A real-time live broadcast auditing task grading and distributing method and device, equipment and medium

Abstract

The application relates to a real-time live broadcast auditing task grading and distribution method and device, equipment and medium. The method comprises the following steps: when an auditor enters an auditing hall for the first time, one-level auditing task distribution is performed based on the hall entering time length and average task number of the auditor; when the auditor starts to perform an auditing task, work state monitoring and verification are sequentially performed by collecting multi-modal feature data of the auditor, and two-level auditing task distribution is performed according to the work state of the auditor, including auditing task incentive distribution of the auditor in a normal work state and auditing task release of the auditor in an off-duty or abnormal work state; on the basis that the one-level auditing task distribution and the two-level auditing task distribution are exclusive distribution, for the auditor in the normal work state, auditing task redundant distribution is performed according to a preset redundancy ratio and a rotation cycle. The method can improve the efficiency and accuracy of live broadcast auditing tasks and ensure the continuous and efficient execution of auditing work.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of task allocation, in particular to a real-time live broadcast auditing task grading and allocation method and device, equipment and medium. BACKGROUND

[0002] At present, with the rapid development of the Internet and mobile Internet technology, network live broadcast has become a widely popular form of content dissemination. In order to ensure the compliance of live broadcast content, live broadcast platforms need to strictly monitor and audit live broadcast content and implement efficient crackdown on illegal and irregular content. Although part of the auditing work can be assisted by machine auditing technology, due to the blind spots and misjudgments of algorithm recognition, it is impossible to achieve 100% coverage, so manual auditing is still an indispensable key link.

[0003] The real-time and high irregularity of live broadcast content pose great challenges to auditing work. Irregular content can appear at any time, and if it is not intercepted in time, it will lead to the spread of bad information and have a negative impact on society and the platform. At present, the manual auditing team needs to work in shifts for 24 hours to deal with uninterrupted live broadcast streams. However, long-term high-intensity work is easy to make auditors tired, resulting in problems such as distraction, slow reaction, and even missed, misjudged or delayed processing. In addition, even if the task has been assigned to the auditor, if he or she fails to maintain concentration, it may still cause regulatory loopholes due to human negligence.

[0004] The existing task allocation and supervision mechanism cannot effectively guarantee the real-time and accuracy of live broadcast auditing. Conventional technology cannot achieve omnidirectional monitoring, especially in the last line of defense of manual auditing, and if scientific management means are lacking, it will directly affect the auditing effect. Therefore, it is necessary to optimize the task allocation strategy and strengthen the supervision mechanism of auditors to ensure the continuous and efficient execution of auditing work and maintain the health and safety of the network environment. SUMMARY

[0005] Therefore, it is necessary to provide a real-time live broadcast auditing task grading and allocation method, device, equipment and medium to solve the above technical problems.

[0006] A real-time live broadcast auditing task grading and allocation method, the method comprising:

[0007] When the auditor enters the auditing hall for the first time, the first-level auditing task is allocated based on the auditor's hall entry time and average number of tasks;

[0008] When the auditor starts to perform the audit task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditor, and the secondary audit task distribution is performed according to the working state of the auditor; wherein the secondary audit task distribution includes: for the auditor in the normal working state, the audit task incentive distribution is performed by calculating the working state ranking of the auditor; for the auditor in the off-duty or abnormal working state, the audit task of the auditor is released and evenly distributed to the auditors in the normal working state;

[0009] On the basis that the primary audit task distribution and the secondary audit task distribution are both exclusive distribution, for the auditor in the normal working state, the audit task redundancy distribution is performed according to the preset redundancy ratio and rotation cycle.

[0010] In one of the embodiments, when the auditor first enters the audit hall, the primary audit task distribution is performed based on the hall entry time length and the average task number of the auditor, including:

[0011] When the auditor first enters the audit hall, the corresponding distribution priority is set according to the hall entry time length of each auditor, the current to-be-distributed task is preferentially distributed to the new auditor of the audit hall, and the primary audit task distribution is performed according to the current to-be-distributed task number and the average task number of the auditor, and the primary audit task distribution logic is:

[0012] The current to-be-distributed task number, the total task number and the total number of auditors in the audit hall are obtained, the average task number is calculated according to the total task number and the total number of auditors;

[0013] If the to-be-distributed task number is less than the average task number, the auditor list is polled to query the current task number, and the difference between the current task number and the average task number of each auditor is taken as the to-be-released task number, if the to-be-released task number is positive, the task number corresponding to the auditor is released as the to-be-distributed task, and the current to-be-distributed task number is re-obtained for task distribution; otherwise, the difference between the average task number and the current task number of each auditor is taken as the to-be-distributed task number, if the to-be-distributed task number of a certain auditor is positive, the corresponding number of to-be-distributed tasks is distributed to the auditor.

