A community electronic screen broadcast information generation system based on a cloud platform

Abstract

The present application relates to the field of electronic screen content broadcasting based on cloud platform, and particularly relates to a community electronic screen broadcasting information generation system based on cloud platform, which periodically collects information through a collection module, cleans and divides priorities through a preprocessing module, generates article information abstract through an article processing module, determines a playing position through a positioning module, formulates a playing form through a content generation module, arranges a playing order through a queue module, is responsible for content transmission through a transmission module, detects abnormalities in real time and timely locates abnormal reasons, corrects abnormal states through a monitoring module, and adjusts parameters of each module according to quality correction of the monitoring module through a control optimization module, so as to guarantee stable and efficient operation of the system. The community electronic screen broadcasting information generation system based on cloud platform realizes automatic acquisition and generation of broadcasting information of multiple community electronic screens, solves the problem that the prior art cannot automatically monitor and repair abnormalities, and thus realizes all-round management of broadcasting information generation, monitoring and optimization.

Description

Technical Field

[0001] This invention relates to the field of electronic screen content broadcasting based on a cloud platform, and specifically to a community electronic screen broadcasting information generation system based on a cloud platform. Background Technology

[0002] A smart community is a new type of community that leverages modern information technologies such as the Internet of Things, big data, cloud computing, and artificial intelligence to integrate various community resources and achieve intelligent community management, convenient services, and comfortable living. By building intelligent infrastructure and service platforms, smart communities can significantly improve community governance, optimize residents' living experience, and enhance their sense of happiness and security.

[0003] The intelligent construction of community electronic screen broadcasts is an important component of smart community development. However, existing community electronic screen broadcasts typically rely on local storage and pre-set programs for automatic playback. While simple to implement, this approach suffers from issues such as untimely content updates and the inability to provide personalized content. Therefore, connecting community electronic screens to the internet, enabling them to periodically retrieve and play the latest content, represents a better option for smart community development.

[0004] However, while this internet connection method solves the problems of content updates and personalized push notifications, when malfunctions occur on community electronic screens, users typically need to report the issue, and then a specialized operating company will dispatch professionals to inspect and repair it. This process is not only time-consuming and labor-intensive, but also seriously affects the efficiency of problem-solving and the user experience for residents.

[0005] Therefore, how to achieve intelligent positioning and generation of information broadcast on community electronic screens, while simultaneously monitoring the working status of these screens, promptly identifying anomalies, quickly locating the causes of anomalies, and automatically repairing them, has become one of the key challenges in building smart communities. Summary of the Invention

[0006] This invention provides a cloud-based community electronic screen broadcast information generation system to overcome the problems of wasted human resources and decreased user experience caused by the inability of existing network-connected community electronic screen broadcast systems to automatically monitor, locate, and repair broadcast anomalies.

[0007] Therefore, this invention provides a community electronic screen broadcast information generation system based on a cloud platform, comprising:

[0008] The data acquisition module periodically collects broadcast information and electronic screen data based on the cloud platform;

[0009] A preprocessing module, which is connected to the acquisition module, is used to preprocess the acquired data and divide the playback priority based on the type of broadcast information;

[0010] An article processing module, which is connected to the preprocessing module, is used to generate article-type broadcast information summaries based on the Transformer algorithm;

[0011] The positioning module, which is connected to the preprocessing module, is used to obtain the broadcast location and broadcast date from the broadcast information based on the pre-trained NER model and locate the corresponding electronic screen.

[0012] A content generation module, which is connected to the preprocessing module, the article processing module and the positioning module, is used to determine the broadcast format of the broadcast information based on the content generation model;

[0013] A queue module, which is connected to the preprocessing module, is used to determine the playback queue based on the forced rotation threshold and the playback priority;

[0014] A transmission module, which is connected to the content generation module and the positioning module, is used to compress the playback content into a playback task and transmit it to the corresponding electronic screen;

[0015] The monitoring module, which is connected to the preprocessing module, is used to determine whether the current playback is abnormal and the cause of the abnormality based on the proportion of the lowest priority unplayed task in the playback queue with resource backlog within the period, the difference between the sending time and the corresponding receiving time of the most recent playback task, and the length of the associated path from each broadcast location to the corresponding broadcast date. Based on the cause of the abnormality, the queue generation parameters, data transmission parameters, the sliding window parameters of the NER model, and the long text processing parameters are corrected.

[0016] The control optimization module, which is connected to the monitoring module, the transmission module, the article processing module, the positioning module, and the queue module, is used to correct the parameters of the corresponding modules based on the instructions issued by the transmission module.

[0017] Furthermore, the content generation module is used to construct a content generation model based on a self-learning algorithm, train the content generation model based on the relationship between historical electronic screen data and broadcast information format, and determine the corresponding electronic screen playback format based on the electronic screen data and the broadcast information after preprocessing and article processing.

[0018] Furthermore, the monitoring module is used to determine the playback coverage rate based on the proportion of the number of playback tasks that are played at least once in the playback queue within the period to the total number of playback tasks in the playback queue, and to determine whether there is resource backlog in the current playback based on the playback coverage rate, and to determine the minimum priority ratio based on the proportion of the lowest priority unplayed task in the playback queue to the total number of unplayed tasks when resource backlog is determined, and to determine whether there is priority starvation based on the minimum priority ratio, and to correct the forced rotation threshold based on the ratio of the minimum priority ratio to the preset minimum priority ratio when priority starvation is determined, or to determine the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recent playback task, or to determine the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recent playback task when resource backlog is determined.

[0019] The most recently played task refers to the task that was sent most recently from the time node of the periodic collection. The forced rotation threshold refers to the number of playback tasks that are executed consecutively with a higher priority than the number of playback tasks that are forced to be executed when the playback task with the longest current waiting time and the lowest priority in the current playback queue is forcibly executed.

[0020] Furthermore, the monitoring module is also used to determine the lowest priority ratio based on the ratio of the lowest priority ratio to the preset lowest priority ratio, and to reduce the forced rotation threshold based on the lowest priority ratio, wherein the reduction of the forced rotation threshold is proportional to the lowest priority ratio.

[0021] Furthermore, the monitoring module is also used to determine the timestamp difference based on the difference between the timestamp sent by the playback task and the timestamp received by the corresponding playback task, and to determine the cause of the abnormality based on the timestamp difference, and to correct the compression ratio of the transmission module based on the ratio of the timestamp difference to the preset timestamp difference when the network is determined to be unstable, or to determine whether there is a playback abnormality based on the path length from the broadcast location to the broadcast date in the original text of the broadcast information.

[0022] The compression ratio refers to the ratio of the compressed data size to the original data size after data compression during data transmission.

[0023] Furthermore, the monitoring module is also used to determine a timestamp difference ratio based on the ratio of the timestamp difference to the preset timestamp difference, and to increase the compression ratio of the transmission module based on the timestamp difference ratio, wherein the increase in the compression ratio of the transmission module is proportional to the timestamp difference ratio.

[0024] Furthermore, the monitoring module is also used to determine the associated path length based on the path length from the broadcast location to the broadcast date in the original broadcast information text, and to determine whether there is an anomaly based on the associated path length, and to adjust the sliding step size of the NER model sliding window based on the ratio of the associated path length to the preset associated path length when the location is determined to be abnormal, or to determine whether there is an anomaly based on the proportion of the number of keywords covered in the broadcast information summary to the total number of keywords covered.

