A system and method for improving multi-tenant 5G message push efficiency

Through intelligent scheduling and AI-driven Chatbot resource allocation, situation-aware permission management and content audit, the problems of uneven resource allocation, low efficiency of permission management and content review in multi-tenant 5G environment are solved, and efficient message push and personalization are achieved Serve.

CN118714522BActive Publication Date: 2024-11-22JIANGXI PALM CENT UNLIMITED TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411178888.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-22
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

In a multi-tenant 5G environment, the existing technology lacks an intelligent network resource scheduling mechanism, resulting in uneven resource allocation, affecting the timeliness of message push and user satisfaction, and the efficiency of permission management and content review is low, making it difficult to deal with complex users demand and security risks.

Method used

Through intelligent scheduling and Chatbot cluster management module dynamically allocate resources, combine AI technology and situational awareness technology for permission management and content auditing, use 5G network slicing and edge computing to optimize resource utilization and review efficiency, and provide personalized multimedia material recommendations.

Benefits of technology

It significantly improves the efficiency and user experience of message push, ensures data transmission security and content quality, and enhances marketing effectiveness and operational efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118714522B_ABST
    Figure CN118714522B_ABST
Patent Text Reader

Abstract

The present invention provides a system and method for improving the efficiency of multi-tenant 5G message push, intelligently schedules Chatbot resources, dynamically allocates the best service resources according to user location and network conditions, and uses 5G network slicing technology to achieve flexible resource management. Through context-aware fine-grained permission management, user permissions are dynamically adjusted according to temporal and spatial changes. The content review module uses AI technology for real-time review and dynamically adjusts the review strictness in combination with the user reputation model. Multimedia material management incorporates user preferences and location information to provide personalized high-quality material recommendations. The operation analysis module uses big data and machine learning to optimize message strategies, thereby solving the problems of low push efficiency, inflexible resource allocation, insufficient security, low content review efficiency, shallow personalized services, and lack of operation analysis in existing 5G message services.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of 5G communication technologies, and in particular to a system and method for improving the efficiency of multi-tenant 5G message push. Background Art

[0002] Although 5G technology promises higher data rates, lower latency and greater connection density, in actual deployment, the resource management and allocation efficiency in a multi-tenant environment has not yet reached the optimal level. Existing technologies often lack intelligent network resource scheduling mechanisms, especially for dynamically changing user needs and network conditions such as changes in signal strength caused by geographic location, network congestion periods and operator service quality. This leads to uneven resource allocation, affecting the timeliness of message push and user satisfaction.

[0003] Traditional permission control systems often rely on static role and rule configurations, and it is difficult to adjust permission levels in real time based on user location changes and time factors. This may result in some users being unable to access necessary resources at a specific time or place, or overly loose permissions leading to increased security risks. For a multi-tenant environment, there may be complex interactions between different tenants, but the existing permission management system does not take into account the impact of these interactions on permission allocation.

[0004] Although 5G messages have the ability to provide highly personalized content and services, existing technologies are insufficient in utilizing user data for in-depth analysis and precise recommendations. There is a lack of effective user behavior analysis, preference learning, and application of contextual awareness technologies, which results in a gap between pushed content and user needs, reducing the level of personalization of the user experience.

[0005] The widespread use of 5G messaging services has increased the requirements for the immediacy and accuracy of content review. The current review mechanism often relies on traditional keyword filtering or simple manual review, which makes it difficult to effectively cope with complex contextual understanding, image recognition, and in-depth analysis of audio and video content. This leads to low review efficiency and is prone to misjudgments or omissions. Especially when processing large-scale data streams, the lack of intelligent review tools has become a major bottleneck. Summary of the invention

[0006] The purpose of the present invention is to provide a system and method for improving the efficiency of multi-tenant 5G message push.

[0007] The problem to be solved by the present invention is: by providing a system and method for improving the efficiency of multi-tenant 5G message push, Chatbot resources are intelligently allocated according to the geographical locations and network conditions of different users, 5G is used to dynamically optimize high bandwidth requirements, encryption and fine-grained permission management are used to dynamically adjust access rights in combination with real-time location information, AI technology is integrated to automatically identify sensitive content, and review strategies are dynamically adjusted in combination with reputation models. Highly relevant multimedia materials are pushed based on user preferences and context-aware technology, resource allocation and billing are automatically completed using smart contracts, and big data and AI technology are used to provide in-depth business insights and guide strategy optimization.

[0008] A system for improving multi-tenant 5G message push efficiency, comprising:

[0009] Intelligent Scheduling and Chatbot Cluster Management Module: The API interface abstracts the differences between different operators' Chatbots to achieve unified management and scheduling. It uses intelligent routing algorithms to dynamically allocate optimal Chatbot resources based on user geographic location, network conditions including latency, packet loss rate, and operator service quality. It has a built-in load balancing strategy and an integrated 5G network slice intelligent selector to dynamically allocate the most appropriate 5G network slice resources based on Chatbot task type and user location. It introduces an encrypted authentication mechanism, uses an AI-driven load prediction model, and combines edge computing to implement a preloading strategy for high-frequency conversation scenarios.

[0010] Multi-tenant SaaS integration and resource optimization module: supports tenants to customize message queues, security policies, and message transmission protocols to meet the needs of different business scenarios. It uses distributed caching technology with the ARC algorithm to dynamically adjust memory allocation, adopts a multi-level isolation storage solution, and combines the key management system KMS to implement end-to-end encryption of tenant data. In the multi-tenant architecture, tenants customize 5G network resource strategies, dynamically adjust slice resources according to business needs, introduce 5G dynamic scheduling algorithms, and automatically switch frequency bands for high-bandwidth demand scenarios;

[0011] Context-aware fine-grained tenant rights management module: Build a role-based access control RBAC model, combined with ABAC-based access control policies, and control tenants' viewing, editing, and deletion rights through ACL. At the same time, it integrates SSL / TLS protocols and data transmission encryption technology, combines 5G positioning and context-aware technology, dynamically adjusts user rights levels, and uses the isolation characteristics of 5G network slices, combined with key distribution technology, to provide dedicated slices for tenant data transmission at different security levels;

[0012] Intelligent content review and supervision module: It integrates natural language processing (NLP), computer vision (CV) and machine learning (ML) technologies to conduct real-time intelligent review of text, image, audio and video content, build a multi-dimensional content scoring system, and set a scoring threshold. When the content score exceeds this threshold, the module will mark it as suspected illegal content. For suspected illegal content, the manual review process will be initiated, and the tenant reputation model will be introduced to dynamically adjust the review strictness based on historical behavior. In addition, a violation warning and penalty mechanism will be implemented. In combination with edge computing resources, the review model will be deployed on the edge nodes of the network to reduce review delays and isolate review tasks using 5G network slicing;

[0013] Multimedia material management and personalized service module: upload, classify, and label various materials through API and user interface, automatically label materials using content recognition technology, support material library permission management, combine user preference analysis, and combine 5G's high bandwidth and positioning capabilities to provide personalized material recommendations for different tenants and users based on user location and time. By switching 5G frequency bands, fast upload and distribution channels are provided for high-quality materials. Context-aware labeling algorithms are introduced to automatically label materials based on content and user context.