[0014] In one of the embodiments, when the auditor starts to perform the audit task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditor, including:

[0015] When an auditor begins performing an audit task, multimodal feature data within a preset time window is collected and input into a pre-built multimodal model. The multimodal feature data is then matched with a preset normal operation feature template in the multimodal model. If the matching degree output by the model is greater than a set threshold, the auditor is considered to be in a normal working state. The multimodal feature data includes mouse operation data, live stream details page viewing behavior data, and audit operation data.

[0016] Otherwise, further verification of the auditor's work status can be conducted through heartbeat detection or random check-in strategies;

[0017] The verification of auditors' work status through heartbeat detection includes: periodically sending data to the heartbeat interface in the audit hall and marking the latest heartbeat time of each auditor to monitor their real-time on-duty status. If no heartbeat data is received within a preset time threshold, the auditor is determined to be off duty and marked as having an abnormal work status; otherwise, the auditor is determined to be on duty and marked as having a normal work status.

[0018] Verifying auditors' work status through a random check-in strategy includes: deploying a dynamic verification engine in the audit lobby or live stream details page. When multimodal feature data indicates an anomaly, a check-in verification pop-up window set by the dynamic verification engine is triggered to verify the auditor's on-duty status in real time. If the auditor fails to complete the check-in verification within the specified time, the auditor is deemed to be off duty and marked as having an abnormal work status; otherwise, it is further determined whether the auditor's daily check-in verification count exceeds the maximum check-in threshold. If it does, the auditor is marked as having an abnormal work status; otherwise, the auditor is marked as having a normal work status.

[0019] In one embodiment, for auditors in a normal working state, the ranking of auditors' working status is calculated, including:

[0020] For auditors in normal working condition, the number of times each auditor views the live stream details page and the number of times they perform audit operations are collected within a preset time window. The total number of views and the total number of audit operations are then calculated. View points are calculated based on the ratio of each auditor's live stream details page views to the total number of views, and audit points are calculated based on the ratio of each auditor's audit operations to the total number of audit operations. The view points and audit points of each auditor are combined and sorted to obtain the ranking of each auditor's working status.

[0021] In one embodiment, the allocation of audit task incentives is performed by calculating the auditors' work status ranking, including:

[0022] Based on the work status ranking from highest to lowest, high-risk tasks are assigned first, and the incentive allocation for review tasks is based on the work status ranking. The logic for allocating review task incentives is as follows:

[0023] For the current working state ranking above the middle value, the number of tasks for incentive distribution = average number of tasks + (maximum number of distribution - average number of tasks) / working state ranking; wherein, the maximum number of distribution = total number of tasks / (total number of auditors - 1);

[0024] For the current working state ranking below the middle value, the number of tasks for incentive distribution = average number of tasks - (average number of tasks - minimum number of distribution) / (total number of auditors - working state ranking); wherein, the minimum number of distribution = total number of tasks / (total number of auditors + 1).

[0025] In one embodiment, for auditors in normal working state, the audit task redundancy distribution is performed according to the preset redundancy ratio and rotation period, including:

[0026] Obtain the redundancy ratio dynamically configured in the audit hall;

[0027] Obtain the auditors currently in normal working state, and poll the task retention time and task risk level of each auditor, calculate the redundancy priority of each task by comprehensively considering the task retention time and the task risk level, and put the task quantity meeting the preset redundancy ratio into the redundancy list in the order from high to low according to the redundancy priority;

[0028] Distribute the redundancy tasks in the redundancy list to mutually exclusive auditors; wherein, the mutually exclusive auditors are auditors whose original tasks are inconsistent with the current required redundancy tasks;

[0029] When the time of audit task redundancy distribution exceeds the preset rotation period, switch to the next batch of non-redundant tasks, and preferentially select the non-redundant task with the longest task retention time into the redundancy list for periodic rolling distribution, so as to ensure that all tasks are subjected to a round of redundancy distribution after a certain period.

[0030] In one embodiment, the above method further includes: configuring an abnormal alarm engine in the audit hall to respectively perform task overflow alarm, auditor abnormal working state alarm and task surge alarm; when the task overflow alarm is triggered, the task distribution degradation logic is triggered at the same time; wherein, the task distribution degradation logic is:

[0031] Calculate the overflow task number, if the overflow task number is less than the product of the total number of tasks and the redundancy ratio, reduce the redundancy ratio to ensure the distribution of all tasks, and recalculate the overflow task number after the task distribution; wherein, the overflow task number = total number of tasks - (total number of auditors x single maximum task distribution number); if there is still overflow, reduce the low-risk task distribution amount, preferentially distribute the high-risk tasks identified by the machine, and ensure that the machine review result is audited in the first time.

