Man-machine collaborative learning system and method based on stream data driving

Abstract

The embodiment of the invention discloses a man-machine collaborative learning system and method based on stream data driving, and belongs to the technical field of man-machine collaborative learning, and the system comprises an initialization module, a calculation module and an evaluation module, the initialization processing comprises data preprocessing, basic model training and task demand analysis; the calculation module is used for selecting a man-machine cooperation mode according to the task type, and dynamically distributing the subtasks to a human calculation unit and a machine calculation unit for corresponding processing according to a task distribution mechanism; and the evaluation module is used for constructing a multi-dimensional evaluation system to evaluate a task processing result and carrying out iterative optimization on low-confidence data, and the low-confidence data is the input stream data of which the task processing accuracy is lower than an expected threshold. According to the invention, the accuracy and efficiency of man-machine collaborative learning based on streaming data can be improved, and the cost can be reduced.

Description

Technical Field

[0001] The present application relates to the technical field of human-computer collaborative learning, and is related to but not limited to a human-computer collaborative learning system and method based on stream data drive. Background Art

[0002] In recent years, with the in-depth research of deep learning, its performance in various practical applications has surpassed most machine learning methods. Although the accuracy of task execution can be improved through the use of neural network architectures in deep learning, in special cases such as a lack of training samples, the output is sometimes unclear and difficult to understand, resulting in poor problem solving.

[0003] To address the shortcomings of deep learning, existing technologies have proposed human-machine collaborative intelligence (HMI) technologies, effectively combining human and machine intelligence to tackle complex tasks in data-driven scenarios. For example, the first team proposed an uncertainty quantification framework that incorporates uncertainty and performance into AI-assisted human decision-making to enhance user trust. The second team developed a medical image retrieval system that leverages visual matching of historically similar cases to assist in diagnosing new patients, reducing misdiagnosis rates and ensuring reliable medical decisions. The third team proposed a hierarchical task allocation model that combines AI and human intelligence in tasks such as object detection and speech recognition to improve detection efficiency in complex scenarios. However, existing HMI technologies often employ a static "machine-first, human-second" division of labor, preventing dynamic role switching and leading to issues such as role stabilization. High latency in human-machine interaction prevents the full potential of parallel processing, resulting in low collaborative efficiency. The lack of visual display of the collaborative process during HMI collaboration can lead to poor user acceptance of hybrid decision-making outcomes and a lack of trust mechanisms.

[0004] Therefore, there is an urgent need for a more comprehensive and reasonable human-computer collaboration method to solve the many problems existing in existing human-computer collaborative intelligent technologies, so as to achieve improvements in the accuracy, flexibility and efficiency of solving complex tasks through human-computer collaboration methods, as well as a reduction in human workload. Summary of the Invention

[0005] The embodiments of the present application provide a human-computer collaborative learning system and method based on stream data drive.

[0006] The technical solution of the embodiment of the present application is implemented as follows: In a first aspect, an embodiment of the present application provides a human-computer collaborative learning system driven by streaming data, the system comprising an initialization module, a calculation module, and an evaluation module, wherein: The initialization module is used to perform initialization processing on the input stream data, and the initialization processing includes data preprocessing, basic model training and task requirement analysis; the calculation module is used to select the human-computer collaboration mode according to the task type, and dynamically allocate subtasks to human computing units and machine computing units for corresponding processing according to the task allocation mechanism; the evaluation module is used to construct a multi-dimensional evaluation system to evaluate the task processing results, and iteratively optimize the low-confidence data, and the low-confidence data is the input stream data whose task processing accuracy is lower than the expected threshold.

[0007] The technical solution provided by the present application initializes the input stream data through the initialization module, and performs data preprocessing on the input stream data to eliminate interference factors in the processing process and obtain clean data, which improves the accuracy of the processing results to a certain extent. By completing the basic model training according to the clean data and continuously optimizing the basic model in a cycle, the human-computer collaborative learning system can improve the accuracy of task processing to a certain extent compared with the traditional machine learning model with less manpower input, and effectively prevent the cliff-like decline of model performance. Task demand analysis is performed based on the clean data to determine the task demand, which facilitates the smooth progress of the subsequent processing process. Through the calculation module, the human-computer collaboration mode is selected according to the task type, and subtasks are dynamically assigned to humans according to the task allocation mechanism. The computing unit and the machine computing unit perform corresponding processing, and dynamically match the optimal processing solution through the collaborative computing mode of the quintuple model, thereby improving the efficiency of task processing. The dynamic allocation of tasks to be processed is flexibly adjusted through different human-machine working modes and human-machine collaborative modes, effectively preventing the occurrence of misjudgments, reducing the error rate of task processing, greatly improving the task accuracy, and improving the utilization rate of machine resources. A multi-dimensional evaluation system is constructed through the evaluation module to evaluate the task processing results. Iterative optimization of low-confidence data in a closed-loop form can maximize the value of data, achieve a breakthrough in the precision and accuracy of task processing results, and reduce the frequency of expert intervention to achieve a reduction in labor costs. The closed-loop optimization method can accelerate model iteration, shorten its training cycle, and achieve continuous feedback of model evolution.