[0014] Operation analysis and statistical optimization module: Build a real-time data stream processing platform based on 5G network, use big data analysis, machine learning algorithms and real-time stream processing technology to monitor the system operation status in all directions, collect and analyze user interaction data, message sending trends, and key information of system performance indicators. Utilize 5G-NR-U technology to provide the operation team with in-depth business insights through customized reports and visual dashboards, assist in decision making, and combine A / B testing and user behavior analysis to optimize message strategies.

[0015] A method for improving the efficiency of multi-tenant 5G message push is based on the above-mentioned system for improving the efficiency of multi-tenant 5G message push, and the technical solution adopted is as follows:

[0016] S1: The multi-tenant SaaS integration and resource optimization module is started to allocate initial resources to each tenant, including custom message queues, security policy settings, and message transmission protocol selection. The 5G network slice resources are dynamically adjusted according to business needs and the 5G dynamic scheduling algorithm.

[0017] S2: Initialize the context-aware fine-grained tenant rights management module, set basic tenant rights based on the RBAC and ABAC models, dynamically adjust the level of rights based on geographic location and time through 5G positioning technology, and enable key distribution technology to ensure data transmission security;

[0018] S3: Users access the system through the 5G network. The intelligent scheduling and Chatbot cluster management module combines the user's geographic location, network conditions including latency, packet loss rate, and operator service quality, and uses intelligent routing algorithms and 5G network slice intelligent selectors to direct user requests to the optimal Chatbot resources.

[0019] S4: Edge computing uses AI-driven load prediction models to implement pre-loading strategies for high-frequency conversation scenarios;

[0020] S5: Chatbot processes tenant requests and generates response content, including text, images, audio and video. The intelligent content review and supervision module intervenes and combines NLP, CV and ML technologies to conduct real-time intelligent review of the generated content. The review tasks are isolated through 5G network slicing. Suspected illegal content triggers the manual review process and dynamically adjusts the review strictness based on the tenant reputation model.

[0021] S6: The multimedia material management and personalized service module uploads materials based on the response content, and automatically tags them using the context-aware tagging algorithm, and recommends personalized materials based on the user's location and preferences;

[0022] S7: Distributed cache and multi-level isolated storage technology for multi-tenant SaaS integration and resource optimization modules. Approved content is pushed to users through intelligent scheduling and Chatbot cluster management modules combined with edge computing.

[0023] S8: The operation analysis and statistical optimization module continuously collects user interaction data, message sending trends and system performance indicators, uses 5G-NR-U technology to expand the scope of data collection, and uses big data analysis and real-time stream processing technology to generate customized reports and visual dashboards to assist the operation team in making decisions and optimize message strategies through A / B testing and user behavior analysis.

[0024] Furthermore, the S1 dynamically adjusts 5G network slice resources according to business needs in combination with the 5G dynamic scheduling algorithm, including:

[0025] S11: The tenant analyzes the business model, user behavior, and message types including instant messaging and rich media messaging, and specifies the required network resource types including bandwidth and latency requirements. Based on the demand analysis results, the tenant customizes the message queue, security policy, and message transmission protocol through the interactive interface provided by the multi-tenant SaaS platform, and submits the resource requirements.

[0026] S12: The resource requirements submitted by the tenants automatically trigger the intelligent matching algorithm. The algorithm analyzes the tenants’ business characteristics including geographic location, message volume, peak traffic, and network conditions including current slice load and historical performance, and recommends the matching 5G network slices.

[0027] The smart matching algorithm function is , , , , , where B is the bandwidth requirement satisfaction, AB is the available bandwidth, RB is the required bandwidth, and L is the delay sensitivity. is the positive adjustment factor, EL is the expected delay, AL is the actual delay, S is the slice stability, HSS is the maximum stability score, MSS is the historical stability score, BST is the comprehensive service characteristic index, γG is the adjustment factor, DNN is the geographical location, PPT is the predicted peak traffic, PTC is the peak traffic capacity, PMV is the predicted message volume, HMV is the historical message volume, is the weight factor;

[0028] S13: For business scenarios that require high bandwidth for the transmission of a large number of rich media messages, the system automatically detects and triggers the 5G dynamic scheduling algorithm, and automatically switches resources based on network load and availability.

[0029] Furthermore, in S2, the permission level is dynamically adjusted according to the geographic location and time through 5G positioning technology, and the key distribution technology is enabled to ensure the security of data transmission, including:

[0030] S21: Integrate high-precision positioning services through 5G network interfaces, combine the collaborative positioning capabilities of 5G base stations and terminals, and obtain location information of tenants and users in real time;

[0031] S22: The system is synchronized with the standard time server to monitor and record the timestamps of tenant interactions in real time, and combine them with geographic location information to form time-space coordinate data points;

[0032] S23: Dynamically calculate the risk score based on the combination of the tenant's geographical location and time, and dynamically adjust the permission level based on the risk score. The risk score formula is: ,in Due to geographical location factors, is the time period factor, is the sigmoid function, φ and θ are adjustment parameters, and is the weight factor, satisfying ;

[0033] Collect and analyze tenants' behavior data in different geographical locations, which are tenant location data and message traffic data. Obtain tenant location data through the 5G network, record the number of messages sent and received by tenants in each geographical location as message traffic data, record the frequency of tenants visiting a specific geographical location as tenant location frequency L, and record the number of messages sent and received by tenants in a specific geographical location as message traffic T. ,in It is the tenant location frequency weight and message traffic weight;

[0034] The time periods of a day are classified into working time, non-working time, and late night time. Working time is from 9 am to 5 pm, late night time is from 11 pm to 6 am, and the rest of the time is non-working time. The weighted average method is used to first combine the working time, non-working time, and late night time of the four seasons of spring, summer, autumn, and winter, and holidays to calculate, and then the calculation results are combined with the working time, non-working time, and late night time of special events to calculate using the weighted average method ,Special events are defined as events where the flow of people per minute in the geographical location of the tenant is greater than 5,000. The geographical location of the tenant is a circular area with a radius of 300m with the tenant as the center;

