A data labeling distribution intelligent scheduling and permission management system

Abstract

The application relates to the technical field of special query processing, in particular to a data labeling distribution intelligent scheduling and permission management system, which comprises a behavior segmentation module, a modal disassembly module, a window mapping module and a query arrangement module, and an instruction generation module; the behavior segmentation module identifies discontinuous modes to construct switching paragraph information set; the modal disassembly module analyzes action type structure to generate operation path comparison set; the window mapping module reconstructs start and end boundaries to generate window section structure list; the query arrangement module filters matching paths to generate task query path calling table; and the instruction generation module constructs query scheduling instruction set. In the application, the operation interval discontinuous mode is identified, the image text voice synchronous position is located and the edge area is analyzed, the sound energy change point section is extracted, the segment boundary is reconstructed based on the action structure, the window section sequence is constructed, the query task field is mapped and the matching path is filtered, the task and the path are encoded and combined, the scheduling rigidity problem is solved, massive data efficient distribution and dynamic permission control are realized, and the task response speed and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of special query processing technology, and in particular to a data labeling, distribution, intelligent scheduling, and permission management system. Background Technology

[0002] Specialized query processing technology involves unique methods for handling complex query tasks in computer database systems, aiming to improve query efficiency and accuracy. This technology primarily includes statistical queries, fuzzy queries, distributed queries, graph queries, and time-series queries. Its core aspects include query request identification and parsing, query statement rewriting and optimization, coordinated access to distributed data nodes, query result aggregation and filtering, the application of fuzzy matching algorithms, and access control and task scheduling for massive heterogeneous data sources. Specialized query processing is typically applied to scenarios requiring in-depth mining and analysis of large-scale unstructured or semi-structured data, and is widely used in search engines, data mining platforms, intelligent recommendation systems, and government and enterprise data platforms. It is a key component supporting intelligent retrieval capabilities in data processing systems. Among these, the traditional data annotation, distribution, intelligent scheduling, and access control management system refers to an information processing system that automatically assigns raw data tasks to appropriate annotators and manages the annotation results through access control and scheduling. These systems typically use preset rules or manual allocation to push data tasks to specific users by category or quantity. Static role-based permission configurations control user access and operation scope. Task scheduling often employs round-robin or targeted distribution strategies, with task redistribution based on task completion progress or user history. Access control records user identity information and controls their data read and operation permissions through configuration files or database tables. However, these methods suffer from insufficient response efficiency, rigid scheduling strategies, and permission conflicts when dealing with large task volumes, complex user distribution, or highly granular access control.

[0003] Traditional data labeling and distribution systems push data tasks to specific users by category using preset rules or manual allocation. They control user access scope through static role permission configuration. In terms of task scheduling, they mostly use polling or targeted distribution strategies, and redistribute tasks based on task completion progress or personnel history. Permission management records user identity information and controls their data operation permissions through configuration files. When facing massive heterogeneous data sources, complex personnel distribution, or fine-grained permission management scenarios, this approach suffers from insufficient response efficiency, rigid scheduling strategies, and permission conflicts. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a data annotation and distribution intelligent scheduling and permission management system. On one hand, a data annotation and distribution intelligent scheduling and permission management system is provided, the system comprising: The behavior segmentation module acquires the time record of each data annotation operation in the annotation terminal, forms a continuous operation sequence according to the time axis, compares the time interval of any two adjacent data annotation operations, identifies discontinuous patterns and verifies the distribution, and generates a set of switching segment information. Based on the time position of the switching paragraph information set, the modal decomposition module locates the synchronization position of image segments, text sentences and speech segments, analyzes the edge regions of the image segments, calculates the frequency of differentiated part-of-speech units in the text sentences, extracts the spectral segments of the sound energy change points in the speech segments, and generates an operation path comparison set. The window mapping module maps the start and end points of adjacent action segments according to the action type structure order of the operation path reference set, reconstructs the start and end boundaries of the action segments, and generates a list of window segment structures. Based on the start and end boundaries of the window segment structure list, the query orchestration module maps the start field of each query task to the dispatch identifier, filters the matching set of window task paths, and generates a task query path call table. The instruction generation module uses the window start and end information and task identifier in the task query path call table to match the undistributed task number and encode and merge it with the query path to generate a query scheduling instruction set.

[0005] As a further embodiment of the present invention, the switching segment information set includes an interval time threshold, an operation sequence identifier, and a time distribution node; the operation path reference set includes image edge action features, text part-of-speech frequency units, and speech energy change structure; the window segment structure list includes action start and end positions, path type labels, and window sorting index; the task query path call table includes field mapping relationships, boundary matching index, and path task number; and the query scheduling instruction set includes scheduling number information, path encoding results, and task distribution identifier.

[0006] As a further aspect of the present invention, the frequency of differentiated part-of-speech units refers to the frequency of occurrence of differentiated part-of-speech units in a text sentence, which measures the differential distribution in the text.

[0007] As a further aspect of the present invention, the start and end points of the mapped adjacent action segments are verified and corresponded to the start and end positions of adjacent action segments according to the action type structure order.

[0008] As a further aspect of the present invention, the behavior segmentation module includes: The data receiving submodule obtains the operation timestamp corresponding to each data annotation operation in the annotation terminal, and assembles the obtained operation timestamp sequence into a continuous operation record sequence according to the time axis order based on the receiving order, generating an operation time sorted sequence. The operation interval calculation submodule, based on the two adjacent timestamps in the operation time sorting sequence, calls the numerical difference between the timestamps respectively, uses the absolute time difference as the calculation basis, performs a subtraction operation item by item, and generates an adjacent operation interval sequence of the same length. The sequence distribution recognition submodule calls the discontinuous mode discrimination rule based on the time interval value in the adjacent operation interval sequence, and judges whether there is a sudden change point with a significant difference from the previous and next values ​​based on the continuous interval interval benchmark value and the sudden change judgment threshold. It marks the index position information corresponding to the sudden change point, and calls the time value corresponding to the index position in the overall sequence to generate a switching segment information set.