[0025] Furthermore, the monitoring module is also used to determine the associated path length ratio based on the ratio of the associated path length to the preset associated path length, and to increase the sliding step size of the NER model sliding window based on the associated path length ratio, wherein the increase in the sliding step size is proportional to the associated path length ratio.

[0026] Furthermore, the monitoring module is also used to determine the sliding step size difference based on the difference in sliding step size of the sliding window of the NER model before and after correction, and to increase the sliding window size of the NER model sliding window based on the sliding step size difference, wherein the increase in the sliding window size is proportional to the sliding step size difference.

[0027] Furthermore, the monitoring module is also used to determine the keyword coverage rate based on the proportion of the number of keywords covered in the article broadcast information summary to the total number of keywords covered, and to determine whether there is an anomaly based on the keyword coverage rate. When it is determined that the electronic screen is working normally, the system parameters are maintained and the detection continues. Alternatively, when it is determined that the broadcast information summary is deviating, the keyword coverage rate difference is determined based on the difference between the keyword coverage rate and the preset coverage rate. The length penalty in the Transformer algorithm is increased based on the keyword coverage rate difference, and the increase in the length penalty is proportional to the keyword coverage rate difference.

[0028] Compared with existing technologies, the beneficial effects of this invention are that the community electronic screen broadcast information generation system of this invention achieves fully automated monitoring and management of the entire process of broadcast information generation, transmission, and playback by integrating acquisition, preprocessing, article processing, positioning, content generation, queue, transmission, monitoring, and control optimization modules. The system can automatically monitor the task execution status in the playback queue, promptly detect various anomalies such as resource backlog, priority starvation, network latency, and positioning abnormalities, and intelligently repair these issues by dynamically adjusting relevant parameters (such as forced rotation threshold, transmission module compression ratio, and NER model sliding window parameters). This automatic monitoring and adaptive adjustment mechanism significantly reduces the waste of human resources caused by anomalies, avoids the cumbersome process of manual feedback and on-site maintenance by professionals required in traditional systems, and greatly improves problem-solving efficiency. Simultaneously, by optimizing parameter adjustments, the system ensures that information is displayed on the community electronic screen in a timely and accurate manner, significantly improving user experience, enhancing residents' attention to and satisfaction with the broadcast content, and providing efficient and intelligent technical support for the construction of smart communities.

[0029] Furthermore, the content generation module constructs a content generation model based on a self-learning algorithm and trains it using the relationship between historical electronic screen data and broadcast information formats. This enables the system to intelligently determine the corresponding playback format based on electronic screen data and pre-processed broadcast information. This self-learning and historical data-driven approach allows for more personalized and diversified broadcast information generation, meeting the specific needs of different community electronic screens, further enhancing the attractiveness and practicality of broadcast information, and increasing residents' attention to and satisfaction with the content broadcast on community electronic screens.

[0030] Furthermore, by analyzing the task execution status in the playback queue, the monitoring module can promptly detect resource backlog and priority starvation, and adjust the forced rotation threshold accordingly. This dynamic adjustment mechanism effectively prevents low-priority tasks from being unable to play for extended periods, ensuring that all types of information are displayed fairly and reasonably, improving the fairness and efficiency of the system, while reducing manual intervention and lowering operating costs.

[0031] Furthermore, by dynamically adjusting the forced rotation threshold based on the lowest priority ratio, the system can achieve scientific, reasonable, flexible, and dynamic adaptive control. This adjustment method can prevent high-priority playback tasks from continuously preempting the playback resources of low-priority playback tasks, while ensuring the timely broadcast of audio and video content and improving the stability and service quality of the community electronic screen broadcasting system.

[0032] Furthermore, the monitoring module determines the cause of the anomaly based on the timestamp difference and adjusts the compression ratio of the transmission module accordingly. This dynamic adjustment mechanism effectively addresses data transmission latency issues caused by network instability, ensuring timely data transmission and smooth playback, thus improving system stability and user experience. Through automated adjustments, the system can promptly detect and resolve anomalies, providing users with a smoother and more stable playback experience.

[0033] Furthermore, the compression ratio of the transmission module is dynamically adjusted based on the timestamp difference ratio, enabling the system to flexibly adapt to different levels of network anomalies. This adaptive adjustment mechanism effectively reduces network load and overall transmission latency, improves system stability and user experience, and ensures timely data transmission and smooth playback.

[0034] Furthermore, the monitoring module determines anomalies by analyzing the associated path length and adjusts the sliding step size of the NER model's sliding window accordingly. This adaptive adjustment mechanism effectively addresses identification anomalies caused by low associated path lengths, improves entity recognition accuracy and broadcast information quality, enhances the community electronic screen broadcasting system's ability to identify and handle anomalies, and improves system intelligence and robustness.

[0035] Furthermore, dynamically adjusting the sliding step size of the NER model's sliding window based on the ratio of associated path lengths helps to make the adjustment flexible and intelligent, preventing over- or under-adjustment and ensuring the stable and efficient operation of the pre-trained NER model. This design improves the accuracy of NER recognition, avoids recognition gaps or entity splitting caused by increasing the step size, adapts to different abnormal situations through a dynamic adjustment mechanism, achieves parameter self-correction and optimization, and improves localization accuracy.

[0036] Furthermore, the monitoring module compensates for the reduced overlap caused by the increased step size by adjusting the sliding window size of the NER model, ensuring that each window contains more context and maintaining recognition accuracy. This dynamic adjustment mechanism can dynamically adjust the window size according to the magnitude of the step size change, matching the compensation intensity with the risk level, preventing over-compensation or under-compensation, thereby improving the system's adaptability and stability.

[0037] Furthermore, the monitoring module judges the quality of the summary by keyword coverage and adjusts the length penalty in the Transformer algorithm accordingly. This adaptive adjustment mechanism can dynamically adjust based on the summary quality, avoiding over- or under-adjustment, ensuring system stability, and achieving refined and intelligent summary optimization. This effectively improves the quality and completeness of the information summaries broadcast on the community electronic screen, enhances the system's ability to perceive and handle abnormalities in the quality of broadcast information, and improves the accuracy of the broadcast content and user experience. Attached Figure Description

[0038] Figure 1 This is a block diagram of a community electronic screen broadcast information generation system based on a cloud platform, as described in an embodiment of the present invention.

[0039] Figure 2 This is a flowchart illustrating the workflow of a community electronic screen broadcast information generation system based on a cloud platform, as described in this embodiment of the invention.

[0040] Figure 3 This is a flowchart illustrating how the working status of an electronic screen is determined based on playback coverage and the lowest priority percentage in an embodiment of the present invention.

[0041] Figure 4 This is a flowchart illustrating the sliding step size of the NER model sliding window based on the ratio of associated path lengths in an embodiment of the present invention. Detailed Implementation

[0042] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0043] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0044] Please see Figure 1 The diagram shown is a block diagram of a community electronic screen broadcast information generation system based on a cloud platform, as described in this embodiment of the invention. This system includes a data acquisition module, an article processing module, a positioning module, a content generation module, a queue module, a transmission module, a monitoring module, and a control optimization module.

