A multi-dimensional feature-based course resource precise adaptation method

Abstract

This invention relates to the field of multimedia data processing technology and discloses a method for precise adaptation of course resources based on multi-dimensional features. The method includes: a server segmenting course resources to generate data objects of different presentation modalities and writing them into course information entropy parameters; a client pulling data for playback and listening for underlying physical hardware interruption events, calculating and generating a real-time cognitive load margin scalar value based on interaction demand factors; when the differential slope of the value triggers a warning threshold, the system preloads low-density presentation modal data into a secondary buffer and changes the physical memory read pointer of the decoding input source at the slice boundary to complete the modal switching; when the cognitive load margin recovers to a safe threshold, the corresponding plain text data is retrieved based on the dimensionality reduction record for rendering compensation. This invention avoids streaming media playback stuttering through underlying parameter scheduling and memory addressing changes, and utilizes a dynamic compensation mechanism to ensure the integrity of course information transmission.

Description

Technical Field

[0001] This invention relates to the field of multimedia data processing technology, specifically to a method for precise adaptation of course resources based on multi-dimensional features. Background Technology

[0002] Existing multimedia course resource playback systems employ adaptive bitrate technology, relying primarily on network transmission speed as a single dimension for dynamic adjustment of streaming media resolution. This mechanism ignores the actual cognitive load of users receiving complex course content. Even when network conditions are good, if users are cognitively overloaded due to frequent interactions, the system continues to push high-information-density media stream data, leading to decreased content delivery efficiency. Furthermore, when the system performs routine image quality downscaling or modal switching operations, it needs to re-initiate network requests and reset buffer data at the application layer, which can cause underlying decoding interruptions and playback pauses. On the other hand, the downscaling operations such as frame skipping adopted by the system to reduce transmission load or playback pressure inevitably result in the loss of some detailed information in the original media stream. After the user interaction pressure is relieved, existing systems lack a structured extraction and asynchronous rendering compensation mechanism for this missing content, leading to omissions of core course knowledge points. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for precise adaptation of course resources based on multi-dimensional features, which solves the problems that existing streaming media playback systems cannot smoothly switch between dimensions based on user cognitive load and lack a content compensation mechanism after dimension reduction.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] This invention provides a method for precise adaptation of course resources based on multi-dimensional features, the method comprising:

[0006] The server cluster performs segmentation operations on the course resource files to generate data objects with different presentation modalities, calculates and obtains the course information entropy density parameters and interaction requirement factors, and writes them into the file header area of ​​each presentation modal data object.

[0007] The client terminal pulls and presents modal data objects into the main buffer for the rendering module to decode and play. The listening module collects the underlying physical hardware interruption events and calculates the actual interaction event rate and human-computer interaction information entropy. The evaluation module combines the interaction demand factors to calculate and generate a real-time cognitive load margin scalar value.

[0008] When the first-order time difference slope of the real-time cognitive load margin scalar value meets the internally set warning threshold condition, lower-density presentation modal data is preloaded into the secondary buffer.

[0009] When the dimensionality reduction threshold condition is met at the slice boundary time node, the memory read pointer of the decoding input source is located to the secondary buffer to achieve mode switching;

[0010] Once the real-time cognitive load margin scalar value recovers to the safe recovery threshold, the corresponding third presentation modal data retrieval and text character rendering and display operations are initiated.

[0011] In this invention, the server modally segments and writes parameters to the media stream, quantifying the complexity of the content into specific metrics. The client listens for underlying hardware interrupt events to obtain peripheral operation data, calculates the entropy of human-computer interaction information, and thus determines the user's cognitive load margin. By performing time difference calculations on this margin, the system can suspend the main network thread when the data attenuation slope triggers a threshold, allocating transmission bandwidth to the secondary buffer to preload low-density data. In the playback decoding stage, the system directly achieves modal switching by changing the physical address of the underlying memory read pointer. When the system assesses that the user's cognitive state has recovered safely, it retrieves the knowledge nodes that were skipped during dimensionality reduction based on log records, asynchronously requests third-level presentation modal data (i.e., plain text data), and renders and displays it in an independent layer to achieve knowledge compensation. This logic combines underlying hardware interrupt status, bandwidth scheduling, and memory pointer addressing, ensuring complete information transmission while preventing playback stuttering.

[0012] Furthermore, the steps of the server cluster to perform segmentation operations on course resource files to generate data objects with different presentation modalities include: retaining the high frame rate video track and complete audio track of the original data stream of the slice sequence entity to generate a first presentation modal data object; extracting the core key frames of the video and mixing them with the original audio track to generate a second presentation modal data object; and using a speech recognition model to map the audio stream data into a core text character stream and encapsulate it to generate a third presentation modal data object.

[0013] Furthermore, the step of calculating and obtaining the course information entropy density parameter and the interaction demand factor and writing them into the file header area of ​​each presentation modality data object includes: calculating the course information entropy density parameter based on the video payload bit rate and audio payload bit rate of the presentation modality data object in the transmission state; setting the interaction demand factor in conjunction with the reference frequency expected by the front-end operator to generate the underlying peripheral hardware interrupt operation; and encapsulating the course information entropy density parameter, the interaction demand factor, and the knowledge node identification code into a joint parameter label and writing it into the file header area to achieve control parameter feature binding.

[0014] Furthermore, the steps of the monitoring module in collecting underlying physical hardware interruption events and calculating the actual interaction event rate and human-computer interaction information entropy include: establishing a sliding time window with a fixed time span parameter to extract the total number of physical hardware interruption events to obtain the actual interaction event rate; extracting the time interval sequence of adjacent physical hardware interruption events and converting continuous mouse coordinate offset data into a spatial trajectory curvature distribution matrix and projecting it into multiple pre-divided discrete quantization intervals; statistically analyzing the probability distribution of feature values ​​falling into each specific quantization interval and substituting it into the Shannon information entropy calculation model to obtain the human-computer interaction information entropy.

[0015] Furthermore, the step of the evaluation module in calculating and generating a real-time cognitive load margin scalar value by combining the interaction demand factor includes: using the interaction demand factor and the absolute resting baseline constant modulation to obtain a dynamic baseline modulated by the presentation modality attribute; generating a resting penalty constraint value based on the difference range between the interaction demand factor and the actual interaction event rate value; and summarizing the dynamic baseline, the human-computer interaction information entropy value, and the resting penalty constraint value to calculate and output the real-time cognitive load margin scalar value based on an algebraic subtraction weighted model.

[0016] Furthermore, the step of preloading lower-density presentation modal data into the secondary buffer when the first-order time difference slope of the real-time cognitive load margin scalar value meets the internally set warning threshold condition includes: when the calculated first-order time difference slope value is less than or equal to the manually configured negative constant parameter or the real-time cognitive load margin scalar value of the current calculation cycle is less than the set absolute warning bottom line, it is determined that the warning threshold condition is met, the data retrieval task thread of the network channel corresponding to the main buffer is suspended and the connection bandwidth resources of the underlying transmission control protocol are released, and the secondary buffer is controlled to write presentation modal data entities with the same timestamp and course information entropy density parameter lower than the target upper limit threshold into the independent memory stack area to complete the data preloading.

[0017] Furthermore, the step of positioning the memory read pointer of the decoding input source to the secondary buffer to achieve modality switching when the dimensionality reduction threshold condition is met at the slice boundary time node includes: monitoring the local graphics rendering clock progress state to capture the boundary time node of the slice sequence entity; determining that the dimensionality reduction threshold condition is met when the scalar value of the course information entropy density parameter of the next pending main modality slice is greater than the product of the real-time cognitive load margin scalar value and the margin scaling constant; and generating a high-priority memory address overwrite instruction carrying the physical base address to control the memory read pointer of the underlying decoding input source to change to the starting physical address of the secondary buffer when the secondary buffer is confirmed to be ready.

