Network live barrage language dynamic early warning and linkage management method and system thereof

Abstract

This invention discloses a method and system for dynamic early warning and coordinated governance of online live streaming barrage comments, belonging to the field of online live streaming content governance and computing resource scheduling technology. The method includes: acquiring the remaining visually visible time and end-to-end processing latency of each barrage comment to be governed in a first live streaming room; determining the barrage-level governance urgency by comparing the two; and further determining the governance urgency and visual time budget of the first live streaming room; generating an early warning signal when the governance urgency reaches a preset early warning threshold; and initiating resource appropriation from a second live streaming room with lower governance urgency to the first live streaming room, with the appropriation amount satisfying a first constraint of symmetry and a second constraint of security. This invention uses the remaining visually visible time as the physical benchmark for governance decisions and, through cross-live streaming room resource appropriation with dual constraint verification, shortens the exposure time of illegal barrage comments and reduces governance costs in high-concurrency scenarios.

Description

Technical Field

[0001] This invention relates to the field of online live streaming content governance and computing resource scheduling technology, and in particular to a method and system for dynamic early warning and coordinated governance of online live streaming barrage language. Background Technology

[0002] Live streaming, as a real-time interactive content dissemination format, has become one of the mainstream media in various fields such as social entertainment, e-commerce, and knowledge dissemination due to its immediacy, interactivity, and wide reach. As the most important interactive medium between live streamers and viewers, bullet comments (danmu) carry the functions of real-time expression, emotional feedback, and community communication for viewers. Due to its high frequency, high density, and high concurrency, it has also become a key aspect of content governance on live streaming platforms. If an inappropriate bullet comment is not effectively blocked before it is displayed on the viewer's end, it will spread directly to a large number of viewers through their screens, causing irreversible consequences. Therefore, the real-time requirements for bullet comment governance are far higher than those for post-event review of ordinary text and image content. Existing bullet comment governance technology systems generally adopt a three-stage server-side pipeline architecture of risk identification, handling decision-making, and instruction issuance. The live streaming platform first identifies risks and generates risk tags for bullet comments after they enter the server using keyword matching, text classification models, and user profile matching. Then, based on these risk tags, it makes handling decisions (including allowing, alerting, blocking, intercepting, and banning). Finally, it sends the handling instructions to the receiving terminal, the broadcaster's terminal, the review terminal, and the risk control terminal, completing the closed loop for managing each bullet comment. This pipeline architecture can basically meet the management needs under normal traffic scenarios.

[0003] For example, Chinese invention patent CN111526374B discloses a method and apparatus for processing bullet comments based on live streaming, as well as a method and apparatus for pulling streams. When the method is applied to a server, the method includes: segmenting the received live video stream and storing the segmented video data fragments in a first container; segmenting the received bullet comment data stream for the live video stream and storing the segmented bullet comment data fragments in a second container; determining whether the data stored in the first or second container triggers its corresponding preset container depth threshold; if so, obtaining N video groups obtained by grouping the video data fragments in the first container and M bullet comment groups obtained by grouping the bullet comment data fragments in the second container; processing the N video groups and M bullet comment groups to obtain a target data packet, and sending the target data packet to the lower-level device in the form of a target data stream.

[0004] For example, Chinese invention patent CN114598899B discloses a web crawler-based method for analyzing bullet screen (danmaku) broadcasts. This method analyzes the bullet screen message transmission protocol of the live stream source, uses a web crawler to collect bullet screens and pushes the data to a message queue, then locally polls the message queue data, performs voice conversion and storage, sequentially reads the voice files for voice broadcasting, and finally analyzes the stored data using data mining algorithms. This enables real-time monitoring and broadcasting of bullet screens, improving live stream popularity and interactivity, while also analyzing user profiles, live stream hotspots, and public opinion evolution to optimize live stream content and operational strategies.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0006] Existing bullet screen (danmaku) management technologies generally employ a server-side pipeline architecture for risk identification, handling decisions, and instruction issuance, aiming to intercept bullet screens before they reach the receiving terminal or before they disappear from view. The success of this pre-emptive interception is heavily constrained by the processing speed of the handling process: when the processing speed is faster than the bullet screen's display speed on the receiving terminal, inappropriate bullet screens can be intercepted promptly; when the processing speed is slower than the display speed, pre-emptive interception fails. Situations where processing speed cannot keep up occur precisely in scenarios with high bullet screen traffic, drastic fluctuations, and high concurrency, such as sudden trending events, popular streamers broadcasting, large-scale event live streams, and e-commerce promotional live streams. In such high-concurrency scenarios, the influx of bullet screens increases dramatically, processing resources are heavily consumed, and the processing time for a single bullet screen is prolonged. Simultaneously, if the platform's pre-emptive interception rules are incomplete or interception resources are limited, the risk of inappropriate bullet screens not being intercepted before they disappear from the viewer's sight increases significantly.

[0007] Current technologies generally focus on optimizing the distribution chain of bullet comments or post-event data collection and analysis, paying insufficient attention to the real-time governance and scheduling of bullet comments before their actual display on the receiving terminal. Existing bullet comment governance server architectures typically rely on internal server-side status indicators such as processing queue level, message queue depth, and violation identification confidence as the basis for governance decisions. However, there is no deterministic correspondence between these internal status indicators and the physical display process of bullet comments on the receiving terminal. An idle server-side processing queue does not necessarily mean timely governance, and a busy server-side processing queue does not necessarily mean governance failure. The only physical benchmark that determines the effectiveness of governance is the remaining visible time of the bullet comments on the receiving terminal. Furthermore, current technologies generally allocate processing resources independently for each live streaming room as a governance unit. To cope with sudden traffic surges, platforms commonly adopt a static allocation strategy that reserves a large amount of redundant processing resources for each live streaming room, resulting in low overall utilization of processing resources and high governance costs. Summary of the Invention

[0008] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for dynamic early warning and coordinated management of online live streaming barrage language. The technical solution is as follows:

[0009] The methods for dynamic early warning and coordinated management of online live streaming barrage language include:

[0010] The remaining visual time and corresponding end-to-end processing latency of each barrage to be managed in the first live broadcast room are obtained. Based on the comparison relationship between the remaining visual time and end-to-end processing latency of each barrage to be managed, the barrage-level management urgency of each barrage to be managed is determined, and the preset failure boundary is determined based on the comparison relationship.

[0011] The urgency of managing each unmanaged comment in the first live stream room is determined based on the comment-level urgency of management.

[0012] Based on the available resources and characteristics of the bullet screen stream in the first live stream room, the visual time budget for the first live stream room is determined.

[0013] When the urgency of governance in the first live broadcast room is greater than or equal to the preset warning threshold, a warning signal is generated for the first live broadcast room. Based on the warning signal, a second live broadcast room with a governance urgency less than the preset warning threshold is selected as the appropriator. The appropriation of disposal resources from the second live broadcast room to the first live broadcast room is initiated. The amount of disposal resources transferred by the appropriation satisfies the first constraint and the second constraint.

