some embodiments of this application introduce the process of slicing the target text data into knowledge points based on the slice big model to obtain the knowledge point segmentation results.

[0135] Based on the large-scale slicing model, there are multiple ways to slice target text data into knowledge points and obtain the knowledge point segmentation results. This embodiment provides the following implementation methods:

[0136] The first implementation method is based on a large slice model, which directly slices the target text data into knowledge points to obtain the knowledge point segmentation results.

[0137] Specifically, firstly, knowledge point segment prompts are constructed based on the target text data. These prompts guide the large-scale slicing model to segment the target text data into knowledge points. Preferably, the prompts also guide the model to generate titles and summaries for the segmented knowledge point fragments. Then, the prompts are input into the large-scale slicing model, which outputs the knowledge point segmentation results. These results can include the start and end times of the knowledge point fragments (i.e., the start and end times of the micro-lessons), the text content of the fragments (i.e., the text content corresponding to the micro-lessons), the titles of the fragments (i.e., the titles of the micro-lessons), and the summaries.

[0138] The second approach is to introduce the knowledge graph and curriculum standards of the target subject to improve the knowledge point slicing effect. That is, based on the large slicing model, combined with the knowledge graph and curriculum standards of the target subject, the target text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0139] Specifically, firstly, knowledge point slicing prompts are constructed based on the target text data, the knowledge graph of the target subject, and the curriculum standards. (These knowledge point slicing prompts are used to prompt the slicing model to perform knowledge point slicing on the target text data in conjunction with the knowledge graph of the target subject and the curriculum standards. Preferably, the slicing model can also prompt the slicing model to generate titles and summaries for the segmented knowledge point fragments.) Then, the knowledge point slicing prompts are input into the slicing model, and the slicing model outputs the knowledge point slicing results.

[0140] The third implementation method is to introduce a list of knowledge points under the chapter to which the target course video belongs, so as to improve the knowledge point slicing effect. That is, based on the slicing model, combined with the list of knowledge points under the chapter to which the target course video belongs, the target text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0141] Specifically, firstly, based on the target text data and the list of knowledge points under the chapter to which the target course video belongs, knowledge point slicing prompts are constructed (these knowledge point slicing prompts are used to prompt the slicing model to perform knowledge point slicing on the target text data in conjunction with the list of knowledge points under the chapter to which the target course video belongs; preferably, the slicing model can also prompt the slicing model to generate titles and summaries for the segmented knowledge point fragments). Then, the knowledge point slicing prompts are input into the slicing model, and the slicing model outputs the knowledge point slicing results.

[0142] The fourth implementation method is as follows: Based on the large slice model, the target text data is initially sliced ​​into knowledge points to obtain preliminary knowledge point segmentation results; based on the large slice model, the preliminary knowledge point segmentation results are optimized according to the knowledge graph of the subject to which the target course video belongs, the curriculum standards, and the knowledge related to the knowledge point slices to obtain the final knowledge point segmentation results.

[0143] Specifically, based on the target text data, knowledge point segmentation prompts are constructed (these prompts guide the large-scale segmentation model to segment the target text data into knowledge points; preferably, they also guide the model to generate titles and summaries for the segmented knowledge point fragments). These prompts are then input into the large-scale segmentation model, which outputs preliminary knowledge point segmentation results. Next, based on the preliminary segmentation results, the target subject's knowledge graph and curriculum standards, and related knowledge, optimized prompts are constructed (these optimized prompts guide the large-scale segmentation model to reflect on the segmentation based on the target subject's knowledge graph, curriculum standards, and related knowledge, thereby adjusting and optimizing the preliminary segmentation results). These optimized prompts are then input into the large-scale segmentation model, which outputs the optimized segmentation results, i.e., the final knowledge point segmentation results.

[0144] The fifth implementation method is as follows: Obtain the course syllabus of the target subject; slice the course syllabus according to teaching links (such as introduction, knowledge point introduction, homework assignment, and class summary) to obtain the syllabus segmentation results; based on the syllabus segmentation results, filter out text fragments in the target text data that are obviously unrelated to the knowledge point segments; based on the large segmentation model, slice the filtered text data into knowledge points to obtain the knowledge point segmentation results.

[0145] Based on the segmentation results of the teaching syllabus, the target text data can be labeled with segments, and then text fragments that are obviously unrelated to the knowledge point segments can be filtered out based on the segmentation results.

[0146] When performing knowledge point slicing on filtered text data based on the large slice model, knowledge point slicing can be performed directly on the filtered text data based on the large slice model, or it can be performed by combining the knowledge graph of the target subject and the curriculum standards, or by combining the knowledge point list under the chapter of the target course video.

[0147] To enrich micro-lesson resources, in some embodiments of this application, the micro-lesson extraction method may further include the following steps a1~a2, and / or the following steps b1~b2, and / or the following steps c1~c2.

[0148] Step a1: Based on the large-scale slicing model and combined with the typical question slicing rules, perform typical question slicing on the target text data to obtain typical question segmentation results that include at least the start and end times of the typical question fragments.

[0149] The rules for slicing typical examples can be preset. Then, based on the large slicing model and combined with the rules for slicing typical examples, the target text data can be sliced ​​into typical examples.

[0150] like Figure 4 As shown, example question slicing prompts can be constructed based on the number of target texts and example question slicing rules. Then, the example question slicing prompts are input into the slicing big model to perform example question slicing (logical slicing).

[0151] In one possible implementation, when slicing the target text data into example questions based on the large-scale slicing model, only the example question slicing rules need to be considered. In another possible implementation, keywords related to example question slicing (such as "example question", "practice", "let's look at this one") can be obtained from the target text data, and the presentation method of the example questions in the classroom (such as blackboard writing, PPT presentation, oral presentation, etc.) can be determined. Therefore, when slicing the target text data into example questions based on the large-scale slicing model, in addition to combining the example question slicing rules, keywords and the presentation method of the example questions in the classroom can also be combined to improve the example question slicing effect.

[0152] The above embodiments mention that during the process of obtaining target text data, text with timestamps and speaker tags will be corrected. During correction, text information extracted from key video frames can be used to correct key information in the text to be corrected. It should be noted that if extracting example questions and micro-lessons, for key video frames involving example questions, it is not enough to simply extract text information; the question content must also be parsed, and then the parsed question content is used to supplement and correct the text to be corrected (supplementing missing content and correcting errors). When parsing the question content, the following can be performed: question stem extraction (using OCR to recognize question text, graphics, and tables), formula extraction (recognizing mathematical, physical, and chemical formulas in the question and converting them to LaTeX format), and condition analysis (extracting known conditions, solution objectives, and constraints).