[0018] Furthermore, the steps of initiating the corresponding third presentation modality data retrieval and text character rendering and display operation after the real-time cognitive load margin scalar value recovers to the safety recovery threshold include: capturing the memory address overwrite instruction, extracting the absolute timestamp sequence of the dimensionality reduction switch and the knowledge node identification code, and persistently writing them into the damaged record mapping table of the local database; extracting the recovery state stability parameter based on the current real-time cognitive load margin scalar value and the safety recovery threshold constant; retrieving the corresponding third presentation modality data entity from the server cluster based on the parameter and splicing it into a complete semantic character stream; and rendering and generating an independent text prompt sidebar with an independent scroll bar control in an independent layer of the front-end graphical user interface.

[0019] Furthermore, the steps of capturing the memory address overwrite instruction, extracting the absolute timestamp sequence of the dimensionality reduction switch and the knowledge node identification code, and persistently writing them into the damaged record mapping table of the local database include: continuously listening to the message bus at the application layer and triggering an interrupt callback function when the memory address overwrite instruction is captured; combining the absolute timestamp sequence and the knowledge node identification code into a structured log object; and calling the underlying file writing interface to write the structured log object into the damaged record mapping table of the local database to perform feature mapping characterization of the skipped nodes of the original presentation modal data.

[0020] Furthermore, the step of rendering and generating an independent text prompt sidebar with an independent scrollbar control in an independent layer of the front-end graphical user interface includes: sending the complete semantic character stream to the local cache queue of the client terminal, detecting whether there is an instantiated independent text prompt sidebar in the front-end graphical user interface, and if not, allocating a graphic overlay layer with independent depth coordinates in the non-overlapping edge safe area of ​​the main media stream layer for transparency blending rendering output, and synchronously calling the underlying file writing interface to update the associated records in the damaged record mapping table to the processed state to block the duplicate request logic.

[0021] This invention provides a method for precise adaptation of course resources based on multi-dimensional features. It has the following beneficial effects:

[0022] 1. This invention calculates the human-computer interaction information entropy by collecting interrupt events of the underlying physical hardware and generates a real-time cognitive load margin scalar value by combining interaction demand factors. This mechanism does not rely on explicit feedback from the operator or operation statistics of the application layer, but directly uses the physical input frequency of the underlying peripheral hardware to quantify the current interaction load state, so that the system can objectively determine whether the user is on the edge of cognitive overload, providing accurate data basis for subsequent streaming media dimensionality reduction scheduling.

[0023] 2. When the first-order time difference slope of the cognitive load margin reaches the warning threshold, this invention releases bandwidth by suspending the main buffer network thread, preloads low-density modal data in a targeted manner by the secondary buffer, and directly changes the physical memory read pointer of the decoding input source at the slice boundary. This processing logic starts from network bandwidth scheduling and underlying memory address redirection, replacing the traditional streaming media process of re-establishing a connection at the application layer, avoiding the buffering stutters common during modal switching, and ensuring the continuity of playback.

[0024] 3. When dimensionality reduction switching occurs, this invention extracts the absolute timestamp sequence and knowledge node identifier and writes them into the mapping table of the local database. When the cognitive load is detected to rise back to a safe threshold, the corresponding plain text data is retrieved from the server based on the record for layer overlay rendering. This processing method can automatically trace back and supplement the content skipped during dimensionality reduction in a lightweight text form after the user's interaction burden is reduced, preventing the omission of core course information due to the system's adaptive reduction of bitrate. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall functional module architecture of the system of the present invention;

[0026] Figure 2 This is a schematic diagram of the overall execution flow of the multimedia resource dimensionality reduction presentation method based on cognitive load of the present invention;

[0027] Figure 3 This is a schematic diagram of the logical architecture for the server-side multimodal media resource preprocessing, segmentation, and dimensionality reduction co-encapsulation of the present invention;

[0028] Figure 4 This is a schematic diagram of the interaction architecture between the client-side multidimensional cognitive load state assessment engine and the underlying hardware of this invention;

[0029] Figure 5 This is a schematic diagram of the timing waveforms for state differential early warning and network bandwidth scheduling in this invention. In this diagram, a is a schematic diagram of the real-time cognitive load margin sequence and absolute bottom line tracking trajectory, b is a schematic diagram of the first-order time differential slope evolution and the underlying threshold trigger level, and c is a schematic diagram of the dynamic reconfiguration state of the network interface primary and secondary buffer bandwidth resources.

[0030] Figure 6 This is a schematic diagram of the underlying hardware memory addressing redirection and graphics pipeline decoding clock control topology of the present invention;

[0031] Figure 7 This is a schematic diagram of the state recovery assessment and discrete integral monitoring waveform of the present invention, wherein a is a schematic diagram of the evolution of continuous cognitive load margin and safety boundary calibration trajectory, b is a schematic diagram of the state timing of the bottom discrete step function Boolean mapping, and c is a schematic diagram of the state monitoring sliding window integral operation and compensation wake-up triggering mechanism.

[0032] Figure 8 This is a schematic diagram of the multimodal continuous adaptive switching evolution process based on the dynamic fluctuation trajectory of cognitive load according to the present invention;

[0033] Figure 9 This diagram illustrates the evaluation results of a comparative experiment between the method of this invention and existing mainstream streaming media playback systems.

[0034] The meanings of the labels in the diagram are as follows:

[0035] 100. Server Cluster; 110. Preprocessing Module; 120. Calibration Module; 130. Storage Node; 200. Client Terminal; 210. Listening Module; 220. Evaluation Module; 230. Scheduling Module; 240. Main Buffer; 250. Secondary Buffer; 260. Matching Module; 270. Rendering Module; 280. Compensation Module. Detailed Implementation

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] See attached document Figure 1 The present invention provides a course resource precise adaptation system based on multi-dimensional features. The underlying hardware environment for executing the method of the present invention includes a server cluster 100 and a client terminal 200 that establish a communication connection through a network communication link.

[0038] The server cluster 100 is deployed on a cloud server physical node and is used to provide a hardware computing power platform for offline data computing and resource allocation for the method of the present invention. The server cluster 100 is internally configured with a preprocessing module 110, a calibration module 120 and a storage node 130. The preprocessing module 110 is a processor process unit with computing power for disassembling and recoding digital multimedia files.

[0039] The calibration module 120 and the preprocessing module 110 are associated through the server's internal data bus. The storage node 130 is composed of an array of persistent storage media, used to store the processed data blocks and to provide the client terminal 200 with a data access interface based on the Hypertext Transfer Protocol.

[0040] The client terminal 200 runs on a physical computing device on the user side. The physical computing device is equipped with standard human-machine interface hardware devices on external interfaces such as the general serial bus. The client terminal 200 is divided into a listening module 210, an evaluation module 220, a scheduling module 230, a main buffer 240, a secondary buffer 250, a matching module 260, a rendering module 270, and a compensation module 280.

[0041] The listening module 210 encapsulates hardware interrupt vector call logic at the operating system kernel level. The main buffer 240 and the secondary buffer 250 are two independently addressable data spaces statically allocated in the physical memory area of ​​the client terminal 200. The rendering module 270 controls the graphics computing unit and display output device of the client terminal 200. The evaluation module 220, the scheduling module 230 and the matching module 260 are configured to control the data flow and memory read / write execution pointers between the main buffer 240, the secondary buffer 250 and the network interface controller.