[0035] The permission levels are divided into three levels: low, medium and high. The low permission level allows access to public information and basic operations, the medium permission level allows access to some restricted information and sensitive operations, and the high permission level allows access to highly sensitive information and high-risk operations. For low authority level, risk score Medium authority level, risk score It is a high authority level;

[0036] S24: a fixed geographical radius of 300m is set for each tenant's current geographical location, and the activities of the tenants in this area are recorded to construct an activity range profile of the corresponding tenant;

[0037] When the activity ranges of multiple tenants overlap, that is, their activity radii intersect, the system evaluates the impact of the intersection on permissions. A tenant whose activity range overlaps more extensively with other tenants is considered to be a more influential tenant. For example, if tenant A often appears in the activity range of tenant B, while tenant B rarely appears in the activity range of A, then tenant B will be considered to be more influenced by tenant A, and tenant B's permissions will be reduced.

[0038] Calculate the influence coefficient based on the degree of overlap between tenants’ activities , This indicates that A's influence on B is greater than B's influence on A, and B's authority will be downgraded. , , ,in and is the factor that tenants A and B are in a specific geographical location, is the intersection area of ​​the activity range of tenant A and tenant B, It is the one with the smaller activity area between Tenant A and Tenant B;

[0039] The risk scoring formula is updated to ,in is the weight factor of the influence coefficient;

[0040] S25: After each permission adjustment, key distribution technology is used to generate new encryption keys for sensitive data transmission between tenants, and the validity period and scope of use of the keys are dynamically adjusted according to the permission level and data sensitivity.

[0041] Furthermore, the activities of the tenants in this area are recorded in S24 to construct an activity range profile of the corresponding tenants, including:

[0042] S241: Record the timestamp of each location data of the tenant accessing the system through 5G messages;

[0043] S242: Create a database of interest points tags containing different location categories, where the location category tags include catering, entertainment, culture and education, retail and shopping, nature and outdoor, accommodation, health and fitness, business and office, transportation, and community services, and the interest point tags include food, coffee, fast food, movies, performances, concerts, exhibitions, reading, learning, shopping, clothing, electronic products, walking, running, accommodation, vacation, exercise, yoga, conference room, printing service, transportation hub, and medical service;

[0044] S243: Matching the current location data and activity data of the tenant with the point of interest tag database to determine the specific location type and activity type of the tenant;

[0045] S244: Aggregate the activity data of the tenants and create a profile for each tenant, including the locations most frequently visited by the tenant and the time period during the day when the tenant at the corresponding location is most active.

[0046] Furthermore, the S3 uses an intelligent routing algorithm and a 5G network slice intelligent selector to direct user requests to the optimal Chatbot resource, including:

[0047] S31: The intelligent routing algorithm builds a comprehensive scoring model based on the network status, geographic location, and the current load of the Chatbot cluster, and assigns weights to each dimension;

[0048] The network status includes latency, packet loss rate, and bandwidth. The geographic location includes distance and location hotspot, i.e., whether it is a high-traffic area. The traffic threshold f is set. If the user is in a high-traffic area, the traffic is greater than f. If the user is in a low-traffic area, the traffic is less than f. The current load of the Chatbot cluster includes CPU usage, memory usage, and task queue length.

[0049] S32: The comprehensive scoring model is total score = ∑(single score × weight), and the weights of network status, geographic location, and current load of the Chatbot cluster are 0.45, 0.3, and 0.25 respectively;

[0050] The latency score in the network condition is , the packet loss rate score is , bandwidth score is ,The total score of network status is the sum of the delay, packet loss rate and bandwidth scores;

[0051] Location score = ,in and is the weight of distance and location hotspot, distance score = ,in is the actual distance between the Chatbot server and the user, The distance covered by the Chatbot server. If the user is in a high-traffic area, the location hotspot score is 0; if the user is in a low-traffic area, the location hotspot score is 1;

[0052] The current load score of the Chatbot cluster = ,in The weights corresponding to CPU usage, memory usage, and task queue length, The maximum acceptable CPU usage threshold, the maximum acceptable memory usage threshold, and the maximum acceptable queue length threshold;

[0053] S33: Combined with the constructed scoring model, all available Chatbot resources are matched and scored. The Chatbot resource with the highest score is the optimal Chatbot resource. For users in high-traffic areas, the slices are automatically switched to direct user requests to the optimal Chatbot resource.

[0054] Furthermore, the edge computing in S4 implements a preloading strategy for high-frequency conversation scenarios based on an AI-driven load prediction model, including:

[0055] S41: Through the 5G network, edge computing nodes collect user interaction data in real time, including message frequency, conversation content category, user active time period, and geographic location distribution;

[0056] S42: Use feature selection algorithms to extract features from raw data that are valuable for load forecasting, including conversation peaks, high-demand periods in specific geographic locations, and user preference patterns;

[0057] S43: Build AI models, continuously learn on edge computing nodes, and use online learning technology to adjust model parameters in real time based on new data;

[0058] S44: Integrates scene recognition algorithms to identify high-frequency conversation scenes and dynamically adjusts the resource preloading priority based on the identified scene types and real-time predicted load conditions;

[0059] S45: Based on AI prediction results, design an adaptive preloading strategy to preload the corresponding Chatbot model, conversation database, and system components on the edge server during the predicted high-load period.

[0060] Furthermore, in S6, personalized materials are recommended based on the user's location and preferences, including:

[0061] S61: Perform content analysis on multimedia materials, extract visual features and audio features, combine with geographic information system data, generate location features, including distance from specific locations and surrounding environment types, and establish user preference models based on historical browsing, likes, and sharing interaction data;

[0062] S62: combining the material features, the user location features, and the current time and weather external context factors, automatically adding context labels to the material through a label generation algorithm;

[0063] S63: Using the tags generated by the context-aware tagging algorithm, combined with the user's current location and preferences, a context-aware recommendation model is constructed. The model is ,in is the recommendation score of material i for a specific user, is the user feature vector, which contains the user's historical behavior and preference information. is the vector representation of the j-th context label, n is the total number of context labels possessed by material i, W is the weight vector, and b is the constant bias term.