[0009] As a further aspect of the present invention, the modal decomposition module includes: The synchronous positioning submodule obtains the time position of the switching segment information, and locates the time frame range of image segments, text sentences and voice segments in the data to be processed according to the time axis correspondence. It also defines the synchronization position of the three types of data according to the modal start and end time overlap interval, and generates the modal synchronization position interval. The modal feature extraction submodule performs image grayscale gradient detection to extract edge pixel distribution density based on image segments, text sentences and speech segments in the modal synchronization position interval, counts the frequency of occurrence of differentiated part-of-speech units in the text, and extracts the frequency band frame range covered by the sound energy change points in the spectrogram to generate a three-modal change index group of image, text and sound. The action structure recognition submodule, based on the feature values ​​corresponding to the modalities in the three-modal change index group of image, text and audio, divides the action occurrence patterns in the synchronous segment according to the combination relationship between the degree of image edge change, the proportion of different parts of speech in the text and the number of frames covered by the speech spectrum segment, establishes the temporal comparison path relationship between modalities, and generates an operation path comparison set.

[0010] As a further aspect of the present invention, the window mapping module includes: The action sequence parsing submodule collects the numbering and time arrangement information of action segments in the continuous processing task based on the action type structure order in the operation path reference set, establishes a sequence index table according to the relationship between action types in the sequence, and generates an action sequence index sequence. The boundary mapping and sorting submodule detects the start and end time values ​​of adjacent action segments based on the action sequence index, performs corresponding mapping and alignment processing on the start and end points, eliminates overlapping and gap segments, and generates action segment boundary mapping intervals. The window sorting construction submodule calls the action fragment boundary mapping interval, performs linear rearrangement of the interval according to the operation path arrangement order, calculates the sequence number and span length of the interval on the time axis, constructs the window segment sorting relationship between action fragments, and generates a window segment structure list.

[0011] As a further aspect of the present invention, the query orchestration module includes: The field mapping matching submodule obtains the start and end boundaries of windows in the window segment structure list, collects the start field and distribution identifier information of all query tasks in the current scheduling request, and performs field-level mapping and positioning of fields one by one according to the structural index relationship between field values ​​and window boundaries, generating field window mapping index values. The boundary overlap determination submodule matches and determines whether the index position corresponding to each query task field falls within the window boundary index range based on the field window mapping index value. According to the window boundary range interval rules, it calculates whether the field index falls within the window segment and generates a field overlap status determination sequence. The path set filtering submodule calls the field overlap status determination sequence to filter all window segment numbers that have a complete overlap with the field index, extracts the corresponding task path structure information, and combines and arranges them according to the order in which the fields appear to generate a task query path call table.

[0012] As a further aspect of the present invention, the instruction generation module includes: The undistributed task screening submodule obtains the window start and end information and task identifier field included in each query path in the task query path call table, detects all task numbers in the scheduled task list, and determines the execution status based on the distribution status identifier of each task number, screens out the set of task numbers that have not yet been distributed, and generates a list of schedulable task numbers. The path task encoding submodule calls the path structure that matches the number in the task query path call table according to the task number in the schedulable task number list, extracts the window start and end values ​​and task identifier fields in the path corresponding to each number, performs sequential concatenation and field compression and merging operations on the path content and number, and generates a task path merge encoding sequence. The scheduling instruction construction submodule calls each group of encoded data in the task path merging encoding sequence, arranges the instruction fields in sequence according to the standard format of the scheduling structure field, adds a unified format identifier header and control tail mark to all merged encoding structures, and generates a query scheduling instruction set.

[0013] As a further aspect of the present invention, the execution status determination of the distribution status identifier refers to determining whether the task has been distributed and the current execution status based on the task's distribution status identifier.

[0014] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: The system acquires and identifies discontinuous patterns in the operation time records of labeled terminals to construct switching segment information. It locates the synchronization positions of image text and speech segments and performs edge region and part-of-speech frequency analysis. It extracts the coverage segments of sound energy change points to generate operation path comparisons. Based on the action type structure, it reconstructs the segment start and end boundaries and constructs a window segment sorting structure. It performs field-level mapping on the query task start field and distribution identifier. It compares the overlap relationship between the query task and the window boundary index to filter matching paths. It merges the schedulable tasks and query paths into an instruction set, solving the problem of rigid scheduling in traditional methods. It achieves efficient and accurate distribution and dynamic access control of massive heterogeneous data, improving the response speed and accuracy of complex query tasks. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the system of the present invention; Figure 2 This is a flowchart of the behavior segmentation module in this invention; Figure 3 This is a flowchart of the modal decomposition module in this invention; Figure 4 This is a flowchart of the window mapping module in this invention; Figure 5 This is a flowchart of the query and arrangement module in this invention; Figure 6 This is a flowchart of the instruction generation module in this invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] This invention provides a data labeling, distribution, intelligent scheduling, and permission management system, such as... Figure 1-2 The diagram shown illustrates a data labeling, distribution, intelligent scheduling, and access control system. This system includes: The behavior segmentation module obtains the time record corresponding to each data annotation operation in the annotation terminal, forms a continuous operation sequence according to the time axis, performs item-by-item comparison on the time interval between any two adjacent operations, identifies the discontinuous pattern in the operation interval, determines the distribution pattern of the sequence position corresponding to the discontinuous pattern on the time axis, and constructs a switching segment information set. Based on the time position included in the switching paragraph information set, the modal decomposition module locates the synchronization position of image segments, text sentences and speech segments in the data to be processed. It performs edge region analysis on the image within the range of each synchronization position, calculates the occurrence frequency of differentiated part-of-speech units for the text, extracts the coverage segment of the sound energy change point in the spectrum for the speech, analyzes the action type structure of each data modality at the synchronization position, and generates an operation path comparison set. The window mapping module maps and organizes the start and end points of adjacent action segments in a continuous processing task based on the action type structure order in the operation path reference set. It reconstructs the start and end boundaries of each segment according to the operation path arrangement order, constructs a window segment sorting structure based on the operation path between action segments, and generates a window segment structure list. The query orchestration module maps the start field and dispatch identifier of each query task in the current scheduling request to the start field and dispatch identifier based on the start and end boundaries of the window segment structure list. It compares the index overlap between the query task fields and the window boundary structure, filters the set of window task paths that can be directly matched, and generates a task query path call table. The instruction generation module uses the window start and end information and task identifier recorded in each matching path in the task query path call table to perform matching screening based on the number of the undistributed task in the scheduling task list, and encodes and merges the schedulable task with the corresponding query path to construct a query scheduling instruction set.