[0042] See attached document Figure 2 This invention provides a method for precise adaptation of course resources based on multi-dimensional features. It relies on the aforementioned hardware system architecture and calls the aforementioned modules for execution, including the following steps:

[0043] S100, the server cluster 100 receives the original audiovisual format course resource file. The preprocessing module 110 performs data frame segmentation operation on the course resource file according to a fixed time step parameter to generate a slice sequence containing continuous timestamps. For the slice entity corresponding to a single timestamp, the information dimensionality reduction and stripping calculation is performed to output the first presentation modality, the second presentation modality, and the third presentation modality. The calibration module 120 obtains the channel transmission capacity occupied by each presentation modality and outputs the course information entropy density parameter through algebraic operation. Combined with the manually set operation frequency expectation benchmark, the interaction demand factor is extracted. The course information entropy density parameter, the interaction demand factor, and the manually preset knowledge node identification code are encapsulated into a joint parameter label and written into the file header area of ​​each presentation modality data object. The storage node 130 writes the data object into its persistent storage medium.

[0044] S200, the scheduling module 230 of the client terminal 200 sends a data retrieval request to the storage node 130 and stores it in the main buffer 240 for the rendering module 270 to perform decoding and playback tasks. The monitoring module 210 continuously collects the mouse coordinate offset numerical sequence and key trigger timestamp sequence of the external standard human-machine interface hardware device through the hardware interrupt pipeline. The evaluation module 220 establishes a sliding time window with a fixed time span and statistically extracts the frequency and time interval distribution probability of physical hardware trigger actions within the sliding time window. It calculates and generates the actual interaction event rate value and the human-machine interaction information entropy value. It reads the interaction demand factor carried in the header area of ​​the current decoded modal file of the rendering module 270, generates a resting penalty term based on the difference range between the interaction demand factor and the actual interaction event rate value, and calculates and outputs the real-time cognitive load margin representing the time period in combination with the human-machine interaction information entropy value.

[0045] S300, the scheduling module 230 obtains the real-time cognitive load margin sequence within multiple consecutive sliding time windows output by the evaluation module 220, calculates and extracts the first-order time difference slope, and when it is determined that the first-order time difference slope meets the internally set warning threshold condition, it blocks the data packet acquisition process of the network channel corresponding to the main buffer 240, allocates the idle connection bandwidth resources of the underlying transmission control protocol to the pull queue corresponding to the secondary buffer 250, and requests presentation modal data with the same timestamp as the next slice to be processed but with a lower course information entropy density parameter from the storage node 130, and stores the acquired data packets into the secondary buffer 250 to complete the preloading.

[0046] S400, when the rendering module 270 advances to the slice boundary time node based on its internal rendering clock, it triggers a state update. The matching module 260 extracts the course information entropy density of the next pending main modality slice in the data queue, and performs a conditional judgment on the course information entropy density and the real-time cognitive load margin of the current time period. When the judgment meets the dimensionality reduction threshold condition, it issues a memory address overwrite instruction to the rendering module 270. The rendering module 270 changes the memory read pointer of its underlying decoding input source from the first address of the main buffer 240 to the first address of the secondary buffer 250 according to the instruction requirements to realize the frame level switching of the output presentation modality.

[0047] S500, after receiving the memory address overwrite instruction, the compensation module 280 extracts the current timestamp sequence and the associated knowledge node identification code, and stores them in the local database in the format of a structured log. The evaluation module 220 continuously calculates and outputs the real-time cognitive load margin. When the real-time cognitive load margin rises above the safety recovery threshold and remains stable within a set period, the compensation module 280 extracts the knowledge node identification code and the associated timestamp sequence from the local database, retrieves the corresponding third presentation modality data based on the timestamp sequence, and completes the text character rendering and display operation at an independent level on the front-end display interface through the rendering module 270.

[0048] See attached document Figure 3 In specific implementations, step S100 provided by the present invention may include the following steps:

[0049] S110, the server cluster 100 obtains the original format multimedia course file and inputs it into the decoding pipeline of the preprocessing module 110. The preprocessing module 110 reads the internally preset fixed time step parameter with a value range. In this embodiment, the value range of the fixed time step parameter is 2 seconds to 10 seconds. The preprocessing module 110 performs equal division and cutting operations on the multimedia course file along the time axis according to the set time span, and then outputs the slice sequence entity arranged continuously in time sequence. The multimedia course file contains independently decoded keyframes and prediction frames that depend on reference frames. The preprocessing module 110 performs frame alignment and cutting operations based on the image group boundary. For the frame alignment and cutting operations of the image group boundary, those skilled in the art can use existing streaming media slice alignment algorithms to construct the corresponding image group boundary constraint logic.

[0050] S120, the preprocessing module 110 starts a multi-threaded architecture to perform parallel information dimensionality reduction and stripping calculations on the slice sequence entity. It generates a first presentation modal data object by retaining the high frame rate video track and complete audio track of the original data stream of the slice sequence entity. The preprocessing module 110 calls the conventional inter-frame difference algorithm to extract the video core key frames of the slice sequence entity, and timestamps the video core key frames and the original audio track to generate a second presentation modal data object containing static images and text and continuous audio.

[0051] The preprocessing module 110 strips the audio track data from the slice sequence entity and extracts its Mel-frequency cepstral coefficients as the input feature sequence. This is then input into a pre-deployed speech recognition model based on an end-to-end full attention transform network architecture. This speech recognition model is configured with an encoder layer and a decoder layer containing a multi-head self-attention mechanism. The encoder layer performs acoustic context encoding on the one-dimensional temporal speech feature sequence to extract the hidden layer feature vector. The decoder layer receives the hidden layer feature vector and uses an autoregressive mechanism combined with historical character states to predict the character probability distribution matrix of the current time step. Finally, it outputs the core text character stream with the highest probability path through a beam search algorithm.

[0052] Before deployment, the aforementioned speech recognition model uses an open-source Chinese speech corpus and manually annotated text labels as training samples. It employs the cross-entropy loss function to calculate the error between the predicted character probability distribution and the actual text labels. The network weight parameters of the encoder and decoder layers are iteratively updated through backpropagation combined with an adaptive moment estimation optimizer until the loss value converges to a set range, enabling the model to map audio stream data into text stream data. The preprocessing module 110 obtains the core text character stream output by the speech recognition model and encapsulates it into a preset binary structured media container format with a file header region, thereby generating a third presentation modality data object.

[0053] S130, the calibration module 120 acquires the generated data objects of each presentation modality and calculates the course information entropy density parameter, which characterizes the amount of audiovisual information, based on channel transmission theory. The calibration module 120 reads the video payload bit rate and audio payload bit rate of the current presentation modality data object in the transmission state, and constructs a calculation model based on the amount of audiovisual information of a specific presentation modality per unit time. The specific calculation formula is as follows:

[0054] ;

[0055] In the formula, This refers to the course information entropy density parameter. The visual information perception weight constant is defined as follows: In this embodiment, the value of the visual information perception weight constant ranges from 0.6 to 0.8. This refers to the real-time video payload bit rate. The upper limit constant for video transmission bandwidth is 2000Kbps to 5000Kbps in this embodiment. The auditory information perception weighting constant is, in this embodiment, the value of the auditory information perception weighting constant ranges from 0.2 to 0.4; For real-time audio payload bit rate; The upper limit constant for audio transmission bandwidth is 128Kbps to 320Kbps in this embodiment.