[0045] The acquisition module periodically collects broadcast information and electronic screen data based on the cloud platform;

[0046] The preprocessing module is connected to the acquisition module and is used to preprocess the acquired data and divide the playback priority based on the type of broadcast information.

[0047] The article processing module is connected to the preprocessing module and is used to generate article-type broadcast information summaries based on the Transformer algorithm.

[0048] The positioning module is connected to the preprocessing module and is used to obtain the broadcast location and broadcast date from the broadcast information based on the pre-trained NER model and locate the corresponding electronic screen.

[0049] The content generation module is connected to the preprocessing module, the article processing module and the positioning module, and is used to determine the broadcast format of the broadcast information based on the content generation model.

[0050] The queue module, which is connected to the preprocessing module, is used to determine the playback queue based on the forced rotation threshold and the playback priority;

[0051] The transmission module is connected to the content generation module and the positioning module, and is used to compress the playback content into a playback task and transmit it to the corresponding electronic screen.

[0052] The monitoring module, which is connected to the preprocessing module, is used to determine whether the current playback is abnormal and the cause of the abnormality based on the proportion of the lowest priority unplayed task in the playback queue with resource backlog within the period, the difference between the sending time and the corresponding receiving time of the most recent playback task, and the length of the associated path from each broadcast location to the corresponding broadcast date. Based on the cause of the abnormality, the queue generation parameters, data transmission parameters, the sliding window parameters of the NER model, and the long text processing parameters are corrected.

[0053] The control optimization module is connected to the monitoring module, the transmission module, the article processing module, the positioning module, and the queue module, and is used to correct the parameters of the corresponding modules based on the instructions issued by the transmission module.

[0054] In principle, there are no restrictions on the method of periodically collecting and broadcasting information based on the cloud platform. The collection method includes, but is not limited to, any one or more of the following: cloud platform timed scheduling service, cloud function polling, building a cloud crawler scheduling platform, and periodically requesting third-party or proprietary broadcasting information API interfaces. This will not be elaborated further.

[0055] The data collected from the community electronic screens is not limited in principle. Technical personnel may choose to collect data information from the community electronic screens according to their needs, including but not limited to the location information of the community electronic screens, the length of the playback queue, the priority of the broadcast task, and the data transmission timestamp, etc., which will not be elaborated here.

[0056] When the preprocessing module preprocesses the collected data, the preprocessing method is not limited in principle, including but not limited to data cleaning and data normalization, which will not be elaborated here.

[0057] The preprocessing module includes the process of prioritizing broadcast information based on its type, which includes:

[0058] The collected information types are classified using manual or pre-trained information classification models;

[0059] The steps for constructing the information classification model include: constructing the information classification model using artificial intelligence algorithms. In principle, the algorithms mentioned here are not limited, and technicians can choose different algorithms to construct the information classification model according to their needs.

[0060] An information classification model is trained based on the relationship between historical broadcast information and broadcast information types. Then, the collected broadcast information is classified using the trained information classification model.

[0061] Among them, safety and emergency information, such as epidemic prevention and control notices and extreme weather warnings, are classified as level one, which is the highest priority information;

[0062] Policy promotion and government information disclosure, such as community notices and policy proposals, are classified as secondary priority.

[0063] Information related to public services and daily life, such as weather forecasts, community event announcements, and fraud prevention cases, is classified as Level 3.

[0064] In the article processing module, broadcast messages with more than 200 characters are recorded as article broadcast messages, and a summary of the article broadcast messages is generated based on the Transformer algorithm. The process of generating the summary of the article broadcast messages based on the Transformer algorithm includes:

[0065] First, the article broadcast information is segmented into sub-words. The methods for segmenting the article broadcast information include, but are not limited to, any one or more of BERT's WordPiece segmentation method, spaCy, and jieba, which will not be elaborated here.

[0066] The segmented text is then converted into a numerical sequence that the model can understand, and markers are added at the beginning and end of the article broadcast information to indicate the start and end of the article;

[0067] Ensure all input sequences are of consistent length; truncate if the text is too long, and pad if the text is too short.

[0068] Choose a pre-trained Transformer model. In principle, there is no limitation on the Transformer model mentioned here. Technical personnel can choose any one of the models, including but not limited to BART, T5 and ATS_CG models. This will not be elaborated further.

[0069] The maximum length ML in the model is set to 80-100 to ensure that the key points of the original article broadcast are accurately disclosed, while also ensuring that the information is concise and clear, making it easy for residents to quickly obtain key information.

[0070] The minimum length (ML) in the model is set to 40-50 to avoid generating summary information that is too short and cannot fully convey the key information in the original article broadcast information;

[0071] The length penalty LP is set to a value between 0.5 and 0.7 to balance the conciseness of the summary with the integrity of the information.

[0072] In principle, the decoding strategy in the model is not limited. The decoding strategy includes, but is not limited to, greedy decoding, beam search and sampling. Taking beam search as an example, when beam search is used as the decoding strategy, the value of the beam width is selected from 2 to 5 in order to generate a more complete and fluent summary.

[0073] Once the settings are complete, input the preprocessed article broadcast information into the model to obtain the summary.

[0074] The process by which the positioning module obtains the broadcast location and broadcast date from the broadcast information based on the pre-trained NER model includes:

[0075] First, select a pre-trained NER model for loading. In principle, there are no restrictions on the pre-trained NER model. Technical personnel can choose according to their needs. The pre-trained NER model includes, but is not limited to, any one of BERT, BERT-base-NER, RoBERTa, BiLSTM, T5, or Spacy.

[0076] Define the entity type to be identified, including the broadcast location and broadcast date;

[0077] The value range for the sliding window size SWS is set to 128 to 512 tokens.

[0078] The sliding step size ST of the NER sliding window is set to a range of 30% - 70% of the sliding window size;

[0079] Set the confidence threshold for entity recognition to 0.85-0.95;

[0080] The article is segmented into words or phrases using a word segmentation tool. Each word is converted into a corresponding code based on the vocabulary of the pre-trained NER model and input into the model. The output includes several broadcast locations and corresponding broadcast dates.

[0081] In the queue module, the community electronic screen plays the received playback information in order of priority and reception, and deletes the corresponding task after the playback task expires. When the community electronic screen receives new playback information, it retrieves the priority of the broadcast information and compares it with the priority of the current playback information.

[0082] If a newly received playback message has a higher priority than the currently played message, then the newly received message will be played after the current message finishes playing, and will be inserted at the end of the queue corresponding to its priority.

[0083] If the priority of a newly received playback message is lower than or equal to the priority of the current playback message, it is inserted at the end of the queue corresponding to its priority.

[0084] To ensure fairness in playback, whenever N tasks with higher priority than the lowest priority in the current playback queue are played consecutively, the task with the longest waiting time and lowest priority in the current playback queue is forcibly played. Here, N is denoted as the forced rotation threshold, which is the number of consecutive playback tasks with higher priority than the task being forcibly executed when the task with the longest waiting time and lowest priority in the current playback queue is forcibly executed. In most commercial and municipal information publishing systems, the forced rotation threshold is usually set to 3-5 to balance information priority and playback fairness.