[0064] The beneficial effects of the present invention are as follows: through intelligent scheduling and Chatbot cluster management, combined with real-time information such as user geographic location and network status, the optimal resources are dynamically allocated, the delay of message push is significantly reduced, and the user experience is improved. At the same time, edge computing and AI-driven preloading strategies are used to reduce processing time and ensure instant response to high-frequency dialogue scenarios;

[0065] The context-aware fine-grained permission management model combines RBAC and ABAC models, as well as key distribution technology, to manage tenant permissions in a refined manner and ensure the security of data transmission, especially in sensitive or high-risk areas. It can automatically adjust permission levels to protect user privacy.

[0066] The multi-tenant SaaS integration and resource optimization module supports dynamic adjustment of 5G network slice resources, achieving efficient resource utilization and maximizing cost-effectiveness. In particular, the 5G dynamic scheduling algorithm provides an effective solution for high-bandwidth demand scenarios, which not only meets business needs but also controls costs.

[0067] The intelligent content review and supervision module uses NLP, CV and ML technologies to ensure content quality and compliance, reducing the burden of manual review. At the same time, the tenant reputation model and dynamic review strategy improve review efficiency and reduce the risk of violations.

[0068] The multimedia material management and personalized service module combines user preferences and contextual awareness to provide personalized recommendations, which enhances the relevance and appeal of messages, improves marketing effectiveness and user satisfaction;

[0069] The operation analysis and statistical optimization module uses big data analysis, real-time stream processing technology and 5G-NR-U technology to provide the operation team with in-depth data insights and decision-making support. Through A / B testing and user behavior analysis, it continuously optimizes message strategies and drives business growth. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A system module diagram for improving the efficiency of multi-tenant 5G message push;

[0071] Figure 2 The present invention is a flow chart of a method for improving the efficiency of multi-tenant 5G message push. DETAILED DESCRIPTION

[0072] The present invention is further clearly and completely described below, but the protection scope of the present invention is not limited thereto.

[0073] A system for improving multi-tenant 5G message push efficiency, comprising:

[0074] Intelligent Scheduling and Chatbot Cluster Management Module: The API interface abstracts the differences between different operators' Chatbots to achieve unified management and scheduling. It uses intelligent routing algorithms to dynamically allocate optimal Chatbot resources based on user geographic location, network conditions including latency, packet loss rate, and operator service quality. It has a built-in load balancing strategy and an integrated 5G network slice intelligent selector to dynamically allocate the most appropriate 5G network slice resources based on Chatbot task type and user location. It introduces an encrypted authentication mechanism, uses an AI-driven load prediction model, and combines edge computing to implement a preloading strategy for high-frequency conversation scenarios.

[0075] Multi-tenant SaaS integration and resource optimization module: supports tenants to customize message queues, security policies, and message transmission protocols to meet the needs of different business scenarios. It uses distributed caching technology with the ARC algorithm to dynamically adjust memory allocation, adopts a multi-level isolation storage solution, and combines the key management system KMS to implement end-to-end encryption of tenant data. In the multi-tenant architecture, tenants customize 5G network resource strategies, dynamically adjust slice resources according to business needs, introduce 5G dynamic scheduling algorithms, and automatically switch frequency bands for high-bandwidth demand scenarios;

[0076] Context-aware fine-grained tenant rights management module: Build a role-based access control RBAC model, combined with ABAC-based access control policies, and control tenants' viewing, editing, and deletion rights through ACL. At the same time, it integrates SSL / TLS protocols and data transmission encryption technology, combines 5G positioning and context-aware technology, dynamically adjusts user rights levels, and uses the isolation characteristics of 5G network slices, combined with key distribution technology, to provide dedicated slices for tenant data transmission at different security levels;

[0077] Intelligent content review and supervision module: It integrates natural language processing (NLP), computer vision (CV) and machine learning (ML) technologies to conduct real-time intelligent review of text, image, audio and video content, build a multi-dimensional content scoring system, and set a scoring threshold. When the content score exceeds this threshold, the module will mark it as suspected illegal content. For suspected illegal content, the manual review process will be initiated, and the tenant reputation model will be introduced to dynamically adjust the review strictness based on historical behavior. In addition, a violation warning and penalty mechanism will be implemented. In combination with edge computing resources, the review model will be deployed on the edge nodes of the network to reduce review delays and isolate review tasks using 5G network slicing;

[0078] Multimedia material management and personalized service module: upload, classify, and label various materials through API and user interface, automatically label materials using content recognition technology, support material library permission management, combine user preference analysis, and combine 5G's high bandwidth and positioning capabilities to provide personalized material recommendations for different tenants and users based on user location and time. By switching 5G frequency bands, fast upload and distribution channels are provided for high-quality materials. Context-aware labeling algorithms are introduced to automatically label materials based on content and user context.

[0079] Operation analysis and statistical optimization module: Build a real-time data stream processing platform based on 5G network, use big data analysis, machine learning algorithms and real-time stream processing technology to monitor the system operation status in all directions, collect and analyze user interaction data, message sending trends, and key information of system performance indicators. Utilize 5G-NR-U technology to provide the operation team with in-depth business insights through customized reports and visual dashboards, assist in decision making, and combine A / B testing and user behavior analysis to optimize message strategies.

[0080] refer to Figure 1 As shown, this is a system module diagram for improving the efficiency of multi-tenant 5G message push.

[0081] A method for improving the efficiency of multi-tenant 5G message push is based on the above-mentioned system for improving the efficiency of multi-tenant 5G message push, and the technical solution adopted is as follows:

[0082] S1: The multi-tenant SaaS integration and resource optimization module is started to allocate initial resources to each tenant, including custom message queues, security policy settings, and message transmission protocol selection. The 5G network slice resources are dynamically adjusted according to business needs and the 5G dynamic scheduling algorithm.

[0083] S2: Initialize the context-aware fine-grained tenant rights management module, set basic tenant rights based on the RBAC and ABAC models, dynamically adjust the level of rights based on geographic location and time through 5G positioning technology, and enable key distribution technology to ensure data transmission security;

[0084] S3: Users access the system through the 5G network. The intelligent scheduling and Chatbot cluster management module combines the user's geographic location, network conditions including latency, packet loss rate, and operator service quality, and uses intelligent routing algorithms and 5G network slice intelligent selectors to direct user requests to the optimal Chatbot resources.