[0056] Based on the algebraic architecture and the data composition structure analysis of each presentation modality, the first presentation modality data object retains all video data, resulting in the calculated course information entropy density parameter value being in the high range. The third presentation modality data object does not contain video and audio media frames, resulting in the corresponding real-time video payload bit rate and real-time audio payload bit rate values ​​being both 0, and the corresponding calculated course information entropy density parameter value being close to 0. The system has completed the quantitative mapping operation of the complexity of multi-dimensional presentation modality data content at the data calculation level.

[0057] S140, the calibration module 120 extracts the interaction demand factor based on the expected operation frequency of the terminal's graphical user interface. This interaction demand factor represents the reference frequency value of the expected operation of the underlying peripheral hardware interruption by the front-end operator under a specific presentation mode. The first presentation mode data object is characterized by highly dense audiovisual information. When the operator receives such continuous dynamic screen information, he tends to maintain visual focus and does not perform physical button intervention. The calibration module 120 assigns a low-frequency interaction demand factor constant with a value close to 0 according to objective operation rules. In this embodiment, the value range of the low-frequency interaction demand factor constant is 0.01 to 0.05.

[0058] The third presentation modal data object is presented as a plain text layout structure. Due to the fixed physical size of the visible area of ​​the physical display device, when the user reads long text, physical peripheral hardware response actions such as dragging the scroll bar or pressing the page-turning key will occur. The calibration module 120 assigns a high-frequency interaction demand factor with a higher value than that of the first presentation modal to the third presentation modal data object through mapping logic. In this embodiment, the value range of the high-frequency interaction demand factor is 0.6 to 0.9. The threshold boundary of the interaction demand factor is dynamically constrained based on the user operation behavior statistical log interval initially preset by the system.

[0059] S150, the calibration module 120 combines the calculated course information entropy density parameter with the interaction demand factor and the manually preset knowledge node identification code to generate a joint parameter label of fixed byte length. It parses the underlying binary file structure of each presentation modality data object and extracts the offset address of the file header area storing media metadata information. It uses memory write instructions to directly inject the joint parameter label into the reserved extended data structure field of the file header area to realize the binding of control parameter features and media entity stream data.

[0060] All presentation modal data objects that have undergone the joint parameter tag injection step are encapsulated into standard transmission protocol message format and transmitted to the remote storage node 130 via the server's internal data bus. The control component of storage node 130 constructs a structured database index association table using a composite primary key constructed from a continuous timestamp sequence, knowledge node identification code, and modal identifier. It then persistently writes each presentation modal data object carrying the joint parameter tag into the underlying array disk physical medium. The open network-side Hypertext Transfer Protocol port establishes a network data response transmission channel for the client terminal, waiting for the application layer program to initiate a data retrieval request.

[0061] See attached document Figure 4 In specific implementations, step S200 provided by the present invention may include the following steps:

[0062] S210, during the operation of the client terminal 200, the scheduling module 230 initiates a Hypertext Transfer Protocol request to the storage node 130 to retrieve the presentation modal data object and store it in the main buffer 240 for the rendering module 270 to execute the local decoding buffer playback process. The listening module 210 calls the global hook application interface of the operating system to build a physical interrupt event interception pipeline for standard human-machine interface hardware devices, intercepting the mouse coordinate offset event identifier transmitted by the underlying hardware driver and the absolute system timestamp sequence data attached to the key trigger action. The evaluation module 220 establishes a sliding time window with a fixed time span parameter in the time domain dimension and extracts the total number of physical hardware interrupt events accumulated within the window. It divides this number by the width of the time window to obtain the actual interaction event rate value. The formula for calculating the actual interaction event rate is:

[0063] ;

[0064] In the formula, The actual value of the interactive event; The total number of physical hardware interrupt events within the sliding time window; This is a constant for the length of the sliding time window.

[0065] The sliding time window length constant is set as a threshold based on the instruction execution throughput of the system processor in a single context switch cycle, and the value range is configured between 500 milliseconds and 2000 milliseconds. The above time window division mechanism establishes a data buffer debouncing pipeline, which transforms the scattered and randomly triggered underlying asynchronous hardware interrupt signals into a continuous scalar characteristic input stream with a stable time domain reference.

[0066] S220, the evaluation module 220 reads the actual interaction event rate value and executes the algorithm branch decision operation. Under the condition that the actual interaction event rate value is greater than 0, it performs probability distribution dimensionality reduction of the physical feature sequence, extracts the time interval sequence of adjacent physical hardware interruption events, and converts the continuous mouse coordinate offset data into a spatial trajectory curvature distribution matrix. The above feature values ​​are projected into multiple discrete quantization intervals pre-divided by the system according to equal step sizes. The probability distribution of the feature values ​​falling into each specific quantization interval is statistically analyzed and substituted into the Shannon information entropy calculation model to obtain the human-computer interaction information entropy value. Under the boundary condition that the actual interaction event rate value is equal to 0, the human-computer interaction information entropy value is directly assigned to 0 to avoid the risk of program abnormality. The calculation formula of this information entropy is:

[0067] ;

[0068] In the formula, The information entropy value for human-computer interaction; This is the index number variable for the quantization interval; A constant used to limit the total number of quantization intervals during spatial dimensionality reduction; For the eigenvalue to fall into the first The probability value within a quantization interval.

[0069] The value of the upper limit constant of the total number of quantization intervals is determined by the data bit width of the input feature sequence. In this embodiment, the value of the upper limit constant of the total number of quantization intervals is in the range of 8 to 32. The probability value of the feature value falling into a specific quantization interval is in the closed interval of 0 to 1. The underlying algorithm that uses information entropy to evaluate the disorder of features maps the irregular physical coordinate trajectory sequence and the tapping time difference sequence into numerical parameters that characterize the stability of the peripheral operating state.

[0070] S230, the evaluation module 220 reads the rendering modal data object currently decoded by the rendering module 270 and extracts the encapsulated interaction requirement factor written in the file header area. Using this interaction requirement factor in conjunction with the absolute resting reference constant during the initial system calibration, the dynamic baseline at the current moment is calculated. The corresponding calculation formula is as follows:

[0071] ;

[0072] In the formula, The output value is a dynamic baseline modulated by the presented modal attributes; The time variable is the current calculation time. In order to combine the absolute resting reference constant calibrated with the device's basic hardware specifications, in this embodiment, the value range of the absolute resting reference constant is 50 to 100. This is the first adjustment coefficient; This is the interaction demand factor parameter at the current moment.

[0073] The first adjustment coefficient ranges from 0.1 to 0.5. The evaluation module 220 assigns an offset compensation value to the evaluation baseline based on the differences in the operator's behavior when facing different media content.

[0074] The evaluation module 220 generates an independent rest penalty parameter based on the difference between the interaction demand factor and the actual interaction event rate. The calculation formula for this rest penalty is as follows:

[0075] ;

[0076] In the formula, This is the rest penalty constraint value; The time variable is the current calculation time. The penalty amplitude coefficient is used in this embodiment, and its value ranges from 1.5 to 3.0. The expected matching coefficient is defined as the coefficient of performance. In this embodiment, the expected matching coefficient ranges from 0.8 to 1.2. The interaction demand factor parameter at the current moment; This represents the actual interaction event rate.

[0077] The resting penalty term constraint value is limited to the non-negative range by mathematical functions. The system uses manually calibrated user operation behavior logs to perform linear regression calculations to determine the specific positive constant value of the penalty magnitude coefficient. The expected matching coefficient is used to compensate for the mapping ratio deviation between the hardware sampling frequency and the expected benchmark of system operation at the numerical level, and to transform the hardware non-response cycle that exceeds the normal response threshold time into the decay characteristic index of the state margin.