[0032] A real-time live broadcast audit task grading and distribution device, the device comprises:

[0033] A first distribution module is configured to perform first-level audit task distribution based on the audit hall entry duration and the average number of tasks of the auditor when the auditor enters the audit hall for the first time.

[0034] A second distribution module is configured to perform work state monitoring and verification by collecting multi-modal feature data of the auditor, and perform second-level audit task distribution according to the work state of the auditor when the auditor starts to perform the audit task. The second-level audit task distribution includes: for the auditor in a normal work state, performing audit task incentive distribution by calculating the work state ranking of the auditor; for the auditor in an off-duty or abnormal work state, releasing the audit task of the auditor and distributing the audit task to auditors in a normal work state.

[0035] A redundancy distribution module is configured to perform audit task redundancy distribution according to a preset redundancy ratio and rotation period for the auditor in a normal work state on the basis that the first-level audit task distribution and the second-level audit task distribution are exclusive distribution.

[0036] A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0037] When the auditor enters the audit hall for the first time, first-level audit task distribution is performed based on the audit hall entry duration and the average number of tasks of the auditor.

[0038] When the auditor starts to perform the audit task, work state monitoring and verification are performed by collecting multi-modal feature data of the auditor, and second-level audit task distribution is performed according to the work state of the auditor. The second-level audit task distribution includes: for the auditor in a normal work state, performing audit task incentive distribution by calculating the work state ranking of the auditor; for the auditor in an off-duty or abnormal work state, releasing the audit task of the auditor and distributing the audit task to auditors in a normal work state.

[0039] On the basis that the first-level audit task distribution and the second-level audit task distribution are exclusive distribution, for the auditor in a normal work state, audit task redundancy distribution is performed according to a preset redundancy ratio and rotation period.

[0040] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the following steps:

[0041] When the auditor enters the audit hall for the first time, first-level audit task distribution is performed based on the audit hall entry duration and the average number of tasks of the auditor.

[0042] When the auditor starts to perform the audit task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditor, and the secondary audit task allocation is performed according to the working state of the auditor; wherein the secondary audit task allocation includes: for the auditor in the normal working state, the audit task incentive allocation is performed by calculating the working state ranking of the auditor; for the auditor in the off-duty or abnormal working state, the audit task is released and evenly distributed to the auditor in the normal working state;

[0043] On the basis that the primary audit task allocation and the secondary audit task allocation are both exclusive allocation, for the auditor in the normal working state, the audit task redundant allocation is performed according to the preset redundancy ratio and rotation cycle.

[0044] The above real-time live audit task grading and allocation method, device, equipment and medium construct a hierarchical audit task grading and allocation framework that changes with the working process of the auditor. When the auditor first enters the audit hall, the primary audit task allocation is performed based on the hall entry time and the average number of tasks of the auditor; when the auditor starts to perform the audit task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditor, so that the secondary audit task allocation strategy can be dynamically adjusted according to the real-time working state of the auditor, the key or high-risk task is preferentially allocated to the appropriate auditor, the situation that the auditor is not in good condition and leads to untimely task response or idling is avoided, and the audit efficiency is improved. In the normal and abnormal working states, the incentive allocation and release redistribution mechanism are respectively adopted, the risk of missing or wrong audit caused by the abnormal state of the auditor is reduced, the accuracy and response speed of the audit are further improved, and the continuous and efficient execution of the audit work can be ensured. Further, on the basis that the primary audit task allocation and the secondary audit task allocation are both exclusive allocation, the redundant allocation based on the redundancy ratio and the rotation cycle is designed, so that different live rooms can be simultaneously supervised by multiple auditors in different time periods, the problem of single auditor blind spot or mistake is effectively solved, and the accuracy of the audit is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 It is a flowchart of a real-time live audit task grading and allocation method in one embodiment;

[0046] Figure 2 It is a flowchart of working state monitoring and verification in one embodiment;

[0047] Figure 3 It is a structure block diagram of a real-time live audit task grading and allocation device in one embodiment;

[0048] Figure 4 It is an internal structure diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0050] In one embodiment, as shown in Figure 1 A real-time live review task grading and distribution method is provided, comprising the following steps:

[0051] Step 1: When the auditor first enters the review hall, the first-level review task distribution is performed based on the auditor's hall entry duration and average task number.

[0052] In step 1, when the auditor first enters the review hall, a check-in request is sent to the background, and a certain time is sent regularly so that the background receives the information of the current auditor entering the review hall, sets the corresponding distribution priority according to the hall entry duration of each auditor, preferentially distributes the current unassigned task to the new auditor of the review hall, and performs the first-level review task distribution according to the current unassigned task number and the average task number of the auditor. The first-level review task distribution logic is:

[0053] First, the current unassigned task number, total task number and total number of auditors in the review hall are obtained, and the average task number is calculated according to the total task number and the total number of auditors.

