An AI-based multi-platform media content optimization and recommendation method and system

Abstract

The present invention relates to the technical field of data recommendation, and particularly to a method and system for optimizing the recommendation of multi-platform media content based on AI. The method includes the following steps: obtaining user device parameters and user platform behavior data; confirming user preferences based on the user platform behavior data to obtain user platform preference data; analyzing the hardware performance of the user's device according to the user device parameters and grading it, and generating a device adaptability matrix through the grading results; extracting the user active time period from the user platform preference data to obtain user active time period feature data; analyzing the content loading time consumption of the device adaptability matrix to generate device time consumption data; extracting format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library. The present invention improves the compatibility and response speed of cross-platform media content recommendation by comprehensively considering user preferences, device performance, time period activity, content adaptability, and network fluctuations.

Description

Technical Field

[0001] The present invention relates to the technical field of data recommendation, and in particular, to an AI-based multi-platform media content optimization recommendation method and system. Background Art

[0002] Initially, traditional recommendation systems mainly relied on content-based filtering and collaborative filtering technologies to make recommendations based on user historical behavior data or item attribute information. However, when faced with large-scale, diverse, and high-dimensional data, these methods often suffer from problems such as low accuracy and poor recommendation quality. The introduction of deep learning technology has brought revolutionary changes to recommendation systems. Through neural network models, recommendation systems can better understand users' interest preferences and can handle complex non-linear relationships. In recent years, with the rise of big data, cloud computing, and edge computing technologies, recommendation systems have gradually developed in the direction of multi-platform and cross-device. AI algorithms are no longer limited to data analysis on a single platform but can achieve cross-platform data sharing and analysis to provide more personalized and accurate recommendation services. However, current existing recommendation systems often ignore the performance differences of devices when recommending content, resulting in slow loading or incompatibility of the recommended content on low-performance devices. At the same time, there are differences in the content loading time-consuming on different devices, which will affect the user experience and further lead to low compatibility and response speed of cross-platform media content recommendation. Summary of the Invention

[0003] Based on this, it is necessary to provide an AI-based multi-platform media content optimization recommendation method and system to solve at least one of the above technical problems.

[0004] To achieve the above object, an AI-based multi-platform media content optimization recommendation method, the method includes the following steps:

[0005] Step S1: Obtain user device parameters and user platform behavior data; confirm user preferences based on the user platform behavior data to obtain user platform preference data;

[0006] Step S2: Analyze the hardware performance of the user's device according to the user device parameters and classify it, and generate a device adaptability matrix through the classification results; extract the user active period from the user platform preference data to obtain user active period feature data; analyze the content loading time-consuming of the device adaptability matrix to generate device time-consuming data;

[0007] Step S3: extracting format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library; calculating the matching degree between the device adaptability matrix and the format adaptation parameters to generate a device compatibility index; generating a time period recommendation weight by combining device time consumption data with user active time period feature data;

[0008] Step S4: Obtain the real-time fluctuation value of the network bandwidth; construct a dynamic recommendation scoring model through the device compatibility index and the time period recommendation weight, and use the dynamic recommendation scoring model based on the real-time fluctuation value of the network bandwidth to perform lightweight recommendation of graphic content on the cross-platform content feature library to perform multi-platform media content optimization recommendation operations.

[0009] The present invention collects user device parameters and platform behavior data, combines the characteristics of user active time periods, accurately depicts the user's content preferences in different time periods and different devices, and realizes the refined construction of personalized recommendation strategies. By grading the device hardware performance and generating the device adaptability matrix, combined with the content format adaptation parameters, the device compatibility index is calculated to ensure the display and operation effect of the content on different devices, avoiding loading failure or jamming problems. The time period recommendation weight constructed by the user's active time period characteristics and the device time consumption data enables the recommendation system to push the most matching content in the time period when the user is most likely to be active, thereby improving the user's click-through rate and satisfaction. Combined with the real-time fluctuation data of network bandwidth, a real-time scoring mechanism is introduced into the recommendation model to realize lightweight processing of graphic content, still ensure content loading and reading experience in a low-bandwidth environment, and significantly optimize the recommendation effect in a weak network environment. The method organically integrates device performance, user behavior, content characteristics and network environment, and provides a general and scalable optimization recommendation framework for cross-platform content distribution systems, which can be applied to various terminal forms such as APP, applets, and web pages. Therefore, the present invention improves the compatibility and response speed of cross-platform media content recommendation by comprehensively considering user preferences, device performance, time period activity, content adaptability and network fluctuations.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: using network protocols to obtain user device parameters and user platform behavior data;

[0012] Step S12: performing usage log analysis and scene feature extraction on the user platform behavior data to generate user behavior feature data, wherein the user behavior feature data includes usage feature data and usage scene feature data;

[0013] Step S13: Analyze the user interaction pattern based on the usage feature data to generate user interaction pattern data;

[0014] Step S14: Perform behavioral clustering analysis on the user interaction pattern data to generate user behavior clustering data; combine the user device parameter data and the user behavior clustering data to perform personalized preference modeling and generate user preference feature data;

[0015] Step S15: Calculate the preference weights of the user in different usage scenarios for the user preference feature data according to the usage scenario feature data to generate user platform preference data.