[0008] Optionally, the initialization module includes a data preprocessing unit, a model training unit and a task analysis unit, wherein: the data preprocessing unit is used to identify and filter noise data in the input stream data to obtain cleaned data; the model training unit is used to complete basic model training based on the cleaned data to obtain initial parameters; the task analysis unit is used to determine task requirements, analyze task complexity and formulate initial strategies based on the cleaned data, and the task requirements include task type, expected task processing accuracy, expected task processing time and task processing cost.

[0009] Optionally, the computing module adopts a quintuple model, which includes the human computing unit, the machine computing unit, the human-computer working mode, the human-computer collaboration mode and performance indicators, wherein: the human-computer working mode is the working mode in which the human computing unit and the machine computing unit complete task processing, and the human-computer working mode includes a parallel working mode and a serial working mode; the human-computer collaboration mode is the collaboration mode when the human computing unit and the machine computing unit collaborate to complete task processing, and the human-computer collaboration mode includes an interactive mode, a feedback mode and an enhanced mode; the performance indicators include task processing accuracy, task processing time and task processing cost, and the multidimensional evaluation system is composed of the performance indicators.

[0010] Optionally, the parallel working mode is that the human computing unit and the machine computing unit simultaneously process different subtasks of the same task and realize complementary computing through real-time interaction; the serial working mode is that after the machine computing unit completes the preliminary calculation, the human computing unit verifies or corrects the low-confidence results; the interactive mode is a collaborative mode in which the human computing unit and the machine computing unit interact and correct intermediate results in real time through the interactive interface; the feedback mode is a collaborative mode in which the human computing unit provides optimization suggestions to the machine computing unit after the machine computing unit completes the calculation; the enhanced mode is a mode in which the human computing unit guides the machine computing unit to adjust parameters to complete collaborative work.

[0011] Optionally, the task allocation mechanism includes: dividing the input stream data into a task set containing multiple tasks, each task including a manual processing subtask and a machine processing subtask; calculating the confidence of each subtask, when the confidence is greater than or equal to a configurable threshold, the indicator variable is 1, and the subtask corresponding to the confidence is the machine processing subtask, when the confidence is less than the configurable threshold, the indicator variable is 0, and the subtask corresponding to the confidence is the manual processing subtask; allocating the manual processing subtasks according to dynamic user portraits, novice users handle low-complexity subtasks, and expert users handle high-complexity and high-risk subtasks, and the dynamic user portrait allocation is dynamically updated through task historical performance data, user feedback quality, task completion rate and error type statistics; when the subtask processing trend does not meet expectations, data denoising, model replacement or manual expert intervention is triggered, and when the processing trend is as expected, machine resource allocation is increased.

[0012] Optionally, the system also includes a conflict resolution module, which includes: a prediction result generation unit, which is used to generate an original decision for processing the machine processing subtask through the machine computing unit, and output the confidence corresponding to the machine processing subtask and the key features of the machine processing subtask that affect the original decision; an expert review unit, which is used to review the original decision through experts. When the expert raises an objection to the original decision, the system displays the corresponding key features of the machine processing subtask to the expert and accepts the expert's revised decision. If a single expert cannot resolve the conflict, a multi-expert comprehensive analysis is initiated to generate a final revised decision; a weight allocation unit, which is used to dynamically allocate weights of the manual processing subtask and the machine processing subtask based on historical performance data and task complexity, assigning high weights to machine processing subtasks in high-precision scenarios, and assigning high weights to manual processing subtasks in low-precision and high-risk scenarios; a human-machine iterative correction unit, which is used to iteratively optimize the basic model using the final revised decision as a training sample, and experts select difficult samples for targeted retraining.

[0013] Optionally, the system also includes a model adjustment module, which includes: a difficult sample collection unit, used to identify erroneous prediction samples output by the basic model, use the erroneous prediction samples as difficult samples, and screen and collect the difficult samples through experts to construct an augmented training set; a multimodal feedback unit, used to correct and score the basic model output through explicit feedback, infer feedback through implicit feedback analysis of user interaction behavior, and guide humans to collect feedback through preset tools through structured annotations to ensure that humans provide high-quality input; a model fine-tuning unit, used to input the augmented training set and multimodal feedback data into the basic model for model fine-tuning.

[0014] In the second aspect, an embodiment of the present application provides a human-computer collaborative learning method driven by stream data, which is applied to a human-computer collaborative learning system, and the method includes: initializing the input stream data, and the initialization processing includes data preprocessing, basic model training and task requirement analysis; selecting a human-computer collaboration mode according to the task type, and dynamically allocating subtasks to human computing units and machine computing units for corresponding processing according to the task allocation mechanism; constructing a multidimensional evaluation system to evaluate the task processing results, and iteratively optimizing low-confidence data, and the low-confidence data is the input stream data whose task processing accuracy is lower than the expected threshold.