[0054] If the unassigned task number is less than the average task number, the auditor list is polled to query the current task number, and the difference between the current task number of each auditor and the average task number is taken as the unassigned task number. If the unassigned task number is positive, the task number corresponding to the auditor is released as the unassigned task, and the current unassigned task number is reacquired for task distribution; otherwise, the difference between the average task number and the current task number of each auditor is taken as the assigned task number, and if the assigned task number of a certain auditor is positive, the corresponding number of unassigned tasks is assigned to the auditor.

[0055] Step 2: When the auditor starts to perform the review task, the working state monitoring and verification are performed in sequence by collecting the multi-modal feature data of the auditor, and the second-level review task distribution is performed according to the working state of the auditor; wherein the second-level review task distribution includes: for the auditor in normal working state, the review task incentive distribution is performed by calculating the working state ranking of the auditor; for the auditor in off-duty or abnormal working state, the review task is released and evenly distributed to the auditors in normal working state.

[0056] The process of the auditor performing the audit task is as follows: the auditor starts to perform the task after receiving the task assigned to him in the audit hall. The audit hall will refresh the latest live frame graph of each live room in real time. The auditor needs to repeatedly browse the latest live frame graph back and forth to determine whether there is suspicious or illegal content. When encountering suspicious or illegal content, the auditor needs to click into the live detail page. The live detail page will display all the frame graphs of the current live room, which can further determine whether the live room has illegal content. At the same time, the audit hall will also display the viewing time difference (current time - last time entering the live detail page) from the last time entering the live detail page, and will be displayed in the front position in priority. If a live room has no live detail page audit record for a long time, the audit hall will trigger a pop-up box to remind the auditor to pay attention to the audit risk.

[0057] The working state monitoring and verification of the auditor in step 2 is as described in Figure 2 , including:

[0058] When the auditor enters the audit hall and starts to perform the audit task, the multi-modal feature data of the auditor in a preset time window (such as 1 minute) is collected and input into a pre-constructed multi-modal model. In the multi-modal model, the multi-modal feature data is matched with the preset normal operation feature template for multi-modal fusion feature matching. If the matching degree output by the model is greater than the set threshold, it is considered that the auditor is in a normal working state. The multi-modal feature data includes mouse operation data (such as the number of times the mouse is scrolled, moved, and clicked), live detail page viewing behavior data (such as the number of times of viewing), and audit operation data.

[0059] Otherwise, further verification of the working state of the auditor is performed through heartbeat detection or random check-in strategy.

[0060] The verification of the working state of the auditor through heartbeat detection includes: sending data to the heartbeat interface at a fixed time (such as once every 5 seconds) in the audit hall, and marking the latest heartbeat time of each auditor for real-time on-duty state monitoring of the auditor. If no heartbeat data is received within a preset time threshold (such as 1 minute), it is determined that the auditor is off-duty and is marked as an abnormal working state. Otherwise, it is determined that the auditor is on-duty and is marked as a normal working state.

[0061] The auditor working state verification through the random check-in strategy includes: deploying a dynamic verification engine in the audit hall or the live detail page, when the multi-modal feature data is judged to be abnormal, triggering the non-periodic check-in verification pop-up window (including but not limited to: random string, random number, sliding verification, etc.) set by the dynamic verification engine to perform real-time on-duty state verification on the auditor, if the auditor does not complete the check-in verification within the specified time (such as 30 seconds), it is determined that the auditor is off-duty and is marked as an abnormal working state; otherwise, it is determined that the auditor is on duty, and further determines whether the number of check-in verifications of the auditor on the same day exceeds the maximum check-in threshold, if it exceeds, the auditor is marked as an abnormal working state; otherwise, the auditor is marked as a normal working state. The maximum check-in threshold is obtained by increasing a certain percentage of the average number of check-in verifications of each normal auditor in history.

[0062] It should be noted that, due to the possibility of page switching (i.e. switching between the audit hall and the live detail page) of the auditor, when check-in verification is required, the non-periodic check-in verification pop-up window is popped up in both pages (because at this time it is not determined which specific page the auditor opens), after the auditor performs check-in verification in any page, the check-in behavior is sent to the background, and the non-periodic check-in verification pop-up window in the other page disappears through window communication.

[0063] It should be understood that the above working state monitoring and verification can effectively avoid the situation that the task is not audited when the auditor leaves the page, and can prompt and correct the auditor state in time when the auditor is abnormal or dozes off (the auditor is on duty at night), and perform check-in verification when abnormal, that is, to avoid the disturbance of invalid pop-up window and fully realize the normal supervision of the auditor on duty, so as to timely discover abnormal state, and can timely synchronize abnormal information to the group leader and the person in charge, and timely adjust manpower or strategy to ensure audit safety.