[0085] Furthermore, the content generation module is used to construct a content generation model based on a self-learning algorithm, train the content generation model based on the relationship between historical electronic screen data and broadcast information format, and determine the corresponding electronic screen playback format based on the electronic screen data and the broadcast information after preprocessing and article processing.

[0086] When constructing a content generation model based on a self-learning algorithm, the self-learning algorithm is not limited in principle. Technical personnel can choose any one of the following according to their needs: Q-Learning, Deep Q-Network, Proximal Policy Optimization, Actor-Critic Methods, SimCLR, Fine-Tuning, and MoCo. This will not be elaborated further.

[0087] Further, please refer to Figure 2 The diagram shown illustrates the workflow of a community electronic screen broadcast information generation system based on a cloud platform, as described in this embodiment of the invention. The workflow of this cloud-based community electronic screen broadcast information generation system includes:

[0088] S1: The acquisition module periodically collects broadcast information and electronic screen data based on the cloud platform;

[0089] S2: The preprocessing module preprocesses the collected data and assigns playback priority based on the type of broadcast information;

[0090] S3: The article processing module generates an article-type broadcast information summary based on the Transformer algorithm;

[0091] S4: The positioning module obtains the broadcast location and broadcast date from the broadcast information based on the pre-trained NER model and locates the corresponding electronic screen;

[0092] S5: The content generation module constructs a content generation model based on a self-learning algorithm, and trains the content generation model based on the relationship between historical electronic screen data and broadcast information format, and determines the corresponding electronic screen playback format based on the electronic screen data and the broadcast information after preprocessing and article processing.

[0093] S6: The queue module determines the playback queue based on the forced rotation threshold and the playback priority;

[0094] S7: The transmission module compresses the playback content into a playback task and transmits it to the corresponding electronic screen;

[0095] S8: The monitoring module determines whether the current playback is abnormal and the cause of the abnormality based on the proportion of the lowest priority unplayed task in the playback queue with resource backlog within the period, the difference between the sending time and the corresponding receiving time of the most recent playback task, and the length of the associated path from each broadcast location to the corresponding broadcast date. Based on the cause of the abnormality, the module corrects the queue generation parameters, data transmission parameters, the NER model sliding window parameters, and long text processing parameters.

[0096] S9: The control optimization module corrects the corresponding module parameters based on the instructions issued by the transmission module.

[0097] Furthermore, the monitoring module is used to determine the playback coverage rate based on the proportion of the number of playback tasks that are played at least once in the playback queue within the period to the total number of playback tasks in the playback queue, and to determine whether there is resource backlog in the current playback based on the playback coverage rate, and to determine the minimum priority ratio based on the proportion of the lowest priority unplayed task in the playback queue to the total number of unplayed tasks when resource backlog is determined, and to determine whether there is priority starvation based on the minimum priority ratio, and to correct the forced rotation threshold based on the ratio of the minimum priority ratio to the preset minimum priority ratio when priority starvation is determined, or to determine the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recent playback task, or to determine the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recent playback task when resource backlog is determined.

[0098] The most recently played task refers to the task that was sent most recently from the time node of the periodic collection. The forced rotation threshold refers to the number of playback tasks that are executed consecutively with a higher priority than the number of playback tasks that are forced to be executed when the playback task with the longest current waiting time and the lowest priority in the current playback queue is forcibly executed.

[0099] Playback coverage refers to the percentage of playback tasks that have been played at least once in the playback queue within a unit of time, out of the total number of playback tasks in the queue. Playback coverage directly reflects the operational status of the playback system: high coverage means that most content has been successfully played, while low coverage may indicate a large backlog of tasks, i.e., resource backlog. Since this invention uses periodic data collection, playback tasks may be received around the collection time. To further determine the playback status, this embodiment of the invention determines the minimum priority percentage by the percentage of the lowest-priority unplayed tasks in the playback queue out of the total unplayed tasks. The minimum priority percentage reflects the proportion of the lowest-priority tasks among the backlogged tasks. When this percentage is high, it indicates that a large number of low-priority playback tasks have been received, while high-priority playback tasks are also jumping the queue, leading to a large backlog of low-priority tasks, i.e., priority starvation. Therefore, to avoid delays caused by low-priority tasks not being played for a long time, the forced rotation threshold is adjusted to speed up the playback process and prevent priority starvation. By combining playback coverage with the minimum priority percentage, this invention can intuitively judge the execution status of playback tasks, thereby quickly monitoring, locating, and optimizing playback anomalies.

[0100] Please see Figure 3 As shown, this is a flowchart illustrating how the electronic screen's working status is determined based on playback coverage and the lowest priority percentage in an embodiment of the present invention. The process by which the monitoring module in this embodiment determines the current playback status based on playback coverage and the lowest priority percentage includes:

[0101] The monitoring module determines the playback coverage rate BR based on the proportion of the number of playback tasks that are played at least once in the playback queue within a period to the total number of playback tasks in the playback queue. The playback coverage rate BR is compared with the preset playback coverage rate BR1. Different application scenarios have different requirements for playback coverage rate. For emergency announcement information, the coverage rate per unit time usually needs to reach more than 80%, while for information such as community activities and weather, the requirement can be slightly lowered and set at around 70%. Therefore, in this invention, the preset playback coverage rate BR1∈[70, 80%] is set.

[0102] If the playback coverage rate BR is less than or equal to the preset playback coverage rate BR1, the monitoring module determines resource backlog. The monitoring module determines the minimum priority ratio LPP based on the proportion of the lowest priority unplayed task in the playback queue to the total unplayed tasks. The minimum priority ratio LPP is compared with the preset minimum priority ratio LPP1. Generally, it is considered that when the proportion of the lowest priority unplayed task to the total unplayed tasks reaches 30%, the starvation phenomenon is already relatively obvious, and when it reaches 50% or more, the current starvation phenomenon is considered to be severe. Therefore, the preset minimum priority ratio LPP1 is set to [0.3, 0.5].

[0103] If the lowest priority percentage (LPP) is less than or equal to the preset lowest priority percentage (LPP1), the monitoring module determines the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recently played task.

[0104] If the lowest priority percentage (LPP) is greater than the preset lowest priority percentage (LPP1), the monitoring module determines that a priority starvation phenomenon has occurred, and the monitoring module adjusts the forced rotation threshold based on the ratio of the lowest priority percentage (LPP) to the preset lowest priority percentage (LPP1).

[0105] If the playback coverage rate BR is greater than the preset playback coverage rate BR1, the monitoring module determines the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recent playback task.

[0106] Furthermore, the monitoring module is also used to determine the lowest priority ratio based on the ratio of the lowest priority ratio to the preset lowest priority ratio, and to reduce the forced rotation threshold based on the lowest priority ratio, wherein the reduction of the forced rotation threshold is proportional to the lowest priority ratio.

[0107] The lowest priority ratio refers to the ratio of the preset lowest priority ratio to the lowest priority ratio. The lowest priority ratio reflects the severity of priority hunger in the current playback queue. Dynamically adjusting the forced rotation threshold of the playback queue based on the lowest priority ratio can prevent high-priority playback tasks from continuously preempting low-priority playback tasks, while ensuring the timely broadcast of audio and video content. By making the threshold reduction proportional to the lowest priority ratio, the system can achieve scientific, reasonable, flexible, and dynamic adaptive control, improving the stability and service quality of the community electronic screen broadcasting system.