[0014] The first constraint is: the decrease in visual time budget caused by misappropriation in the second live broadcast room shall not be greater than the increase in visual time budget caused by misappropriation in the first live broadcast room.

[0015] The second constraint is: the remaining visual time budget of the second live broadcast room after the misappropriation execution shall not be less than the minimum visual time budget corresponding to the preset failure boundary.

[0016] In addition, a dynamic early warning and coordinated management system for online live streaming barrage language is also provided, including:

[0017] The data acquisition module is used to acquire the remaining visual time of each barrage to be managed in the first live broadcast room and the corresponding end-to-end processing latency. Based on the comparison relationship between the remaining visual time of each barrage to be managed and the end-to-end processing latency, the barrage-level management urgency of each barrage to be managed is determined, and the preset failure boundary is determined based on the comparison relationship.

[0018] The governance urgency module is used to determine the governance urgency of the first live broadcast room based on the barrage-level governance urgency of each barrage to be governed in the first live broadcast room.

[0019] The visual time budget module is used to determine the visual time budget of the first live broadcast room based on the available resources and the characteristics of the barrage stream of the first live broadcast room.

[0020] The resource misappropriation module is used to generate an early warning signal for the first live broadcast room when the governance urgency of the first live broadcast room is greater than or equal to a preset early warning threshold. Based on the early warning signal, the module selects the second live broadcast room with a governance urgency less than the preset early warning threshold as the misappropriator and initiates resource misappropriation from the second live broadcast room to the first live broadcast room. The amount of misappropriated resources satisfies the first constraint and the second constraint.

[0021] The first constraint is: the decrease in visual time budget caused by misappropriation in the second live broadcast room shall not be greater than the increase in visual time budget caused by misappropriation in the first live broadcast room.

[0022] The second constraint is: the remaining visual time budget of the second live broadcast room after the misappropriation execution shall not be less than the minimum visual time budget corresponding to the preset failure boundary.

[0023] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0024] 1. The method for dynamic early warning and coordinated governance of online live streaming bullet screen language provided by this invention introduces the visually visible remaining time as a benchmark physical quantity for governance decisions. The visually visible remaining time directly originates from the actual display progress of bullet screens on the receiving terminal, reflecting the objective time window remaining before inappropriate bullet screens disappear from the viewer's field of vision. This invention anchors governance decisions to this physically measurable time benchmark. In high-concurrency bullet screen scenarios, when the platform's pre-interception is insufficient to handle inappropriate bullet screens, it can accurately determine whether governance is timely based on this physical benchmark, minimizing the exposure time of inappropriate bullet screens in the viewer's field of vision and significantly reducing governance failure rate.

[0025] 2. This invention constructs a dynamic resource allocation mechanism across live streaming rooms. In high-concurrency scenarios such as sudden trending events, popular streamers going live, large-scale event live streams, and e-commerce promotional live streams, live streaming rooms with limited resources can receive real-time support from live streaming rooms with spare capacity, and idle resources can be instantly allocated to where they are truly needed. This allocation mechanism can handle sudden traffic surges without reserving a large amount of redundant resources for each live streaming room, significantly improving the overall utilization rate of resources, reducing platform governance costs, and greatly enhancing the flexibility of resource allocation. Furthermore, this invention is superimposed on the platform's existing bullet screen recognition and blocking functions without altering the existing recognition and blocking processes, making it easy to integrate and implement on existing bullet screen management systems.

[0026] 3. This invention fundamentally avoids the reverse scheduling problem caused by existing cross-live-stream resource scheduling schemes that use the nominal quantity of resources disposed of as the sole scheduling criterion by simultaneously introducing a dual verification mechanism of symmetry constraints and security constraints in the misappropriation decision-making process. The symmetry constraint ensures that the value of resources lost by the misappropriating party does not exceed the value of resources gained by the misappropriated party, and takes into account the actual differences in resource scarcity among different live-stream rooms, avoiding the transfer of resources from the party that needs them more to the party that needs them less. The security constraint ensures that the misappropriating party can still maintain its own governance without falling into governance failure after surrendering resources. These two constraints are independent yet complementary, ensuring that the misappropriation action alleviates the governance scarcity of the misappropriated party without triggering reverse governance failure on the misappropriating party's own part, thus providing dual protection for the symmetry and security of the misappropriation action. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A schematic diagram of the process for dynamic early warning and coordinated management of online live streaming barrage language provided in this application embodiment.

[0029] Figure 2 A schematic diagram of a network live streaming barrage language dynamic early warning and linkage management system module provided in an embodiment of this application.

[0030] Figure 3 A schematic diagram illustrating the logic flow of dynamic early warning and coordinated management of online live streaming barrage language provided in this application embodiment. Detailed Implementation

[0031] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present disclosure are shown in the drawings, it should be understood that embodiments of the present disclosure may be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure.

[0032] It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure. In the description of the embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "this embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects.

[0033] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0034] Please see Figure 1 As shown, the methods for dynamic early warning and coordinated management of online live streaming barrage language include:

[0035] like Figure 3 The diagram shows the logical flowchart of the dynamic early warning and linkage governance method for online live streaming barrage language. The main arrow strictly indicates the direction of data flow, starting from the input end of the barrage entering the system, processing layer by layer until the output end of the misappropriation action. The positions of the decision branches and parallel constraint boxes precisely correspond to the two key decision nodes in the method—the triggering judgment of the misappropriation action and the dual constraint verification of the misappropriation amount.

[0036] It should be noted that the dynamic early warning and coordinated governance method for online live streaming barrage language provided by this invention is applicable to real-time governance and scheduling scenarios under high-concurrency barrage, especially suitable for situations with large barrage traffic, drastic fluctuations, and high concurrency. Typical triggering scenarios include sudden hot events, popular anchors going live, large-scale event live streaming, e-commerce promotion live streaming, etc.

[0037] The first live stream room also includes the identification of the first live stream room, specifically including:

[0038] It should be noted that in the practice of managing bullet comments, a single bullet comment is often not illegal on the surface, but when a large number of bullet comments with similar meanings appear in a short period of time, it may constitute a coordinated negative behavior such as controlling comments, inciting conflict, and cyberbullying.

[0039] Several live streaming rooms on the platform are monitored in real time to obtain the bullet comments in each live streaming room within a preset time window. The bullet comments are clustered according to their semantic similarity to obtain at least one semantically similar bullet comment cluster.

[0040] Real-time monitoring refers to continuously collecting the bullet comment streams of each live streaming room currently broadcasting on the platform. A preset time window is a sliding time interval used to capture and analyze bullet comments. Its length is preset by technical personnel based on the typical emergence rhythm of bullet comments on the platform. The window slides forward over time and triggers an analysis periodically. Semantically similar bullet comment clusters refer to the set of bullet comments within this window that are semantically close to each other. In practice, each bullet comment within the window is semantically represented, and then clustered according to the closeness of their semantic representations. Bullets with semantic closeness reaching a preset clustering threshold are grouped into the same cluster. The preset clustering threshold is also preset by technical personnel, and its value can be determined by trial clustering on historical bullet comment data, using the standard that bullet comments with the same origin variants exactly cluster together. A window may yield one or more semantically similar bullet comment clusters, or it may not yield any effective clusters due to the dispersion of bullet comments. In this case, the live streaming room is not identified as the first live streaming room in this window.