[0020] The switching segment information set includes the interval time threshold, operation sequence identifier, and time distribution nodes. The operation path comparison set includes image edge action features, text part-of-speech frequency units, and speech energy change structure. The window segment structure list includes the start and end positions of actions, path type labels, and window sorting index. The task query path call table includes field mapping relationships, boundary matching index, and path task number. The query scheduling instruction set includes scheduling number information, path encoding results, and task distribution identifier.

[0021] Specifically, such as Figure 2 , 3 As shown, the behavior segmentation module includes: The data receiving submodule obtains the operation timestamp corresponding to each data annotation operation in the annotation terminal, and assembles the obtained operation timestamp sequence into a continuous operation record sequence according to the time axis order based on the receiving order, generating an operation time sorted sequence. First, a high-precision time acquisition interface is initialized, configured with millisecond-level time resolution, to interface with the annotation terminal's input / output controller. When the annotation terminal detects user interaction, such as mouse clicks, keyboard keystrokes, or touchscreen swipes, a hardware interrupt signal is triggered. The submodule immediately captures the system clock value generated by this interrupt signal. This clock value uses Unix timestamp format, i.e., the number of milliseconds calculated from January 1st, 00:00:00 UTC. The submodule establishes a dynamically long first-in-first-out queue as a buffer, pushing the acquired raw timestamps into the queue sequentially. Subsequently, the submodule calls the quicksort logic, selecting the first timestamp in the queue as the base value, comparing the values ​​of subsequent timestamps with the base value, and rearranging all elements in the queue according to an ascending order of numerical values. During the sorting process, if duplicate timestamps with identical values ​​are encountered, the submodule checks their corresponding operation type identifier ID, performing a secondary sort based on the ASCII code value of the operation type ID to ensure strict monotonicity of the sequence. After sorting, the submodule traverses the entire queue, assigning a unique, incrementing sequence number to each timestamp, and binding the timestamp to its sequence number to construct a structured linear linked list. This linked list is the sorted sequence for the operation time. For example, if the system collects a set of original timestamps 1678886400050, 1678886400010, and 1678886400100, after sorting, the generated sequence will be 1678886400010, 1678886400050, and 1678886400100.

[0022] The operation interval calculation submodule is based on two adjacent timestamps in the operation time sorting sequence. It calls the numerical difference between the timestamps respectively, uses the absolute time difference as the calculation basis, performs the item-by-item subtraction operation, and generates an adjacent operation interval sequence of the same length. A traversal pointer is initiated, pointing to the first element of the time-ordered sequence. The submodule simultaneously reads the timestamp value at index i, and the next timestamp value at index i+1. Internally, the submodule contains a high-precision arithmetic logic unit (ALU). This unit receives these two timestamp values ​​as input operands and performs a subtraction operation: subtracting the timestamp value at index i from the timestamp value at index i+1. Since the sequence is pre-sorted, the result is naturally non-negative. However, to ensure robust data processing, the submodule enforces an absolute value function on the result. This absolute value represents the pause time between two consecutive user actions. The pointer then automatically moves forward one position, repeating the reading and calculation process until the second-to-last element of the sequence is reached. After each calculation, the submodule stores the calculated interval value into a new floating-point array, whose index order strictly corresponds to the operation order of the original timestamp sequence. For example, for the aforementioned sorted timestamp sequence, the submodule calculates the difference between the second item (1678886400050) and the first item (1678886400010), obtaining 40 milliseconds; it calculates the difference between the third item (1678886400100) and the second item (1678886400050), obtaining 50 milliseconds. The final generated sequence of adjacent operation intervals is 40 and 50.

[0023] The sequence distribution recognition submodule calls the discontinuous mode discrimination rule based on the time interval value in the adjacent operation interval sequence, and judges whether there is a sudden change point with a significant difference from the previous and next values ​​based on the continuous interval interval benchmark value and the sudden change judgment threshold. It marks the index position information corresponding to the sudden change point, and calls the time value corresponding to the index position in the whole sequence to generate the switching segment information set. Based on the time interval values ​​in the adjacent operation interval sequence, the discontinuous mode discrimination rule is invoked. Using the continuous interval baseline value and the abrupt change judgment threshold, it is determined whether there are abrupt changes with significant differences between the interval values ​​and the preceding and following values. The index position information corresponding to the abrupt change point is marked, and the time value corresponding to the index position is retrieved in the overall sequence to generate a switching paragraph information set. The sequence distribution recognition submodule first loads a preset parameter configuration file, from which it reads the continuous interval baseline value and the abrupt change judgment threshold. These two parameters are calculated using the statistical distribution characteristics of historical behavior data. Specifically, the submodule uses the average interval of historical normal continuous operations (such as continuous typing) as the baseline value, and sets three times the baseline value as the abrupt change judgment threshold. The submodule iterates through each interval value in the adjacent operation interval sequence, comparing the current interval value with the abrupt change judgment threshold. If the current interval value is greater than or equal to the abrupt change judgment threshold, the logic judge determines that a break or switch in the behavior mode has occurred at that position, i.e., a abrupt change point exists. At this time, the submodule records the index number of the abrupt change point in the interval sequence and marks it as a "segment end identifier". Subsequently, the submodule uses the index number to perform a reverse lookup of the operation time sorting sequence, extracts the specific timestamp value corresponding to the index position as the end time of the previous row paragraph, and extracts the timestamp of the next index position as the start time of the new row paragraph. The submodule encapsulates all identified mutation point indices, corresponding start and end timestamps, and paragraph numbers into a structure object, and compiles them into a switching paragraph information set.