[0015] In a third aspect, an embodiment of the present application provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, it implements the steps in the above-mentioned human-computer collaborative learning method driven by streaming data.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-mentioned human-computer collaborative learning method driven by streaming data.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present application include at least: The present application provides a human-machine collaborative learning system and method based on stream data drive, which initializes the input stream data through the initialization module, and performs data preprocessing on the input stream data to eliminate interference factors in the processing process and obtain clean data, thereby improving the accuracy of the processing results to a certain extent. By completing the basic model training based on the clean data and continuously optimizing the basic model in a cycle, the human-machine collaborative learning system can improve the accuracy of task processing to a certain extent compared with the traditional machine learning model with less manpower input, and effectively prevent the cliff-like decline of model performance. Task demand analysis is performed based on the clean data to determine the task demand, which facilitates the smooth progress of the subsequent processing process. Through the calculation module, the human-machine collaboration mode is selected according to the task type, and the sub-tasks are assigned according to the task allocation mechanism. Tasks are dynamically assigned to human computing units and machine computing units for corresponding processing. The optimal processing solution is dynamically matched through the collaborative computing mode of the quintuple model, which improves the efficiency of task processing. The dynamic allocation of tasks to be processed is flexibly adjusted through different human-machine working modes and human-machine collaborative modes, effectively preventing the occurrence of misjudgments, reducing the error rate of task processing, greatly improving task accuracy, and improving machine resource utilization. A multi-dimensional evaluation system is constructed through the evaluation module to evaluate the task processing results. Iterative optimization of low-confidence data in a closed-loop form can maximize the value of data, achieve a breakthrough in the precision and accuracy of task processing results, and reduce the frequency of expert intervention to achieve a reduction in labor costs. The closed-loop optimization method can accelerate model iteration, shorten its training cycle, and achieve continuous feedback of model evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present application. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 A schematic diagram of a human-computer collaborative learning system driven by streaming data provided in an embodiment of the present application; Figure 2 A flow chart of a human-computer collaborative learning method based on stream data drive provided in an embodiment of the present application; Figure 3A hardware entity diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. The following examples are used to illustrate the present application, but are not intended to limit the scope of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present application are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0022] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as generally understood by those skilled in the art in the art to which the embodiments of the present application belong. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0023] The embodiments of the present application will be further described below with reference to the accompanying drawings.

[0024] In view of the current problems in the field of human-computer collaborative learning technology with respect to human-computer collaborative learning based on stream data drive, the embodiments of the present application provide a human-computer collaborative learning system and method based on stream data drive.

[0025] The technical solution of the present application is introduced below, and first the system embodiment of the present application is introduced.

[0026] Please refer to Figure 1 , which shows a schematic diagram of a human-computer collaborative learning system based on stream data drive provided by an embodiment of the present application, such as Figure 1As shown, the system includes an initialization module 01, a calculation module 02 and an evaluation module 03. The initialization module 01, the calculation module 02 and the evaluation module 03 are connected in sequence. The initialization module 01 is used to perform initialization processing on the input stream data, and the initialization processing includes data preprocessing, basic model training and task requirement analysis; the calculation module 02 is used to select the human-computer collaboration mode according to the task type, and dynamically allocate subtasks to human computing units and machine computing units for corresponding processing according to the task allocation mechanism; the evaluation module 03 is used to construct a multi-dimensional evaluation system to evaluate the task processing results, and iteratively optimize the low-confidence data, which is the input stream data whose task processing accuracy is lower than the expected threshold.

[0027] In an embodiment of the present application, initialization module 01 includes a data preprocessing unit, a model training unit, and a task analysis unit. The data preprocessing unit is used to identify and filter noise data in the input stream data to obtain cleaned data. Specifically, because the input stream data often contains more or less noise data, which will affect subsequent processing stages, the data preprocessing unit identifies and filters the noise data in the input stream data (for example, through wavelet transform denoising and 3σ outlier detection to identify and filter the noise data in the input stream data) to complete preprocessing, thereby eliminating interference factors in the processing process and obtaining cleaned data, which helps to smoothly carry out the subsequent processing process and improves the accuracy of the processing results to a certain extent.

[0028] In an embodiment of the present application, the model training unit is used to complete basic model training based on the cleaned data and obtain initial parameters. Basic model training is an important part of the initialization module of the human-computer collaborative learning system. During the basic model training process (for example, using a pre-trained residual network (ResNet) backbone network as the basic model and updating the fully connected layer through incremental learning to adapt to specific scenarios), the pre-processed cleaned data is used as the data input of the model to complete the basic model training and obtain the initial parameters. Furthermore, the human-machine collaborative learning system also includes a model adjustment module, which includes a difficult sample collection unit for identifying erroneous prediction samples (or low-confidence samples) output by the basic model, treating the erroneous prediction samples as difficult samples, and constructing an augmented training set by experts to screen, collect and annotate the difficult samples. In addition, difficult samples (such as severely occluded targets and high-frequency sound mixtures, etc.) are collected simultaneously during the training process to enhance the robustness of the model. The model adjustment module also includes a multimodal feedback unit for correcting and scoring the output of the basic model through explicit feedback, and inferring feedback through implicit feedback analysis of user interaction behavior. The user interaction behavior includes selecting operations, coverage and usage patterns, and guiding humans to collect feedback through preset tools through structured annotations to ensure that humans provide high-quality and unbiased input. The model adjustment module also includes a model fine-tuning unit for inputting the augmented training set, difficult samples and multimodal feedback data into the basic model for targeted fine-tuning to ensure consistency with the desired results without overfitting to specific feedback instances, thereby significantly improving the generalization ability of the basic model in complex scenarios (such as intersection over union in pedestrian re-identification tasks). The technical solution provided in the embodiments of the present application performs iterative optimization through a model adjustment module, continuously improving the basic model. This can enable the human-machine collaborative learning system to improve the accuracy by 12-15% compared to traditional machine learning models with no more than 20% human input, and effectively prevent a cliff-like decline in model performance.