[0064] Further, specifically, for the above working state monitoring and verification, the auditors in abnormal working state and the auditors in normal working state are screened out, the tasks of the auditors in abnormal working state are released and distributed to the auditors in normal working state. At the same time, for the auditors in normal working state, the working state ranking of the auditors is calculated to perform audit task incentive distribution, including:

[0065] First, for the auditors in normal working state, the live detail page viewing times and the audit operation times of each auditor in a preset time window are collected, and the total viewing times and the total audit operation times are obtained, the viewing points are calculated according to the ratio of the live detail page viewing times of each auditor to the total viewing times, the audit points are calculated according to the ratio of the audit operation times of each auditor to the total audit operation times, the viewing points and the audit points of each auditor are integrated and sorted to obtain the working state ranking of each auditor. For example, set the total points to 100, the audit points and the viewing points of a certain auditor account for 3:7, if the audit operation times of the current auditor account for 10% of the total audit operation times, the audit points are , and the live detail page viewing times account for 20% of the total viewing times, the viewing points are , and the total points of the current auditor are 13.

[0066] Then, according to the order from high to low of the working state ranking, high-risk tasks are preferentially allocated, that is, the higher the ranking, the more high-risk tasks are allocated, the lower the ranking, the less tasks are allocated, and at the same time, the less high-risk tasks are allocated, and the audit task incentive allocation is allocated based on the working state ranking, and the logic of the audit task incentive allocation is:

[0067] For the current working state ranking above the middle value, the number of tasks for incentive allocation = average number of tasks + (maximum number of allocation - average number of tasks) / working state ranking; wherein, the maximum number of allocation = total number of tasks / (total number of auditors -1).

[0068] For the current working state ranking below the middle value, the number of tasks for incentive allocation = average number of tasks - (average number of tasks - minimum number of allocation) / (total number of auditors - working state ranking); wherein, the minimum number of allocation = total number of tasks / (total number of auditors +1).

[0069] Step 3, on the basis of exclusive allocation of the first-level audit task allocation and the second-level audit task allocation, for the auditors in normal working state, the audit task redundancy allocation is performed according to the preset redundancy ratio and the rotation period.

[0070] Among them, exclusive allocation refers to that each task is only allocated to a single auditor. Due to the real-time and variability of live broadcast auditing, exclusive allocation can easily lead to the fact that a single auditor is difficult to immediately find content violations. Therefore, on the basis of exclusive allocation, the present scheme further performs redundant allocation of auditing tasks according to a preset redundancy ratio (such as 20%, 30%) and a rotation period, that is, a certain proportion of tasks in the hands of each auditor is allocated to other auditors according to a certain redundancy ratio. In this way, while a certain auditor is processing other tasks, a certain proportion of tasks in his hands are processed by other auditors at the same time, reducing the risk of missed and wrong auditing by a single auditor. And the redundant tasks will be replaced according to the rotation period, and the high-risk tasks are given priority, and the time is rotated, so that all live rooms can be redundant for a period of time after a certain period of time, and after both auditors confirm that there is no violation, the live room is marked as a low-risk live room. The logic of redundant allocation is as follows:

[0071] (1) Obtain the redundancy ratio dynamically configured by the auditing hall. Among them, the redundancy ratio can be configured with different ratios according to different periods (such as Qinglang activities, key festivals, etc.).

[0072] (2) Obtain the auditors currently in normal working state, and poll the task retention time and task risk level of each auditor. The redundancy priority of each task is calculated by comprehensively considering the task retention time and the task risk level, and the tasks meeting the preset redundancy ratio are put into the redundancy list in order from high to low according to the redundancy priority.

[0073] (3) Distribute the redundant tasks in the redundancy list to mutually exclusive auditors; among them, mutually exclusive auditors are auditors whose original tasks are inconsistent with the current required redundant tasks.

[0074] (4) When the time of redundant allocation of auditing tasks exceeds the preset rotation period (such as every 5 minutes), switch to the next batch of non-redundant tasks, and preferentially select the non-redundant task with the longest task retention time into the redundancy list for periodic rolling distribution, to ensure that all tasks are subjected to a round of redundant allocation after a certain period of time.

[0075] Further, the above method further comprises: configuring an abnormal alarm engine in the auditing hall to perform task overflow alarm, auditor abnormal working state alarm and task surge alarm respectively.

[0076] Among them, the task overflow alarm is when the total number of tasks is too much and the online auditor is insufficient, at this time there will be unassigned tasks, which can be considered as unreviewed state, thereby triggering the alarm. The auditor abnormal working state alarm is when a certain auditor triggers the sign-in verification multiple times, or the detail review timeout behavior exists multiple times, at this time an abnormal alarm is needed, and the auditor's superior needs to coordinate the state of the auditor. The task surge alarm is that the current time period is compared with the average number of tasks in the same time period in the past 7 days, and when it exceeds the maximum fluctuation value or reaches the maximum task upper limit of all online auditors, a pre-warning and prompt are given to timely coordinate the auditor manpower.