[0041] For semantically similar bullet screen clusters, the degree of focus is determined based on the convergence of the referents of each bullet screen within the cluster; the degree of content deviation is determined based on the overall semantic deviation of the semantically similar bullet screen cluster relative to the live content features of the live room; and the degree of emergence anomaly is determined based on the distribution characteristics and sending time rhythm of the sending accounts corresponding to each bullet screen within the cluster.

[0042] This embodiment characterizes whether a semantically similar danmaku cluster has collaborative negativity from three independent perspectives: focus, content deviation, and emergence anomaly. The determination process for each perspective is executed by the live streaming platform's server. All three perspectives are obtained by statistically calculating relevant data of the danmaku cluster and are converted into values ​​between 0 and 1. The larger the value, the stronger the collaborative negative tendency in that perspective.

[0043] The concentration of references reflects whether a cluster of bullet comments collectively points to the same object. Collaborative negative behavior often targets a specific person, viewpoint, or group, while normal group resonance is usually diffuse or directed towards the live stream content itself. In practice, the referential components of each bullet comment within a cluster are extracted to obtain the referential object of each bullet comment to be addressed. Referential objects with consistent semantics are grouped into the same referential object, and the bullet comments within the cluster are grouped according to the referential object. The number of bullet comments corresponding to each referential object is counted. The referential object with the most bullet comments is selected, and its corresponding number of bullet comments is divided by the total number of bullet comments in the bullet comment cluster. The resulting ratio is the concentration of references. When the vast majority of bullet comments in a bullet comment cluster point to the same object, this ratio is close to 1. When the bullet comments are dispersed, this ratio is close to 0.

[0044] Content deviation reflects whether the bullet comment cluster is disconnected from the current live stream content. Normal group resonance is instantly triggered by the live stream content and is highly correlated with the live stream screen or the streamer's words and actions at that moment. Collaborative negative behavior often lacks relevance to the live stream content because it is pre-organized. Live stream content features refer to a general representation of the content broadcast in the live stream within that time window, which can be extracted from at least one of the live stream screen or the streamer's speech transcription. In practice, the text of all bullet comments within a cluster is aggregated and mapped to a bullet comment cluster semantic vector, and the live stream content features are mapped to a live stream content vector located in the same semantic space. The semantic similarity between the two vectors is calculated, and the resulting similarity value is between 0 and 1. The difference between 1 and this similarity value is taken as the content deviation. The more relevant the bullet comment cluster is to the live stream content, the closer the similarity is to 1, and the closer the content deviation is to zero. The more disconnected the cluster is, the closer the content deviation is to 1.

[0045] Emergence anomaly reflects the naturalness of the emergence pattern of bullet comment clusters. Normal group resonance is generated by a large number of dispersed viewers, with sending accounts scattered and sending times staggered. Collaborative negative behavior is often generated by relatively concentrated accounts, sending at similar rhythms. In practice, statistics are collected from two aspects: account concentration and time regularity. Regarding account concentration, the number of different sending accounts involved in the bullet comment cluster is counted. The total number of bullet comments in the cluster is divided by the number of different sending accounts to obtain the average number of bullet comments sent per account. The higher the average number of bullet comments sent per account, the more bullet comments a few accounts contribute, and the more concentrated the accounts are. This value is then converted into an account concentration component between 0 and 1. Regarding time regularity, the bullet comments within the cluster are sorted by sending time, and the sending time interval between adjacent bullet comments is calculated. The dispersion of these time intervals is counted. The more consistent the time intervals are, the more regular the emergence rhythm is. This is then converted into a time regularity component between 0 and 1. Finally, the smaller value between the account concentration component and the time regularity component is taken as the emergence anomaly, so that the emergence anomaly is high only when both account concentration and time regularity are met.

[0046] It should be noted that the conversion relationship between the statistical results and the values ​​between zero and one is preset by the technical staff. Its purpose is to make the value range of the three perspectives consistent and facilitate subsequent comprehensive judgment. The reference level used for the conversion can be determined by the technical staff after examining the statistical distribution of normal group resonance samples and collaborative negative behavior samples on historical barrage data.

[0047] It should also be noted that all three perspectives are indispensable: focusing solely on a single point of view might mean that viewers unanimously praise the same brilliant move; deviating solely from the content might mean that viewers are chatting casually; and appearing only in a regular manner might mean that the platform's activities are guiding a neat and orderly flood of posts. No single perspective is sufficient to determine collaborative negative behavior; all three must be considered together.

[0048] Based on the concentration of direction, the deviation of content, and the degree of emergence anomaly, a collaborative negative index is determined for semantically similar bullet screen clusters.

[0049] Considering that all three perspectives are indispensable, this embodiment adopts a strongly coupled approach for integration, rather than a simple summation. In specific implementation, the orientation concentration, content deviation, and emergence anomaly are first compared with their respective preset dimension thresholds. The three preset dimension thresholds represent the critical levels at which each perspective is considered too high. Technical personnel examine the value distribution of normal group resonance samples and collaborative negative behavior samples at this perspective on historical data, and take the level that can roughly separate the two types of samples as the threshold.

[0050] When the focus concentration, content deviation, and emergence anomaly are all not lower than their respective preset dimensional thresholds, the focus concentration, content deviation, and emergence anomaly are multiplied together, and the product is used as the synergistic negative index. Since all three are values ​​between zero and one, a low value in any one of them will significantly reduce the product; the synergistic negative index is only high when all three are high.

[0051] If any one of the three dimensions falls below its corresponding preset threshold, the collaborative negative indicator will be set to a level that will not trigger subsequent judgments. This avoids misjudgments caused by a single dimension being excessively high.

[0052] Live stream rooms with semantically similar bullet screen clusters that exhibit negative collaborative indicators and meet preset collaborative anomaly conditions are identified as the first live stream room.

[0053] Collaborative negative behavior refers to a series of negative expressions in the live stream that echo each other semantically and appear in a concentrated manner over time, thus collectively targeting the same object or intention. In contrast, normal group resonance refers to a concentrated expression spontaneously generated by viewers regarding the live stream content, which is semantically similar but does not have a collaborative intent to attack.

[0054] The remaining visually visible time includes:

[0055] Based on the display physical parameters and scrolling motion parameters of each bullet screen to be managed on the receiving terminal, the remaining time required for each bullet screen to move from the starting position of its track to the ending position of its track is determined.