[0085] S4: Edge computing uses AI-driven load prediction models to implement pre-loading strategies for high-frequency conversation scenarios;

[0086] S5: Chatbot processes tenant requests and generates response content, including text, images, audio and video. The intelligent content review and supervision module intervenes and combines NLP, CV and ML technologies to conduct real-time intelligent review of the generated content. The review tasks are isolated through 5G network slicing. Suspected illegal content triggers the manual review process and dynamically adjusts the review strictness based on the tenant reputation model.

[0087] S6: The multimedia material management and personalized service module uploads materials based on the response content, and automatically tags them using the context-aware tagging algorithm, and recommends personalized materials based on the user's location and preferences;

[0088] S7: Distributed cache and multi-level isolated storage technology for multi-tenant SaaS integration and resource optimization modules. Approved content is pushed to users through intelligent scheduling and Chatbot cluster management modules combined with edge computing.

[0089] S8: The operation analysis and statistical optimization module continuously collects user interaction data, message sending trends and system performance indicators, uses 5G-NR-U technology to expand the scope of data collection, and uses big data analysis and real-time stream processing technology to generate customized reports and visual dashboards to assist the operation team in making decisions and optimize message strategies through A / B testing and user behavior analysis.

[0090] refer to Figure 2 The figure shows a flow chart of a method for improving the efficiency of multi-tenant 5G message push.

[0091] Furthermore, the S1 dynamically adjusts 5G network slice resources according to business needs in combination with the 5G dynamic scheduling algorithm, including:

[0092] S11: The tenant analyzes the business model, user behavior, and message types including instant messaging and rich media messaging, and specifies the required network resource types including bandwidth and latency requirements. Based on the demand analysis results, the tenant customizes the message queue, security policy, and message transmission protocol through the interactive interface provided by the multi-tenant SaaS platform, and submits the resource requirements.

[0093] S12: The resource requirements submitted by the tenants automatically trigger the intelligent matching algorithm. The algorithm analyzes the tenants’ business characteristics including geographic location, message volume, peak traffic, and network conditions including current slice load and historical performance, and recommends the matching 5G network slices.

[0094] The smart matching algorithm function is , , , , , where B is the bandwidth requirement satisfaction, AB is the available bandwidth, RB is the required bandwidth, and L is the delay sensitivity. is the positive adjustment factor, EL is the expected delay, AL is the actual delay, S is the slice stability, HSS is the maximum stability score, MSS is the historical stability score, BST is the comprehensive service characteristic index, γG is the adjustment factor, DNN is the geographical location, PPT is the predicted peak traffic, PTC is the peak traffic capacity, PMV is the predicted message volume, HMV is the historical message volume, is the weight factor;

[0095] S13: For business scenarios that require high bandwidth for the transmission of a large number of rich media messages, the system automatically detects and triggers the 5G dynamic scheduling algorithm, and automatically switches resources based on network load and availability.

[0096] Furthermore, in S2, the permission level is dynamically adjusted according to the geographic location and time through 5G positioning technology, and the key distribution technology is enabled to ensure the security of data transmission, including:

[0097] S21: Integrate high-precision positioning services through 5G network interfaces, combine the collaborative positioning capabilities of 5G base stations and terminals, and obtain location information of tenants and users in real time;

[0098] S22: The system is synchronized with the standard time server to monitor and record the timestamps of tenant interactions in real time, and combine them with geographic location information to form time-space coordinate data points;

[0099] S23: Dynamically calculate the risk score based on the combination of the tenant's geographical location and time, and dynamically adjust the permission level based on the risk score. The risk score formula is: ,in Due to geographical location factors, is the time period factor, is the sigmoid function, φ and θ are adjustment parameters, and is the weight factor, satisfying ;

[0100] Collect and analyze tenants' behavior data in different geographical locations, which are tenant location data and message traffic data. Obtain tenant location data through the 5G network, record the number of messages sent and received by tenants in each geographical location as message traffic data, record the frequency of tenants visiting a specific geographical location as tenant location frequency L, and record the number of messages sent and received by tenants in a specific geographical location as message traffic T. ,in It is the tenant location frequency weight and message traffic weight;

[0101] For example, if the tenant is in the bank area, the tenant location frequency L is quantified as 0.9 for rare visits, 0.5 for occasional visits, and 0.1 for frequent visits, and the message flow T is quantified as 0.9 for few messages, 0.5 for moderate messages, and 0.1 for a large number of messages;

[0102] For example, if the tenant is in the bookstore area, the tenant location frequency L is quantified as rarely visited 0.2, occasionally visited 0.5, and frequently visited 0.7, and the message flow T is quantified as few messages 0.2, medium messages 0.5, and a large number of messages 0.7;

[0103] For example, if the tenant is in a shopping mall area, the tenant location frequency L is quantified as rarely visited 0.25, occasionally visited 0.5, and frequently visited 0.75, and the message flow T is quantified as few messages 0.25, medium messages 0.5, and a large number of messages 0.75;

[0104] The time periods of a day are classified into working time, non-working time, and late night time. Working time is from 9 am to 5 pm, late night time is from 11 pm to 6 am, and the rest of the time is non-working time. The weighted average method is used to first combine the working time, non-working time, and late night time of the four seasons of spring, summer, autumn, and winter, and holidays to calculate, and then the calculation results are combined with the working time, non-working time, and late night time of special events to calculate using the weighted average method ,Special events are defined as events where the flow of people per minute in the geographical location of the tenant is greater than 5,000. The geographical location of the tenant is a circular area with a radius of 300m with the tenant as the center;

[0105] The permission levels are divided into three levels: low, medium and high. The low permission level allows access to public information and basic operations, the medium permission level allows access to some restricted information and sensitive operations, and the high permission level allows access to highly sensitive information and high-risk operations. For low authority level, risk score Medium authority level, risk score It is a high authority level;

[0106] S24: a fixed geographical radius of 300m is set for each tenant's current geographical location, and the activities of the tenants in this area are recorded to construct an activity range profile of the corresponding tenant;

[0107] When the activity ranges of multiple tenants overlap, that is, their activity radii intersect, the system evaluates the impact of the intersection on permissions. A tenant whose activity range overlaps more extensively with other tenants is considered to be a more influential tenant. For example, if tenant A often appears in the activity range of tenant B, while tenant B rarely appears in the activity range of A, then tenant B will be considered to be more influenced by tenant A, and tenant B's permissions will be reduced.