[0077] When the above abnormal alarm is triggered to cause task overflow, the task allocation degradation logic is triggered at the same time to redistribute tasks, focusing on allocating high-risk tasks, and only a certain amount of low-risk tasks is allowed to overflow, so as to avoid the problem of auditor human negligence, and timely remind and timely alarm when personnel negligence, guarantee the continuous, safe and reliable of the review work. Among them, the high-risk tasks are: machine-identified tasks with violation risk, tasks with long time since the last live detailed page comprehensive review, etc.

[0078] Among them, the task allocation degradation logic is:

[0079] Calculate the overflow task number, if the overflow task number is less than the product of the total task number and the redundancy ratio, reduce the redundancy ratio to ensure all task allocation, and recalculate the overflow task number after task allocation; among them, the overflow task number = total task number - (total number of auditors x single maximum task allocation number); if there is still overflow, reduce the low-risk task allocation amount, and preferentially allocate machine-identified high-risk tasks to ensure that the machine review result is audited in the first time.

[0080] In the real-time live broadcast auditing task grading and distribution method, a hierarchical auditing task grading and distribution framework changing with the workflow of auditors is constructed, when the auditors enter the auditing hall for the first time, the first-level auditing task distribution is performed based on the hall entering time length and the average task number of the auditors; when the auditors start to perform the auditing task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditors, so that the second-level auditing task distribution strategy can be dynamically adjusted according to the real-time working state of the auditors, the key or high-risk task can be preferentially distributed to the appropriate auditor, the state of the auditor is avoided to be poor, the task response is not timely or idle, the auditing efficiency is improved, the incentive distribution and release redistribution mechanism are respectively adopted in the normal and abnormal working states, the risk of missing and wrong auditing caused by the abnormal state of the auditor is reduced, the accuracy and response speed of the auditing are further improved, and the continuous and efficient execution of the auditing work can be ensured. Further, on the basis that the first-level auditing task distribution and the second-level auditing task distribution are exclusive distribution, the redundant distribution based on the redundancy ratio and the rotation cycle is designed, so that different live broadcast rooms can be simultaneously supervised by multiple auditors in different time periods, the single auditor blind spot or mistake problem is effectively solved, and the accuracy of the auditing is significantly improved.

[0081] In one embodiment, as shown in Figure 3 A real-time live broadcast auditing task grading and distribution apparatus is provided, comprising:

[0082] A first distribution module is configured to perform first-level auditing task distribution based on the hall entering time length and the average task number of the auditors when the auditors enter the auditing hall for the first time.

[0083] A second distribution module is configured to perform working state monitoring and verification by collecting multi-modal feature data of the auditors when the auditors start to perform the auditing task, and to perform second-level auditing task distribution according to the working state of the auditors; wherein the second-level auditing task distribution comprises: for the auditors in normal working state, performing auditing task incentive distribution by calculating the working state ranking of the auditors; for the auditors in off-duty or abnormal working state, releasing the auditing task and equally distributing it to the auditors in normal working state.

[0084] A redundant distribution module is configured to, on the basis that the first-level auditing task distribution and the second-level auditing task distribution are exclusive distribution, perform auditing task redundant distribution according to a preset redundancy ratio and rotation cycle for the auditors in normal working state.

[0085] The specific limitation of the real-time live broadcast auditing task grading and distributing device can refer to the limitation of the real-time live broadcast auditing task grading and distributing method, which will not be repeated here. Each module in the real-time live broadcast auditing task grading and distributing device can be realized by software, hardware, and a combination thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to call and execute the operations corresponding to each module by the processor.

[0086] In one embodiment, a computer device, which can be a terminal, has an internal structure diagram as shown in Figure 4 The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a real-time live broadcast auditing task grading and distributing method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0087] Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0088] In one embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the following steps:

[0089] When the auditor enters the auditing hall for the first time, the first-level auditing task is assigned based on the auditor's hall entry duration and the average number of tasks;

[0090] When the auditor starts to perform the audit task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditor, and the secondary audit task distribution is performed according to the working state of the auditor; wherein the secondary audit task distribution includes: for the auditor in the normal working state, the audit task incentive distribution is performed by calculating the working state ranking of the auditor; for the auditor in the off-duty or abnormal working state, the audit task of the auditor is released and evenly distributed to the auditors in the normal working state;

[0091] On the basis that the primary audit task distribution and the secondary audit task distribution are both exclusive distribution, for the auditor in the normal working state, the audit task redundancy distribution is performed according to the preset redundancy ratio and rotation cycle.