[0153] Step a2: Extract the video segments of the classic examples from the target course video according to the start and end times of the classic example problem segments to obtain the classic example problem video micro-lessons.

[0154] After obtaining the start and end times of the example problem segment, such as Figure 4 As shown, video segments of typical examples (i.e., physics slices) can be extracted from the target course video based on the start and end times of the typical example problem segments, thus obtaining video micro-lessons of typical examples.

[0155] Optionally, solution process data for exemplary problems can be obtained from excerpts of exemplary problems to link with video micro-lessons on exemplary problems. When obtaining this data, each step explained by the teacher (e.g., analysis-thought process-calculation-verification), the solution method used (e.g., formula method, graphical method, reverse thinking, etc.), and key points (e.g., common mistakes, techniques, turning points in thinking, etc.) can be identified. In other words, the final solution process data can include the solution steps, methods, and key points of the exemplary problems.

[0156] It should be noted that the segmentation results of the exemplary questions include not only the start and end times of the exemplary question segments, but also the titles of the exemplary question segments, i.e., the titles of the exemplary question video micro-lessons (e.g., "[Example] Knowledge Point Name - Question Type").

[0157] Metadata can be constructed for exemplary problem video micro-lessons, including information such as title, knowledge point tags, difficulty level, applicable grade level, and solution time. This metadata, along with content-related information (such as the example problem, its solution process, explanations of the knowledge points involved, and common mistakes), can then be packaged with the exemplary problem video micro-lessons (or only the metadata can be packaged with the video micro-lessons) to obtain complete exemplary problem video micro-lesson resources. These complete resources can then be associated with corresponding knowledge points in the knowledge graph of the target subject. Furthermore, exemplary problem video micro-lesson resources can be associated with related knowledge point video micro-lesson resources and / or accompanying practice questions.

[0158] Step b1: Based on the large-scale slicing model and combined with the scenario material slicing rules, the target text data is sliced ​​into scenario materials to obtain scenario material segmentation results that include at least the start and end times of the scenario material segments.

[0159] Contextual materials refer to materials used in the teaching process to create specific situations, stimulate students' interest, guide inquiry, and promote knowledge construction. The core function of contextual materials is to connect abstract knowledge with real life, scientific inquiry, or social events, helping students perceive problems in specific contexts, stimulate interest, think actively, and achieve knowledge transfer and application.

[0160] In this embodiment, scenario material slicing rules can be preset, and then, based on the large slicing model and combined with the scenario material slicing rules, scenario material slicing can be performed on the target text data.

[0161] like Figure 5 As shown, scenario material slicing prompts can be constructed based on target text data and scenario material slicing rules. Then, the scenario material slicing prompts are input into the slicing big model to perform scenario material slicing (logical slicing).

[0162] In one possible implementation, the scenario type of the contextual material involved in the target text data can be determined. Then, based on the large-scale slicing model, and combining the scenario type with the slicing rules corresponding to the scenario type in the scenario material slicing rules, the target text data can be sliced ​​into scenario material segments. It should be noted that the scenario material slicing rules include slicing rules corresponding to several scenario types. After determining the scenario type of the contextual material involved in the target text data, the target text data can be sliced ​​into scenario material segments by combining the determined slicing rules corresponding to the scenario type.

[0163] The scenario types of scenario materials can include, but are not limited to, experimental demonstrations, scenario introductions, and case analyses. Among them, experimental demonstrations involve experimental equipment, operational actions, and phenomenon observations; scenario introductions involve life scenes, storylines, and problem situations; and case analyses involve case backgrounds, analysis processes, and conclusion summaries.

[0164] Step b2: Extract scenario material video clips from the target course video based on the start and end times of the scenario material clips to obtain scenario material video micro-lessons.

[0165] After obtaining the start and end times of the scenario material fragment, such as Figure 5 As shown, scenario material video segments (i.e., physical slices) can be extracted from the target course video based on the start and end times of the scenario material segments, thus obtaining scenario material video micro-lessons.

[0166] Optionally, key elements of the scenario material can be extracted from the scenario material clips to link with the scenario material video micro-lessons. For example, for experimental scenario material, elements such as equipment list, operation steps, experimental phenomena, data recording, and precautions can be extracted. For situational scenario material, elements such as scene description, character dialogue, question posing, and thinking guidance can be extracted. For case-based scenario material, elements such as case background, key information, analysis perspective, and key conclusions can be extracted. In addition, visual elements such as key video frames, charts, animations, and annotation information can be extracted from the scenario material video clips.

[0167] In addition, the teaching intentions of the scenario material segments can be analyzed so that the analyzed teaching intentions can be physically or logically linked to the scenario material video micro-lessons. Specifically, the objectives can be identified (such as whether it is to introduce new knowledge, verify theories, cultivate abilities or stimulate interest), the design ideas can be analyzed (such as why the teacher chose this scenario and how to organize the teaching), and the interaction methods can be identified (such as teacher-student interaction, group discussion, observation and thinking, etc.).

[0168] In addition to the start and end times of the scenario material segments, the segmentation results can also include the title of the scenario material (e.g., "[Scenario / Experiment / Case] ​​Knowledge Point Name - Scenario Theme").

[0169] After obtaining the scenario-based video micro-lessons, metadata can be constructed for them, including information such as title, knowledge point tags, and applicable scenarios (e.g., new lesson introduction / knowledge consolidation / ability development). This metadata, along with micro-lesson content-related information (e.g., scenario presentation, process recording, analysis and explanation, knowledge connections, and extended thinking), can then be packaged with the scenario-based video micro-lessons (or only the metadata can be packaged with the micro-lessons) to obtain complete scenario-based video micro-lesson resources. These resources can then be associated with corresponding knowledge points in the target subject's knowledge graph. Furthermore, the scenario-based video micro-lesson resources can be linked with related knowledge point video micro-lessons and / or accompanying exercises.

[0170] Step c1: Based on the large-scale slicing model and combined with the classroom activity slicing rules, the target text data is sliced ​​into classroom activities to obtain a classroom activity segmentation result that includes at least the start and end times of the classroom activity segments.

[0171] Classroom activity slicing rules can be preset, and then, based on the slicing model and combined with the classroom activity slicing rules, the target text data can be sliced ​​into classroom activities.

[0172] like Figure 6 As shown, classroom activity slicing prompts can be constructed based on the target text data and classroom activity slicing rules. Then, the classroom activity slicing prompts are input into the slicing big model to perform classroom activity slicing (logical slicing).

[0173] In one possible implementation, the activity types of the classroom activities involved in the target text data can be determined. Then, based on the large-scale slicing model, and combining the activity types with the corresponding slicing rules in the classroom activity slicing rules, the target text data is sliced ​​into classroom activities. It should be noted that the classroom activity slicing rules can include slicing rules corresponding to several activity types. After determining the activity types of the classroom activities involved in the target text data, the target text data can be sliced ​​into classroom activities by combining the determined slicing rules corresponding to the activity types.