[0108] Specifically, the process by which the monitoring module reduces the forced rotation threshold based on the lowest priority ratio includes:

[0109] The monitoring module determines the minimum priority ratio A based on the ratio of the preset minimum priority ratio to the minimum priority ratio, and compares the minimum priority ratio A with the set first preset minimum priority ratio A1 and second preset minimum priority ratio A2. The first preset minimum priority ratio A1 is set to (1, 1.7), and the second preset minimum priority ratio A2 is set to [1.7, 3]. In actual playback, when the minimum priority ratio LPP reaches 50%, the current hunger phenomenon is already very serious, and when it reaches 70%, it will directly affect the timeliness of residents obtaining information. Therefore, in this scheme, the first preset minimum priority ratio A1 is set to (1, 1.7), and the second preset minimum priority ratio A2 is set to [1.7, 3].

[0110] If the lowest priority ratio A is less than or equal to the first preset lowest priority ratio A1, then the monitoring module uses the first rotation correction threshold α1 to correct the forced rotation threshold QLT. The corrected forced rotation threshold... The first queue rotation correction threshold α1 is set to 0.98.

[0111] If the lowest priority ratio A is greater than the first preset lowest priority ratio A1 and less than or equal to the second preset lowest priority ratio A2, then the monitoring module uses the second queue rotation correction threshold α2 to correct the forced rotation threshold QLT. The corrected forced rotation threshold... In this case, the second queue rotation correction threshold α2 is set to 0.95;

[0112] If the lowest priority ratio A is greater than the second preset lowest priority ratio A2, then the monitoring module uses the third queue rotation correction threshold α3 to correct the forced rotation threshold QLT. The corrected forced rotation threshold... The third queue rotation correction threshold α3 is set to 0.9.

[0113] Furthermore, the monitoring module is also used to determine the timestamp difference based on the difference between the timestamp sent by the playback task and the timestamp received by the corresponding playback task, and to determine the cause of the abnormality based on the timestamp difference, and to correct the compression ratio of the transmission module based on the ratio of the timestamp difference to the preset timestamp difference when the network is determined to be unstable, or to determine whether there is a playback abnormality based on the path length from the broadcast location to the broadcast date in the original text of the broadcast information.

[0114] The compression ratio refers to the ratio of the compressed data size to the original data size during data transmission. Generally, the compression ratio is set between 2:1 and 10:1 during cloud platform data transmission, so the compression ratio CR∈[10, 50%].

[0115] The timestamp difference is the time interval between the timestamp of the server sending the playback task and the timestamp of the corresponding playback task receiving the task on the electronic screen. It directly reflects the transmission delay of the playback task from the server to the terminal. Under normal network and system operation conditions, the timestamp difference is usually small and stable. If the timestamp difference increases abnormally, it often means that there is a delay in the transmission or processing of the task. Common causes include poor network transmission, bandwidth bottlenecks, or data packet loss and retransmission. As an objective quantitative indicator, the timestamp difference can accurately quantify the task delay and avoid subjective judgment. Using the timestamp difference for judgment not only improves the accuracy of anomaly detection but also enhances the overall performance and reliability of the system. Through automatic adjustment, the system can detect and resolve anomalies in a timely manner, providing users with a smoother and more stable playback experience.

[0116] Specifically, the process by which the monitoring module determines the cause of the anomaly based on the timestamp difference includes:

[0117] The monitoring module determines the timestamp difference TD based on the difference between the timestamp of the playback task being sent and the timestamp of the corresponding playback task being received, and compares the timestamp difference TD with the preset timestamp difference TD1. In cloud platform-based tasks, for real-time interactive tasks, the preset timestamp difference is usually less than the 300ms threshold to ensure the timeliness of transmission. However, for the community electronic screen broadcast information generation system based on the cloud platform, the real-time requirement is not high, so it can be relaxed to 500ms. Therefore, in this invention, the preset timestamp difference TD1 is set to [300, 500ms].

[0118] If the timestamp difference TD is less than or equal to the preset timestamp difference TD1, the monitoring module determines whether there is a playback abnormality based on the path length from the broadcast location to the broadcast date in the original broadcast information.

[0119] If the timestamp difference TD is greater than the preset timestamp difference TD1, the monitoring module determines that the network is unstable, and the monitoring module adjusts the compression ratio of the transmission module based on the ratio of the timestamp difference to the preset timestamp difference.

[0120] Furthermore, the monitoring module is also used to determine a timestamp difference ratio based on the ratio of the timestamp difference to the preset timestamp difference, and to increase the compression ratio of the transmission module based on the timestamp difference ratio, wherein the increase in the compression ratio of the transmission module is proportional to the timestamp difference ratio.

[0121] The timestamp difference ratio refers to the ratio of the timestamp difference to the preset timestamp difference. This ratio quantifies the network transmission latency experienced between the time a task is sent from the server to the community electronic screen. The main function of data compression is to reduce the size of the original data, resulting in smaller data packets to be transmitted. This reduction in data size decreases the amount of data transmitted per unit of time using limited bandwidth, thereby reducing network load and overall transmission latency. This avoids the impact of poor network conditions on transmission. Dynamically adjusting the compression ratio of the transmission module based on the timestamp difference ratio effectively addresses data transmission latency issues caused by network instability. Furthermore, the design, where the increase is proportional to the difference ratio, allows the system to flexibly adapt to different levels of network anomalies, maintaining timely data transmission and smooth playback, thus improving system stability and user experience.

[0122] Specifically, the process by which the monitoring module increases the compression ratio of the transmission module based on the timestamp difference ratio includes:

[0123] The monitoring module determines the timestamp difference ratio B based on the ratio of the timestamp difference to the preset timestamp difference, and compares the timestamp difference ratio B with the first preset timestamp difference ratio B1 and the second preset timestamp difference ratio B2. In the process of transmission based on the cloud platform, when the transmission timestamp difference exceeds 500ms, it is generally considered that the network jitter is serious, and when the timestamp difference exceeds 1000ms, it is generally considered that the network congestion is serious. Therefore, the first preset timestamp difference ratio B1 is set to (1, 2), and the second preset timestamp difference ratio B2 is set to [2, 3.5].

[0124] If the timestamp difference ratio B is less than or equal to the first preset timestamp difference ratio B1, the monitoring module uses the first compression ratio correction threshold β1 to correct the transmission module compression ratio CR. The corrected transmission module compression ratio CR' = CR × β1, where the first compression ratio correction threshold β1 is set to 1.03.

[0125] If the timestamp difference ratio B is greater than the first preset timestamp difference ratio B1 and less than or equal to the second preset timestamp difference ratio B2, then the monitoring module uses the second compression ratio correction threshold β2 to correct the transmission module compression ratio CR. The corrected transmission module compression ratio CR' = CR × β2, where the second compression ratio correction threshold β2 is set to 1.07.

[0126] If the timestamp difference ratio B is greater than the second preset timestamp difference ratio B2, the monitoring module uses the third compression ratio correction threshold β3 to correct the compression ratio CR of the transmission module. The corrected compression ratio CR' of the transmission module is CR×β3, where the third compression ratio correction threshold β3 is set to 1.12.