[0029] In an embodiment of the present application, a task analysis unit is used to determine task requirements, analyze task complexity, and formulate an initial strategy based on clean data. Task requirements include task type, expected task processing accuracy, expected task processing time, and task processing cost. In an optional embodiment, the task type (e.g., classification task, recognition task, generation task), expected task processing accuracy (e.g., accuracy greater than or equal to 95%), expected task processing time (e.g., maximum latency less than or equal to 260 milliseconds), and task processing cost (e.g., processing cost capped at three experts) are automatically extracted based on the clean data; task complexity is calculated using a multivariate weighted formula; and an initial strategy is formulated using a strategy generator. For example, when the expected task processing time requirement is a latency less than or equal to 60 milliseconds, an emergency response mode is initiated. In this case, the initial strategy is to initiate parallel computing and perform real-time interactive corrections.

[0030] In the embodiment of the present application, the calculation module 02 adopts the quintuple model , the quintuple model includes human computational units , machine computing unit , human-machine working mode , human-machine collaboration mode and performance indicators Specifically, the Human Computing Unit The human computing unit may include an expert knowledge base, optionally, an expert A who is familiar with a specific field and has rich experience, or a group of workers or experts ABC; the machine computing unit The deep learning algorithm library is integrated in the machine, and each machine has its own performance parameters; human-machine working mode The working mode in which the human computing unit and the machine computing unit complete task processing includes parallel working mode and serial working mode. The parallel working mode is that the human computing unit and the machine computing unit simultaneously process different subtasks of the same task and realize complementary calculations through real-time interaction. The serial working mode is that after the machine computing unit completes the preliminary calculation, the human computing unit verifies or corrects the low-confidence results. The expression formula of the human-machine working mode is expressed by the following formula (1): Formula (1); Where, Indicates the human-machine working mode; Human Computational Unit and machine computing units Working in parallel mode, humans and machines simultaneously process different subtasks of the same task, achieving complementary computations through real-time interaction, and with experts observing the intermediate results of the machine processing tasks; Human Computational Unit and machine computing units Working in serial mode, when the machine completes the initial calculation task, humans take relevant measures to verify or correct the low-confidence results.

[0031] Further, human-machine collaboration mode The human-machine collaboration mode is a collaborative mode in which the human computing unit and the machine computing unit collaborate to complete task processing. The human-machine collaboration mode includes an interactive mode, a feedback mode, and an enhanced mode. The interactive mode is a collaborative mode in which the human computing unit and the machine computing unit interact and correct the intermediate results in real time through the interactive interface (the expert performs corresponding operations based on the intermediate output of the interactive interface to achieve real-time correction of the intermediate results). The feedback mode is a collaborative mode in which the human computing unit provides optimization suggestions to the machine computing unit after the machine computing unit completes the calculation (humans can learn new knowledge from the machine and then provide optimization suggestions). The enhanced mode is a mode in which the human computing unit guides the machine computing unit to adjust parameters to complete collaborative work (human experience can provide new methods to guide the machine to adjust parameters to complete the complex situation being solved). The human-machine collaboration mode expression formula is expressed by the following formula (2): Formula (2); Where, Indicates the human-machine collaboration mode; Represents interaction patterns, i.e. human computational units and machine computing units Interactive work: In this mode, an interactive interface is designed, through which experts observe the intermediate output and perform corresponding operations (such as marking, zooming in, and zooming out), thus achieving real-time correction of intermediate results. Indicates feedback mode, that is, when the machine computing unit After completing the calculation, the human computing unit Can be calculated from the machine unit Learn new knowledge from the calculation results and provide feedback on the calculation results (i.e., give optimization suggestions) to promote model construction. For example, in the pedestrian re-identification task, humans find more overlapping parts in some crowded scenes, so the IoU can be set to a smaller value to further reduce occlusion); Represents the enhanced mode, i.e. the human computational unit and machine computing units Working together to improve performance, human computing units Instructing machine computing units Adjust parameters to complete collaborative work, that is, human experience can provide new methods to guide machines to adjust parameters to complete tasks currently performed only by machine computing units. Complex situations are being addressed, for example, in sound detection tasks, where multiple sounds are mixed in a short period of time and the frequencies of each sound are different, so at this time, humans can guide the machine's enhanced mode to use multiple frequencies to distinguish coarse-grained categories, thereby better completing the sound detection task.

[0032] In the embodiment of this application, the performance index Including task processing accuracy, task processing time and task processing cost, the multi-dimensional evaluation system is composed of the above performance indicators. For a specified task, its corresponding execution performance expression formula is expressed by the following formula (3): Formula (3); Where, Indicates the execution performance corresponding to a specified task; Indicates the task processing accuracy corresponding to the task; Indicates the task processing time corresponding to the task; Indicates the task processing cost corresponding to the task.

[0033] In the embodiment of the present application, subtasks are dynamically assigned to human computing units and machine computing units for corresponding processing according to the task allocation mechanism. The specific process is as follows: the input stream data is divided into a task set containing multiple tasks. , Represents the number of tasks, where each task Includes multiple subtasks , each subtask It contains multiple manual processing subtasks and machine processing subtasks, such as speech recognition tasks , which includes the voice data cleaning subtask , feature extraction subtask and speech data classification subtask For each task , which includes subtasks The total expression formula is expressed by the following formula (4): Formula (4); Where, Indicates a task Subtasks included in total; Indicates the total number of manual processing subtasks assigned to the manual processing unit; Indicates the total number of machine processing subtasks assigned to the machine processing unit.