[0174] Classroom activities can include, but are not limited to, group discussions, interactive Q&A sessions, inquiry-based activities, games, and competitions. Among these, group discussions involve grouping, discussion topics, and presentations; interactive Q&A sessions involve asking questions, students answering, and teachers commenting; inquiry-based activities involve posing questions, the inquiry process, and the formation of conclusions; and games and competitions involve explaining rules, the competition process, and the evaluation of results.

[0175] Step c2: Extract classroom activity video segments from the target course video based on the start and end times of the classroom activity segments to obtain classroom activity video micro-lessons.

[0176] After obtaining the start and end times of a segment of a classroom activity, such as Figure 6 As shown, classroom activity video segments (i.e., physics segments) can be extracted from the target course video based on the start and end times of the classroom activity segments, thus obtaining classroom activity video micro-lessons.

[0177] Optionally, activity flow analysis results can be obtained from classroom activity segments to establish a physical or logical connection with the classroom activity video micro-lessons. The activity flow analysis results include some or all of the following information: activity organization format information (e.g., whole-class activity, group activity, individual activity), activity participant role information (e.g., teacher role (organizer / facilitator / evaluator)), student role (participant / speaker / observer)), activity process information (e.g., teacher-student interaction, student-student interaction, group collaboration), and time allocation information for each stage of the activity (e.g., 5 minutes for preparation, 10 minutes for discussion, 8 minutes for presentation).

[0178] Optionally, teaching objectives can be derived from classroom activity segments to establish a physical or logical connection between them and the classroom activity video micro-lessons. Specifically, knowledge objectives (knowledge points to be mastered through the activities), ability objectives (thinking skills, expression skills, collaboration skills, etc. to be cultivated), affective objectives (interest to be stimulated, attitudes to be cultivated, values ​​to be formed, etc.), and expected outcomes (the effects that teachers hope to achieve by designing activities).

[0179] Optionally, teacher-student interaction records can be constructed based on classroom activity segments to physically or logically link them to the classroom activity video micro-lessons. These records can include complete teacher-student dialogues (with speaker tags), as well as information on thought processes (such as students' thought processes, intellectual exchanges, and clashes of viewpoints) and teacher guidance information (such as teachers' questioning techniques, guidance methods, and evaluation language). Additionally, they can include annotations of insightful remarks, typical misconceptions, and breakthroughs in thinking.

[0180] In addition to the start and end times of the classroom activity segments, the segmentation results can also include the title of the classroom activity (e.g., "[Classroom Activity] Knowledge Point Name - Activity Type").

[0181] After obtaining classroom activity video micro-lessons, metadata containing information such as titles and knowledge point tags can be constructed for them. This metadata, along with relevant information about the micro-lesson content (such as activity flow analysis results, teacher-student interaction records, etc.), can then be packaged with the classroom activity video micro-lessons (or only the metadata can be packaged with the micro-lessons) to obtain complete classroom activity video micro-lesson resources. These complete resources can then be associated with corresponding knowledge points in the target subject's knowledge graph. Additionally, classroom activity video micro-lessons can be associated with related knowledge point video micro-lessons.

[0182] Please see Figure 7 This document illustrates a flowchart of a process for processing target classroom videos to obtain micro-lessons on knowledge points, exemplary questions, scenario materials, and classroom activities. Users can upload target course videos. After obtaining the target course videos, the system can perform comprehensive and multimodal analysis to obtain target text data. Then, a large-scale slicing model is invoked to perform logical slicing of the target text data on knowledge points, exemplary questions, scenario materials, and classroom activities. Finally, based on the logical slicing results, the target classroom video is physically sliced ​​to obtain micro-lessons on knowledge points, exemplary questions, scenario materials, and classroom activities.

[0183] like Figure 8 As shown, in addition to slicing the target classroom video, a resource annotation process, as well as PPT frame extraction and key information extraction processes, can be performed simultaneously. The resource annotation process is used to obtain metadata for knowledge point video micro-lessons, typical example problem video micro-lessons, scenario material video micro-lessons, and classroom activity video micro-lessons, such as knowledge point tags, keywords, subject, grade, and chapter. For typical example problem video micro-lessons, the question stem information can also be obtained. The PPT frame extraction and key information extraction processes are used to extract PPT video frames from the target classroom video and extract key information from the PPT video frames based on a multimodal large model, thereby obtaining the key information of the typical example problem video micro-lessons. The slicing process, resource annotation process, and PPT frame extraction and key information extraction processes can be executed in parallel to improve efficiency.

[0184] In some embodiments of this application, the micro-lesson extraction method may further include:

[0185] Step d1: Obtain the lesson plan document corresponding to the target course video.

[0186] The lesson plan document corresponding to the target course video can be, but is not limited to, a PDF document.

[0187] Step d2: Parse the lesson plan document to obtain the document parsing results.

[0188] Figure 9The flowchart illustrates the process of slicing knowledge points from a PDF lesson plan document. After obtaining the PDF lesson plan document, it can be processed into a long PDF image, which can then be parsed. Specifically, it can perform text extraction (extracting text content such as body text, titles, and annotations), image extraction (extracting illustrations, charts, and diagrams (preserving position information)), table recognition (identifying table structure and extracting table data), formula extraction (identifying formulas and converting them to LaTeX format), and layout analysis (identifying headers, footers, multi-column layouts, text boxes, etc.).

[0189] Step d3: Based on the document parsing results, identify the document structure of the lesson plan document to obtain structured text data.

[0190] When identifying the document structure of lesson plan documents, it is possible to perform chapter identification (such as identifying first-level headings, second-level headings, third-level headings, etc.), hierarchical relationship construction (constructing the tree structure of the document (such as chapter-section-subsection)), paragraph classification (such as distinguishing between main text, examples, exercises, annotations, summaries, etc.), and logical relationship identification (such as identifying logical relationships such as cause and effect, progression, and transition).

[0191] Step d4: Based on the document parsing results and document structure, obtain structured lesson plan text data, and slice the structured lesson plan text data into knowledge points to obtain knowledge point documents.

[0192] It can slice structured lesson plan text data into knowledge points based on the slice model, and can also generate summaries and extract keywords.

[0193] Step d5: Associate the knowledge point document with the corresponding knowledge point micro-lesson, and / or associate the knowledge point document with the corresponding knowledge point in the knowledge graph of the target subject.