[0024] Serial Number Raw timestamp (milliseconds) Sorted timestamps (milliseconds) Adjacent interval value (milliseconds) Mutation determination results 1 1000 1000 0 Starting point 2 1050 1050 50 continuous 3 1200 1200 150 continuous 4 3500 3500 2300 Mutation point 5 3600 3600 100 continuous As shown in Table 1, this table displays the entire process data from obtaining the original timestamp to determining the final mutation point. The interval value corresponding to sequence number 4 is 2300 milliseconds. Specific parameter setting logic and examples are introduced here: the baseline value for the continuous interval is set to 100 milliseconds, based on the average frequency of normal continuous keystrokes by the user; the mutation judgment threshold is set to 500 milliseconds, which is set based on 5 times the baseline value to distinguish interruptions in the operation flow. The specific logical derivation is as follows: the interval value of 2300 milliseconds corresponding to sequence number 4 is read, and the mutation judgment threshold of 500 milliseconds is read simultaneously. The submodule performs a comparison operation, determining that 2300 is greater than 500, and the condition is met. Therefore, the system identifies a significant difference at this position and determines it as a mutation point. This means that there is an operation interruption between the timestamp 1200 milliseconds and 3500 milliseconds. The system will generate a switching segment information accordingly, recording 1200 milliseconds as the end of the previous segment and 3500 milliseconds as the start of the new segment. This logic accurately separates continuous intensive operations from intermittent thinking or switching behaviors.

[0025] Specifically, such as Figure 2 , 4 As shown, the modal decomposition module includes: The synchronous positioning submodule obtains the time position of the switching segment information. Based on the time axis correspondence, it locates the time frame range of image segments, text sentences and voice segments in the data to be processed. It delineates the synchronization position of the three types of data according to the modal start and end time overlap interval and generates the modal synchronization position interval. First, the module parses the segment information set to extract the start timestamp T_start and end timestamp T_end for each segment. For image modality, the submodule reads the video stream's frame rate metadata (e.g., 30 frames / second) and converts the timestamps into frame sequence numbers using a formula: the start frame number equals T_start multiplied by the frame rate, and the end frame number equals T_end multiplied by the frame rate, thus locking the frame range of the image segment. For text modality, the submodule iterates through log files or subtitle files with timestamps, retrieving all text lines whose timestamps fall within the closed interval of T_start to T_end, determining the start and end line numbers of the text statements. For speech modality, the submodule calculates the corresponding sampling point positions based on the audio sampling rate (e.g., 44100 Hz): the start sampling point equals T_start divided by 1000 multiplied by the sampling rate, and the end sampling point equals T_end divided by 1000 multiplied by the sampling rate, thus extracting speech segments. After completing the independent localization of the three modalities, the submodule executes the intersection operation logic, comparing the actual coverage time ranges of the three modal segments and taking the intersection of their time ranges. If a modality is missing in the current segment, the time overlap area of ​​the remaining existing modalities is used as the reference. The submodule uses the start and end points of the calculated final overlap time range as boundaries and marks the corresponding pointer positions in the three data streams. These aligned pointer positions constitute the modal synchronization position interval. For example, if the behavior segment is 1000 to 5000 milliseconds, the image coverage is 1000 to 5000 milliseconds, the speech coverage is 1500 to 5500 milliseconds, and the text coverage is 1200 to 4800 milliseconds, then the submodule calculates the intersection as 1500 to 4800 milliseconds and sets this interval as the synchronization position.

[0026] The modal feature extraction submodule performs image grayscale gradient detection to extract edge pixel distribution density based on image segments, text sentences and speech segments in the modal synchronization position interval, counts the frequency of occurrence of differential part-of-speech units in the text, and extracts the frequency band frame range covered by the sound energy change points in the spectrogram to generate a three-modal change index group of image, text and sound. Three processing threads are launched in parallel. In the image processing thread, the submodule first converts the color image frame to grayscale to eliminate color interference. Then, the Sobel operator is used to calculate the grayscale gradients in the horizontal and vertical directions. The submodule sums the squared horizontal and vertical gradients of each pixel and takes the square root to obtain the gradient magnitude. The submodule sets a gradient threshold (e.g., a grayscale value of 50) and marks pixels with magnitudes greater than this threshold as edge points. Next, the submodule calculates the proportion of edge points in the total number of pixels, i.e., the edge pixel distribution density. In the text processing thread, the submodule calls a natural language processing (NLP) word segmentation tool to segment and tag the text within the synchronization interval. The submodule predefines a set of differentiated parts of speech (e.g., verbs, adjectives), counts the total number of times these specific parts of speech appear in the text segment, divides it by the total number of words, and obtains the frequency of occurrence of the differentiated part-of-speech units. In the speech processing thread, the submodule performs a Short-Time Fourier Transform (STFT) on the audio data, setting the window length to 2048 sampling points and the overlap rate to 50%, generating a spectrogram. The submodule calculates the energy value of each frame's spectrum and the spectral flux between adjacent frames. It sets a threshold for acoustic energy change and counts the proportion of frames with spectral flux exceeding this threshold, representing the coverage area of ​​acoustic energy change points. Finally, the submodule combines the edge density values ​​of the image, the part-of-speech frequency values ​​of the text, and the proportion of acoustic energy change in the speech data to construct a three-dimensional vector, namely the image-text-speech three-modal change index group.