[0108] Calculate the influence coefficient based on the degree of overlap between tenants’ activities , This indicates that A's influence on B is greater than B's influence on A, and B's authority will be downgraded. , , ,in and is the factor that tenants A and B are in a specific geographical location, is the intersection area of ​​the activity range of tenant A and tenant B, It is the one with the smaller activity area between Tenant A and Tenant B;

[0109] The risk scoring formula is updated to ,in is the weight factor of the influence coefficient;

[0110] S25: After each permission adjustment, key distribution technology is used to generate new encryption keys for sensitive data transmission between tenants, and the validity period and scope of use of the keys are dynamically adjusted according to the permission level and data sensitivity.

[0111] Furthermore, the activities of the tenants in this area are recorded in S24 to construct an activity range profile of the corresponding tenants, including:

[0112] S241: Record the timestamp of each location data of the tenant accessing the system through 5G messages;

[0113] S242: Create a database of interest points tags containing different location categories, where the location category tags include catering, entertainment, culture and education, retail and shopping, nature and outdoor, accommodation, health and fitness, business and office, transportation, and community services, and the interest point tags include food, coffee, fast food, movies, performances, concerts, exhibitions, reading, learning, shopping, clothing, electronic products, walking, running, accommodation, vacation, exercise, yoga, conference room, printing service, transportation hub, and medical service;

[0114] For example, for shopping malls, the location category label is retail and shopping, and the point of interest label is shopping, dining, and entertainment;

[0115] For example, for hotels, the location category label is accommodation, and the point of interest labels are business center and gym;

[0116] For example, for parks, the location category label is nature and outdoor, and the interest point label is walking, sightseeing, and parent-child activities;

[0117] For example, for a cafe, the location category label is catering, and the point of interest labels are coffee, breakfast, and free Wi-Fi;

[0118] S243: Matching the current location data and activity data of the tenant with the point of interest tag database to determine the specific location type and activity type of the tenant;

[0119] S244: Aggregate the activity data of the tenants and create a profile for each tenant, including the locations most frequently visited by the tenant and the time period during the day when the tenant at the corresponding location is most active.

[0120] Furthermore, the S3 uses an intelligent routing algorithm and a 5G network slice intelligent selector to direct user requests to the optimal Chatbot resource, including:

[0121] S31: The intelligent routing algorithm builds a comprehensive scoring model based on the network status, geographic location, and the current load of the Chatbot cluster, and assigns weights to each dimension;

[0122] The network status includes latency, packet loss rate, and bandwidth. The geographic location includes distance and location hotspot, i.e., whether it is a high-traffic area. The traffic threshold f is set to 3MB / s. If the user is in a high-traffic area, the traffic is greater than f. If the user is in a low-traffic area, the traffic is less than f. The current load of the Chatbot cluster includes CPU usage, memory usage, and task queue length.

[0123] S32: The comprehensive scoring model is total score = ∑(single score × weight), and the weights of network status, geographic location, and current load of the Chatbot cluster are 0.45, 0.3, and 0.25 respectively;

[0124] The latency score in the network condition is , the packet loss rate score is , bandwidth score is ,The total score of network status is the sum of the delay, packet loss rate and bandwidth scores;

[0125] Location score = ,in and is the weight of distance and location hotspot, distance score = ,in is the actual distance between the Chatbot server and the user, The distance covered by the Chatbot server. If the user is in a high-traffic area, the location hotspot score is 0; if the user is in a low-traffic area, the location hotspot score is 1;

[0126] The current load score of the Chatbot cluster = ,in The weights corresponding to CPU usage, memory usage, and task queue length, The maximum acceptable CPU usage threshold, the maximum acceptable memory usage threshold, and the maximum acceptable queue length threshold;

[0127] S33: Combined with the constructed scoring model, all available Chatbot resources are matched and scored. The Chatbot resource with the highest score is the optimal Chatbot resource. For users in high-traffic areas, the slices are automatically switched to direct user requests to the optimal Chatbot resource.

[0128] Furthermore, the edge computing in S4 implements a preloading strategy for high-frequency conversation scenarios based on an AI-driven load prediction model, including:

[0129] S41: Through the 5G network, edge computing nodes collect user interaction data in real time, including message frequency, conversation content category, user active time period, and geographic location distribution;

[0130] S42: Use feature selection algorithms to extract features from raw data that are valuable for load forecasting, including conversation peaks, high-demand periods in specific geographic locations, and user preference patterns;

[0131] S43: Build AI models, continuously learn on edge computing nodes, and use online learning technology to adjust model parameters in real time based on new data;

[0132] S44: Integrates scene recognition algorithms to identify high-frequency conversation scenes and dynamically adjusts the resource preloading priority based on the identified scene types and real-time predicted load conditions;

[0133] S45: Based on AI prediction results, design an adaptive preloading strategy to preload the corresponding Chatbot model, conversation database, and system components on the edge server during the predicted high-load period.

[0134] Furthermore, in S6, personalized materials are recommended based on the user's location and preferences, including:

[0135] S61: Perform content analysis on multimedia materials, extract visual features and audio features, combine with geographic information system data, generate location features, including distance from specific locations and surrounding environment types, and establish user preference models based on historical browsing, likes, and sharing interaction data;

[0136] S62: combining the material features, the user location features, and the current time and weather external context factors, automatically adding context labels to the material through a label generation algorithm;

[0137] S63: Using the tags generated by the context-aware tagging algorithm, combined with the user's current location and preferences, a context-aware recommendation model is constructed. The model is ,in is the recommendation score of material i for a specific user, is the user feature vector, which contains the user's historical behavior and preference information. is the vector representation of the j-th context label, n is the total number of context labels possessed by material i, W is the weight vector, and b is the constant bias term.

[0138] The present invention provides a system and method for improving the efficiency of multi-tenant 5G message push, intelligently schedules Chatbot resources, dynamically allocates the best service resources according to user location and network conditions, and uses 5G network slicing technology to achieve flexible resource management. Through context-aware fine-grained permission management, user permissions are dynamically adjusted according to temporal and spatial changes. The content review module uses AI technology for real-time review and dynamically adjusts the review strictness in combination with the user reputation model. Multimedia material management incorporates user preferences and location information to provide personalized high-quality material recommendations. The operation analysis module uses big data and machine learning to optimize message strategies, thereby solving the problems of low push efficiency, inflexible resource allocation, insufficient security, low content review efficiency, shallow personalized services, and lack of operation analysis in existing 5G message services.