[0194] Preferably, metadata (such as knowledge point tags, summaries, keywords, etc.) can be constructed for the knowledge point document. This metadata is then packaged with the knowledge point document to obtain a complete knowledge point document resource. This complete knowledge point document resource can then be associated with corresponding knowledge point video micro-lessons, and / or with corresponding knowledge points in the knowledge graph of the target subject. Additionally, the knowledge point document resource can also be associated with related audio resources (such as explanation audio, reading audio), related courseware (such as PPT courseware, mind maps, etc.), and related exercises.

[0195] In some embodiments of this application, the micro-lesson extraction method may further include:

[0196] Step e1: Generate teaching scripts for the target knowledge points.

[0197] The target knowledge point can be the knowledge point corresponding to the knowledge point video micro-lesson.

[0198] When generating teaching scripts for target knowledge points, such as Figure 10 As shown, based on the target knowledge points and the instructional design information for those target knowledge points, prompts for generating instructional scripts can be constructed. These prompts are used to guide the large language model in analyzing the target knowledge points (e.g., extracting core concepts, key points, and common mistakes), and generating instructional scripts for those target knowledge points based on their instructional design information. The prompts are then input into the large language model to generate the instructional scripts, resulting in the instructional scripts for the target knowledge points output by the large language model.

[0199] The instructional design information for the target knowledge points may, but is not limited to, include some or all of the following: dialogue design logic (designing interactive dialogues between students and teachers: question-answer-comment-summary), plot arrangement information (designing teaching scenarios, introducing cases, creating suspense, etc.), content organization information (introduction stage: reviewing old knowledge and introducing new knowledge; explanation stage: explaining concepts, analyzing examples, and summarizing methods; interactive stage: asking questions and thinking, students answering, and correcting mistakes; summary stage: summarizing knowledge, emphasizing key points, and assigning homework), and role design and allocation information, etc.

[0200] The character design and configuration information can include character design information (e.g., teacher role, student role), character configuration information, and dialogue allocation information (allocating the dialogue volume and appearance timing for each character). An example of character configuration information is as follows:

[0201] The configuration information for the teacher role is as follows:

[0202] Voice characteristics: mature, approachable, and professional;

[0203] Language style: clear, organized, and adept at guiding the conversation;

[0204] Emotional expression: encouragement, patience, enthusiasm;

[0205] The configuration information for multiple student roles is as follows:

[0206] Xiaoming: Lively and curious, loves to ask questions, occasionally careless;

[0207] Xiao Hong: meticulous, good at summarizing, and has clear logic;

[0208] Other: Design more student roles as needed.

[0209] Step e2: Based on the teaching script for the target knowledge points, generate the teaching audio for the target knowledge points to obtain audio micro-lessons for the knowledge points.

[0210] like Figure 10As shown, after obtaining the teaching script for the target knowledge point, speech synthesis can be performed sentence by sentence according to the teaching script to obtain the teaching audio of the target knowledge point, that is, the audio micro-lesson of the knowledge point.

[0211] Specifically, the dialogue subjects in the teaching script for the target knowledge points are analyzed, and appropriate timbres are matched according to their roles (such as teacher and student). The speech rate, pauses, stress, and intonation are marked on the teaching script. Based on this, the teaching script is optimized for naturalness, transforming written language into more conversational language that is closer to daily teaching. Emotional tags (such as question, surprise, sudden realization, etc.) are predefined for key dialogue segments to ensure that the corresponding emotions can be accurately injected during speech synthesis. Finally, according to the matched timbre, and based on the marked speech rate, pauses, stress, intonation, and emotional tags, the speech is synthesized sentence by sentence to generate teaching audio with role differentiation and emotional expressiveness.

[0212] Preferably, after generating the teaching audio, it can be subjected to emotional calibration and tone fine-tuning (for example, when asking questions, use an upward tone and pause to wait; when answering, use an affirmative tone and clear expression; when commenting, use an encouraging tone and appropriate emphasis). In addition to emotional calibration and tone fine-tuning of the generated teaching audio, classroom ambient sounds (such as slight background noise, the sound of turning pages, etc.) can also be added to enhance the sense of immersion in the scene.

[0213] Preferably, for the teaching audio after fine-tuning and adding classroom ambient sound, further optimization processes can be performed, such as volume equalization (unifying the volume of each character and avoiding sudden changes in volume), noise reduction (removing noise and artifacts from the synthesized audio), mixing (mixing dialogues from multiple characters), and adding background music (e.g., adding soft background music), ultimately generating teaching audio in standard audio formats such as MP3 or AAC, i.e., knowledge point audio micro-lessons. Optionally, a single teaching audio can be controlled to be 5 to 10 minutes long.

[0214] Step e3: Associate the audio micro-lessons of knowledge points with the corresponding knowledge points in the knowledge graph of the target subject.

[0215] By linking audio micro-lessons on knowledge points with corresponding knowledge points in the knowledge graph of the target subject, these micro-lessons are no longer isolated resources but are integrated into the overall knowledge network, supporting retrieval and personalized recommendations based on knowledge points. When users search by knowledge point, they can simultaneously obtain both video and audio micro-lessons. This means that two different forms of media teaching resources can be presented around the same knowledge point. Video micro-lessons are primarily visual with supplementary audio, suitable for demonstrating processes, conducting experiments, and creating dynamic diagrams—content requiring visual assistance. Audio micro-lessons, on the other hand, are primarily auditory, suitable for scenarios where viewing a screen is inconvenient, such as commuting or exercising. They can also be used for review and consolidation or simple lecture-based learning. The two formats complement each other, together forming a multimedia learning package for the knowledge point. Users can flexibly choose or combine them according to their own learning scenarios and preferences to achieve multimodal learning and improve learning effectiveness.

[0216] This application also provides a retrieval method, which includes:

[0217] Step f1: Obtain the query information related to the knowledge points input by the user.

[0218] The search query information entered by the user can be, but is not limited to, user-input keywords, natural language query statements, knowledge point codes, etc.

[0219] Keyword searches, such as "quadratic function" and "circumference of a circle," are concise and clear. Natural language queries, such as "how to find the circumference of a circle" and "how to draw the graph of a quadratic function," are closer to everyday expressions. Knowledge point encoding allows users to directly input the knowledge point ID, making it easier for professional users to accurately locate information.

[0220] Step f2: Retrieve resources that match the query information from the pre-built resource library.

[0221] The resource library includes resources associated with knowledge points in the subject knowledge graph. The resources in the resource library may include video micro-lessons extracted using the video micro-lesson extraction method described above (such as knowledge point video micro-lessons, scenario material video micro-lessons, and classroom activity video micro-lessons), audio micro-lessons generated using the audio micro-lesson generation method described above (such as knowledge point audio micro-lessons), and knowledge point documents obtained in the above embodiments.