[0034] Furthermore, the confidence of each subtask is calculated , when the confidence Greater than or equal to a configurable threshold When the indicator variable is 1, that is, when hour, , at this time, the confidence Corresponding subtasks For machine processing subtasks, the machine processing subtasks are assigned to the machine processing unit for the next step of processing; when the confidence Less than a configurable threshold When the indicator variable is 0, that is, when hour, , at this time, the confidence Corresponding subtasks For manual processing subtasks, the manual processing subtask is assigned to the manual processing unit for further processing.

[0035] Furthermore, due to the imprecision of human capabilities, it is necessary to accurately quantify the human skills required for tasks. However, accurately quantifying the human skills required for tasks may place a heavy burden on the system. Therefore, a fuzzy rule-based quality assessment strategy is adopted to ensure the initial allocation of specific tasks. Manual processing subtasks are assigned based on dynamic user profiles. Dynamic user profile assignments are dynamically updated based on historical task performance data, user feedback quality, task completion rate, and error type statistics. The hierarchical task allocation mechanism is crucial for assigning different subtasks to users of different levels. Novice users handle low-complexity subtasks (such as labeling in annotation tasks), while expert users handle high-complexity and high-risk subtasks (such as medical diagnosis). Furthermore, when the subtask processing trend does not meet expectations, data denoising, model replacement, or human expert intervention is triggered. When the processing trend meets expectations, machine resource allocation is increased.

[0036] In an embodiment of the present application, the human-machine collaborative learning system also includes a conflict resolution module, which includes a prediction result generation unit, an expert review unit, a weight allocation unit, and a human-machine iterative correction unit. Specifically, the prediction result generation unit is used to generate an original decision for processing a machine processing subtask through a machine calculation unit, and output the confidence corresponding to the machine processing subtask and the key features of the machine processing subtask that affect the original decision; the expert review unit is used to review the original decision through an expert. When the expert raises an objection to the original decision, the system displays the corresponding key features of the machine processing subtask to the expert and accepts the expert's revised decision. If a single expert cannot resolve the conflict, a multi-expert comprehensive analysis is initiated to generate a final revised decision; the weight allocation unit is used to dynamically allocate weights of manual processing subtasks and machine processing subtasks based on historical performance data and task complexity, assigning high weights to machine processing subtasks in high-precision scenarios, and assigning high weights to manual processing subtasks in low-precision and high-risk scenarios; the human-machine iterative correction unit is used to use the final revised decision as a training sample to iteratively optimize the basic model, and the expert selects difficult samples for targeted retraining.

[0037] In the embodiment of the present application, the evaluation module 03 is used to construct a multidimensional evaluation system to evaluate the task processing results and iteratively optimize the low-confidence data, which is the input stream data whose task processing accuracy is lower than the expected threshold. Specifically, the evaluation module is an intelligent quality inspection center in the human-computer collaborative learning system. It quantifies the task processing results in real time through the multidimensional evaluation system, thereby performing corresponding evaluations. For input stream data whose task processing accuracy is lower than the expected threshold, it is iteratively optimized. The multidimensional evaluation system is composed of performance indicators, including task processing accuracy, task processing time, and task processing cost. The task processing cost is generally obtained by converting the consumption of CPU, GPU, and human resources. In one embodiment, an expected threshold value for task processing accuracy is set. When the current task processing accuracy is greater than or equal to the expected threshold value, it indicates that the current task processing accuracy has met the index requirement of the task processing accuracy rate; the task processing time is required to reach millisecond-level processing delay, that is, when the stream data window length is less than or equal to 1 second, it indicates that the current task processing time has met the index requirement of the task processing time.

[0038] In an optional embodiment, data confidence is calculated in real time by a confidence monitoring engine, and low-confidence data is marked through dual-threshold redundancy check. For example, if the task processing accuracy of task A is less than the expected threshold and the task processing time is greater than the expected threshold, task A will be marked as low-confidence data. The confidence of the input stream data is calculated in real time. For high-confidence data, the corresponding processing results are directly output; for low-confidence data, there are two processing methods: for simple input stream data, automatic denoising and enhancement processing is performed on it, which can be achieved through a denoising autoencoder; for complex input stream data, expert correction processing is performed on it, which can be completed with the help of a real-time annotation platform for WebSockets. The complexity of the input stream data can be obtained by processing the complexity classifier of the lightweight gradient boosting machine (LightGBM). After the low-confidence data is processed, it is added to the augmented dataset for model fine-tuning, and the corresponding task processing is re-performed and the processing results are evaluated. For high-confidence data that meets the standard, its processing results are directly output; for low-confidence data that does not meet the standard, it is iteratively processed again using the above method until it becomes high-confidence data that meets the standard, and its processing results are directly output. The above is the process of iterative optimization of low-confidence data. Using this closed-loop approach to iteratively process low-confidence data maximizes data value, significantly improving the precision and accuracy of task processing results. It also reduces the frequency of expert intervention, lowering labor costs. This closed-loop optimization approach accelerates model iteration, shortens training cycles, and enables continuous feedback loops for model evolution. It's important to note that the evaluation module and conflict resolution module operate in deep collaboration, automatically triggering a multi-expert review process when low-confidence data streams are continuously detected.