Claims

1. A method for improving the efficiency of multi-tenant 5G message push, characterized in that: Processing is based on a system that improves the efficiency of multi-tenant 5G message push, including: A system for improving the efficiency of multi-tenant 5G message push includes: intelligent scheduling and Chatbot cluster management module, multi-tenant SaaS integration and resource optimization module, context-aware fine-grained tenant authority management module, intelligent content review and supervision module, multimedia material management and personalized service module, and operation analysis and statistical optimization module; Intelligent Scheduling and Chatbot Cluster Management Module: The API interface abstracts the differences between different operators' Chatbots to achieve unified management and scheduling. It uses intelligent routing algorithms to dynamically allocate optimal Chatbot resources based on user geographic location, network conditions including latency, packet loss rate, and operator service quality. It has a built-in load balancing strategy and an integrated 5G network slice intelligent selector to dynamically allocate the most appropriate 5G network slice resources based on Chatbot task type and user location. It introduces an encrypted authentication mechanism, uses an AI-driven load prediction model, and combines edge computing to implement a preloading strategy for high-frequency conversation scenarios. Multi-tenant SaaS integration and resource optimization module: supports tenants to customize message queues, security policies, and message transmission protocols to meet the needs of different business scenarios. It uses distributed caching technology with the ARC algorithm to dynamically adjust memory allocation, adopts a multi-level isolation storage solution, and combines the key management system KMS to implement end-to-end encryption of tenant data. In the multi-tenant architecture, tenants customize 5G network resource strategies, dynamically adjust slice resources according to business needs, introduce 5G dynamic scheduling algorithms, and automatically switch frequency bands for high-bandwidth demand scenarios; Context-aware fine-grained tenant rights management module: Build a role-based access control RBAC model, combined with ABAC-based access control policies, and control tenants' viewing, editing, and deletion rights through ACL. At the same time, it integrates SSL / TLS protocols and data transmission encryption technology, combines 5G positioning and context-aware technology, dynamically adjusts user rights levels, and uses the isolation characteristics of 5G network slices, combined with key distribution technology, to provide dedicated slices for tenant data transmission at different security levels; Intelligent content review and supervision module: It integrates natural language processing (NLP), computer vision (CV) and machine learning (ML) technologies to conduct real-time intelligent review of text, image, audio and video content, build a multi-dimensional content scoring system, and set a scoring threshold. When the content score exceeds this threshold, the module will mark it as suspected illegal content. For suspected illegal content, the manual review process will be initiated, and the tenant reputation model will be introduced to dynamically adjust the review strictness based on historical behavior. In addition, a violation warning and penalty mechanism will be implemented. In combination with edge computing resources, the review model will be deployed on the edge nodes of the network to reduce review delays and isolate review tasks using 5G network slicing; Multimedia material management and personalized service module: upload, classify, and label various materials through API and user interface, automatically label materials using content recognition technology, support material library permission management, combine user preference analysis, and combine 5G's high bandwidth and positioning capabilities to provide personalized material recommendations for different tenants and users based on user location and time. By switching 5G frequency bands, fast upload and distribution channels are provided for high-quality materials. Context-aware labeling algorithms are introduced to automatically label materials based on content and user context. Operation analysis and statistical optimization module: Build a real-time data stream processing platform based on 5G network, use big data analysis, machine learning algorithms and real-time stream processing technology to monitor the system operation status in all aspects, collect and analyze user interaction data, message sending trends, key information of system performance indicators, and use 5G-NR-U technology to provide in-depth business insights for the operation team through customized reports and visual dashboards to assist decision-making, and combine A / B testing and user behavior analysis to optimize message strategies; A method for improving multi-tenant 5G message push efficiency, comprising: S1: The multi-tenant SaaS integration and resource optimization module is started to allocate initial resources to each tenant, including custom message queues, security policy settings, and message transmission protocol selection. The 5G network slice resources are dynamically adjusted according to business needs and the 5G dynamic scheduling algorithm. Tenants analyze business models, user behaviors, and message types, including instant messaging and rich media messaging, to identify the types of network resources they need, including bandwidth and latency requirements. Based on the demand analysis results, tenants customize message queues, security policies, and message transmission protocols through the interactive interface provided by the multi-tenant SaaS platform, and submit resource requirements. The resource requirements submitted by tenants automatically trigger the intelligent matching algorithm, which analyzes the tenants’ business characteristics, including geographic location, message volume, peak traffic, and network conditions, including current slice load and historical performance, and recommends matching 5G network slices. The smart matching algorithm function is , , , , , where B is the bandwidth requirement satisfaction, AB is the available bandwidth, RB is the required bandwidth, and L is the delay sensitivity. is the positive adjustment factor, EL is the expected delay, AL is the actual delay, S is the slice stability, HSS is the maximum stability score, MSS is the historical stability score, BST is the comprehensive service characteristic index, γG is the adjustment factor, DNN is the geographical location, PPT is the predicted peak traffic, PTC is the peak traffic capacity, PMV is the predicted message volume, HMV is the historical message volume, is the weight factor; For business scenarios that require high bandwidth for the transmission of a large number of rich media messages, the system automatically detects and triggers the 5G dynamic scheduling algorithm to automatically switch resources based on network load and availability; S2: Initialize the context-aware fine-grained tenant rights management module, set basic tenant rights based on the RBAC and ABAC models, dynamically adjust the level of rights based on geographic location and time through 5G positioning technology, and enable key distribution technology to ensure data transmission security; Integrate high-precision positioning services through 5G network interfaces, combine the collaborative positioning capabilities of 5G base stations and terminals, and obtain location information of tenants and users in real time; The system is synchronized with the standard time server to monitor and record the timestamps of tenant interactions in real time, and combines the geographic location information to form time-space coordinate data points; The risk score is dynamically calculated based on the combination of the tenant's geographical location and time. Based on the risk score, the permission level is dynamically adjusted. The risk score formula is: ,in Due to geographical location factors, is the time period factor, is the sigmoid function, φ and θ are adjustment parameters, and is the weight factor, satisfying ; Collect and analyze tenants' behavior data in different geographical locations, which are tenant location data and message traffic data. Obtain tenant location data through the 5G network, record the number of messages sent and received by tenants in each geographical location as message traffic data, record the frequency of tenants visiting a specific geographical location as tenant location frequency L, and record the number of messages sent and received by tenants in a specific geographical location as message traffic T. ,in It is the tenant location frequency weight and message traffic weight; The time periods of a day are classified into working time, non-working time, and late night time. Working time is from 9 am to 5 pm, late night time is from 11 pm to 6 am, and the rest of the time is non-working time. The weighted average method is used to first combine the working time, non-working time, and late night time of the four seasons of spring, summer, autumn, and winter, and holidays to calculate, and then the calculation results are combined with the working time, non-working time, and late night time of special events to calculate using the weighted average method ,Special events are defined as events where the flow of people per minute in the geographical location of the tenant is greater than 5,000. The geographical location of the tenant is a circular area with a radius of 300m with the tenant as the center; The permission levels are divided into three levels: low, medium and high. The low permission level allows access to public information and basic operations, the medium permission level allows access to some restricted information and sensitive operations, and the high permission level allows access to highly sensitive information and high-risk operations. For low authority level, risk score Medium authority level, risk score It is a high authority level; A fixed geographical radius of 300m is set for each tenant's current geographical location, and the activities of tenants in this area are recorded to build an activity range profile for the corresponding tenants. The timestamp of each location data when the tenant accesses the system through 5G messages is recorded, and a point of interest tag database containing different location categories is created. The tenant's current location data and activity data are matched with the point of interest tag database to determine the specific location type and activity type of the tenant, summarize the tenant's activity data, and create a profile for each tenant, including the tenant's most frequently visited locations and the most active time period of the day for the corresponding tenants; When the activity ranges of multiple tenants overlap, that is, their activity radii intersect, the system evaluates the impact of the intersection on permissions; Calculate the influence coefficient based on the degree of overlap between tenants’ activities , , , ,in and is the factor that tenants A and B are in a specific geographical location, is the intersection area of ​​the activity range of tenant A and tenant B, It is the one with the smaller activity area between Tenant A and Tenant B; The risk score formula is updated to ,in is the weight factor of the influence coefficient; After each permission adjustment, key distribution technology is used to generate new encryption keys for sensitive data transmission between tenants, and the validity period and scope of use of the keys are dynamically adjusted according to the permission level and data sensitivity; S3: Users access the system through the 5G network. The intelligent scheduling and Chatbot cluster management module combines the user's geographic location, network conditions including latency, packet loss rate, and operator service quality, and uses intelligent routing algorithms and 5G network slice intelligent selectors to direct user requests to the optimal Chatbot resources. The intelligent routing algorithm builds a comprehensive scoring model based on the network status, geographic location, and the current load of the Chatbot cluster, assigning weights to each dimension; The network status includes latency, packet loss rate, and bandwidth. The geographic location includes distance and location hotspot, i.e., whether it is a high-traffic area. The traffic threshold f is set. If the user is in a high-traffic area, the traffic is greater than f. If the user is in a low-traffic area, the traffic is less than f. The current load of the Chatbot cluster includes CPU usage, memory usage, and task queue length. The comprehensive scoring model is total score = ∑ (single score × weight), and the weights of network status, geographical location, and current load of the Chatbot cluster are 0.45, 0.3, and 0.25 respectively; The latency score in the network condition is , the packet loss rate score is , bandwidth score is ,The total score of network status is the sum of the delay, packet loss rate and bandwidth scores; Location score = ,in and is the weight of distance and location hotspot, distance score = ,in is the actual distance between the Chatbot server and the user, The distance covered by the Chatbot server. If the user is in a high-traffic area, the location hotspot score is 0; if the user is in a low-traffic area, the location hotspot score is 1; The current load score of the Chatbot cluster = ,in The weights corresponding to CPU usage, memory usage, and task queue length, The maximum acceptable CPU usage threshold, the maximum acceptable memory usage threshold, and the maximum acceptable queue length threshold; Combined with the constructed scoring model, all available Chatbot resources are matched and scored. The Chatbot resource with the highest score is the optimal Chatbot resource. For users in high-traffic areas, the slice is automatically switched to direct user requests to the optimal Chatbot resource. S4: Edge computing uses AI-driven load prediction models to implement pre-loading strategies for high-frequency conversation scenarios; S5: Chatbot processes tenant requests and generates response content, including text, images, audio and video. The intelligent content review and supervision module intervenes and combines NLP, CV and ML technologies to conduct real-time intelligent review of the generated content. The review tasks are isolated through 5G network slicing. Suspected illegal content triggers the manual review process and dynamically adjusts the review strictness based on the tenant reputation model. S6: The multimedia material management and personalized service module uploads materials based on the response content, and automatically tags them using the context-aware tagging algorithm, and recommends personalized materials based on the user's location and preferences; S7: Distributed cache and multi-level isolated storage technology for multi-tenant SaaS integration and resource optimization modules. Approved content is pushed to users through intelligent scheduling and Chatbot cluster management modules combined with edge computing. S8: The operation analysis and statistical optimization module continuously collects user interaction data, message sending trends and system performance indicators, uses 5G-NR-U technology to expand the scope of data collection, and uses big data analysis and real-time stream processing technology to generate customized reports and visual dashboards to assist the operation team in making decisions and optimize message strategies through A / B testing and user behavior analysis.