[0222] When acquiring matching resources, the query information can be preprocessed first, such as word segmentation, stop word removal, and synonym expansion. Then, semantic understanding and intent recognition can be performed on the preprocessed query information (an intent recognition model tailored to the characteristics of educational resources can be trained to perform semantic understanding and intent recognition on the query information) to identify the user's true query intent (e.g., what type of resources are needed). At the same time, the query information can be rewritten, for example, by converting colloquial expressions into standard knowledge point descriptions. Then, matching resources can be retrieved from the resource database based on the identified query intent and the rewritten query information.

[0223] During resource retrieval, based on the query intent and the rewritten query information, the target knowledge point is accurately located in the subject knowledge graph, and then matching resources are obtained from resources associated with the target knowledge point. In addition, in some scenarios, the retrieval can be expanded to include resources associated with the preceding and / or subsequent knowledge points of the target knowledge point, and the difficulty level can be matched according to the user's learning stage to ensure that the difficulty of the resources is appropriate.

[0224] When conducting a search, the resource library can be searched from some or all dimensions based on the query intent and the rewritten query information. These multiple dimensions may include, but are not limited to, knowledge point dimensions, resource type dimensions, teaching scenario dimensions, and quality dimensions. Specifically, if the user specifies a knowledge point, related resources will be retrieved based on that knowledge point; if a resource type is specified (e.g., video micro-lessons / audio micro-lessons / document resources), resources will be filtered by type; if a teaching scenario is specified (e.g., new lesson introduction / knowledge explanation / example analysis / consolidation exercises), resources for that scenario will be matched; if resource quality requirements are specified, high-quality, reviewed resources will be prioritized to ensure professionalism and reliability.

[0225] After obtaining the search results, each resource in the search results can be scored from dimensions such as knowledge point matching degree, content completeness, user learning stage matching degree, and resource quality score. Then, the scores of the same resource in each dimension are weighted and summed (for example, the weight of knowledge point matching degree is 0.4, the weight of user learning stage matching degree is 0.2, the weight of content completeness degree is 0.2, and the weight of resource quality degree is 0.2) to obtain the score of each resource in the search results. Then, the resources in the search results are sorted according to the obtained scores. Optionally, personalized recommendations can be made, such as recommending resources based on the user's historical learning records, similar users' learning paths, and the user's mastery of knowledge points, while controlling the diversity of results to ensure a balanced display of different types of resources.

[0226] Search results can be displayed in a structured manner, such as using a card layout, including thumbnails, titles, summaries, and knowledge point tags. They can also display knowledge point trees, presenting relevant knowledge points in the form of knowledge graphs, and provide recommended learning paths.

[0227] In addition, after displaying the search results, users can filter the results based on their filtering actions. Users can filter by resource type (e.g., video / audio / document), difficulty level (e.g., basic / intermediate / advanced), duration (e.g., within 3 minutes / within 5 minutes / within 10 minutes), and teaching scenario (e.g., new lesson / practice / review). Furthermore, the system supports preview functions, including video preview displaying keyframes, fast playback with speed adjustment, content summary displaying the core content of the micro-lesson, and batch operations such as favorites, adding to the learning list, and sharing.

[0228] This application also provides a micro-lesson extraction device, such as... Figure 11 As shown, the micro-lesson extraction device may include: a course video acquisition module 1101, a course video parsing module 1102, a knowledge point slicing module 1103, and a knowledge point video micro-lesson extraction module 1104.

[0229] The course video acquisition module 1101 is used to acquire target course videos for the target subject.

[0230] The course video parsing module 1102 is used to parse the content of the target course video and generate corresponding target text data with timestamps.

[0231] The knowledge point slicing module 1103 is used to slice target text data into knowledge points based on the large slicing model, and obtain the knowledge point segmentation results.

[0232] The knowledge point segmentation results include at least the start and end times of the knowledge point segments. The large-scale slicing model uses training text data labeled with the actual knowledge point segmentation results, supplemented by thinking process data that obtains the actual knowledge point segmentation results, to train the general large-scale model base. The training text data is time-stamped text data corresponding to the course video samples of the target subject.

[0233] The knowledge point video micro-lesson extraction module 1104 is used to extract knowledge point video segments from the target course video based on the start and end times of the knowledge point segments, and obtain knowledge point video micro-lessons.

[0234] In one possible implementation, the course video parsing module 1102, when parsing the content of the target course video and generating corresponding, timestamped target text data, is specifically used for:

[0235] The target course video is subjected to audio-video separation to obtain audio data and silent video data;

[0236] By combining the video content of the silent video data, the audio data is processed into standardized, correct, time-stamped, and speaker-tagged structured text data to obtain the target text data corresponding to the target course video.

[0237] In one possible implementation, the course video parsing module 1102, when combining the video content of the silent video data with the audio data to process the audio data into standardized, correct, time-stamped, and speaker-tagged structured text data, is specifically used for:

[0238] Speech recognition is performed on the audio data to obtain the recognized text with timestamps. By distinguishing between the speech of teachers and students in the recognized text, text with timestamps and speaker tags is obtained.

[0239] Extract key video frames from silent video data and extract text information from the key video frames;

[0240] Based on a large language model, combined with text information extracted from key video frames, the text with timestamps and speaker tags is corrected semantically, grammatically, and in terms of key information. The corrected text is then processed by long sentence segmentation, formula standardization, and semantic-based paragraph division to obtain standardized, correct, time-stamped, and speaker-tagged structured text.

[0241] In one possible implementation, the knowledge point slicing module 1103, when performing knowledge point slicing on the target text data based on the large slicing model to obtain the knowledge point segmentation results, is specifically used for:

[0242] Based on the large-scale slicing model, combined with the knowledge graph and curriculum standards of the target subject, the target text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results;

[0243] Alternatively, based on the large-scale slicing model, and combined with the knowledge point list of the chapter to which the target course video belongs, the target text data can be sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0244] In one possible implementation, the knowledge point slicing module 1103, when performing knowledge point slicing on the target text data based on the large slicing model to obtain the knowledge point segmentation results, is specifically used for:

[0245] Based on the large-scale slicing model, the target text data is initially sliced ​​into knowledge points to obtain preliminary knowledge point segmentation results.

[0246] Based on the aforementioned large-scale model of knowledge points, the preliminary knowledge point segmentation results are optimized according to the knowledge graph and curriculum standards of the target subject, as well as the knowledge related to the knowledge point segments, to obtain the final knowledge point segmentation results.