[0039] In summary, the embodiment of the present application provides a human-machine collaborative learning system based on stream data drive, which initializes the input stream data through the initialization module, and performs data preprocessing on the input stream data to eliminate interference factors in the processing process and obtain clean data, thereby improving the accuracy of the processing results to a certain extent. By completing the basic model training according to the clean data and continuously optimizing the basic model in a cycle, the human-machine collaborative learning system can improve the accuracy of task processing to a certain extent compared with the traditional machine learning model with less manpower input, and effectively prevent the cliff-like decline of model performance. Task demand analysis is performed based on the clean data to determine the task demand, which facilitates the smooth progress of the subsequent processing process. Through the calculation module, the human-machine collaboration mode is selected according to the task type, and according to the task allocation mechanism Subtasks are dynamically assigned to human computing units and machine computing units for corresponding processing. The optimal processing solution is dynamically matched through the collaborative computing mode of the quintuple model, which improves the efficiency of task processing. The dynamic allocation of tasks to be processed is flexibly adjusted through different human-machine working modes and human-machine collaborative modes, effectively preventing the occurrence of misjudgments, reducing the error rate of task processing, greatly improving task accuracy, and improving machine resource utilization. A multi-dimensional evaluation system is constructed through the evaluation module to evaluate the task processing results. Iterative optimization of low-confidence data in a closed-loop form can maximize the value of data, achieve a breakthrough in the precision and accuracy of task processing results, and reduce the frequency of expert intervention to achieve a reduction in labor costs. The closed-loop optimization method can accelerate model iteration, shorten its training cycle, and achieve continuous feedback of model evolution.

[0040] The above is an introduction to the system embodiment of the present application. Based on the aforementioned embodiment, the method embodiment of the present application is introduced below.

[0041] Please refer to Figure 2 , which shows a flow chart of a human-computer collaborative learning method based on stream data driven provided by an embodiment of the present application, the method is applied to Figure 1 The human-machine collaborative learning system shown is based on stream data drive. For details not disclosed in the method embodiment, please refer to the system embodiment. Figure 2 As shown, the startup method includes the following steps S210 to S230.

[0042] Step S210: Initialize the input stream data. The initialization process includes data preprocessing, basic model training, and task requirement analysis.

[0043] In an embodiment of the present application, clean data is obtained by identifying and filtering noise data in the input stream data; initial parameters are obtained by completing basic model training based on the clean data; task requirements are determined based on the clean data, task complexity is analyzed, and initial strategies are formulated, wherein the task requirements include task type, expected task processing accuracy, expected task processing time, and task processing cost.

[0044] Step S220 , selecting a human-machine collaboration mode according to the task type, and dynamically allocating subtasks to the human computing unit and the machine computing unit for corresponding processing according to the task allocation mechanism.

[0045] In an embodiment of the present application, task processing is completed by adopting a quintuple model, which includes a human computing unit, a machine computing unit, a human-computer working mode, a human-computer collaboration mode and performance indicators, wherein the human computing unit contains multiple expert knowledge bases; the machine computing unit integrates a deep learning algorithm library; the human-computer working mode is the working mode in which the human computing unit and the machine computing unit complete task processing, and the human-computer working mode includes a parallel working mode and a serial working mode; the human-computer collaboration mode is the collaboration mode when the human computing unit and the machine computing unit collaborate to complete task processing, and the human-computer collaboration mode includes an interactive mode, a feedback mode and an enhanced mode; the performance indicators include task processing accuracy, task processing time and task processing cost, and the multidimensional evaluation system is composed of the performance indicators.

[0046] In an embodiment of the present application, the parallel working mode is that the human computing unit and the machine computing unit simultaneously process different subtasks of the same task and realize complementary calculations through real-time interaction; the serial working mode is that after the machine computing unit completes the preliminary calculation, the human computing unit verifies or corrects the low-confidence results; the interactive mode is a collaborative mode in which the human computing unit and the machine computing unit interact and correct the intermediate results in real time through the interactive interface; the feedback mode is a collaborative mode in which the human computing unit provides optimization suggestions to the machine computing unit after the machine computing unit completes the calculation; the enhanced mode is a mode in which the human computing unit guides the machine computing unit to adjust parameters to complete collaborative work.

[0047] In an embodiment of the present application, the task allocation mechanism includes: dividing the input stream data into a task set containing multiple tasks, each task including a manual processing subtask and a machine processing subtask; calculating the confidence of each subtask, when the confidence is greater than or equal to a configurable threshold, the indicator variable is 1, and the subtask corresponding to the confidence is the machine processing subtask, when the confidence is less than the configurable threshold, the indicator variable is 0, and the subtask corresponding to the confidence is the manual processing subtask; allocating the manual processing subtasks according to dynamic user profiles, novice users handle low-complexity subtasks, and expert users handle high-complexity and high-risk subtasks, and the dynamic user profile allocation is dynamically updated through task historical performance data, user feedback quality, task completion rate and error type statistics; when the subtask processing trend does not meet expectations, data denoising, model replacement or manual expert intervention is triggered, and when the processing trend is in line with expectations, machine resource allocation is increased.