[0027] The action structure recognition submodule is based on the feature values ​​corresponding to the modal in the three-modal change index group of image, text and audio. According to the combination relationship between the degree of image edge change, the proportion of different parts of speech in the text and the number of frames covered by the speech spectrum segment, it divides the action occurrence pattern in the synchronous segment, establishes the temporal comparison path relationship between modalities, and generates an operation path comparison set. A rule-based decision tree logic or a predefined lookup table is introduced. The submodule reads three feature values ​​from a trimodal change index group: image edge density D_img, text part-of-speech frequency D_txt, and speech change ratio D_aud. The submodule first compares each feature value with a preset classification threshold, discretizing continuous values ​​into state levels. For example, if D_img is greater than 0.3, it is marked as "high visual change," otherwise as "low visual change"; if D_aud is greater than 0.5, it is marked as "high audio activity," otherwise as "low audio activity." Subsequently, the submodule performs pattern matching based on a combination relationship rule base: Rule 1: If "high visual change" and "low audio activity" are present, and D_txt displays high-frequency verbs, the system determines the current action as a "silent operation mode" (such as drawing or reading). Rule 2: If "low visual change" and "high audio activity" are present, the system determines it as an "interactive dialogue mode." Rule 3: If all three are high values, it is determined as a "high-frequency mixed operation mode" (such as gaming or video editing). The submodule assigns a recognized action pattern label to each synchronization segment and links the pattern labels of each segment in chronological order to form a logical path describing the evolution of the action. The submodule then associates and stores this path with the original time segment information to generate an operation path lookup set.

[0028] Sample number Image edge density Text verb frequency Speech energy change rate Visual judgment threshold Hearing threshold Identify action patterns 1 0.45 0.12 0.05 0.30 0.20 Vision-driven operation 2 0.10 0.05 0.65 0.30 0.20 Voice interaction mode 3 0.55 0.25 0.70 0.30 0.20 Hybrid high-frequency operation Table 2 lists the specific feature values ​​and judgment results. Here, we explain the settings for the "visual judgment threshold" and the "auditory judgment threshold": The visual judgment threshold is set to 0.30. The process for obtaining this value is as follows: The system collects 1000 image samples from standard static reading scenes, calculating their average edge density to be 0.15; it also collects 1000 samples from standard video switching scenes, with an average edge density of 0.45. The system selects the median value of these two values ​​(0.15 plus 0.45 divided by 2), i.e., 0.30, as the benchmark for distinguishing between static and dynamic vision. The auditory judgment threshold is set to 0.20. Similarly, the system analyzes environmental noise samples (average change rate 0.05) and clear voice command samples (average change rate 0.35), taking their weighted average to determine the threshold as 0.20. A specific logical deduction example (taking sample 3 as an example): The system obtains an image edge density of 0.55, a text verb frequency of 0.25, and a speech energy change rate of 0.70. First, the image value of 0.55 is compared with the visual threshold of 0.30. Since 0.55 is greater than 0.30, it is judged as "high visual." Second, the speech value of 0.70 is compared with the auditory threshold of 0.20. Since 0.70 is greater than 0.20, it is judged as "high auditory." Finally, the system searches the combination rule base and finds that the combination of "high visual" and "high auditory" corresponds to the "mixed high-frequency operation" mode. Based on this, the system outputs a judgment result, which indicates that during the time period of sample 3, the user was performing a complex task with highly active visual and auditory senses. This recognition result will serve as a key basis for subsequent window mapping classification.

[0029] Specifically, such as Figure 2 , 5 As shown, the window mapping module includes: The action sequence parsing submodule collects the numbering and time arrangement information of action segments in continuous processing tasks based on the action type structure order in the operation path reference set, establishes a sequence index table according to the sequential relationship of action types in the sequence, and generates an action sequence index sequence. First, the submodule accesses the operation path lookup set and reads all the action fragment objects it contains. For each action fragment, the submodule extracts its internal attributes, including the unique ID of the action (e.g., Action_001), the action type label (e.g., "visual-driven operation"), and the timestamp range of its occurrence (Start_Time, End_Time). The submodule creates a doubly linked list data structure to store these action objects. Based on the value of Start_Time, the submodule inserts all action fragments into the linked list in chronological order. During insertion, the submodule checks the logical continuity of adjacent nodes; if the End_Time of a previous node is greater than the Start_Time of a subsequent node, it is marked as overlapping and needs to be processed. After the linked list is built, the submodule traverses the linked list and assigns a logical order index value (Index_0, Index_1...) to each node. Simultaneously, the submodule constructs a hash map (HashMap) using the action type ID as the key and its position index list in the linked list as the value to create an inverted index. Finally, the submodule integrates the time-sorted linked list structure with the inverted index table to output an action order index sequence containing the complete topological relationships. For example, if the sequence is A->B->C, the index sequence generated by the submodule explicitly records that A is the predecessor of B, C is the successor of B, and the start time of B immediately follows the end time of A.