[0247] In one possible implementation, the knowledge point slicing module 1103, when performing knowledge point slicing on the target text data based on the large slicing model to obtain the knowledge point segmentation results, is specifically used for:

[0248] Obtain the curriculum syllabus for the target subject;

[0249] The course syllabus is segmented according to the teaching stages to obtain the syllabus segmentation results;

[0250] Based on the teaching syllabus segmentation results, text fragments that are irrelevant to the knowledge point segments in the target text data are filtered out;

[0251] Based on the aforementioned large-scale slicing model, the filtered text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results.

[0252] In one possible implementation, the knowledge point segmentation result also includes: the title of the knowledge point micro-lesson, the summary of the knowledge point micro-lesson, and the text content of the knowledge point micro-lesson.

[0253] The micro-lesson extraction device may also include a micro-lesson data packaging module.

[0254] The micro-lesson data packaging module is used for:

[0255] Construct metadata that includes the title of the micro-lesson for the knowledge point, the summary of the micro-lesson for the knowledge point, and the tags for the knowledge point;

[0256] The metadata and the text content of the knowledge point micro-lessons are packaged and encapsulated with the knowledge point video micro-lessons to obtain complete knowledge point video micro-lesson resources;

[0257] The video micro-lesson resources for the knowledge points are associated with the corresponding knowledge points in the knowledge graph of the target subject.

[0258] In one possible implementation, the micro-lesson extraction device may further include: a typical example problem slice module and a typical example problem video micro-lesson extraction module; and / or, a scenario material slice module and a scenario material video micro-lesson extraction module; and / or, a classroom activity slice module and a classroom activity video micro-lesson extraction module.

[0259] The Classic Example Question Slicing Module is used to slice target text data based on the slicing big model and in combination with classic example question slicing rules, and obtain the segmentation result that includes at least the start and end times of the classic example question fragments.

[0260] The module for extracting exemplary problem video micro-lessons is used to extract exemplary problem video segments from the target course video based on the start and end times of the exemplary problem segments, thus obtaining exemplary problem video micro-lessons.

[0261] The scenario material slicing module is used to slice target text data based on the slicing model and scenario material slicing rules, and obtain the segmentation results that include at least the start and end times of the scenario material fragments.

[0262] The scenario material video micro-lesson extraction module is used to extract scenario material video segments from the target course video based on the start and end times of the scenario material segments, and obtain scenario material video micro-lessons.

[0263] The Classroom Activity Slicing Module is used to slice target text data into classroom activity segments based on the large slicing model and classroom activity slicing rules, and obtain segmentation results that include at least the start and end times of classroom activity segments.

[0264] The classroom activity video micro-lesson extraction module is used to extract classroom activity video segments from the target course video based on the start and end times of the classroom activity segments, and obtain classroom activity video micro-lessons.

[0265] In one possible implementation, the example question slicing module, based on the slicing model and combined with example question slicing rules, performs example question slicing on the target text data specifically for:

[0266] Based on the target text data, keywords related to the typical example questions were obtained, and the presentation method of the typical example questions in the classroom was determined.

[0267] Based on the large-scale slicing model, and combining the slicing rules of typical examples, keywords, and the way typical examples are presented in the classroom, the target text data is sliced ​​into typical examples.

[0268] In one possible implementation, the contextual content slicing module, when slicing target text data based on the large slicing model and in accordance with contextual content slicing rules, is specifically used for:

[0269] Determine the scenario type of the scenario material involved in the target text data;

[0270] Based on the large-scale slicing model, and combining the scenario type and the slicing rules corresponding to the scenario type in the scenario material slicing rules, the target text data is sliced ​​into scenario material slices.

[0271] In one possible implementation, the classroom activity slicing module, when slicing target text data into classroom activity segments based on the large slicing model and in accordance with classroom activity slicing rules, is specifically used for:

[0272] Determine the activity types of the classroom activities involved in the target text data;

[0273] Based on the large-scale slicing model, and combining the activity types and the slicing rules corresponding to the activity types in the classroom activity slicing rules, the target text data is sliced ​​into classroom activity segments.

[0274] In one possible implementation, the micro-lesson extraction device may further include: a module for acquiring and processing data related to classic example micro-lessons, and / or a module for acquiring and processing data related to scenario material micro-lessons, and / or a module for acquiring and processing data related to classroom activity micro-lessons.

[0275] The module for acquiring and processing data related to exemplary problem micro-lessons is used to obtain data on the solution process of exemplary problems based on the exemplary problem segments, and to associate the data on the solution process of exemplary problems with the exemplary problem video micro-lessons.

[0276] The module for acquiring and processing data related to scenario-based micro-lessons is used to obtain key elements and / or teaching intentions of scenario-based materials based on scenario-based material segments, and to associate the key elements and / or teaching intentions with scenario-based material video micro-lessons;

[0277] The module for acquiring and processing data related to classroom activity micro-lessons is used to obtain the activity process analysis results, and / or teaching objectives, and / or teacher-student interaction information based on classroom activity segments, and to associate the activity process analysis results, and / or teaching objectives, and / or teacher-student interaction information with the classroom activity video micro-lessons. The activity process analysis results include some or all of the following information: activity organization form information, activity participant role information, activity process information, and time allocation information for each stage of the activity.

[0278] In one possible implementation, the micro-lesson extraction device may also include a lesson plan document processing module.

[0279] The lesson plan document processing module is used for:

[0280] Obtain the lesson plan document corresponding to the target course video;

[0281] The lesson plan document is parsed to obtain the document parsing results;

[0282] Based on the document parsing results, identify the document structure of the lesson plan document;

[0283] Based on the document parsing results and document structure, structured lesson plan text data is obtained, and knowledge point slices are made from the structured lesson plan text data to obtain knowledge point documents.

[0284] Associate knowledge point documents with corresponding knowledge point video micro-lessons, and / or associate knowledge point documents with corresponding knowledge points in the knowledge graph of the target subject.

[0285] In one possible implementation, the micro-lesson extraction device may also include: a module for generating and processing audio micro-lessons of knowledge points.

[0286] The knowledge point audio micro-lesson generation and processing module is used for:

[0287] Generate a teaching script for the target knowledge point, where the target knowledge point is the knowledge point corresponding to the video micro-lesson of the stated knowledge point;

[0288] Based on the teaching script for the target knowledge points, generate teaching audio for the target knowledge points to obtain audio micro-lessons for the knowledge points;

[0289] Link the audio micro-lessons on knowledge points with the corresponding knowledge points in the knowledge graph of the target subject.

[0290] The knowledge point audio micro-lesson generation and processing module is specifically used for generating teaching scripts for target knowledge points when:

[0291] Based on the target knowledge points and the instructional design information for those target knowledge points, a teaching script is constructed to generate prompts. The instructional design information includes some or all of the following: dialogue design logic, plot arrangement information, content organization information, and character design and configuration information.