[0092] In one embodiment, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to implement the following steps:

[0093] When the auditor initially enters the audit hall, the primary audit task distribution is performed based on the hall entry duration and the average task number of the auditor;

[0094] When the auditor starts to perform the audit task, the working state monitoring and verification are sequentially performed by collecting the multi-modal feature data of the auditor, and the secondary audit task distribution is performed according to the working state of the auditor; wherein the secondary audit task distribution includes: for the auditor in the normal working state, the audit task incentive distribution is performed by calculating the working state ranking of the auditor; for the auditor in the off-duty or abnormal working state, the audit task of the auditor is released and evenly distributed to the auditors in the normal working state;

[0095] On the basis that the primary audit task distribution and the secondary audit task distribution are both exclusive distribution, for the auditor in the normal working state, the audit task redundancy distribution is performed according to the preset redundancy ratio and rotation cycle.

[0096] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0097] Any combination of the technical features of the above embodiments can be made, and in order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0098] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the present application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for hierarchical allocation of real-time live streaming review tasks, characterized in that, The method includes: When an auditor first enters the audit hall, primary audit tasks are assigned based on the auditor's time spent in the hall and the average number of tasks. When an auditor begins to perform an audit task, the auditor's work status is monitored and verified sequentially by collecting multimodal feature data, and secondary audit tasks are allocated according to the auditor's work status. The secondary audit task allocation includes: for auditors in normal working status, audit task incentives are allocated by calculating the auditor's work status ranking; for auditors who are absent from their posts or in abnormal working status, their audit tasks are released and evenly distributed to auditors in normal working status. Based on the fact that both the first-level and second-level audit task allocations are exclusive allocations, for auditors in normal working status, audit task redundancy allocation is carried out according to the preset redundancy ratio and rotation cycle. For auditors in normal working condition, audit task redundancy is allocated according to a preset redundancy ratio and rotation cycle, including: Obtain the redundancy ratio dynamically configured in the review hall; Obtain the auditors who are currently working normally, and poll each auditor's task retention time and task risk level. Calculate the redundancy priority of each task based on the task retention time and task risk level, and put the number of tasks that meet the preset redundancy ratio into the redundancy list in order of redundancy priority from high to low. The redundant tasks in the redundant list are assigned to mutually exclusive auditors; wherein, the mutually exclusive auditor is the auditor whose original task is inconsistent with the currently required redundant task. When the time for redundant allocation of audit tasks exceeds the preset rotation cycle, switch to the next batch of non-redundant tasks, and prioritize the non-redundant task with the longest task retention time to enter the redundancy list for periodic rolling allocation, so as to ensure that all tasks have undergone a round of redundancy allocation after a certain period.

2. The method for hierarchical allocation of real-time live streaming review tasks according to claim 1, characterized in that, When an auditor first enters the audit hall, primary audit tasks are assigned based on the auditor's length of time in the hall and average number of tasks, including: When an auditor first enters the audit hall, a corresponding allocation priority is set based on the auditor's time spent in the hall. New auditors are given priority in being assigned tasks currently pending. First-level audit task allocation is then performed based on the current number of tasks pending and the auditor's average number of tasks. The logic for first-level audit task allocation is as follows: Get the current number of tasks to be assigned, the total number of tasks, and the total number of auditors in the audit hall. Calculate the average number of tasks based on the total number of tasks and the total number of auditors. If the number of tasks to be assigned is less than the average number of tasks, the auditor list is polled to check the current number of tasks. The difference between the current number of tasks for each auditor and the average number of tasks is taken as the number of tasks to be released. If the number of tasks to be released is positive, the number of tasks corresponding to the auditor is released as tasks to be assigned, and the current number of tasks to be assigned is retrieved again for task allocation. Otherwise, the difference between the average number of tasks and the current number of tasks for each auditor is taken as the number of tasks to be assigned. If the number of tasks to be assigned for an auditor is positive, the corresponding number of tasks to be assigned is assigned to that auditor.

3. The method for hierarchical allocation of real-time live streaming review tasks according to claim 1, characterized in that, When an auditor begins performing an audit task, their work status is monitored and verified sequentially by collecting multimodal characteristic data, including: When an auditor begins performing an audit task, multimodal feature data within a preset time window is collected and input into a pre-constructed multimodal model. The multimodal feature data is then matched against a preset normal operation feature template using multimodal fusion features. If the matching degree output by the model is greater than a set threshold, the auditor is considered to be in a normal working state. The multimodal feature data includes mouse operation data, live stream details page viewing behavior data, and audit operation data. Otherwise, further verification of the auditor's work status can be conducted through heartbeat detection or random check-in strategies; The verification of auditors' work status through heartbeat detection includes: periodically sending data to the heartbeat interface in the audit hall and marking the latest heartbeat time of each auditor to monitor their real-time on-duty status. If no heartbeat data is received within a preset time threshold, the auditor is determined to be off duty and marked as having an abnormal work status; otherwise, the auditor is determined to be on duty and marked as having a normal work status. Verifying auditors' work status through a random check-in strategy includes: deploying a dynamic verification engine in the audit lobby or live stream details page. When multimodal feature data indicates an anomaly, a check-in verification pop-up window set by the dynamic verification engine is triggered to verify the auditor's on-duty status in real time. If the auditor fails to complete the check-in verification within the specified time, the auditor is deemed to be off duty and marked as having an abnormal work status; otherwise, it is further determined whether the auditor's daily check-in verification count exceeds the maximum check-in threshold. If it does, the auditor is marked as having an abnormal work status; otherwise, the auditor is marked as having a normal work status.