[0127] Furthermore, the monitoring module is also used to determine the associated path length based on the path length from the broadcast location to the broadcast date in the original broadcast information text, and to determine whether there is an anomaly based on the associated path length, and to adjust the sliding step size of the NER model sliding window based on the ratio of the associated path length to the preset associated path length when the location is determined to be abnormal, or to determine whether there is an anomaly based on the proportion of the number of keywords covered in the broadcast information summary to the total number of keywords covered.

[0128] The associated path length refers to the path length from the broadcast location to the broadcast date in the original text of the broadcast information. It reflects the strength of the reasonable semantic relationship between the broadcast location and broadcast date entities. For pre-trained NER models, when there are multiple broadcast location and broadcast date information in the broadcast information used for identification, there may be cases where the broadcast location information and broadcast date information fail to correspond, resulting in abnormal positioning of the corresponding community electronic screen. Therefore, in this invention, the correlation between the broadcast location and broadcast date in the original text is judged by the associated path length, and the positioning is judged to be abnormal when its associated path length is greater than the preset associated path length. By using the associated path length, this invention can effectively judge the playback abnormality caused by entity extraction error, thereby improving the system's recognition and correction capabilities and ensuring the accuracy and effectiveness of the broadcast information of the community electronic screen.

[0129] The length of the associated path is calculated based on the syntactic relevance.

[0130] Specifically, the process by which the monitoring module determines whether an anomaly is present based on the length of the associated path includes:

[0131] The monitoring module determines the associated path length (LR) based on the path length from the broadcast location to the broadcast date in the original broadcast information. The path length from the broadcast location to the broadcast date is calculated by syntactic dependency degree. The syntactic dependency analysis tool is not limited in principle, but includes, but is not limited to, Stanford NLP, Spacy or HanLP. This will not be elaborated further.

[0132] In this embodiment, the Stanford NLP tool is used to calculate syntactic relevance. The specific steps include:

[0133] Configure the tokenize, ssplit, pos, ner, and depparse modules in the Stanford NLP parameters, and input the preprocessed broadcast information into the Stanford NLP tool for processing. After processing, the dependency syntax tree is output. Traverse the dependency syntax tree and output the path length (LR) between the broadcast location and the broadcast date.

[0134] During the traversal, the process starts from the index of the broadcast date, searches upwards for all parent nodes, and records the path.

[0135] Starting from the index of the broadcast location corresponding to the broadcast date, the parent node path is also recorded;

[0136] Find the Lowest Common Ancestor (LCA) node on both paths;

[0137] Path length LR = Distance from broadcast date to LCA + Distance from broadcast location to LCA;

[0138] The associated path length LR is compared with the preset associated path length LR1. In this invention, syntactic dependency relation is used to measure the correlation between the broadcast location and the corresponding broadcast date. Syntactic dependency relation refers to the strength of the association between two entities (such as address and date) based on the syntactic dependency tree, that is, the number of edges on the connecting path between the broadcast location and the corresponding broadcast date in the syntactic dependency tree. Generally, it is considered that if the path length between two entities is less than or equal to 2, they are considered to be strongly correlated. When the path length between two entities is 3, they are still considered to belong to the same semantic unit, and the relationship is reasonable and reliable. Therefore, in this invention, the preset associated path length LR1 is set to [1, 3].

[0139] If the associated path length LR is less than or equal to the preset associated path length LR1, the monitoring module determines whether it is abnormal based on the proportion of the number of keywords covered in the article broadcast information summary to the total number of keywords covered.

[0140] If the associated path length LR is greater than the preset associated path length LR1, the monitoring module determines that the positioning is abnormal, and the monitoring module adjusts the sliding step size of the NER sliding window based on the ratio of the associated path length to the preset associated path length.

[0141] Furthermore, the monitoring module is also used to determine the associated path length ratio based on the ratio of the associated path length to the preset associated path length, and to increase the sliding step size of the NER model sliding window based on the associated path length ratio, wherein the increase in the sliding step size is proportional to the associated path length ratio.

[0142] When performing NER on text, a sliding window mechanism is typically used to process text segments sequentially. The sliding step size determines the distance the window slides each time, affecting the overlap of the same text region being covered and recognized. When the system detects a location anomaly, it indicates that the extracted location is not strongly associated with the date entity, resulting in low recognition accuracy and unreasonable entity segmentation. In this case, increasing the sliding step size increases the window sliding distance, reducing the risk of overly fine entity segmentation or ambiguity. The association path length ratio is the ratio of the association path length to the preset association path length. The larger the ratio, the greater the gap between the current recognition effect and the expectation, and the higher the degree of anomaly. Therefore, increasing the sliding step size of the NER model's sliding window based on the association path length ratio, with the increase in the sliding step size being proportional to the association path length ratio, helps to make the adjustment flexible and intelligent, prevents over- or under-adjustment, and ensures the stable and efficient operation of the pre-trained NER model. This anomaly adaptive adjustment mechanism can effectively solve the recognition anomaly caused by low association path length, improve entity recognition accuracy and broadcast information quality, enhance the community electronic screen broadcasting system's ability to identify and handle anomalies, and improve the system's intelligence and robustness.

[0143] Please see Figure 4 The diagram shows a flowchart of increasing the sliding step size of the NER model sliding window based on the ratio of associated path lengths in an embodiment of the present invention. The process of increasing the sliding step size of the NER model sliding window based on the ratio of associated path lengths in an embodiment of the present invention includes:

[0144] The monitoring module determines the association path length ratio C based on the ratio of the association path length to the preset association path length, and compares the association path length ratio C with the first preset association path length ratio C1 and the second preset association path length ratio C2. In syntactic dependency relations, when the path length between two entities is 4-5, the path is relatively long and the syntactic association gradually weakens. When it is greater than 5, the path is considered too long, and the entities may not belong to the same event or component. Therefore, in this invention, the first preset association path length ratio C1 is set to [1.2, 1.7] and the second preset association path length ratio C2 is set to [1.7, 4].

[0145] If the ratio of the associated path lengths C is less than or equal to the first preset ratio of the associated path lengths C1, then the monitoring module uses a first sliding step size correction threshold λ1 to correct the sliding step size ST of the NER model sliding window. Wherein, the first sliding step size correction threshold λ1 is set to 1.02;

[0146] If the ratio of the associated path lengths C is greater than the first preset ratio of the associated path lengths C1 and less than or equal to the second preset ratio of the associated path lengths C2, then the monitoring module uses the second sliding step size correction threshold λ2 to correct the sliding step size ST of the NER model sliding window. Wherein, the second sliding step size correction threshold λ2 is set to 1.05;

[0147] If the ratio of the associated path lengths C is greater than the second preset ratio of the associated path lengths C2, then the monitoring module uses a third sliding step size correction threshold λ3 to correct the sliding step size ST of the NER model sliding window. The third sliding step size correction threshold λ3 is set to 1.1.

[0148] Furthermore, the monitoring module is also used to determine the sliding step size difference based on the difference in sliding step size of the sliding window of the NER model before and after correction, and to increase the sliding window size of the NER model sliding window based on the sliding step size difference, wherein the increase in the sliding window size is proportional to the sliding step size difference.