[0016] The present invention automatically collects user device parameters and platform behavior data through network protocols. By combining usage log analysis and scenario feature extraction, the system can continuously obtain high-quality and multi-dimensional usage data of users under the condition of being unaware, effectively supporting subsequent modeling. Based on the user interaction pattern analysis of usage feature data, the typical operation paths and behavior preferences of users in the platform can be identified, providing a reference basis closer to the real operation habits of users for the recommendation system. Through user behavior clustering analysis, different types of user behaviors can be classified and modeled, further supporting personalized content push, interface customization, and dynamic strategy adjustment, and improving the operation efficiency of the platform. Using the fusion of device parameters and user behavior clustering results for personalized modeling not only considers the interaction behaviors of users themselves but also introduces the device environment context, and the generated preference feature data is more complete and authentic. By associating and calculating the usage scenario features and user preference features, the preference weights of users in different usage scenarios are generated, enabling the recommendation system to have the "context awareness" ability and realizing intelligent switching recommendations of content in various usage scenarios such as commuting, working, and resting. The generation process of platform preference data considers the comprehensive influence of user behavior, device performance, and scenario variables, enabling the recommendation system to adapt to different user groups and device types, and having stronger generalization ability and real-time response ability.

[0017] Preferably, step S15 includes the following steps:

[0018] Step S151: Perform high-dimensional vector mapping transformation on the user preference feature data to generate user multi-dimensional preference vector data; perform time-series dynamic clustering on the usage scenario feature data to generate a scenario dynamic feature matrix;

[0019] Step S152: Perform cross-scenario preference projection based on the user multi-dimensional preference vector data and the scenario dynamic feature matrix to generate a user scenario projection parameter set;

[0020] Step S153: Use the user scenario projection parameter set to perform adaptive preference weight allocation calculation to generate a user scenario adaptive weight matrix;

[0021] Step S154: Perform deep feature fusion on the user scenario projection parameter set according to the user scenario adaptive weight matrix, and perform multi-level normalization processing to calculate the final preference weights of the user in different usage scenarios, and generate user platform preference data.

[0022] In the present invention, by performing high-dimensional vector mapping on user preference feature data to construct multi-dimensional preference vector data, the user preference has stronger expression ability, which is convenient for subsequent high-dimensional feature association and calculation, and significantly improves the modeling accuracy. The time-series dynamic clustering technology is used to process the usage scenario feature data to generate a dynamic feature matrix reflecting the scenario change trend, enabling the recommendation system to perceive the fluctuations of the user usage environment in the time dimension and realizing real-time modeling of complex scenarios. Based on the cross-scenario preference projection mechanism between the multi-dimensional preference vector and the dynamic scenario matrix, the mapping relationship between the user preference and the scenario context is effectively established, realizing the ability to switch the recommendation logic from static preference to dynamic preference. The scenario projection parameter set is used for adaptive weight allocation to generate a user scenario adaptive weight matrix, ensuring that the recommended content can flexibly adjust the display strategy in different usage scenarios and effectively match the actual needs of users. Through deep feature fusion and multi-level normalization processing, the calculation result of the final preference weight has stronger non-linear expression ability and normalization, improving the recognition and adaptation ability of the recommendation system for micro-differentiated users and marginal scenarios. The user platform preference data, as the final output result, comprehensively considers the multi-dimensional behaviors of users, device characteristics, usage scenarios and dynamic changes in the context, provides high-quality input for the subsequent recommendation model, and helps to build a more intelligent, refined and context-aware recommendation system.

[0023] Preferably, step S2 includes the following steps:

[0024] Step S21: Evaluate the device computing power of the user device parameters to generate user device performance evaluation data; perform hierarchical clustering analysis of the hardware computing capabilities of the user device parameters through the user device performance evaluation data to generate a device performance grading label set;

[0025] Step S22: Extract the device driver compatibility and rendering capabilities of the user device parameters, and construct an adaptation matrix for the device performance grading label set to generate a device adaptation matrix;

[0026] Step S23: Perform time series pattern recognition on the user platform preference data, extract the user interaction peak features to generate user active period trend data; perform time series distribution calculation on the user active period trend data to generate user active period feature data;

[0027] Step S24: Simulate the content loading path of the device adaptation matrix, calculate the loading delay of different devices in different scenarios, and generate device time-consuming data.