[0078] S240, the evaluation module 220 summarizes the dynamic baseline data generated in the current operation cycle, as well as the human-computer interaction information entropy value and resting penalty term parameter. Based on the algebraic subtraction weighted model, it calculates and outputs the real-time cognitive load margin index representing this time domain cycle. The formula for calculating the real-time cognitive load margin is:

[0079] ;

[0080] In the formula, To enable real-time monitoring of load margin scalar values; The time variable is the current calculation time. The output value is a dynamic baseline modulated by the presented modal attributes; The empirical mapping constant is calibrated based on the local device's graphics rendering frequency. In this embodiment, the value range of the empirical mapping constant is 0.5 to 1.5. The information entropy value for human-computer interaction; This is the constraint value for the resting penalty term.

[0081] The evaluation module 220 encapsulates the real-time cognitive load margin scalar value generated by continuous calculation along the time axis into a finite-length data sequence with a structured definition and writes it into the shared memory bus of the operating system. The scheduling module 230 uses the data sequence in the shared memory bus to execute the scheduling and determination process of the underlying network request link.

[0082] See attached document Figure 5 In specific implementations, step S300 provided by the present invention may include the following steps:

[0083] S310, the scheduling module 230 reads the real-time cognitive load margin sequence output by multiple consecutive sliding time windows residing in the shared memory bus of the operating system, and uses the data nodes of adjacent calculation cycles to construct a difference equation to calculate and extract the first-order time difference slope. The specific calculation formula is as follows:

[0084] ;

[0085] In the formula, This represents the first-order time difference slope value at the current calculation time. The time variable is the current calculation time. This represents the real-time cognitive load margin scalar value at the current calculation moment. The real-time cognitive load margin scalar value for the previous time step; The time step span constant between two computing nodes is 1 second to 5 seconds. In this embodiment, the value of the time step span constant ranges from 1 second to 5 seconds.

[0086] The system relies on a first-order difference operation mechanism to convert static state margin data into dynamic slope characteristic indicators that characterize the state decay rate.

[0087] S320, the scheduling module 230 executes the comparison and judgment logic between the first-order time difference slope and the internally set warning threshold condition. The warning threshold condition is a manually configured negative slope constant parameter with a value range of -5 to -20. When the calculated first-order time difference slope value is less than or equal to the negative constant parameter or the real-time cognitive load margin scalar value of the current calculation cycle is less than the set absolute warning bottom line, the absolute warning bottom line is a manually configured positive constant based on the absolute resting reference constant, and in this embodiment, its value range is set between 10 and 30. When it is determined that any of the above conditions are met, the system's underlying logic determines that the physical operation terminal has an abnormal fluctuation operation stagnation state or a continuous high load state, and then generates a low-level trigger level signal for network transmission link redirection. When it is determined that no of the above conditions are met, the scheduling module 230 maintains the default data packet acquisition state of the network channel corresponding to the main buffer 240.

[0088] S330, after receiving the underlying trigger level signal, the scheduling module 230 takes over the network interface controller component at the hardware level of the computer device. It locates the data retrieval task thread currently being executed by the main buffer 240 in the kernel space, maintains the reception of the currently transmitted slice data packets until the end of the current slice data boundary, and then extracts the file header area data of the next slice to be processed. Subsequently, it calls the network control interface of the operating system to suspend the data retrieval task thread to pause the Hypertext Transfer Protocol response data stream of the subsequent slices to be processed. After the data retrieval request of the main buffer 240 is suspended, the underlying network protocol stack of the system synchronously releases the corresponding connection bandwidth resources.

[0089] S340, the scheduling module 230 takes over the control of the released connection bandwidth and allocates the idle connection bandwidth resource quota to the network data retrieval task queue corresponding to the secondary buffer 250. It extracts the absolute timestamp of the next slice to be processed in the main buffer queue and the corresponding knowledge node identification code as retrieval parameters. It also uses the product of the real-time cognitive load margin scalar value and the margin scaling constant of the current time period as the target upper limit threshold and encapsulates them together into a new network request message body. The scheduling module 230 sends a presentation modality data request instruction to the storage node 130 on the server side based on the absolute timestamp and with a lower course information entropy density parameter. After receiving the instruction, the storage node 130 queries the structured database index association table established inside it.

[0090] The system selects presentation modal data entities whose course information entropy density parameter under the corresponding absolute timestamp is less than or equal to the target upper limit threshold as the distribution targets. If there are multiple modal data entities that meet the conditions, the one with the largest density parameter is selected for distribution. If there are no modal data entities that meet the target upper limit threshold conditions, the third presentation modal data entity with the lowest course information entropy density parameter under the current absolute timestamp is forcibly selected as the distribution target to perform a fallback dimensionality reduction operation. When the main buffer 240 was processing the first presentation modal data, the system initiated the second or third presentation modal data retrieval process of the corresponding time node to the storage node 130 according to the instruction. This asymmetric data request scheduling logic reshapes the fixed resource consumption mode of single-line sequential loading of conventional streaming media clients.

[0091] S350, after receiving the directed request instruction, storage node 130 sends a presentation modal data packet that meets the dimensionality reduction conditions to client terminal 200. The scheduling module 230 configures the direct memory access controller of the underlying operating system so that the network data read and write mechanism bypasses the instruction scheduling link of the central processor. The data packet entities captured by the network interface controller are sequentially written into the independent memory stack area allocated by the secondary buffer 250 in the manner of direct physical memory addressing. The system synchronously promotes the physical memory preloading process of the backup data according to the consumption cycle of the residual data in the main buffer 240.

[0092] See attached document Figure 6 In specific implementations, step S400 provided by the present invention may include the following steps:

[0093] S410, when the rendering module 270 performs local decoding of multimedia files and graphics output calculation tasks, it maintains an independent rendering clock based on the hardware crystal oscillator beat. The matching module 260 captures the boundary time node of the slice sequence entity in the time domain by listening to the progress state of the count value of the rendering clock. When the count value of the rendering clock is equal to the sum of the timestamp of the currently being decoded slice and the fixed time step of the slice, the underlying hardware of the system triggers the boundary to reach an interrupt signal. This hardware-level interrupt signal indicates in kernel mode that the rendering module 270 is about to exhaust the effective frame data load of the current media slice and is preparing to enter the physical memory loading operation cycle of the next sequence data entity.

[0094] S420, after capturing the boundary interrupt signal, the matching module 260 determines whether there is any unprocessed slice data in the main buffer 240 data queue. If not, it reports a playback end signal to the system main control process and terminates the subsequent determination process; if it exists, it extracts the course information entropy density parameter encapsulated in the header region of the next unprocessed main modality slice file in the main buffer 240 data queue, and substitutes it with the real-time cognitive load margin scalar value output by the evaluation module 220 through the shared memory bus in the current time period into the preset threshold comparator model for condition determination. The calculation formula for this condition determination is:

[0095] ;

[0096] In the formula, The course information entropy density parameter scalar for the next master mode slice to be processed; This is the margin scaling constant; To enable real-time monitoring of load margin scalar values; This is the time variable for the current calculation moment.

[0097] The margin scaling constant is calibrated in conjunction with the rendering throughput performance of the local device's graphics processing unit, and its value is set between 0.5 and 2.0.