[0292] Input the prompt words for generating the teaching script into the large language model to generate the teaching script, and obtain the teaching script for the target knowledge point.

[0293] Since the micro-lesson extraction device provided in this application embodiment automatically extracts knowledge point micro-lessons from existing course videos based on the slice large model without manual intervention, the extraction cost of knowledge point micro-lessons is low, the extraction efficiency of knowledge point micro-lessons is high, and it can make full use of existing digital education resources. In addition, since the slice large model is trained on the general large model base by using training text data (text data corresponding to course video samples) labeled with real knowledge point segmentation results, and supplemented by thinking process data to obtain real knowledge point segmentation results, the slice large model can accurately slice the target text data into knowledge points. Therefore, the knowledge point segmentation results based on the slice large model can extract accurate knowledge point video micro-lessons.

[0294] This application also provides an electronic device, which may include at least one processor and a memory connected to the processor.

[0295] The processor may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application; the memory may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage.

[0296] The memory is used to store computer programs, and the processor is used to execute the computer programs so that the electronic device can implement the micro-lesson extraction method or the resource retrieval method provided in the above embodiments.

[0297] This application also provides a computer storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the micro-lesson extraction method or the resource retrieval method provided in the above embodiments.

[0298] This application also provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the micro-lesson extraction method provided in the above embodiments, or to implement the resource retrieval method provided in the above embodiments.

[0299] This application also provides a resource processing management system, such as... Figure 12 As shown, the resource processing management system may include terminals and servers.

[0300] The terminal is used to receive the target course video uploaded by the user and send the target course video to the server;

[0301] The server is used to extract micro-lesson resources from target course videos using the micro-lesson extraction method provided in the above embodiments.

[0302] It should be noted that the terminal is responsible for interacting with the user, while the server is mainly responsible for resource processing (such as video processing, document parsing, resource annotation, etc.). The server has some or all of the following resource processing capabilities: document parsing, video slicing, knowledge point slicing, audio transcription, image parsing, PPT extraction, original text normalization, key video frame extraction, course syllabus slicing, etc.

[0303] The server's resource processing pipeline has a four-layer architecture: access layer, processing layer, analysis layer, and storage layer. Specifically:

[0304] The access layer is used to access multi-source resources. It supports video formats (such as MP4, AVI, FLV, MOV and other mainstream formats) and can access course video resources. It also supports document formats (such as PDF, PPT, Word and other formats) and can access teaching documents. The access layer can also collect metadata (such as course information, teacher information, subject and grade level) and perform resource quality pre-checks (such as resolution detection, audio quality detection, duration verification and other functions).

[0305] The processing layer is used for multimodal information extraction, and it can perform ASR transcription, speaker separation, timestamp alignment, OCR recognition (such as text extraction from courseware, whiteboard recognition, subtitle extraction, etc.), video parsing (such as key video frame extraction, scene transition detection, screen content analysis, etc.), and formula recognition (such as extraction of mathematical, physical and chemical formulas, LaTeX format conversion, etc.).

[0306] The analysis layer is used for intelligent semantic analysis, which can perform text post-processing (such as sentence segmentation optimization, punctuation correction, terminology standardization, formula normalization, etc.), semantic understanding (such as paragraph segmentation, topic identification, logical relationship analysis, etc.), knowledge point location (such as knowledge point boundary identification by combining subject knowledge graphs and curriculum standards), and micro-lesson slicing (such as segmentation, integrity verification, title generation, summary generation, etc.).

[0307] The storage layer is used to store data. Specifically, it can store raw resources (such as raw videos, raw audio, raw documents, etc.), processing results (such as audio-to-text transcription, OCR results, key video frames, etc.), micro-lesson resources (such as video micro-lessons, micro-lesson metadata, etc.), and relational data (such as knowledge point associations, resource reference relationships, resource quality scores, etc.).

[0308] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0309] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0310] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0311] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for extracting micro-lessons, characterized in that, include: Obtain target course videos for the target subject; The content of the target course video is parsed to generate corresponding target text data with timestamps; Based on the slice-based large model, the target text data is sliced ​​into knowledge points to obtain knowledge point segmentation results. The knowledge point segmentation results include at least the start and end times of the knowledge point segments. The slice-based large model is trained on a general large model base using training text data labeled with real knowledge point segmentation results, supplemented by thinking process data that obtains the real knowledge point segmentation results. The training text data is time-stamped text data corresponding to course video samples of the target subject. The thinking process data includes the thinking path, derivation steps, typical misconceptions, and error correction process in the teaching process, which helps the slice-based large model to have teaching and research level thinking ability. Based on the start and end times of the knowledge point segments, knowledge point video segments are extracted from the target course video to obtain knowledge point video micro-lessons.

2. The micro-lesson extraction method according to claim 1, characterized in that, The process of parsing the content of the target course video and generating corresponding, timestamped target text data includes: The target course video is subjected to audio-video separation to obtain audio data and silent video data; By combining the video content of the silent video data, the audio data is processed into standardized, correct, time-stamped, and speaker-tagged structured text data to obtain the target text data corresponding to the target course video.

3. The micro-lesson extraction method according to claim 2, characterized in that, The process of combining the video content with the silent video data and processing the audio data into standardized, accurate, time-stamped, and speaker-tagged structured text data includes: The audio data is subjected to speech recognition to obtain time-stamped recognized text. By distinguishing between teacher and student speech in the recognized text, text with time stamp and speaker label is obtained. Extract key video frames from the silent video data, and extract text information from the key video frames; Based on a large language model, and combined with text information extracted from the key video frames, the text with timestamps and speaker tags is semantically, grammatically, and with key information corrected. The corrected text is then subjected to long sentence segmentation, formula standardization, and semantic-based paragraph division to obtain standardized, correct, time-stamped, and speaker-tagged structured text.

4. The micro-lesson extraction method according to claim 1, characterized in that, The slice-based large model performs knowledge point slicing on the target text data to obtain knowledge point segmentation results, including: Based on the large-scale slicing model, combined with the knowledge graph and curriculum standards of the target subject, the target text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results; Alternatively, based on the large-scale slicing model, and combined with the knowledge point list of the chapter to which the target course video belongs, the target text data can be sliced ​​into knowledge points to obtain the knowledge point segmentation results.

5. The micro-lesson extraction method according to claim 1, characterized in that, The slice-based large model performs knowledge point slicing on the target text data to obtain knowledge point segmentation results, including: Based on the large-scale slicing model, the target text data is initially sliced ​​into knowledge points to obtain preliminary knowledge point segmentation results. Based on the aforementioned large-scale model of knowledge points, the preliminary knowledge point segmentation results are optimized according to the knowledge graph and curriculum standards of the target subject, as well as the knowledge related to the knowledge point segments, to obtain the final knowledge point segmentation results.