2. A method for improving multi-tenant 5G message push efficiency as claimed in claim 1, characterized in that: The edge computing in S4 is based on an AI-driven load prediction model and implements a preloading strategy for high-frequency conversation scenarios, including: S41: Through the 5G network, edge computing nodes collect user interaction data in real time, including message frequency, conversation content category, user active time period, and geographic location distribution; S42: Use feature selection algorithms to extract features from raw data that are valuable for load forecasting, including conversation peaks, high-demand periods in specific geographic locations, and user preference patterns; S43: Build AI models, continuously learn on edge computing nodes, and use online learning technology to adjust model parameters in real time based on new data; S44: Integrates scene recognition algorithms to identify high-frequency conversation scenes and dynamically adjusts the resource preloading priority based on the identified scene types and real-time predicted load conditions; S45: Based on AI prediction results, design an adaptive preloading strategy to preload the corresponding Chatbot model, conversation database, and system components on the edge server during the predicted high-load period.

3. A method for improving multi-tenant 5G message push efficiency as claimed in claim 1, characterized in that: The S6 combines the user's location and preferences to recommend personalized materials, including: S61: Perform content analysis on multimedia materials, extract visual features and audio features, combine with geographic information system data, generate location features, including distance from specific locations and surrounding environment types, and establish user preference models based on historical browsing, likes, and sharing interaction data; S62: combining the material features, the user location features, and the current time and weather external context factors, automatically adding context labels to the material through a label generation algorithm; S63: Using the tags generated by the context-aware tagging algorithm, combined with the user's current location and preferences, a context-aware recommendation model is constructed. The model is ,in is the recommendation score of material i for a specific user, is the user feature vector, which contains the user's historical behavior and preference information. is the vector representation of the j-th context label, n is the total number of context labels possessed by material i, W is the weight vector, and b is the constant bias term.