[0098] S430, the matching module 260 executes the algebraic inequality judgment logic and selects the corresponding control execution branch based on the underlying Boolean output result. When the judgment result does not meet the dimensionality reduction threshold condition, the matching module 260 determines whether the data retrieval task thread of the main buffer 240 is in a suspended state. If so, it sends a network retrieval task termination instruction to the scheduling module 230 to terminate the current data acquisition process of the secondary buffer 250 and forcibly reclaim the corresponding connection bandwidth resources.

[0099] Subsequently, a wake-up command is sent to the scheduling module 230 to resume the execution of the data retrieval task thread in the main buffer 240, and the rendering module 270 is instructed to enter a buffer waiting state until the main buffer 240 completes the acquisition of the next main modal slice data. After acquisition, the matching module 260 positions the memory read pointer of the system's underlying decoding input source to the starting physical address of the next slice to be processed in the main buffer 240. Then, the rendering module 270 checks whether the underlying decoding core is in a clock suspension state. If it is in this state, it issues a low-level interrupt command to release the clock suspension state and resume the decoding clock operation. After the state is ready, the main buffer 240 is retrieved. The data corresponding to the starting physical address of the next slice to be processed within 40 is used as the decoding input source to execute the regular streaming media playback calculation pipeline; if the determination result is that the data retrieval task thread of the main buffer 240 is not in a suspended state, the matching module 260 directly positions the memory read pointer of the system's underlying decoding input source to the starting physical address of the next slice to be processed in the main buffer 240, and the rendering module 270, after confirming that the underlying decoding core is out of the clock suspension state, directly extracts the data corresponding to the starting physical address of the next slice to be processed in the main buffer 240 as the decoding input source to execute the regular streaming media playback calculation pipeline;

[0100] When the determination result meets the dimensionality reduction threshold condition, the matching module 260 reads the underlying ready status bit of the secondary buffer 250, and generates a high-priority memory address overwrite instruction carrying the physical base address of the secondary buffer 250 and sends the underlying control parameter to the rendering module 270, after confirming that the secondary buffer 250 has completed data packet preloading.

[0101] S440, if the matching module 260 determines the result to be true but reads that the secondary buffer 250 is in an unready state, it determines whether the scheduling module 230 has allocated a network data retrieval task queue for the secondary buffer 250. If so, it triggers a timeout waiting clock and maintains the read pointer of the main buffer 240. If the data is ready during the timeout waiting clock, it interrupts the wait and triggers the issuance of a memory address overwrite instruction. If the timeout threshold is reached and the data is still not ready, it abandons the dimensionality reduction switch of the current slice and executes the above-mentioned judgment and control execution branch when the dimensionality reduction threshold condition is not met. If it determines that the scheduling module 230 has not allocated a network data retrieval task queue, it directly abandons the dimensionality reduction switch of the current slice and executes the above-mentioned judgment and control execution branch when the dimensionality reduction threshold condition is not met.

[0102] Under the condition that the data is ready, the rendering module 270 captures and parses the incoming memory address overwrite instruction in the kernel space. It first parses the file header area of ​​the preloaded data in the secondary buffer 250 to determine the modal attribute. If it belongs to the second rendering modal, it directly performs a pointer offset rewrite operation in the register configuration stage of the underlying decoding input source. It overwrites the memory read pointer of the underlying decoding input source from the starting physical address value of the next slice to be processed in the main buffer 240 to the starting physical address value of the preloaded slice in the secondary buffer 250. This direct addressing change mechanism enables the rendering module 270 to directly extract the preloaded low-density rendering modal data in the secondary buffer 250 and send it into the decoding core.

[0103] If it is the third presentation mode, the underlying decoding core is controlled to enter the clock suspension state and maintain the output of the last frame of the current screen. The pointer offset rewrite operation is not performed, and the plain text data in the secondary buffer 250 is directly routed to the two-dimensional rendering pipeline of the graphics computing unit to perform character drawing operation.

[0104] After the underlying processing operations of any of the above modalities are completed, the rendering module 270 sends a wake-up command to the scheduling module 230 to resume the execution of the main buffer 240 data retrieval task thread. Before resuming execution, the network control interface is called to clear the currently suspended Hypertext Transfer Protocol Request context state of the main buffer 240, and its retrieval timestamp parameter is pushed forward by a fixed time step span, so as to control the main buffer 240 to skip the current dimensionality reduction slice and directly initiate the main modal slice retrieval request for the next timestamp sequence.

[0105] S450: The system uses a hardware bus-level data source addressing redirection mechanism to achieve continuous switching of the output presentation mode on the terminal display screen without re-initiating a network handshake connection. For the audio and video picture synchronization control pipeline of the decoding core when handling the switching of different media stream modes, those skilled in the art can call the existing media presentation timestamp alignment algorithm framework to implement it. Its media frame alignment synchronization scheduling mechanism that uses timestamp offset to compensate for the relative delay between the audio track and the video track is a well-known technology in the field.

[0106] See attached document Figure 7 In specific implementations, step S500 provided by the present invention may include the following steps:

[0107] S510, the compensation module 280 continuously listens to the internal message bus of the system at the application layer and triggers an interrupt callback function when it captures the memory address overwrite instruction issued by the matching module 260. Inside the callback function, it parses the media metadata and extracts the absolute timestamp sequence corresponding to the dimensionality reduction switching action and the knowledge node identification code associated with the slice entity. The compensation module 280 combines the timestamp sequence and the knowledge node identification code into a structured log object according to the preset data table structure. It calls the underlying file writing interface of the operating system to persist the log object to the damaged record mapping table of the local database. The system executes the feature mapping characterization of the original presentation modal data skipped by the physical storage layer through the underlying file writing instruction.

[0108] S520, the evaluation module 220 continuously calculates and outputs continuous real-time cognitive load margin scalar values ​​in the system background and transmits them to the threshold monitoring pipeline of the compensation module 280. The compensation module 280 constructs a state monitoring sliding window with a set time span based on the numerical input stream and performs quantitative judgment calculation of the recovery state. Internally, it configures the recovery state stability evaluation formula according to the integral distribution law of historical window data.

[0109] ;

[0110] In the formula, To restore the state stability parameter; This is a constant representing the total number of discrete computational samples included within a given period. The index number variable for discrete calculation samples; It is a discrete step function; For the first Real-time cognitive load margin scalar value of each computing node; To safely restore the threshold constant.

[0111] The value of the stability parameter is constrained to a closed interval between 0 and 1. The constant of the total number of discrete calculation samples, combined with the sampling frequency of the device, fixes its value range between 100 and 500. The safe recovery threshold constant is separately calibrated and assigned by the system by extracting the average residual value within the historical stable interaction period. In this embodiment, the value range of the safe recovery threshold constant is 60 to 80. The discrete step function outputs a value of 1 when the independent variable is greater than or equal to 0, and outputs a value of 0 otherwise. This mathematical model constructs a calculation channel that transforms discrete fluctuation characteristic data into deterministic Boolean judgment signals.

[0112] S530, the compensation module 280 executes the formula to calculate and obtain the stability parameter of the recovery state and executes the conditional branch judgment instruction along the control pipeline. When the value of the parameter is determined to be less than 1, the underlying logic of the system determines that the current state is in the fluctuation range and controls the compensation module 280 to enter the sleep waiting period until the next sliding time window arrives. When the value of the parameter is determined to be equal to 1 and there are unprocessed records in the damaged record mapping table of the local database, the compensation module 280 extracts the unprocessed knowledge node identification code and its associated absolute timestamp sequence from the damaged record mapping table of the local database.