4. The method for hierarchical allocation of real-time live streaming review tasks according to claim 3, characterized in that, For auditors in a normal working state, calculate the auditor's work status ranking, including: For auditors in normal working condition, the number of times each auditor views the live stream details page and the number of times they perform audit operations are collected within a preset time window. The total number of views and the total number of audit operations are then calculated. View points are calculated based on the ratio of each auditor's live stream details page views to the total number of views, and audit points are calculated based on the ratio of each auditor's audit operations to the total number of audit operations. The view points and audit points of each auditor are combined and sorted to obtain the ranking of each auditor's working status.

5. The method for hierarchical allocation of real-time live streaming review tasks according to claim 4, characterized in that, Incentives for auditing tasks are allocated based on the ranking of auditors' work performance, including: Based on the work status ranking from highest to lowest, high-risk tasks are assigned first, and the incentive allocation for review tasks is based on the work status ranking. The logic for the incentive allocation for review tasks is as follows: For cases where the current work status ranking is above the median, the number of tasks to be allocated as incentives = average number of tasks + (maximum allocation number - average number of tasks) / work status ranking; where, the maximum allocation number = total number of tasks / (total number of auditors - 1). For cases where the current work status ranking is below the median, the number of tasks allocated as incentives = average number of tasks - (average number of tasks - minimum allocation) / (total number of auditors - work status ranking); where the minimum allocation = total number of tasks / (total number of auditors + 1).

6. The method for hierarchical allocation of real-time live streaming review tasks according to claim 1, characterized in that, The method further includes: An anomaly alarm engine is configured in the audit hall to trigger alarms for task overflow, abnormal auditer work status, and task surge. When a task overflow alarm is triggered, task allocation degradation logic is also triggered. The task allocation degradation logic is as follows: Calculate the number of overflow tasks. If the number of overflow tasks is less than the product of the total number of tasks and the redundancy ratio, reduce the redundancy ratio to ensure all tasks are allocated, and recalculate the number of overflow tasks after task allocation. The overflow task number is calculated as follows: Overflow task number = Total number of tasks - (Total number of auditors × Maximum number of tasks allocated per person). If overflow still exists, reduce the allocation of low-risk tasks and prioritize the allocation of high-risk tasks identified by the machine to ensure that the machine review results are reviewed as soon as possible.

7. A real-time live streaming review task hierarchical allocation device, characterized in that, The device includes: The first-level allocation module is used to allocate first-level audit tasks based on the auditor's time spent in the audit hall and the average number of tasks when the auditor first enters the audit hall. The secondary allocation module is used to monitor and verify the work status of auditors by collecting their multimodal characteristic data when they begin to perform audit tasks, and to allocate secondary audit tasks according to the auditors' work status. The secondary audit task allocation includes: for auditors in normal working status, audit task incentive allocation is carried out by calculating the auditors' work status ranking; for auditors who are absent from their posts or in abnormal working status, their audit tasks are released and evenly distributed to auditors in normal working status. The redundancy allocation module is used to redundantly allocate audit tasks to auditors in normal working condition, based on the fact that both the first-level audit task allocation and the second-level audit task allocation are exclusive allocations, according to a preset redundancy ratio and rotation cycle. For auditors in normal working condition, audit task redundancy is allocated according to a preset redundancy ratio and rotation cycle, including: Obtain the redundancy ratio dynamically configured in the review hall; Obtain the auditors who are currently working normally, and poll each auditor's task retention time and task risk level. Calculate the redundancy priority of each task based on the task retention time and task risk level, and put the number of tasks that meet the preset redundancy ratio into the redundancy list in order of redundancy priority from high to low. The redundant tasks in the redundant list are assigned to mutually exclusive auditors; wherein, the mutually exclusive auditor is the auditor whose original task is inconsistent with the currently required redundant task. When the time for redundant allocation of audit tasks exceeds the preset rotation cycle, switch to the next batch of non-redundant tasks, and prioritize the non-redundant task with the longest task retention time to enter the redundancy list for periodic rolling allocation, so as to ensure that all tasks have undergone a round of redundancy allocation after a certain period.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images