[0056] It should be noted that "bullet comments awaiting processing" refers to bullet comments that have entered the receiving terminal's display, are scrolling on their assigned tracks, have not yet scrolled out of the receiving terminal's field of view, and have entered the processing procedure. The remaining visually visible time refers to the remaining time for a bullet comment to continue scrolling from its current position until it completely scrolls out of the viewer's field of view. It depicts the objective time window in which the processing action must be completed before it runs out, otherwise each bullet comment awaiting processing will continue to be seen by the viewer.

[0057] The physical parameters displayed specifically include the screen display size of the receiving terminal, the nominal rendering frame rate of the receiving terminal, and the instantaneous rendering frame rate of the receiving terminal.

[0058] The nominal rendering frame rate refers to the target refresh rate set by the receiving terminal under ideal smooth conditions, while the instantaneous rendering frame rate refers to the actual refresh rate reached by the receiving terminal at the current moment. The two may be inconsistent due to fluctuations in device performance. In this embodiment, the nominal rendering frame rate is 60 frames per second, and the instantaneous rendering frame rate varies with the real-time state of the device, typically ranging from 30 frames per second to 60 frames per second.

[0059] The scrolling motion parameters specifically include the bullet screen's track allocation rules, the bullet screen's nominal scrolling speed, the bullet screen's current cumulative scrolling distance, and the bullet screen's total track length.

[0060] The nominal scrolling speed refers to the set scrolling speed of the bullet comments under ideal and smooth conditions, which is taken as 200 pixels per second in this embodiment. The total track length refers to the complete pixel span of the bullet comments on the assigned track from the starting point to the ending point, and the current cumulative scrolling distance refers to the pixel distance that the bullet comments have scrolled from the starting point to the current moment.

[0061] Based on the cumulative rolling distance and the total track length in the rolling motion parameters, the remaining rolling distance is generated by subtracting the total track length from the cumulative rolling distance.

[0062] Obtain the instantaneous rendering frame rate in the display physical parameters and the nominal scroll speed in the scroll motion parameters. Calculate the ratio of the instantaneous rendering frame rate to the nominal rendering frame rate in the display physical parameters. Multiply the nominal scroll speed by the ratio to generate the instantaneous scroll speed.

[0063] It's important to note that when the actual rendering frame rate decreases, the rendering time for each frame increases, causing the originally set nominal scrolling speed to decay in real time. If the scrolling speed is still estimated based on the nominal speed, the actual movement speed of the comments will be overestimated, and the remaining duration underestimated. Therefore, the ratio of the instantaneous rendering frame rate to the nominal rendering frame rate is used as a decay coefficient to correct the nominal scrolling speed, resulting in a more accurate instantaneous scrolling speed. For example, if a comment's nominal scrolling speed is 200 pixels per second, the nominal rendering frame rate is 60 frames per second, and the current instantaneous rendering frame rate is 45 frames per second, then the ratio is 0.75, and the instantaneous scrolling speed is corrected to 150 pixels per second.

[0064] Divide the remaining scrolling distance by the instantaneous scrolling speed to generate the visually visible remaining duration of each bullet comment to be managed at the current moment.

[0065] Continuing with the previous example, if the remaining scrolling distance of a certain bullet screen to be addressed is 720 pixels and the instantaneous scrolling speed is 150 pixels per second, then the visually visible remaining duration is 4.8 seconds.

[0066] Repeat the above steps according to the preset update cycle to refresh the remaining visual time of each bullet comment to be managed until each bullet comment is removed from the receiving terminal.

[0067] The preset update cycle is the time interval for refreshing the remaining visual time. Its value needs to be determined by balancing tracking timeliness and computational overhead. In this embodiment, the preset update cycle is 100 milliseconds, while in other embodiments it can be between 50 milliseconds and 500 milliseconds.

[0068] Determine the urgency of managing each type of bullet comment, specifically including:

[0069] Obtain a sample of the actual processing time of the bullet comments that have been processed within the preset statistical window. Use the statistical value of the actual processing time sample as the end-to-end processing delay of each bullet comment to be processed. Obtain the elapsed time of each bullet comment from the time it enters the processing process to the current time. Subtract the end-to-end processing delay from the elapsed time to generate the remaining processing time.

[0070] It should be noted that the remaining visual time reflects the physical process on the bullet screen side, while whether the management action can be completed before the bullet screen scrolls out of view also depends on the physical process on the handling side. End-to-end handling latency refers to the total time taken from entering the handling process to issuing the handling instruction for handling a bullet screen, and is a measure of the time consumed on the handling side.

[0071] End-to-end processing latency should ideally be measured after the bullet comments have been processed, but governance decisions require this quantity to be obtained in real time while the bullet comments are still in the processing process, thus creating a timing contradiction. To resolve this contradiction, this embodiment adopts a prediction method based on historical samples: obtaining a sample of the actual processing time of bullet comments that have been processed within a preset statistical window, and using the statistical value of the actual processing time sample as the end-to-end processing latency of each bullet comment to be governed.

[0072] The preset statistical window is a time interval used to collect samples of the actual processing time of completed comments. Its value needs to balance the stability of the statistics with the timeliness of the response. In this embodiment, the preset statistical window is 3 seconds, while in other embodiments, it can be between 1 second and 10 seconds.

[0073] The elapsed time refers to the waiting time that each pending comment has experienced from the time it entered the processing process to the current moment. The remaining processing time indicates how long it is expected to take to complete the processing of each pending comment.

[0074] Obtain the remaining visually visible duration at the current moment, divide the remaining processing time by the remaining visually visible duration, and generate the output ratio.

[0075] The output ratio is assigned as the urgency level of each barrage to be managed.

[0076] The boundary corresponding to the output ratio being equal to 1 is recorded as the preset failure boundary.

[0077] When the output ratio is less than 1, it is determined that the urgency of the barrage-level governance is below the preset failure boundary.

[0078] When the output ratio is greater than 1, it is determined that the urgency of the barrage-level governance has exceeded the preset failure boundary.

[0079] The urgency of managing each unmanaged comment in the first live stream room is determined based on the comment-level urgency of management.

[0080] The set of all unmanaged bullet comments in the first live broadcast room during the period covered by the end-to-end processing delay is recorded as the set of bullet comments participating in the synthesis.

[0081] It should be noted that the governance urgency is a live-stream-level indicator synthesized from the barrage-level governance urgency, representing the overall governance urgency of the first live-stream room. The barrage-level governance urgency of a single barrage only reflects the urgency of that barrage to be regulated, while the platform's warnings and resource allocation target the entire live-stream room. Therefore, it is necessary to combine the barrage-level governance urgency of multiple barrages within a live-stream room into a single live-stream-level quantity.

[0082] For a newly entered barrage of comments to be addressed within the time period covered by the end-to-end processing delay, an entry observation timer is started. When the barrage of comments to be addressed continues to exist for no less than the preset entry observation time within the time period, the barrage of comments to be addressed is included in the barrage set participating in the synthesis.