[0030] The boundary mapping and sorting submodule detects the start and end time values ​​of adjacent action segments based on the action sequence index, performs corresponding mapping and alignment processing on the start and end points, eliminates overlapping and gap segments, and generates action segment boundary mapping intervals. A boundary scan algorithm is initiated, starting from the head of the action sequence index. For each pair of adjacent action segments (denoted as the previous segment Prev and the current segment Curr), the submodule extracts the end time T_end_prev of Prev and the start time T_start_curr of Curr. The submodule executes the following alignment logic: Case 1 (Gap Elimination): If T_start_curr is greater than T_end_prev, it indicates that there is an undefined blank time window between them. The submodule calculates the midpoint T_mid between them, i.e., (T_end_prev + T_start_curr) / 2. Subsequently, the submodule extends the end point of Prev to T_mid and advances the start point of Curr to T_mid, thereby filling the gap and achieving seamless connection. Case 2 (Overlap Elimination): If T_start_curr is less than T_end_prev, it indicates that there is a time conflict between them. The submodule also calculates the midpoint T_mid, forcibly truncating the end point of Prev at T_mid and delaying the start point of Curr to T_mid, thereby segmenting the overlapping region. After the above processing, the submodule updates the time boundary attributes of all action segments to ensure that all time points are continuous and mutually exclusive on the timeline. These corrected start and end time point pairs constitute the standardized action segment boundary mapping interval.

[0031] The window sorting construction submodule calls the action fragment boundary mapping interval, performs linear rearrangement of the interval according to the operation path arrangement order, calculates the sequence number and span length of the interval on the time axis, constructs the window segment sorting relationship between action fragments, and generates a window segment structure list. An empty dynamic array is created to store the final window objects. The submodule sequentially reads each corrected time interval from the action fragment boundary mapping interval. For each interval, the submodule calculates its span length, which is the end time minus the start time. The submodule encapsulates this span length, the corrected start time, the end time, and the original action type ID into a "window unit" object. Subsequently, the submodule performs a linear rearrangement check to reconfirm that all window units are arranged in strictly ascending order of their start times. The submodule assigns a globally unique window number (Window_ID) to each window unit in the array, starting from 1 and incrementing. Furthermore, the submodule calculates the relative positional relationships between windows, such as the distance between the center points of the Nth window and the (N+1)th window, as a density index between windows. Finally, the submodule serializes the set of window unit objects containing window numbers, start and end times, span lengths, and density indices to generate a list of window segment structures. For example, one piece of data in the list might be described as: "Window ID: 5, Start: 10500ms, End: 12000ms, Duration: 1500ms, Type: Text Input".

[0032] Specifically, such as Figure 2 , 6 As shown, the query orchestration module includes: The field mapping matching submodule obtains the start and end boundaries of windows in the window segment structure list, collects the start field and distribution identifier information of all query tasks in the current scheduling request, and performs field-level mapping and positioning of each field according to the structural index relationship between field values ​​and window boundaries, generating field window mapping index values. First, the window segment structure list is loaded and transformed into a fast retrieval structure based on an IntervalTree, where each node represents a window's time range [W_start, W_end]. Simultaneously, the submodule parses the current scheduling request, extracting a task list containing multiple query tasks. For each query task in the list, the submodule reads its defined key time field, namely the task start field T_task_start. The submodule uses T_task_start as the query key to search in the IntervalTree. The search logic checks if T_task_start meets the condition: W_start <= T_task_start <= W_end. If a tree node meeting the condition is found, the window number to which the task belongs is locked. If multiple windows meet the condition (theoretically, this shouldn't happen after previous processing, but as a redundancy check), the submodule selects the window with the smallest span as the matching object. After successful location, the submodule establishes a key-value mapping (Key: Task_ID, Value: Window_ID) between the task's dispatch identifier ID and the found window number. This mapping relationship is recorded and output as the field window mapping index value. For example, if the time field of task T-101 points to time point 11000ms, interval tree search finds that this time point is within window W-05 (10500ms-12000ms), and the system generates the mapping index "T-101->W-05".

[0033] The boundary overlap determination submodule matches the index position corresponding to each query task field with the window boundary index range based on the field window mapping index value. According to the window boundary range interval rules, it calculates whether the field index falls within the window segment and generates a field overlap status determination sequence. The submodule performs secondary verification logic. It iterates through each record in the field window mapping index. For each record (Task_ID, Window_ID), the submodule retrieves the complete task time period [T_start, T_end] corresponding to Task_ID and the window boundary [W_start, W_end] corresponding to Window_ID. The submodule calculates the coverage rate, which is the intersection length of the task time period and the window time period divided by the total length of the task time period. The submodule sets an overlap rule: if the coverage rate is 1.0 (100%), it is considered "completely overlapping"; if the coverage rate is greater than 0 and less than 1.0, it is considered "partially overlapping"; if the coverage rate is 0, it is considered "detached". The submodule encodes the judgment result (completely overlapping / partially overlapping / detached) into a status byte and binds it to the task ID. This process is executed cyclically for all tasks, and the final ordered set of states is the field overlap status judgment sequence.

[0034] The path set filtering submodule calls the field overlap status judgment sequence, filters all window segment numbers that have a complete overlap with the field index, extracts the corresponding task path structure information, and combines and arranges them according to the order of field appearance to generate a task query path call table. The application uses filter logic to iterate through the field overlap status determination sequence. The submodule examines the status byte of each entry, retaining only records with a "complete overlap" status and discarding records with "partial overlap" or "detached" status. For each valid record, the submodule extracts its associated window number. Then, based on these window numbers, the submodule backtracks through the query window segment structure list to obtain detailed path information corresponding to that window (such as associated original modal features and action type). The submodule sorts this extracted path information according to the original order of occurrence of the task IDs. The submodule constructs a two-dimensional table structure, where each row represents a valid query task, and the columns include the task ID, the matching window ID, the window start time, the window end time, and the action path description. This structured table is the task query path call table, providing a precise and validated data source for subsequent instruction generation.