[0113] The extracted records are aggregated and deduplicated based on the knowledge node identifier code. The earliest knowledge node identifier code sorted by timestamp sequence is extracted from the aggregation result. This first knowledge node identifier code is used as the macro-retrieval primary key. The file writing interface is immediately called to mark the status of the batch of related records in the local database damaged record mapping table as "data retrieval in progress" to lock and block concurrent polling. Subsequently, a hypertext transfer protocol request is sent to the server storage node 130 along with the continuous absolute timestamp sequence interval attached to it. The storage node 130 matches and retrieves all continuous third presentation modality data entities contained in the knowledge node within the timestamp sequence interval in its persistent physical medium and performs text concatenation operation on the server. The concatenated complete semantic character stream is sent to the local cache queue of the client terminal 200. If it is determined that the parameter value is equal to 1 but there are no unprocessed records in the local database damaged record mapping table, the current system state is maintained and the current data compensation retrieval process is skipped.

[0114] S540, the rendering module 270 extracts the plain text character stream corresponding to the third presentation modality data entity from the local cache queue and calls the two-dimensional rendering pipeline of the graphics computing unit to perform screen drawing tasks. It first checks whether an instantiated independent text prompt sidebar already exists in the front-end graphical user interface. If it exists, it directly calls the text append interface of the instance to insert the plain text character stream to the end and updates the scroll bar state. If it does not exist, it allocates a graphics overlay layer with independent depth coordinates in the non-overlapping edge safe area of ​​the main media content layer currently being played in the front-end graphical user interface. The rendering module 270 uses the transparency blending channel of the display core to render and output the plain text character stream to the independent graphics overlay layer, forming an independent text prompt sidebar with an independent scroll bar control that is displayed in parallel with the main media stream, and binds a layer destruction callback event triggered by standard human-machine interface hardware to the sidebar.

[0115] After the sidebar text rendering operation of any of the above branches is completed, the rendering module 270 calls the underlying file writing interface of the operating system to synchronously update the record status of all associated timestamps under the knowledge node identifier code in the local database damaged record mapping table to the processed status in order to block the duplicate request logic. For the depth coordinate allocation and transparency pixel blending algorithm in a multi-layer environment, those skilled in the art can call the existing graphics application programming interface to build its underlying rendering logic and apply it in a conventional manner in this embodiment.

[0116] Application Examples:

[0117] To better understand the technical solution of this invention, the following uses the application scenario of playing online multimedia courses on a computer as an example for application implementation and test:

[0118] Before the course data is distributed, the preprocessing module 110 of the server cluster 100 slices the original course video at a fixed time step to generate a video slice sequence in the first presentation modality. It extracts key frames from the video and audio to generate a text-to-audio slice sequence in the second presentation modality. It uses a speech recognition model to convert the audio into a plain text slice sequence in the third presentation modality. The calibration module 120 calculates that the course information entropy density parameter in the first presentation modality is greater than that in the third presentation modality. The course information entropy density parameter, along with the knowledge node identification code, is written into the header area of ​​the file and persistently stored in the storage node 130.

[0119] The user starts playing course data through the client terminal 200. The scheduling module 230 requests the first presentation modality data from the server for buffered playback. The evaluation module 220 calculates and generates the actual interaction event rate value and the real-time cognitive load margin scalar value based on the underlying physical hardware interruption events captured by the monitoring module 210. When the user does not generate a physical hardware operation signal, the actual interaction event rate value is equal to zero, and the real-time cognitive load margin scalar value is kept within the preset safe value range.

[0120] When the course content enters a high information density period, the user generates continuous hardware interrupt operation signals. The evaluation module 220 calculates that the actual interaction event rate increases and the real-time cognitive load margin scalar value decreases. The scheduling module 230 detects that the first-order time difference slope of the real-time cognitive load margin scalar value meets the internally set negative warning threshold condition. The system determines that it meets the dimensionality reduction threshold condition and suspends the data retrieval process of the main buffer 240. It uses the secondary buffer 250 to request the second presentation modal data or the third presentation modal data with the same timestamp from the server storage node 130. When the slice boundary time node is reached, the rendering module 270 changes the memory read pointer of the underlying decoding input source to the first address of the secondary buffer 250. The client terminal then outputs the second presentation modal or the third presentation modal screen sequence.

[0121] The compensation module 280 records the knowledge node identification code corresponding to the aforementioned dimensionality reduction switching action in the damaged record mapping table of the local database. When the frequency of user physical operation behavior decreases, causing the real-time cognitive load margin scalar value output by the evaluation module 220 to rise above the safe recovery threshold and remain stable within a set period, the compensation module 280 extracts the unprocessed knowledge node identification code and absolute timestamp sequence from the local database to initiate a data compensation request. The server storage node 130 sends the corresponding third presentation modality plain text character stream to the local cache queue. The rendering module 270 generates an independent text prompt sidebar with scroll bar control in an independent layer of the front-end graphical user interface and directly renders and displays the plain text character stream to complete the structured data compensation operation.

[0122] Test example:

[0123] In this embodiment, a network interaction test environment was set up, and a total of 120 testers with the same testing background were selected and divided into three groups for a control experiment. All testers operated client devices to play the same 60-minute high-density information video sequence in a test environment with injected random network latency parameters. The first group, as the control group, used a standard streaming media playback system without adaptive processing mechanism. The second group, as the control group, used a bitrate adaptive system based on network transmission bandwidth status. The third group, as the experimental group, used the course resource accurate adaptation system based on multi-dimensional features provided by this invention and enabled cross-modal dimensionality reduction and data compensation execution pipeline.

[0124] The experiment used system background logs and front-end operation records to obtain the average playback stuttering rate and the proportion of cognitive overload for each group of testers. The average playback stuttering rate was the ratio of the total time the player spent in data buffering waiting state to the total duration of the course file. The proportion of cognitive overload was the ratio of the cumulative time the real-time cognitive load margin scalar value output by the system was lower than the set warning threshold constant to the total duration of the course file. After the test process was completed, a standardized closed-book test question bank of knowledge nodes was distributed to all testers, and the average score of knowledge point mastery with a percentage scale was statistically output.

[0125] The statistical values ​​of the experimental test results for each system scheme are shown in the table below:

[0126] Group 1 Standard streaming media playback system 14.6% 38.5% 68.4 Group 2 Adaptive Bitrate System 4.2% 35.1% 72.5 Group 3 This invention provides a precise adaptation system. 1.1% 12.4% 86.7

[0127] By comparing the statistical results in the above table and the attached... Figure 8 and attached Figure 9It can be seen that the average playback stuttering rate of the third group using the solution of the present invention is lower than that of the first group using the standard streaming media playback system and the second group using the bitrate adaptive system. The percentage of cognitive overload in the third group is lower than that of the first and second groups. The average score of knowledge point mastery in the third group is higher than that of the first and second groups. The objectively obtained experimental data results verify that the system of the present invention has the operational ability to reduce the probability of buffer stuttering in multimedia data transmission, the ability to adjust the cognitive overload load caused by high-dimensional content, and the optimization ability to improve the effectiveness and acceptance of target knowledge flow node transmission.

[0128] Appendix Figure 8 This is a graph showing the distribution of the time-series changes in the real-time cognitive load margin scalar value and the presentation mode switching indicator parameter of the client terminal within a fixed operating cycle in an embodiment of the present invention. The horizontal axis represents the operating time step parameter advancing along the time axis, and the vertical axis region is distributed with continuous scalar data lines representing the fluctuation of the real-time cognitive load margin and discrete step graph lines representing the changes in the presentation mode level. The coordinate position of the continuous scalar data line crossing the warning threshold reference line downwards corresponds to the state reversal node of the discrete step graph line from the first presentation mode indicator value to the second or third presentation mode indicator value. The coordinate position of the continuous scalar data line crossing the safety recovery threshold reference line upwards corresponds to the system timestamp coordinate of the rendering compensation event triggered by the independent text prompt sidebar.