[0048] In an embodiment of the present application, the machine processing subtask is processed by the machine computing unit and a corresponding original decision is generated, and the confidence corresponding to the machine processing subtask and the key features of the machine processing subtask that affect the original decision are output; the original decision is reviewed by an expert, and when the expert raises an objection to the original decision, the system displays the corresponding key features of the machine processing subtask to the expert and accepts the expert's revised decision. If a single expert cannot resolve the conflict, a multi-expert comprehensive analysis is initiated to generate a final revised decision; the weights of the manual processing subtask and the machine processing subtask are dynamically allocated based on historical performance data and task complexity, and high weights are allocated to machine processing subtasks in high-precision scenarios, and high weights are allocated to manual processing subtasks in low-precision and high-risk scenarios; the final revised decision is used as a training sample to iteratively optimize the basic model, and experts select difficult samples for targeted retraining.

[0049] In an embodiment of the present application, erroneous prediction samples output by the base model are identified, and the erroneous prediction samples are used as difficult samples. Experts screen and collect the difficult samples to construct an augmented training set. The base model output is corrected and scored through explicit feedback, and feedback is inferred by analyzing user interaction behavior through implicit feedback. Structured annotations are used to guide humans to collect feedback through preset tools to ensure that humans provide high-quality input. The augmented training set and multimodal feedback data are input into the base model for model fine-tuning.

[0050] Step S230 , constructing a multi-dimensional evaluation system to evaluate the task processing results, and iteratively optimizing the low-confidence data, wherein the low-confidence data is the input stream data whose task processing accuracy is lower than the expected threshold.

[0051] In summary, the embodiment of the present application provides a human-machine collaborative learning method based on stream data drive, which initializes the input stream data through the initialization module, and performs data preprocessing on the input stream data to eliminate interference factors in the processing process and obtain clean data, thereby improving the accuracy of the processing results to a certain extent. By completing the basic model training according to the clean data and continuously optimizing the basic model in a cycle, the human-machine collaborative learning system can improve the accuracy of task processing to a certain extent compared with the traditional machine learning model with less manpower investment, and effectively prevent the cliff-like decline of model performance. Task demand analysis is performed based on the clean data to determine the task demand, which facilitates the smooth progress of the subsequent processing process. Through the calculation module, the human-machine collaboration mode is selected according to the task type, and according to the task allocation mechanism Subtasks are dynamically assigned to human computing units and machine computing units for corresponding processing. The optimal processing solution is dynamically matched through the collaborative computing mode of the quintuple model, which improves the efficiency of task processing. The dynamic allocation of tasks to be processed is flexibly adjusted through different human-machine working modes and human-machine collaborative modes, effectively preventing the occurrence of misjudgments, reducing the error rate of task processing, greatly improving task accuracy, and improving machine resource utilization. A multi-dimensional evaluation system is constructed through the evaluation module to evaluate the task processing results. Iterative optimization of low-confidence data in a closed-loop form can maximize the value of data, achieve a breakthrough in the precision and accuracy of task processing results, and reduce the frequency of expert intervention to achieve a reduction in labor costs. The closed-loop optimization method can accelerate model iteration, shorten its training cycle, and achieve continuous feedback of model evolution.

[0052] It should be noted that in the embodiments of the present application, if the above-mentioned human-computer collaborative learning method based on stream data driving is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can be essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiments of the present application are not limited to any specific combination of hardware and software.

[0053] Correspondingly, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the aforementioned methods for human-computer collaborative learning based on stream data drive. Correspondingly, an embodiment of the present application also provides a computer program product, which, when executed by a processor of an electronic device, is used to implement the steps of any of the aforementioned methods for human-computer collaborative learning based on stream data drive.

[0054] Based on the same technical concept, an embodiment of the present application provides an electronic device for implementing a human-computer collaborative learning method based on stream data drive as described in the above method embodiment. Figure 3 A hardware entity diagram of an electronic device provided in an embodiment of the present application is shown as follows: Figure 3 As shown, the electronic device 300 includes a memory 310 and a processor 320, and the memory 310 stores a computer program that can be run on the processor 320. When the processor 320 executes the program, it implements the steps of any one of the human-computer collaborative learning methods based on stream data drive described in the embodiments of the present application.

[0055] The memory 310 is configured to store instructions and applications executable by the processor 320, and can also cache data to be processed or processed by the processor 320 and various modules in the electronic device (for example, image data, audio data, voice communication data, and video communication data), which can be implemented through flash memory (FLASH) or random access memory (RAM).

[0056] When the processor 320 executes the program, the processor 320 implements any of the steps of the above-mentioned human-machine collaborative learning method based on stream data driving. The processor 320 generally controls the overall operation of the electronic device 300.

[0057] The processor may be at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, and a microprocessor. It is understood that the electronic device that implements the functions of the processor may also be other electronic devices, and the embodiments of the present application are not specifically limited thereto.

[0058] The above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface storage device, an optical disc, or a compact disc read-only memory (CD-ROM); it can also be various electronic devices that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0059] It should be noted that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0060] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0061] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0062] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0063] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0064] In addition, all functional units in the embodiments of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0065] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling the automatic test line of the device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks or optical disks.

[0066] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0067] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0068] The above is merely an embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A human-computer collaborative learning system driven by streaming data, characterized in that: The system comprises: An initialization module is used to perform initialization processing on the input stream data, which includes data preprocessing, basic model training and task requirement analysis; The computing module is used to select the human-machine collaboration mode according to the task type and dynamically allocate subtasks to the human computing unit and the machine computing unit for corresponding processing according to the task allocation mechanism; The evaluation module is used to construct a multi-dimensional evaluation system to evaluate the task processing results and iteratively optimize the low-confidence data, where the low-confidence data is the input stream data whose task processing accuracy is lower than the expected threshold.