[0028] Through the present invention, the computing power of the user device parameters is evaluated, and a hierarchical clustering algorithm is used to generate a device performance grading label set, which not only improves the accuracy of device ability evaluation, but also provides accurate hardware basic data support for subsequent content adaptation and recommendation strategies. On the basis of extracting the device driver compatibility and rendering ability, a device adaptability matrix is constructed through the label set, enabling the system to achieve fine-grained adaptation of multi-level elements such as graphic content and interactive special effects according to the true performance ability of the device, and ensuring the consistency of the user experience. By means of time series pattern recognition and peak feature extraction, the high-frequency areas and time periods of user interaction are refined, and combined with time series distribution modeling, feature data of user active time periods are formed, providing key inputs for accurate content push and recommendation rhythm control. Through the simulation of the content loading path driven by the device adaptability matrix, the loading delay is predicted in advance under different scenarios and device combinations, generating device time-consuming data, and improving the forward-looking and adaptive ability of the content distribution mechanism, and significantly optimizing the end-side response speed and user waiting experience. Considering the device performance evaluation, adaptability modeling, user behavior rhythm and loading time-consuming characteristics comprehensively, the recommendation system can still maintain high-quality content distribution and dynamic optimization scheduling in the face of diverse devices, changing user active behaviors and complex network environments. The feature data of user active time periods and device time-consuming data are used as the core inputs of the time period recommendation weight and content adaptation weight in the subsequent recommendation model, ensuring that the recommendation strategy not only focuses on the matching degree of the content itself, but also accurately perceives the association mode among the user-device-time.

[0029] Preferably, step S24 includes the following steps:

[0030] Step S241: Extract the device characteristics of the device adaptability matrix to generate device performance data; match the device performance data with the network conditions to generate network status data;

[0031] Step S242: Analyze the storage medium performance of the network status data, and evaluate the system overhead of the device adaptability matrix based on the storage medium performance to generate system load data;

[0032] Step S243: Formulate a content chunking strategy for the system load data to generate chunking scheme data; sort the chunking scheme data by priority to generate loading sequence data;

[0033] Step S244: Analyze the dependency relationship of the loading sequence data to generate dependency graph data; optimize the loading path of the dependency graph data to generate path scheme data;

[0034] Step S245: Perform multi-dimensional loading delay calculation on the path scheme data to obtain device time-consuming data, where the multi-dimensional loading delay calculation includes transmission delay, storage read / write delay, decoding and rendering delay, and system scheduling delay.

[0035] The present invention completes the device performance modeling under network conditions by extracting the device features of the device adaptability matrix and combining the current network state, providing basic performance data that is more suitable for the real environment for formulating subsequent loading strategies. Analyze the performance of the device storage medium using network state data, and evaluate the system overhead accordingly to generate system load data, thereby effectively predicting the load intensity of each system resource unit (such as memory, IO, CPU, etc.) in a specific content loading process. Based on the system load data, formulate a content chunking strategy and prioritize the loading order of each chunk to improve the loading agility of important resource content in the system, and implement a recommended scheduling mechanism of on-demand distribution and key priority. Perform dependency analysis on the loading sequence data and generate a dependency graph, so that there is a clear dependency logic in the content loading process, avoiding invalid loading and path conflicts, and fundamentally improving the system loading efficiency. Optimize the path scheme of the dependency graph to maximize the use of system resources, and at the same time introduce multiple dimensions such as transmission delay, storage read / write delay, decoding and rendering delay, and system scheduling delay to perform comprehensive delay calculation on the loading path to generate high-precision device time-consuming data. The finally generated device time-consuming data accurately reflects the content loading efficiency under the influence of multiple factors such as specific device performance, network state, system load, and content dependencies, effectively supporting the implementation of the dynamic optimization strategy of the recommendation system. Utilizing the loading path optimization and delay prediction capabilities, the loading strategy and presentation method of recommended content can be adaptively adjusted under different device conditions to ensure that low-performance devices can also obtain a smooth and accurate recommendation experience, significantly expanding the coverage ability and user satisfaction of the system.