6. The micro-lesson extraction method according to claim 1, characterized in that, The slice-based large model performs knowledge point slicing on the target text data to obtain knowledge point segmentation results, including: Obtain the curriculum syllabus for the target subject; The course syllabus is segmented according to the teaching stages to obtain the syllabus segmentation results; Based on the teaching syllabus segmentation results, text fragments that are irrelevant to the knowledge point segments in the target text data are filtered out; Based on the aforementioned large-scale slicing model, the filtered text data is sliced ​​into knowledge points to obtain the knowledge point segmentation results.

7. The micro-lesson extraction method according to claim 1, characterized in that, The knowledge point segmentation results also include: the title of the knowledge point micro-lesson, the summary of the knowledge point micro-lesson, and the text content of the knowledge point micro-lesson; The micro-lesson extraction method also includes: Construct metadata that includes the title of the micro-lesson for the knowledge point, the summary of the micro-lesson for the knowledge point, and the tags for the knowledge point; The metadata and the text content of the knowledge point micro-lessons are packaged and encapsulated with the knowledge point video micro-lessons to obtain complete knowledge point video micro-lesson resources; The video micro-lesson resources for the knowledge points are associated with the corresponding knowledge points in the knowledge graph of the target subject.

8. The micro-lesson extraction method according to claim 1, characterized in that, Also includes: Based on the aforementioned large-scale slicing model and combined with the typical example question slicing rules, the target text data is sliced ​​into typical example questions to obtain a segmentation result that includes at least the start and end times of the typical example question segments. Then, typical example question video segments are extracted from the target course video according to the start and end times of the typical example question segments to obtain typical example question video micro-lessons. And / or, based on the slice big model and combined with the scenario material slicing rules, the target text data is sliced ​​into scenario materials to obtain a segmentation result that includes at least the start and end times of the scenario material segments, and scenario material video segments are extracted from the target course video according to the start and end times of the scenario material segments to obtain scenario material video micro-lessons; And / or, based on the slice model and combined with the classroom activity slice rules, the target text data is sliced ​​into classroom activities to obtain a segmentation result that includes at least the start and end times of the classroom activity segments. Classroom activity video segments are then extracted from the target course video according to the start and end times of the classroom activity segments to obtain classroom activity video micro-lessons.

9. The micro-lesson extraction method according to claim 8, characterized in that, The step of slicing the target text data into typical example questions based on the slice model and the typical example question slicing rules includes: Based on the target text data, keywords related to the typical example questions are obtained, and the presentation method of the typical example questions in the classroom is determined; based on the segmentation model, combined with the typical example question segmentation rules, the keywords, and the presentation method of the typical example questions in the classroom, the target text data is segmented into typical example questions. The step of slicing the target text data into contextual elements based on the slice model and in conjunction with contextual element slicing rules includes: Determine the scenario type of the scenario material involved in the target text data; based on the slice model, and combining the scenario type with the slice rules corresponding to the scenario type in the scenario material slicing rules, slice the target text data into scenario material slices; The step of slicing the target text data into classroom activity segments based on the slice model and combined with classroom activity slicing rules includes: Determine the activity type of the classroom activities involved in the target text data; based on the slice model, and combining the activity type with the slice rules corresponding to the activity type in the classroom activity slice rules, slice the target text data into classroom activities.

10. The micro-lesson extraction method according to claim 9, characterized in that, Also includes: The solution process data of the example problems is obtained from the example problem fragments, and the solution process data of the example problems is associated with the example problem video micro-lessons; Based on the scenario material fragments, obtain the key elements and / or teaching intentions of the scenario material, and associate the key elements and / or teaching intentions with the scenario material video micro-lessons; Based on the classroom activity segments, obtain the activity process analysis results, and / or teaching objectives, and / or teacher-student interaction information, and associate the activity process analysis results, and / or the teaching objectives, and / or the teacher-student interaction information with the classroom activity video micro-lesson. The activity process analysis results include some or all of the following information: activity organization form information, activity participant role information, activity process information, and time allocation information for each stage of the activity.

11. The micro-lesson extraction method according to claim 1, characterized in that, Also includes: Obtain the lesson plan document corresponding to the target course video; The lesson plan document is parsed to obtain the document parsing results; Based on the document parsing results, identify the document structure of the lesson plan document; Based on the document parsing results and the document structure, structured lesson plan text data is obtained, and the structured lesson plan text data is sliced ​​into knowledge points to obtain knowledge point documents. Associate the knowledge point document with the corresponding knowledge point video micro-lesson, and / or associate the knowledge point document with the corresponding knowledge point in the knowledge graph of the target subject.

12. The micro-lesson extraction method according to claim 1, characterized in that, Also includes: Generate a teaching script for the target knowledge point, where the target knowledge point is the knowledge point corresponding to the video micro-lesson of the knowledge point; Based on the teaching script for the target knowledge point, generate the teaching audio for the target knowledge point to obtain the audio micro-lesson for the knowledge point; The audio micro-lessons on the knowledge points are associated with the corresponding knowledge points in the knowledge graph of the target subject.

13. The micro-lesson extraction method according to claim 12, characterized in that, The teaching script for generating the target knowledge points includes: Based on the target knowledge points and the instructional design information of the target knowledge points, a teaching script is constructed to generate prompts. The instructional design information includes some or all of the following: dialogue design logic, plot arrangement information, content organization information, and character design and configuration information. The teaching script generation prompts are input into a large language model to generate the teaching script, thus obtaining the teaching script for the target knowledge point.

14. A resource retrieval method, characterized in that, include: Obtain query information related to knowledge points input by the user; Retrieve resources that match the query information from a pre-built resource library, wherein the resource library includes resources associated with knowledge points in a subject knowledge graph, and the resources in the resource library include micro-lesson resources extracted using the micro-lesson extraction method as described in any one of claims 1 to 13.

15. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program to enable the electronic device to implement the micro-lesson extraction method as described in any one of claims 1 to 13, or to implement the resource retrieval method as described in claim 14.

16. A computer storage medium, characterized in that, The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the micro-lesson extraction method as described in any one of claims 1 to 13, or the resource retrieval method as described in claim 14.

17. A computer program product, characterized in that, It includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the micro-lesson extraction method as described in any one of claims 1 to 13, or to implement the resource retrieval method as described in claim 14.

18. A resource processing management system, characterized in that, include: Terminals and servers; The terminal is used to receive target course videos uploaded by users and upload the target course videos to the server; The server is used to extract micro-lessons from the target course video using the micro-lesson extraction method as described in any one of claims 1 to 13.

Images