[0149] The sliding step size determines the number of text units the window slides in each iteration. A larger step size results in faster processing speed but may miss contextual dependencies. The sliding window size determines the length of text covered by each window; a larger size contains more contextual information, helping to capture long-distance dependencies. Increasing the sliding step size widens the window spacing and reduces text overlap, potentially leading to contextual discontinuities or missing information. Therefore, increasing the sliding window size compensates for this, ensuring each window contains more context and maintaining recognition accuracy, especially for long texts or cross-sentence entities, thus improving NER accuracy. The sliding step size difference reflects the magnitude of parameter adjustment; a larger difference increases the risk of missing information. Therefore, the increase in window size is proportional to the step size difference, allowing for dynamic adjustment of the window size based on step size changes. This ensures the compensation level matches the risk level, preventing over-compensation or under-compensation, thereby improving the system's adaptability and stability. This design improves NER recognition accuracy, avoids recognition gaps or entity splitting caused by increased step size, adapts to different abnormal situations through a dynamic adjustment mechanism, achieves parameter self-correction and optimization, and improves localization accuracy.

[0150] Specifically, the process by which the monitoring module increases the sliding window size of the NER model sliding window based on the sliding step size difference includes:

[0151] The monitoring module is also used to determine the sliding step size difference D of the sliding window of the NER model before and after correction, and compare the sliding step size difference D with the set first preset sliding step size difference D1 and second preset sliding step size difference D2. In order to compensate for the reduction of overlap caused by the increase of step size, the window size usually needs to be increased accordingly to maintain a certain overlap ratio or length. Since the sliding step size in the pre-trained NER model is usually set to 30%-70% of the window size, the difference after correction is between 0.6%-7%. In order to ensure the effect before and after correction to a greater extent, the first preset sliding step size difference D1∈[0.5, 4%) and the second preset sliding step size difference D2∈[5, 7%] are set.

[0152] If the sliding step size difference D is less than or equal to the first preset sliding step size difference D1, then the monitoring module uses the first sliding window correction threshold η1 to correct the sliding window size SWS of the NER model sliding window. Wherein, the first sliding window correction threshold η1 is set to 1.02;

[0153] If the sliding step difference D is greater than the first preset sliding step difference D1 and less than or equal to the second preset sliding step difference D2, then the monitoring module uses the second sliding window correction threshold η2 to correct the sliding window size SWS of the NER model sliding window. The second sliding window correction threshold η2 is set to 1.06.

[0154] If the sliding step size difference D is greater than the second preset sliding step size difference D2, then the monitoring module uses the third sliding window correction threshold η3 to correct the sliding window size SWS of the NER model sliding window. The third sliding window correction threshold η3 is set to 1.11.

[0155] Furthermore, the monitoring module is also used to determine the keyword coverage rate based on the proportion of the number of keywords covered in the article broadcast information summary to the total number of keywords covered, and to determine whether there is an anomaly based on the keyword coverage rate. When it is determined that the electronic screen is working normally, the system parameters are maintained and the detection continues. Alternatively, when it is determined that the broadcast information summary is deviating, the keyword coverage rate difference is determined based on the difference between the keyword coverage rate and the preset coverage rate. The length penalty in the Transformer algorithm is increased based on the keyword coverage rate difference, and the increase in the length penalty is proportional to the keyword coverage rate difference.

[0156] Keyword coverage is the ratio of the number of keywords covered in the summary to the total number of keywords in the original broadcast information. It directly reflects the degree of matching between the summary and the key information in the original text. A low coverage rate indicates that the summary omits key content, resulting in bias and potentially information omissions or distortions. Therefore, keyword coverage can be used to determine if a summary is abnormal. When judging summary bias, the length penalty is increased based on the difference in keyword coverage (i.e., the difference between the actual coverage and the preset ideal coverage). This is because the larger the difference, the more severe the summary bias, requiring a more significant adjustment of the length penalty to generate a longer and more comprehensive summary, thereby improving keyword coverage. The increase in length penalty is proportional to the difference in keyword coverage. This adaptive adjustment mechanism can dynamically adjust according to the summary quality, avoiding over- or under-adjustment, ensuring system stability, and achieving refined and intelligent summary optimization. This design, combining key content coverage assessment with adaptive generation control, effectively improves the quality and completeness of the community electronic screen broadcast information summaries, enhances the system's ability to perceive and handle broadcast information quality anomalies, and improves the accuracy of the broadcast content and user experience.

[0157] Specifically, the process by which the monitoring module determines whether something is abnormal based on keyword coverage includes:

[0158] There are no restrictions on the methods for obtaining keywords from the abstracted article. Technical personnel can obtain the corresponding keywords from the original text based on any one of TF-IDF, TextRank, or RAKE.

[0159] The monitoring module determines the keyword coverage rate KC based on the proportion of keyword coverage in the article broadcast information summary to the total number of keyword coverages. It then compares the keyword coverage rate KC with the preset keyword coverage rate KC. When summarizing the article broadcast information using the Transformer algorithm, if the coverage is less than 70%, it is considered that the summary has failed to effectively extract the core content. For articles such as government announcements, stricter requirements are required. Therefore, the preset keyword coverage rate KC is set to [70, 80%].

[0160] If the keyword coverage rate KC is less than or equal to the preset keyword coverage rate KC, the monitoring module determines that the broadcast information summary is biased, and the monitoring module corrects the length penalty in the Transformer algorithm based on the difference between the keyword coverage rate and the preset coverage rate.

[0161] If the keyword coverage rate KC is greater than the preset keyword coverage rate KC, the monitoring module determines that the community electronic screen is working normally, maintains the system parameters, and continues to monitor.

[0162] Specifically, the process by which the monitoring module corrects the length penalty in the Transformer algorithm based on the difference between the keyword coverage rate and the preset coverage rate includes:

[0163] The monitoring module determines the keyword coverage difference E based on the difference between the keyword coverage rate and the preset coverage rate, and compares the first preset keyword coverage difference E1 and the second preset keyword coverage difference E2. When the coverage of the abstract to the key information (such as entities, events, and conclusions) of the original text is less than 60%, it is generally considered that the abstract has a large deviation from the original text. When it is less than 50%, it is considered that the abstract is unsuccessful. Therefore, the first preset keyword coverage difference E1 is set to [10, 20] and the second preset keyword coverage difference E2 is set to [20, 30].

[0164] If the keyword coverage difference E is less than or equal to the first preset keyword coverage difference E1, the monitoring module uses the first length penalty correction threshold θ1 to correct the length penalty LP in the Transformer algorithm. The corrected LP' = LP × θ1, where the first length penalty correction threshold θ1 is set to 1.02.

[0165] If the keyword coverage difference E is greater than the first preset keyword coverage difference E1 and less than or equal to the second preset keyword coverage difference E2, then the monitoring module uses the second length penalty correction threshold θ2 to correct the length penalty LP in the Transformer algorithm. The corrected LP' = LP × θ2, where the second length penalty correction threshold θ2 is set to 1.05.