[0129] Appendix Figure 9 This is a bar chart comparing the statistical values ​​of three different system schemes in the main control experiment of the test example of this invention on three quantitative indicators: average playback stuttering rate, proportion of cognitive overload state, and average score of knowledge point mastery. The horizontal axis of the chart has three independent experimental group classification parameters. The left vertical axis is used to mark the percentage value boundary interval of the probability parameters, the average playback stuttering rate and the proportion of cognitive overload state. The right vertical axis is used to mark the absolute value boundary interval of the average score of knowledge point mastery. The vertical physical height parameter of each bar element in the chart represents the objective statistical value of its corresponding measurement indicator.

[0130] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for precise adaptation of course resources based on multi-dimensional features, characterized in that, The method includes: The server cluster performs segmentation operations on the course resource files to generate data objects with different presentation modalities, calculates and obtains the course information entropy density parameters and interaction requirement factors, and writes them into the file header area of ​​each presentation modal data object. The client terminal pulls and presents modal data objects into the main buffer for the rendering module to decode and play. The listening module collects the underlying physical hardware interruption events and calculates the actual interaction event rate and human-computer interaction information entropy. The evaluation module combines the interaction demand factors to calculate and generate a real-time cognitive load margin scalar value. When the first-order time difference slope of the real-time cognitive load margin scalar value meets the internally set warning threshold condition, lower-density presentation modal data is preloaded into the secondary buffer. When the dimensionality reduction threshold condition is met at the slice boundary time node, the memory read pointer of the decoding input source is located to the secondary buffer to achieve mode switching; Once the real-time cognitive load margin scalar value recovers to the safe recovery threshold, the corresponding third presentation modal data retrieval and text character rendering and display operations are initiated.

2. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The steps by which the server cluster performs segmentation operations on course resource files to generate data objects with different presentation modalities include: The high frame rate video track and complete audio track of the original data stream of the slice sequence entity are retained to generate the first presentation modality data object. The core key frames of the video are extracted and mixed with the original audio track to generate the second presentation modality data object. The audio stream data is mapped to the core text character stream using a speech recognition model and encapsulated to generate the third presentation modality data object.

3. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The steps of calculating and obtaining the course information entropy density parameter and interaction demand factor and writing them into the file header area of ​​each presentation modality data object include: The course information entropy density parameter is calculated based on the video payload bit rate and audio payload bit rate of the presented modal data object in the transmission state. The interaction demand factor is set by the reference frequency expected by the front-end operator to generate the underlying peripheral hardware interrupt operation. The course information entropy density parameter, the interaction demand factor and the knowledge node identification code are encapsulated as a joint parameter label and written into the file header area to realize the binding of control parameter features.

4. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The steps of the monitoring module in collecting underlying physical hardware interrupt events and calculating the actual interaction event rate and human-computer interaction information entropy include: A sliding time window with a fixed time span parameter is established to extract the total number of physical hardware interrupt events and obtain the actual interaction event rate value. The time interval sequence of adjacent physical hardware interrupt events is extracted, and the continuous mouse coordinate offset data is converted into a spatial trajectory curvature distribution matrix and projected into multiple pre-divided discrete quantization intervals. The probability distribution of statistical feature values ​​falling into each specific quantization interval is substituted into the Shannon information entropy calculation model to obtain the human-computer interaction information entropy value.

5. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The steps of the assessment module in calculating and generating a real-time cognitive load margin scalar value by combining interaction demand factors include: By combining the interaction demand factor with the absolute resting baseline constant modulation, a dynamic baseline modulated by the presentation modality attribute is obtained. The resting penalty term constraint value is generated based on the difference range between the interaction demand factor and the actual interaction event rate value. The dynamic baseline, the human-computer interaction information entropy value and the resting penalty term constraint value are summarized and the real-time cognitive load margin scalar value is calculated and output based on the algebraic subtraction weighted model.

6. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The step of preloading lower-density presentation modal data into the secondary buffer when the first-order time difference slope of the real-time cognitive load margin scalar value meets the internally set warning threshold condition includes: When the calculated first-order time difference slope is less than or equal to the manually configured negative constant parameter or the real-time cognitive load margin scalar value of the current calculation cycle is less than the set absolute warning threshold, it is determined that the warning threshold condition is met. The data retrieval task thread of the network channel corresponding to the main buffer is suspended and the connection bandwidth resources of the underlying transmission control protocol are released. The secondary buffer is controlled to request presentation modal data entities with the same timestamp and course information entropy density parameter lower than the target upper limit threshold to be written into the independent memory stack area to complete the data preloading.

7. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The step of positioning the memory read pointer of the decoding input source to the secondary buffer to achieve mode switching when the dimensionality reduction threshold condition is met at the slice boundary time node includes: The system monitors the local graphics rendering clock's progress and captures the boundary time nodes of entities in the slice sequence. When the scalar value of the course information entropy density parameter of the next pending main modality slice is greater than the product of the real-time cognitive load margin scalar value and the margin scaling constant, it determines that the dimensionality reduction threshold condition is met. Under the condition that the secondary buffer is ready, it generates a high-priority memory address overwrite instruction carrying the physical base address to control the memory read pointer of the underlying decoding input source to change to the starting physical address of the secondary buffer.

8. The method for precise adaptation of course resources based on multi-dimensional features according to claim 1, characterized in that, The steps for initiating the corresponding third presentation modality data retrieval and text character rendering and display operations after the real-time cognitive load margin scalar value recovers to the safety recovery threshold include: The absolute timestamp sequence of the dimension reduction switch and the knowledge node identification code of the captured memory address overwrite instruction are extracted and persistently written into the damaged record mapping table of the local database. Based on the scalar value of the current real-time cognitive load margin and the constant of the safe recovery threshold, the stability parameter of the recovery status is extracted. According to the parameter, the corresponding third presentation modality data entity is retrieved from the server cluster and concatenated into a complete semantic character stream. The independent text prompt sidebar with independent scroll bar control is rendered in an independent layer of the front-end graphical user interface.

9. The method for precise adaptation of course resources based on multi-dimensional features according to claim 8, characterized in that, The steps of capturing memory address overwrite instructions, extracting the absolute timestamp sequence of the dimensionality reduction switch, and persistently writing the knowledge node identifier code into the damaged record mapping table of the local database include: The application layer continuously listens to the message bus and triggers an interrupt callback function when it captures a memory address overwrite instruction. It combines the absolute timestamp sequence with the knowledge node identification code and encapsulates it into a structured log object. It calls the underlying file writing interface to write the structured log object into the damaged record mapping table of the local database to perform feature mapping characterization of the skipped nodes of the original presentation modal data.

10. The method for precise adaptation of course resources based on multi-dimensional features according to claim 8, characterized in that, The steps of rendering and generating an independent text tooltip sidebar with an independent scrollbar control in an independent layer of the front-end graphical user interface include: The complete semantic character stream is sent to the local cache queue of the client terminal. The system checks whether there is an instantiated independent text prompt sidebar in the front-end graphical user interface. If not, a graphic overlay layer with independent depth coordinates is allocated in the non-overlapping edge safe area of ​​the main media stream layer for transparency blending rendering output. The underlying file writing interface is called synchronously to update the associated records in the damaged record mapping table to the processed state to block the duplicate request logic.

Images