[0083] Because the barrage of comments in the live stream keeps entering and leaving, if the set of comments participating in the synthesis increases or decreases in real time, the urgency of governance obtained by the synthesis will fluctuate drastically due to the instantaneous entry and exit of the comments. Therefore, an entry observation and exit observation mechanism is set up for changes in the members of the set.

[0084] For a barrage of comments that exits during the time period covered by the end-to-end processing delay, an exit observation timer is started. When the barrage of comments that needs to be managed continues to exist outside the time period for no less than the preset exit observation time, the barrage of comments that needs to be managed is removed from the barrage set participating in the synthesis. The preset exit observation time is longer than the preset entry observation time.

[0085] The preset entry observation duration is used to filter out short-lived and unrepresentative bullet comments, while the preset exit observation duration is used to prevent bullet comments near the time period boundary from being repeatedly removed and included due to brief entry and exit. The preset exit observation duration is set to be greater than the preset entry observation duration to further suppress jitter in the set members. Both are determined through offline calibration: collecting records of bullet comments entering and exiting near the time period boundary over a period of time, and statistically analyzing the typical dwell and departure durations of bullet comments near the boundary, selecting the duration that can filter out the vast majority of instantaneous jitter as the standard. In this embodiment, the preset entry observation duration is 200 milliseconds, and the preset exit observation duration is 500 milliseconds. In other embodiments, the preset entry observation duration can be between 100 milliseconds and 300 milliseconds, and the preset exit observation duration can be between 300 milliseconds and 800 milliseconds, and the preset exit observation duration is always kept greater than the preset entry observation duration.

[0086] Extract the barrage-level urgency of governance for each barrage to be governed from the set of barrages participating in the synthesis, sort them from high to low value, and extract the barrage-level urgency of governance at a preset proportion of the top as the synthesis input.

[0087] It should be noted that the urgency level of a live stream depends primarily on the most pressing few comments, rather than the average level of all comments. Therefore, after sorting by urgency from highest to lowest, only a certain percentage of the top comments are selected for compositing. In this embodiment, the preset percentage is 20%, while in other embodiments, it can be between 10% and 30%.

[0088] The proportion of each barrage-level governance urgency value in the total value of the synthetic input is used as the normalization weight. The weighted average value of the synthetic input is calculated based on the normalization weight, and the weighted average value is assigned as the governance urgency value of the first live broadcast room.

[0089] It should be noted that the weight is taken as the proportion of the governance urgency value of each barrage level. This means that the barrage with the higher value has a greater weight in the synthesis. As a result, the weighted average is tilted towards the barrage with the most urgent urgency, so that the governance urgency can highlight the most critical situation in the live broadcast room.

[0090] By suppressing the jitter of the tandem set of comments participating in the synthesis through the entry and exit observation mechanism, and highlighting the most urgent situation by cropping the proportion of the head and weighting it with the urgency value itself, a live room-level indicator that is both stable and sensitive to the real urgency of the live room is obtained, thus avoiding the drastic fluctuation of the synthesis result due to the instantaneous entry and exit of comments, or the dilution by a large number of non-urgent comments.

[0091] Based on the available resources and characteristics of the bullet screen stream in the first live stream room, the visual time budget for the first live stream room is determined.

[0092] It's important to note that visual time budget is a livestream-level metric representing the total visually visible time a livestream room can handle within a certain future period, given its current available processing resources. It measures the livestream room's capacity to handle subsequent comments. Available processing resources refer to the livestream room's current available processing capacity, while comment flow characteristics refer to features reflecting the influx of subsequent comments, such as the arrival rate of comments. A larger visual time budget indicates a more relaxed future processing capacity for the livestream room. A smaller visual time budget indicates a tighter future processing capacity for the livestream room.

[0093] The specific process for obtaining the visual time budget is as follows: Based on the processing rate of the currently available processing resources in the first live stream room and the arrival rate of the bullet comments in the first live stream room, determine the number of subsequent bullet comments that the first live stream room can process within the future time period. Use the statistical value of the visually visible remaining time of the bullet comments already observed in the first live stream room as the estimated value of the visually visible remaining time of each subsequent bullet comment. Sum the estimated values ​​of the visually visible remaining time of each subsequent bullet comment, and the sum is the visual time budget of the first live stream room.

[0094] When the urgency of governance in the first live broadcast room is greater than or equal to the preset warning threshold, a warning signal is generated for the first live broadcast room. Based on the warning signal, a second live broadcast room with a governance urgency less than the preset warning threshold is selected as the appropriator. The appropriation of disposal resources from the second live broadcast room to the first live broadcast room is initiated. The amount of disposal resources transferred by the appropriation satisfies the first constraint and the second constraint.

[0095] It should be noted that the preset warning threshold is used to determine whether a live-streaming room is under severe pressure and requires external resource support. The preset warning threshold is determined through offline calibration: data on governance failures in multiple live-streaming rooms at different levels of governance urgency is collected, and the inflection point at which the governance failure rate begins to increase significantly with rising governance urgency is statistically analyzed. The governance urgency corresponding to this inflection point is used as the preset warning threshold. Live-streaming rooms whose governance urgency reaches the preset warning threshold are identified as the first live-streaming room and receive a warning signal. One live-streaming room whose governance urgency is still below the preset warning threshold is selected as the second live-streaming room, i.e., the misappropriator.

[0096] Before performing a misappropriation, determine the type of resource that the misappropriation is targeting.

[0097] According to the preset classification rules, the resources to be processed are divided into at least two types of resources, and the end-to-end processing delay is correspondingly divided into component durations belonging to each resource type.

[0098] It should be noted that the resources being processed are not a single, homogeneous whole, but rather comprised of resources corresponding to different stages in the processing flow. The preset classification rules are set by technical personnel based on the nature of different stages in the processing flow, ensuring that each resource type corresponds to a relatively independent stage. The end-to-end processing latency is correspondingly broken down into component durations for each resource type, with each component representing the time consumed by the comment in the stage corresponding to that resource type.

[0099] The urgency of governance in the first live broadcast room is decomposed into sub-urgency states corresponding to each resource type. Each sub-urgency state is derived based on the component duration of its corresponding resource type in the end-to-end processing latency.

[0100] Sub-urgency status is a measure reflecting the urgency of a specific resource type after decomposing the urgency of governance by resource type. The longer the duration of the component corresponding to a certain resource type, the more time the bullet comments spend on that stage, the more urgent that resource type is, and the higher its sub-urgency status.

[0101] Compare the values ​​of each sub-urgency state and determine the resource type corresponding to the sub-urgency state with the highest value as the target resource type.

[0102] The misappropriation of transferred resources is limited to the target resource type.

[0103] The first and second constraints are validated independently under the target resource type, and visual time budget changes between different resource types are not combined. That is, the two constraints are validated only for the target resource type, and visual time budget changes caused by different resource types are not offset or combined.

[0104] Determining the amount of misappropriated or transferred resources involves the following specific steps:

[0105] The period during which the misappropriated resources are continuously available in the first live broadcast room is recorded as the misappropriation duration.