[0166] If the keyword coverage difference E is greater than the second preset keyword coverage difference E2, the monitoring module uses the third length penalty correction threshold θ3 to correct the length penalty LP in the Transformer algorithm. The corrected LP' = LP × θ3, where the third length penalty correction threshold θ3 is set to 1.11.

[0167] In summary, this invention proposes a cloud-based community electronic screen broadcast information generation system. Through intelligent module collaboration, it achieves comprehensive management of broadcast information generation, monitoring, and optimization. The system possesses real-time anomaly monitoring and adaptive adjustment capabilities, automatically identifying anomalies in playback, positioning, and summary generation stages, and optimizing them by dynamically adjusting relevant parameters. This effectively solves problems such as playback starvation and network instability, significantly improving the quality and completeness of broadcast information. Simultaneously, the system, through a self-learning algorithm and a content generation model trained on historical data, can generate personalized broadcast content to meet the diverse needs of community electronic screens. Furthermore, the system optimizes resource utilization and transmission efficiency, reduces manual intervention, and enhances system stability and robustness, providing efficient and intelligent technical support for the construction of smart communities.

[0168] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A cloud platform-based community electronic screen broadcast information generation system, characterized in that, include, The data acquisition module periodically collects broadcast information and electronic screen data based on the cloud platform; A preprocessing module, which is connected to the acquisition module, is used to preprocess the acquired data and divide the playback priority based on the type of broadcast information; An article processing module, which is connected to the preprocessing module, is used to generate article-type broadcast information summaries based on the Transformer algorithm; The positioning module, which is connected to the preprocessing module, is used to obtain the broadcast location and broadcast date from the broadcast information based on the pre-trained NER model and locate the corresponding electronic screen. A content generation module, which is connected to the preprocessing module, the article processing module and the positioning module, is used to determine the broadcast format of the broadcast information based on the content generation model; A queue module, which is connected to the preprocessing module, is used to determine the playback queue based on the forced rotation threshold and the playback priority; A transmission module, which is connected to the content generation module and the positioning module, is used to compress the playback content into a playback task and transmit it to the corresponding electronic screen; The monitoring module, which is connected to the preprocessing module, is used to determine whether the current playback is abnormal and the cause of the abnormality based on the proportion of the lowest priority unplayed task in the playback queue with resource backlog within the period, the difference between the sending time and the corresponding receiving time of the most recent playback task, and the length of the associated path from each broadcast location to the corresponding broadcast date. Based on the cause of the abnormality, the queue generation parameters, data transmission parameters, the sliding window parameters of the NER model, and the long text processing parameters are corrected. The control optimization module, which is connected to the monitoring module, the transmission module, the article processing module, the positioning module, and the queue module, is used to correct the parameters of the corresponding modules based on the instructions issued by the transmission module.

2. The system for generating a community electronic screen broadcasting message according to claim 1, wherein The content generation module is used to construct a content generation model based on a self-learning algorithm, train the content generation model based on the relationship between historical electronic screen data and broadcast information format, and determine the corresponding electronic screen playback format based on the electronic screen data and the broadcast information after preprocessing and article processing.

3. The system for generating a community electronic screen broadcasting message according to claim 2, wherein The monitoring module is used to determine the playback coverage rate based on the proportion of the number of playback tasks that are played at least once in the playback queue within a period to the total number of playback tasks in the playback queue; and to determine whether there is resource backlog in the current playback based on the playback coverage rate. When resource backlog is determined, the module determines the minimum priority ratio based on the proportion of the lowest priority unplayed task in the playback queue to the total number of unplayed tasks. The module also determines whether priority starvation occurs based on the minimum priority ratio. When priority starvation occurs, the module corrects the forced rotation threshold based on the ratio of the minimum priority ratio to the preset minimum priority ratio. Alternatively, the module determines the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recently played task. Or, when no resource backlog is determined, the module determines the cause of the abnormality based on the difference between the sending time and the corresponding receiving time of the most recently played task. The most recently played task refers to the task that was sent most recently from the time node of the periodic collection. The forced rotation threshold refers to the number of playback tasks that are executed consecutively with a higher priority than the number of playback tasks that are forced to be executed when the playback task with the longest current waiting time and the lowest priority in the current playback queue is forcibly executed.

4. The community electronic screen broadcast information generation system according to claim 3, characterized in that, The monitoring module is also used to determine the lowest priority ratio based on the ratio of the lowest priority ratio to the preset lowest priority ratio, and to reduce the forced rotation threshold based on the lowest priority ratio, wherein the reduction of the forced rotation threshold is proportional to the lowest priority ratio.

5. The system for generating a community electronic screen broadcasting message according to claim 3, wherein The monitoring module is also used to determine the timestamp difference based on the difference between the timestamp sent by the playback task and the timestamp received by the corresponding playback task, and to determine the cause of the abnormality based on the timestamp difference, and to correct the compression ratio of the transmission module based on the ratio of the timestamp difference to the preset timestamp difference when the network is determined to be unstable, or to determine whether there is a playback abnormality based on the path length from the broadcast location to the broadcast date in the original text of the broadcast information. The compression ratio refers to the ratio of the compressed data size to the original data size after data compression during data transmission.

6. The system for generating a community electronic screen broadcasting message according to claim 5, wherein The monitoring module is also used to determine a timestamp difference ratio based on the ratio of the timestamp difference to the preset timestamp difference, and to increase the compression ratio of the transmission module based on the timestamp difference ratio, wherein the increase in the compression ratio of the transmission module is proportional to the timestamp difference ratio.

7. The system for generating a community electronic screen broadcasting message according to claim 5, wherein The monitoring module is also used to determine the associated path length based on the path length from the broadcast location to the broadcast date in the original broadcast information, and to determine whether there is an anomaly based on the associated path length, and to adjust the sliding step size of the NER model sliding window based on the ratio of the associated path length to the preset associated path length when the location is determined to be abnormal, or to determine whether there is an anomaly based on the proportion of the number of keywords covered in the broadcast information summary to the total number of keywords covered.

8. The system for generating a community electronic screen broadcasting message according to claim 7, wherein The monitoring module is also used to determine the associated path length ratio based on the ratio of the associated path length to the preset associated path length, and to increase the sliding step size of the NER model sliding window based on the associated path length ratio, wherein the increase in the sliding step size is proportional to the associated path length ratio.

9. The system for generating a community electronic screen broadcast message according to claim 8, wherein, The monitoring module is also used to determine the sliding step size difference based on the difference in sliding step size of the sliding window of the NER model before and after correction, and to increase the sliding window size of the NER model sliding window based on the sliding step size difference, wherein the increase in the sliding window size is proportional to the sliding step size difference.

10. The community electronic screen broadcast information generation system according to claim 7, characterized in that, The monitoring module is also used to determine the keyword coverage rate based on the proportion of the number of keywords covered in the article broadcast information summary to the total number of keywords covered, and to determine whether there is an anomaly based on the keyword coverage rate. When it is determined that the electronic screen is working normally, the system parameters are maintained and the monitoring continues. Alternatively, when it is determined that the broadcast information summary is deviating, the keyword coverage rate difference is determined based on the difference between the keyword coverage rate and the preset coverage rate. The length penalty in the Transformer algorithm is increased based on the keyword coverage rate difference, and the increase in the length penalty is proportional to the keyword coverage rate difference.

Images