[0035] Specifically, such as Figure 2 , 6 As shown, the instruction generation module includes: The undistributed task screening submodule obtains the window start and end information and task identifier field included in each query path in the task query path call table, detects all task numbers in the scheduled task list, and determines the execution status based on the distribution status identifier of each task number, screens out the set of task numbers that have not yet been distributed, and generates a list of schedulable task numbers. First, the global task list is loaded, which maintains the lifecycle status of all pending tasks in the system. The list contains a boolean status bit "Is_Distributed". The submodule simultaneously reads all task IDs from the task query path call table. The submodule performs a logical AND-NOT operation: iterating through each task ID in the call table and querying the corresponding "Is_Distributed" status bit in the global list. If the status bit is "False" (0), it indicates that the task has not yet been dispatched and meets the scheduling conditions; if it is "True" (1), it is skipped. The submodule extracts all task IDs with a "False" status bit and stores them in a new dynamic list. To prevent concurrent conflicts, the submodule briefly locks the memory address when reading the status bit. After filtering, the IDs in the list represent the tasks that need to generate instructions at the current moment; this list is the list of schedulable task numbers.

[0036] The path task encoding submodule calls the path structure that matches the number in the task query path call table based on the task number in the schedulable task number list. It extracts the window start and end values ​​and task identifier fields from the path corresponding to each number, performs sequential concatenation and field compression and merging operations on the path content and number, and generates a task path merge encoding sequence. For each ID in the list of schedulable task numbers, an encoding process is executed. First, the submodule retrieves all data items corresponding to that ID from the call table: window start time (T_s), window end time (T_e), and window type code (Type_Code). Second, the submodule performs data format conversion, converting T_s and T_e to hexadecimal strings and Type_Code to a predefined short code (e.g., "visual operation" is converted to "V01"). Then, the submodule performs a sequential concatenation operation, concatenating the fields into a long string according to the format "ID_Hex+T_s_Hex+T_e_Hex+Type_Code". Next, to optimize transmission efficiency, the submodule applies a lossless compression algorithm (such as a simplified variant of run-length encoding, RLE) to compress and merge the concatenated string, removing redundant consecutive zeros. The resulting compact binary stream or string sequence is the task path merging encoding sequence. For example, the original data concatenated as "ID01-T1000-T2000-V01" may be compressed and encoded into the format "I1:T1k:T2k:V1".

[0037] Byte offset Field Name Field length (bits) Content Description Example data (Hex) 0 Frame header identifier 8 Fixed start code AA 1 Task ID 16 Unique Task Number 010A 3 Window start 32 Timestamp value 00105000 7 Window terminated 32 Timestamp value 001055DC 11 Action type 8 Behavioral Classification Code 05 12 Checksum 8 Accumulated check value E4 13 Frame end identifier 8 Fixed end code FF Table 3 shows the detailed binary structure of the final instruction.

[0038] The scheduling instruction construction submodule calls each group of encoded data in the task path merging encoding sequence, arranges the instruction fields in the standard format of the scheduling structure field, adds a unified format identifier header and control tail mark to all merged encoding structures, and generates a query scheduling instruction set. Based on the protocol standards defined in Table 3, an instruction builder is instantiated. For each set of encoded data in the sequence, the submodule performs the following encapsulation steps: First, write the frame header identifier. The submodule writes a fixed hexadecimal number 0xAA at the beginning of the memory buffer as the start signal of the instruction packet. Second, fill the data payload. The submodule writes the decoded task ID, window start time, window end time, and action type code sequentially to the corresponding byte offset positions. For example, the task ID is written to bytes 1 and 2, and the timestamp is written to bytes 3 to 10. Third, calculate the checksum. The submodule performs an accumulation summation operation on all bytes from the frame header to the action type code, and takes the result modulo 256 to obtain an 8-bit checksum, which is written to the 12th byte position. Fourth, write the frame tail identifier. The submodule writes 0xFF at the end of the buffer to indicate the end of the instruction packet. Fifth, generate the set. The submodule appends the constructed single instruction packet to the output buffer. After all tasks are processed, the complete binary data stream in the output buffer is the query scheduling instruction set. Logical deduction and example: Assume a task ID of 266 (hexadecimal 010A), a start time of 1069056 milliseconds (hexadecimal 00105000), an end time of 1070556 milliseconds (hexadecimal 001055DC), and action type 5 (hexadecimal 05). Checksum calculation logic: The system performs an accumulation operation: 0xAA + 0x01 + 0x0A + 0x00 + 0x10 + 0x50 + 0x00 + 0x00 + 0x10 + 0x55 + 0xDC + 0x05. Assume the accumulated sum is 5348 (decimal). The system performs a modulo 256 operation on 5348: 5348 MOD256 = 228. The hexadecimal number corresponding to 228 is E4. Therefore, the system fills E4 into the check bit. The advantage of this computational logic lies in its ability to ensure the integrity of instructions during transmission by introducing a checksum mechanism. Any bit flip will cause a mismatch between the checksum calculated at the receiving end and the transmitted checksum, triggering a retransmission mechanism and guaranteeing the accuracy of scheduling instruction execution. The final generated instruction set will be a strictly ordered, self-checking control data stream that can be directly parsed by the underlying executor.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. A data labeling, distribution, intelligent scheduling, and permission management system, characterized in that, The system includes: The behavior segmentation module acquires the time record of each data annotation operation in the annotation terminal, forms a continuous operation sequence according to the time axis, compares the time interval of any two adjacent data annotation operations, identifies discontinuous patterns and verifies the distribution, and generates a set of switching segment information. Based on the time position of the switching paragraph information set, the modal decomposition module locates the synchronization position of image segments, text sentences and speech segments, analyzes the edge regions of the image segments, calculates the frequency of differentiated part-of-speech units in the text sentences, extracts the spectral segments of the sound energy change points in the speech segments, and generates an operation path comparison set. The window mapping module maps the start and end points of adjacent action segments according to the action type structure order of the operation path reference set, reconstructs the start and end boundaries of the action segments, and generates a list of window segment structures. The query orchestration module maps the start field of each query task to the dispatch identifier based on the start and end boundaries of the window segment structure list, filters the matching set of window task paths, and generates a task query path call table. The instruction generation module uses the window start and end information and task identifier in the task query path call table to match the undistributed task number and encode and merge it with the query path to generate a query scheduling instruction set.

2. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that: The switching segment information set includes an interval time threshold, operation sequence identifier, and time distribution nodes. The operation path reference set includes image edge action features, text part-of-speech frequency units, and speech energy change structure. The window segment structure list includes action start and end positions, path type labels, and window sorting index. The task query path call table includes field mapping relationships, boundary matching index, and path task number. The query scheduling instruction set includes scheduling number information, path encoding results, and task distribution identifier.

3. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that: The frequency of differentiated parts of speech refers to the frequency of occurrence of differentiated parts of speech in a text sentence, and measures the differential distribution in the text.

4. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that: The start and end points of the mapped adjacent action segments are verified and correspond to the start and end positions of adjacent action segments according to the action type structure order.

5. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that, The behavior segmentation module includes: The data receiving submodule obtains the operation timestamp corresponding to each data annotation operation in the annotation terminal, and assembles the obtained operation timestamp sequence into a continuous operation record sequence according to the time axis order based on the receiving order, generating an operation time sorted sequence. The operation interval calculation submodule, based on the two adjacent timestamps in the operation time sorting sequence, calls the numerical difference between the timestamps respectively, uses the absolute time difference as the calculation basis, performs a subtraction operation item by item, and generates an adjacent operation interval sequence of the same length. The sequence distribution recognition submodule calls the discontinuous mode discrimination rule based on the time interval value in the adjacent operation interval sequence, and judges whether there is a sudden change point with a significant difference from the previous and next values ​​based on the continuous interval interval benchmark value and the sudden change judgment threshold. It marks the index position information corresponding to the sudden change point, and calls the time value corresponding to the index position in the overall sequence to generate a switching segment information set.

6. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that, The modal decomposition module includes: The synchronous positioning submodule obtains the time position of the switching segment information, and locates the time frame range of image segments, text sentences and voice segments in the data to be processed according to the time axis correspondence. It also defines the synchronization position of the three types of data according to the modal start and end time overlap interval, and generates the modal synchronization position interval. The modal feature extraction submodule performs image grayscale gradient detection to extract edge pixel distribution density based on image segments, text sentences and speech segments in the modal synchronization position interval, counts the frequency of occurrence of differentiated part-of-speech units in the text, and extracts the frequency band frame range covered by the sound energy change points in the spectrogram to generate a three-modal change index group of image, text and sound. The action structure recognition submodule, based on the feature values ​​corresponding to the modalities in the three-modal change index group of image, text and audio, divides the action occurrence patterns in the synchronous segment according to the combination relationship between the degree of image edge change, the proportion of different parts of speech in the text and the number of frames covered by the speech spectrum segment, establishes the temporal comparison path relationship between modalities, and generates an operation path comparison set.

7. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that, The window mapping module includes: The action sequence parsing submodule collects the numbering and time arrangement information of action segments in the continuous processing task based on the action type structure order in the operation path reference set, establishes a sequence index table according to the relationship between action types in the sequence, and generates an action sequence index sequence. The boundary mapping and sorting submodule detects the start and end time values ​​of adjacent action segments based on the action sequence index, performs corresponding mapping and alignment processing on the start and end points, eliminates overlapping and gap segments, and generates action segment boundary mapping intervals. The window sorting construction submodule calls the action fragment boundary mapping interval, performs linear rearrangement of the interval according to the operation path arrangement order, calculates the sequence number and span length of the interval on the time axis, constructs the window segment sorting relationship between action fragments, and generates a window segment structure list.

8. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that, The query orchestration module includes: The field mapping matching submodule obtains the start and end boundaries of windows in the window segment structure list, collects the start field and distribution identifier information of all query tasks in the current scheduling request, and performs field-level mapping and positioning of fields one by one according to the structural index relationship between field values ​​and window boundaries to generate field window mapping index values. The boundary overlap determination submodule matches and determines whether the index position corresponding to each query task field falls within the window boundary index range based on the field window mapping index value. According to the window boundary range interval rules, it calculates whether the field index falls within the window segment and generates a field overlap status determination sequence. The path set filtering submodule calls the field overlap status determination sequence to filter all window segment numbers that have a complete overlap with the field index, extracts the corresponding task path structure information, and combines and arranges them according to the order in which the fields appear to generate a task query path call table.

9. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 1, characterized in that, The instruction generation module includes: The undistributed task screening submodule obtains the window start and end information and task identifier field included in each query path in the task query path call table, detects all task numbers in the scheduled task list, and determines the execution status based on the distribution status identifier of each task number, screens out the set of task numbers that have not yet been distributed, and generates a list of schedulable task numbers. The path task encoding submodule calls the path structure that matches the number in the task query path call table according to the task number in the schedulable task number list, extracts the window start and end values ​​and task identifier fields in the path corresponding to each number, performs sequential concatenation and field compression and merging operations on the path content and number, and generates a task path merge encoding sequence. The scheduling instruction construction submodule calls each group of encoded data in the task path merging encoding sequence, arranges the instruction fields in sequence according to the standard format of the scheduling structure field, adds a unified format identifier header and control tail mark to all merged encoding structures, and generates a query scheduling instruction set.

10. The data labeling, distribution, intelligent scheduling, and permission management system according to claim 9, characterized in that: The execution status determination of the distribution status identifier refers to determining whether the task has been distributed and its current execution status based on the task's distribution status identifier.

Images