[0106] Based on the first and second constraints, determine the real number range of the amount of resources transferred for disposal that have been misappropriated, and extract the upper bound of the real number range.

[0107] It should be noted that the first constraint and the second constraint together limit the feasible range of misappropriation. Taking the upper bound of this range is the maximum value that can be misappropriated without violating the two constraints.

[0108] Determine the minimum allocatable granularity corresponding to the target resource type based on the target resource type.

[0109] The minimum allocatable granularity refers to the smallest unit of allocation for a target resource type that cannot be further subdivided. In a real-world system, resources can only be allocated in whole units according to a certain minimum unit. This minimum unit is the minimum allocatable granularity, which is determined by technical personnel based on the actual allocation method for the target resource type.

[0110] The upper bound of the real number range is rounded down to the smallest distributable granularity, and the rounded result is assigned as the amount of disposal resources that have been misappropriated and transferred.

[0111] It should be noted that since resources can only be allocated at the integer level, the upper bound is rounded down to the smallest allocatable granularity. The use of rounding down instead of rounding up is to ensure that the actual amount misappropriated does not exceed the upper bound of the real number range, thus avoiding violations of the first and second constraints.

[0112] Calculate the difference between the upper bound of the real number range and the result after rounding down, and compare the difference with the preset fragmentation threshold.

[0113] Rounding down leaves a margin less than the minimum allocatable granularity, which is the difference between the upper bound and the rounded result. The preset fragmentation threshold is the critical level for determining whether this margin is worth retaining. In this embodiment, the preset fragmentation threshold is half the minimum allocatable granularity.

[0114] When the difference exceeds the preset fragmentation threshold, the processing resources corresponding to the difference will be retained in the second live broadcast room as processing resources reserved by the second live broadcast room during the period of misappropriation.

[0115] When the difference is not greater than the preset fragmentation threshold, the processing resources corresponding to the difference are discarded.

[0116] The first constraint is: the decrease in visual time budget caused by misappropriation in the second live broadcast room shall not be greater than the increase in visual time budget caused by misappropriation in the first live broadcast room.

[0117] The second constraint is: the remaining visual time budget of the second live broadcast room after the misappropriation execution shall not be less than the minimum visual time budget corresponding to the preset failure boundary.

[0118] The verification steps for the first constraint include:

[0119] The difference between the visual time budget of the first live broadcast room after the misappropriation and the visual time budget of the first live broadcast room before the misappropriation is recorded as the increase in visual time budget.

[0120] The increase in visual time budget refers to the increase in visual time budget for the first live broadcast room due to the acquisition of resources.

[0121] The difference between the visual time budget of the second live stream room before the misappropriation and the visual time budget of the second live stream room after the misappropriation is recorded as the decrease in visual time budget. The decrease in visual time budget is the reduction in visual time budget of the second live stream room due to the surrender of resources.

[0122] Calculate the ratio of the visual time budget of the second live streaming room after the misappropriation to the visual time budget of the second live streaming room before the misappropriation, and generate the marginal utility factor of the second live streaming room.

[0123] Calculate the ratio of the visual time budget of the first live streaming room before the misappropriation to the visual time budget of the first live streaming room after the misappropriation, and generate the marginal utility factor of the first live streaming room.

[0124] It should be noted that the marginal utility factor is used to reflect the difference in the actual value of the same change in visual time budget across different live streaming rooms. For the second live streaming room, the marginal utility factor is the ratio of its budget after the diversion to its budget before the diversion; the tighter the budget after the diversion, the smaller this ratio. For the first live streaming room, the marginal utility factor is the ratio of its budget before the diversion to its budget after the diversion; the tighter the budget before the diversion, the smaller this ratio. The direction of the two marginal utility factors ensures that the budget change corresponding to the tighter budget receives a greater weight in subsequent comparisons.

[0125] Multiply the decrease in visual time budget by the marginal utility factor of the second live broadcast room to generate an equivalent value for the decrease in visual time budget.

[0126] Multiply the increase in visual time budget by the marginal utility factor of the first live broadcast room to generate the equivalent value of the increase in visual time budget.

[0127] The equivalent value is the actual utility value after taking into account the difference in the shortage of visual time budget.

[0128] The first constraint is deemed valid when the equivalent value of the decrease in visual time budget is not greater than the equivalent value of the increase in visual time budget.

[0129] It should be noted that the reason for introducing the marginal utility factor is that simply comparing the nominal values ​​of changes in visual time budget will ignore the fact that the same amount of budget has different meanings for a well-off live streaming room and a tight live streaming room.

[0130] The verification steps for the second constraint include:

[0131] The governance urgency of the second live broadcast room at multiple historical sampling moments within a preset observation window, as well as the visual time budget corresponding to each historical sampling moment, are obtained to form multiple urgency-budget sample pairs.

[0132] It should be noted that the minimum visual time budget required for the second constraint refers to the visual time budget corresponding to the situation where the governance urgency of the second live broadcast room approaches the preset failure boundary, serving as the safety baseline for the second live broadcast room. Since the governance urgency and visual time budget are obtained from different inputs, there is no directly solvable analytical relationship between them. Therefore, this minimum visual time budget is difficult to calculate directly. This embodiment uses a method of back-calculation based on historical data to obtain it. The preset observation window is the time interval used to collect the above historical sample pairs. The governance urgency and the corresponding visual time budget at each historical sampling moment constitute an urgency-budget sample pair. In this embodiment, the preset observation window is 10 minutes; in other embodiments, it can be between 5 and 30 minutes.

[0133] From the urgency-budget sample pairs, select a predetermined number of sample pairs with the smallest difference between governance urgency and the preset failure boundary. Take the statistical value of the visual time budget in the predetermined number of sample pairs as the minimum visual time budget.

[0134] The preset quantity refers to the number of sample pairs participating in the statistics. Its value needs to balance the representativeness and robustness of the statistics. In this embodiment, the preset quantity is 20, while in other embodiments it can be between 10 and 50.

[0135] The urgency factor is generated by dividing the urgency of governance of the second live broadcast room before its misappropriation by the preset failure boundary.

[0136] The minimum visual time budget is subtracted from the visual time budget of the second live broadcast room before the misappropriation is executed to generate a margin benchmark value.

[0137] The urgency ratio factor is multiplied by the margin benchmark value to generate the preset safety margin.

[0138] It should be noted that the preset safety margin is obtained by multiplying the urgency scaling factor by the margin benchmark value. The product form means that the closer the second live broadcast room is to the preset failure boundary, the larger the urgency scaling factor, and the larger the required preset safety margin. The farther the second live broadcast room is from the preset failure boundary, the smaller the preset safety margin.

[0139] The sum of the minimum visual time budget and the preset safety margin is used as the lower bound for the second constraint. The lower bound is the level that the remaining visual time budget of the second live broadcast room after the misappropriation execution must not fall below.

[0140] The second constraint is deemed valid when the remaining visual time budget of the second live broadcast room after the misappropriation execution is not lower than the lower bound of the judgment.