2. The system according to claim 1, wherein: The initialization module includes a data preprocessing unit, a model training unit and a task analysis unit, wherein: The data preprocessing unit is used to identify and filter noise data in the input stream data to obtain cleaned data; The model training unit is used to complete basic model training according to the cleaned data to obtain initial parameters; The task analysis unit is used to determine task requirements, analyze task complexity and formulate initial strategies based on the cleansing data. The task requirements include task type, expected task processing accuracy, expected task processing time and task processing cost.

3. The system according to claim 1, wherein: The computing module adopts a five-tuple model, which includes the human computing unit, the machine computing unit, the human-machine working mode, the human-machine collaboration mode and the performance index, wherein: The human-machine working mode is a working mode in which the human computing unit and the machine computing unit complete task processing, and the human-machine working mode includes a parallel working mode and a serial working mode; The human-machine collaboration mode is a collaboration mode in which the human computing unit and the machine computing unit collaborate to complete task processing, and the human-machine collaboration mode includes an interactive mode, a feedback mode, and an enhanced mode; The performance indicators include task processing accuracy, task processing time and task processing cost, and the multi-dimensional evaluation system is composed of the performance indicators.

4. The system according to claim 3, characterized in that The parallel working mode is that the human computing unit and the machine computing unit process different subtasks of the same task simultaneously and realize complementary computing through real-time interaction; The serial working mode is that after the machine computing unit completes the preliminary calculation, the human computing unit verifies or corrects the low-confidence results; The interactive mode is a collaborative mode in which the human computing unit and the machine computing unit interact with each other and correct intermediate results in real time through an interactive interface; The feedback mode is a collaborative mode in which the human computing unit provides optimization suggestions to the machine computing unit after the machine computing unit completes the calculation; The enhanced mode is a mode in which the human computing unit instructs the machine computing unit to adjust parameters to complete collaborative work.

5. The system according to claim 4, characterized in that The task allocation mechanism includes: Splitting the input stream data into a task set comprising a plurality of tasks, each task including a manual processing subtask and a machine processing subtask; Calculate the confidence level of each subtask. When the confidence level is greater than or equal to a configurable threshold, the indicator variable is 1, and the subtask corresponding to the confidence level is the machine-processed subtask. When the confidence level is less than the configurable threshold, the indicator variable is 0, and the subtask corresponding to the confidence level is the manual-processed subtask. Assigning the manual processing subtasks according to dynamic user profiles, with novice users handling low-complexity subtasks and expert users handling high-complexity and high-risk subtasks. The dynamic user profile assignment is dynamically updated based on historical task performance data, user feedback quality, task completion rate, and error type statistics. When the subtask processing trend does not meet expectations, data denoising, model replacement, or human expert intervention is triggered. When the processing trend meets expectations, machine resource allocation is increased.

6. The system according to claim 5, characterized in that The system further includes a conflict resolution module, wherein the conflict resolution module includes: a prediction result generating unit, configured to generate an original decision for processing the machine processing subtask by the machine computing unit, and output a confidence level corresponding to the machine processing subtask and key features of the machine processing subtask that affect the original decision; An expert review unit is used to review the original decision through experts. When an expert raises an objection to the original decision, the system displays the key features of the corresponding machine processing subtask to the expert and accepts the expert's revised decision. If a single expert cannot resolve the conflict, a multi-expert comprehensive analysis is initiated to generate the final revised decision; a weight allocation unit, configured to dynamically allocate weights of the manual processing subtask and the machine processing subtask based on historical performance data and task complexity, allocating a high weight to the machine processing subtask in high-precision scenarios, and allocating a high weight to the manual processing subtask in low-precision and high-risk scenarios; The human-machine iterative correction unit is used to iteratively optimize the basic model using the final correction decision as a training sample, and experts select difficult samples for targeted retraining.

7. The system according to claim 6, characterized in that The system further includes a model adjustment module, wherein the model adjustment module includes: A difficult sample collection unit is used to identify the erroneous prediction samples output by the basic model, use the erroneous prediction samples as difficult samples, and screen and collect the difficult samples through experts to construct an augmented training set; A multimodal feedback unit, which is used to correct and score the output of the basic model through explicit feedback, infer feedback by analyzing user interaction behavior through implicit feedback, and guide humans to collect feedback through preset tools through structured annotations to ensure that humans provide high-quality input; A model fine-tuning unit is used to input the augmented training set and multimodal feedback data into the basic model for model fine-tuning.

8. A human-computer collaborative learning method based on stream data driving, characterized in that: Applied to a human-machine collaborative learning system, the method includes: Initialize the input stream data, including data preprocessing, basic model training, and task requirement analysis; Select the human-machine collaboration mode according to the task type, and dynamically allocate subtasks to human computing units and machine computing units for corresponding processing according to the task allocation mechanism; A multi-dimensional evaluation system is constructed to evaluate the task processing results, and low-confidence data is iteratively optimized. The low-confidence data is the input stream data whose task processing accuracy is lower than the expected threshold.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the program, the steps in the method of claim 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 8 are implemented.

Images