[0141] The online live streaming bullet screen language dynamic early warning and coordinated management system specifically includes:

[0142] The data acquisition module is used to acquire the remaining visual time of each barrage to be managed in the first live broadcast room and the corresponding end-to-end processing latency. Based on the comparison relationship between the remaining visual time of each barrage to be managed and the end-to-end processing latency, the barrage-level management urgency of each barrage to be managed is determined, and the preset failure boundary is determined based on the comparison relationship.

[0143] The governance urgency module is used to determine the governance urgency of the first live broadcast room based on the barrage-level governance urgency of each barrage to be governed in the first live broadcast room.

[0144] The visual time budget module is used to determine the visual time budget of the first live broadcast room based on the available resources and the characteristics of the barrage stream of the first live broadcast room.

[0145] The resource misappropriation module is used to generate an early warning signal for the first live broadcast room when the governance urgency of the first live broadcast room is greater than or equal to a preset early warning threshold. Based on the early warning signal, the module selects the second live broadcast room with a governance urgency less than the preset early warning threshold as the misappropriator and initiates resource misappropriation from the second live broadcast room to the first live broadcast room. The amount of misappropriated resources satisfies the first constraint and the second constraint.

[0146] The first constraint is: the decrease in visual time budget caused by misappropriation in the second live broadcast room shall not be greater than the increase in visual time budget caused by misappropriation in the first live broadcast room.

[0147] The second constraint is: the remaining visual time budget of the second live broadcast room after the misappropriation execution shall not be less than the minimum visual time budget corresponding to the preset failure boundary.

[0148] This embodiment does not replace the existing bullet screen recognition and blocking functions of the live streaming platform, but rather superimposes them on top of the existing functions. In normal traffic scenarios, the platform's existing pre-screening and blocking functions are sufficient to handle illegal bullet screens before they disappear from the viewer's field of vision. However, in high-concurrency scenarios, and when the platform's pre-screening blocking rules are incomplete or blocking resources are limited, some illegal bullet screens cannot be blocked in time. This embodiment addresses such illegal bullet screens that cannot be blocked in time by using the remaining visually visible time as the physical benchmark for governance decisions, enabling governance scheduling to be anchored to the actual display process of the bullet screens on the receiving terminal. Furthermore, by dynamically allocating processing resources across live streaming rooms, processing resources can be promptly allocated to busy live streaming rooms without relying on statically reserving a large amount of redundant resources. This minimizes the exposure time of illegal bullet screens in the viewer's field of vision, reduces governance failure rate, and improves the overall utilization rate of processing resources.

[0149] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the above functions can be divided into different functional modules to complete all or part of the functions described above.

[0150] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0151] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for dynamic early warning and coordinated management of online live streaming barrage language, characterized in that, include: The remaining visual time and corresponding end-to-end processing latency of each barrage to be managed in the first live broadcast room are obtained. Based on the comparison relationship between the remaining visual time and end-to-end processing latency of each barrage to be managed, the barrage-level management urgency of each barrage to be managed is determined, and a preset failure boundary is determined based on the comparison relationship. The urgency of managing each unmanaged barrage in the first live stream room is determined based on the barrage-level urgency of managing that barrage. Based on the available resources and characteristics of the bullet screen stream in the first live stream room, determine the visual time budget for the first live stream room; When the urgency of governance of the first live broadcast room is greater than or equal to the preset warning threshold, a warning signal is generated for the first live broadcast room, and a second live broadcast room with a governance urgency less than the preset warning threshold is selected as the appropriator based on the warning signal. The appropriation of disposal resources is initiated from the second live broadcast room to the first live broadcast room. The amount of disposal resources transferred by the appropriation satisfies the first constraint and the second constraint. The first constraint is: the decrease in visual time budget of the second live streaming room due to the misappropriation is not greater than the increase in visual time budget of the first live streaming room due to the misappropriation; The second constraint is that the remaining visual time budget of the second live broadcast room after the misappropriation is executed is not less than the minimum visual time budget corresponding to the preset failure boundary.

2. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The first live streaming room also includes the identification of the first live streaming room, specifically including: Several live streaming rooms on the platform are monitored in real time to obtain the bullet comments of each live streaming room within a preset time window. The bullet comments are clustered according to the degree of semantic similarity to obtain at least one semantically similar bullet comment cluster. For the semantically similar bullet screen clusters, the direction concentration is determined based on the convergence of the referents of each bullet screen within the cluster, the content deviation is determined based on the overall semantic deviation of the semantically similar bullet screen cluster relative to the live content features of the live room, and the emergence anomaly is determined based on the distribution characteristics and sending time rhythm of the sending accounts corresponding to each bullet screen within the cluster. The collaborative negative index of the semantically similar bullet screen clusters is determined based on the direction concentration, the content deviation, and the emergence anomaly. Live stream rooms with semantically similar bullet screen clusters that exhibit collaborative negative indicators that meet preset collaborative anomaly conditions are identified as the first live stream room.

3. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The remaining visually visible duration specifically includes: Based on the display physical parameters and scrolling motion parameters of each barrage to be managed on the receiving terminal, determine the remaining time required for each barrage to move from the starting position of its track to the ending position of its track. The display physical parameters specifically include the screen display size of the receiving terminal, the nominal rendering frame rate of the receiving terminal, and the instantaneous rendering frame rate of the receiving terminal. The scrolling motion parameters specifically include the barrage's track allocation rules, the barrage's nominal scrolling speed, the barrage's current cumulative scrolling distance, and the barrage's total track length; Based on the cumulative rolling distance in the rolling motion parameters and the total track length, the remaining rolling distance is generated by subtracting the total track length from the cumulative rolling distance; Obtain the instantaneous rendering frame rate in the display physical parameters and the nominal scroll speed in the scroll motion parameters, calculate the ratio of the instantaneous rendering frame rate to the nominal rendering frame rate in the display physical parameters, and multiply the nominal scroll speed by the ratio to generate the instantaneous scroll speed; Divide the remaining scrolling distance by the instantaneous scrolling speed to generate the visually visible remaining duration of each bullet screen to be managed at the current moment; Repeat the above steps according to the preset update cycle to refresh the remaining visual time of each bullet comment to be managed until each bullet comment is removed from the receiving terminal.

4. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The determination of the urgency of managing each barrage of comments specifically includes: Obtain a sample of the actual processing time of the bullet comments that have been processed within the preset statistical window. Use the statistical value of the actual processing time sample as the end-to-end processing delay of each bullet comment to be processed. Obtain the elapsed time of each bullet comment from the time it enters the processing process to the current time. Calculate the difference between the end-to-end processing delay and the elapsed time to generate the remaining processing time. Obtain the remaining visually visible duration at the current moment, divide the remaining processing time by the remaining visually visible duration, and generate an output ratio. The output ratio is assigned as the urgency level of each barrage to be managed; The boundary corresponding to the output ratio being equal to 1 is recorded as the preset failure boundary; When the output ratio is less than 1, it is determined that the urgency of the barrage-level governance is below the preset failure boundary; When the output ratio is greater than 1, it is determined that the urgency of the barrage-level governance exceeds the preset failure boundary.

5. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The determination of the urgency of managing the first live stream based on the urgency of managing each unmanaged barrage in the first live stream specifically includes: The set of all unprocessed bullet comments in the first live broadcast room during the time period covered by the end-to-end processing delay is denoted as the set of bullet comments participating in the synthesis. For a newly entered barrage to be managed within the time period covered by the end-to-end processing delay, start the entry observation timer. When the barrage to be managed continues to exist for no less than the preset entry observation time within the time period, the barrage to be managed is included in the barrage set participating in the synthesis. For a barrage of comments to be addressed that exits during the time period covered by the end-to-end processing delay, an exit observation timer is started. When the barrage of comments to be addressed continues to exist outside the time period for a period of not less than a preset exit observation time, the barrage of comments to be addressed is removed from the barrage set participating in the synthesis. The preset exit observation time is longer than the preset entry observation time. Extract the barrage-level urgency of each barrage to be addressed from the barrage set participating in the synthesis, sort them from high to low, and extract a preset proportion of the barrage-level urgency of the head as the synthesis input; The proportion of each barrage-level governance urgency value in the total value of the synthesized input is used as the normalization weight. Based on the normalization weight, a weighted average value is calculated for the synthesized input, and the weighted average value is assigned as the governance urgency value of the first live broadcast room.

6. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The aforementioned initiation of resource misappropriation from the second live streaming room to the first live streaming room specifically includes: The processing resources are classified into at least two resource types according to a preset classification rule, and the end-to-end processing delay is correspondingly divided into component durations belonging to each of the resource types. The governance urgency of the first live broadcast room is decomposed into sub-urgency states corresponding to each of the resource types. Each sub-urgency state is derived based on the duration of the component belonging to its corresponding resource type in the end-to-end processing delay. Compare the values ​​of each sub-urgency state and determine the resource type corresponding to the sub-urgency state with the highest value as the target resource type; The misappropriated resources are defined as belonging to the target resource type. The first constraint and the second constraint are validated independently under the target resource type, and the visual time budget variations between different resource types are not combined.

7. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 6, characterized in that: Determining the amount of misappropriated or transferred resources involves the following specific steps: The period during which the misappropriated resources are continuously available in the first live broadcast room is recorded as the misappropriation duration. Based on the first constraint and the second constraint, determine the real number range of the amount of disposal resources transferred by the misappropriation, and extract the upper bound of the real number range; Determine the minimum allocatable granularity corresponding to the target resource type based on the target resource type; The upper bound of the real number range is rounded down according to the smallest distributable granularity, and the rounded result is assigned as the amount of disposal resources transferred by the misappropriation. Calculate the difference between the upper bound of the real number range and the result after rounding down, and compare the difference with a preset fragmentation threshold; When the difference is greater than the preset fragmentation threshold, the processing resources corresponding to the difference are retained in the second live broadcast room as processing resources reserved by the second live broadcast room during the misappropriation period. When the difference is not greater than the preset fragmentation threshold, the processing resources corresponding to the difference are discarded.

8. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The verification steps for the first constraint include: The difference between the visual time budget of the first live streaming room after the misappropriation is executed and the visual time budget of the first live streaming room before the misappropriation is executed is recorded as the increase in visual time budget. The difference between the visual time budget of the second live streaming room before the misappropriation is executed and the visual time budget of the second live streaming room after the misappropriation is executed is recorded as the decrease in visual time budget. Calculate the ratio of the visual time budget of the second live streaming room after the misappropriation is executed to the visual time budget of the second live streaming room before the misappropriation is executed, and generate the marginal utility factor of the second live streaming room; Calculate the ratio of the visual time budget of the first live streaming room before the misappropriation is executed to the visual time budget of the first live streaming room after the misappropriation is executed, and generate the marginal utility factor of the first live streaming room. Multiply the decrease in visual time budget by the marginal utility factor of the second live streaming room to generate an equivalent value of the decrease in visual time budget; Multiply the increase in visual time budget by the marginal utility factor of the first live streaming room to generate an equivalent value of the increase in visual time budget. The first constraint is determined to be valid when the equivalent value of the decrease in visual time budget is not greater than the equivalent value of the increase in visual time budget.

9. The method for dynamic early warning and coordinated management of online live streaming barrage language as described in claim 1, characterized in that: The verification steps for the second constraint include: The governance urgency of the second live broadcast room at multiple historical sampling moments within a preset observation window, and the visual time budget corresponding to each historical sampling moment, are obtained to form multiple urgency-budget sample pairs; From the urgency-budget sample pairs, select a preset number of sample pairs with the smallest difference between governance urgency and the preset failure boundary, and take the statistical value of the visual time budget in the preset number of sample pairs as the minimum visual time budget; Divide the governance urgency of the second live broadcast room before the misappropriation is executed by the preset failure boundary to generate an urgency ratio factor; The minimum visual time budget is subtracted from the visual time budget of the second live broadcast room before the misappropriation is executed to generate a margin benchmark value. Multiply the urgency ratio factor by the margin benchmark value to generate a preset safety margin; The sum of the minimum visual time budget and the preset safety margin is used as the lower bound for the second constraint. The second constraint is determined to be valid when the remaining visual time budget of the second live broadcast room after the misappropriation is not lower than the lower bound of the determination.

10. A dynamic early warning and coordinated management system for online live streaming barrage language, characterized by: Specifically, it includes: The data acquisition module is used to acquire the remaining visual time of each barrage to be managed in the first live broadcast room and the corresponding end-to-end processing delay. Based on the comparison relationship between the remaining visual time of each barrage to be managed and the end-to-end processing delay, the barrage-level management urgency of each barrage to be managed is determined, and a preset failure boundary is determined based on the comparison relationship. The governance urgency module is used to determine the governance urgency of the first live broadcast room based on the barrage-level governance urgency of each barrage to be governed in the first live broadcast room; The visual time budget module is used to determine the visual time budget of the first live broadcast room based on the currently available resources and the characteristics of the bullet screen stream of the first live broadcast room. The resource misappropriation module is used to generate an early warning signal for the first live broadcast room when the governance urgency of the first live broadcast room is greater than or equal to a preset early warning threshold, and select a second live broadcast room with a governance urgency less than the preset early warning threshold as the misappropriator based on the early warning signal, and initiate resource misappropriation from the second live broadcast room to the first live broadcast room. The amount of the misappropriated resources satisfies the first constraint and the second constraint. The first constraint is: the decrease in visual time budget of the second live streaming room due to the misappropriation is not greater than the increase in visual time budget of the first live streaming room due to the misappropriation; The second constraint is that the remaining visual time budget of the second live broadcast room after the misappropriation is executed is not less than the minimum visual time budget corresponding to the preset failure boundary.

Images