[0036] Preferably, calculating the matching degree between the device adaptability matrix and the format adaptation parameter in step S3 to generate a device compatibility index includes:

[0037] Perform eigenvalue decomposition on the device adaptability matrix, extract key performance adaptation factors, and obtain a device adaptation feature vector set, where the formula for eigenvalue decomposition is as follows: Among them, is the device adaptation feature vector set, are the singular value decomposition components of the device adaptability matrix respectively; normalize the format adaptation parameter and convert it into a high-dimensional vector representation to obtain the format adaptation vector : Construct a format adaptation parameter matrix : Calculate the matching degree score between the device adaptation feature vector and the format adaptation parameter matrix to obtain an adaptation matching degree score matrix, where the calculation formula is as follows: Calculate the adaptation matching degree score matrix Normalized matching score: Extract the computing power vector of the devices in the device adaptability matrix Calculate the device adaptation weight distribution: Calculate the weighted matching score: For the weighted matching score Perform fuzzy clustering analysis and calculate the device adaptation stability index : Wherein, is the index number of the device, is the total number of devices, is the fuzzy membership function: In the formula, is the index number of the device, is the total number of devices, is the fuzzy membership function: In the formula, is the smoothing parameter, is the adaptation threshold; Calculate the device compatibility index through the device adaptation stability index, weighted matching score, and normalized matching score to obtain the device compatibility index The calculation formula of the device compatibility index is as follows: In the formula, is the weight parameter.

[0038] In the present invention, by performing singular value decomposition (SVD) on the device adaptability matrix to extract key performance adaptation factors, the potential adaptation structure of the device under multi-dimensional performance indicators is effectively restored, enabling the system to accurately evaluate the compatibility basis between the device and the content format. The content format parameters are processed using normalization and high-dimensional matrices to avoid offsets in the matching calculation caused by scale differences in the format parameters, and improve the measurement accuracy of the device compatibility ability under different format dimensions. Fuzzy clustering analysis is performed based on the weighted matching score, and the adaptation stability index is calculated through the membership function, effectively solving the problem that hard scoring cannot reflect the boundary ambiguity, and enhancing the robustness of the recommendation strategy to performance fluctuations of heterogeneous devices. Calculate the device adaptation weight using the device computing power distribution, introduce a weighted scoring mechanism to make the system pay more attention to the matching performance of high-performance devices, and combine the stability index and the normalized score into a unified compatibility index, effectively improving the overall adaptability of the system in the device recommendation scenario. The ratio among the matching degree, stability, and basic score can be flexibly controlled by adjusting the weight parameters α, β, γ, so that the recommendation strategy can be dynamically adapted to various business objectives such as high-performance device optimization recommendation or low-performance device compatibility optimization. The obtained device compatibility index As the core evaluation basis of the recommendation engine, it can provide strong data support for content distribution across multiple platforms, ensuring that the system can still achieve personalized, lightweight, and precise recommendations in the face of device differentiation scenarios. The designed calculation process is based on linear algebra, matrix transformation, and fuzzy logic, and can be easily embedded into deep learning models, graph neural networks, or embedded recommendation modules, with high engineering feasibility and computational scalability.

[0039] Preferably, the generation of time period recommendation weights by combining device time-consuming data with user active time periods in step S3 includes:

[0040] Perform time interval segmentation on the device time-consuming data to generate multi-time period device performance response data;

[0041] Perform density modeling on the user active time period feature data to generate time period active density distribution data;

[0042] Perform time period mapping alignment on the multi-time period device performance response data and the time period active density distribution data to generate time period performance active mapping data;

[0043] Perform weighted normalization on the time period performance active mapping data to generate a basic time period adaptation coefficient; perform reverse fitting of the adjustment coefficient on the basic time period adaptation coefficient in combination with the cross-platform content feature library to generate a corrected time period adaptation weight;

[0044] Perform exponential smoothing and outlier correction on the corrected time period adaptation weight to generate a smoothed time period weight curve;

[0045] Perform structured compression on the smoothed time period weight curve to generate time period recommendation weights.

[0046] The present invention divides the device time-consuming data into time intervals, combines it with the feature data of the user active period for density modeling, constructs a refined "device response - user active" mapping relationship, and significantly improves the perception accuracy and matching ability of the recommendation system in the time dimension. By introducing a period performance active mapping mechanism and weighted normalization processing, the integrated evaluation of the device performance and the user active trend is realized, effectively reducing the problem of the decline in recommendation accuracy caused by differences in user usage habits or inconsistent device states. By aligning and fitting the basic period adaptation coefficient with the content feature library, the recommendation tendency is adjusted to be consistent with the content attributes, further enhancing the context adaptation ability of the content distribution strategy and the consistency of content performance. By performing exponential smoothing and outlier correction on the corrected period weights, the problem of large fluctuations in the original data and susceptibility to extreme behaviors is solved, making the finally generated period weight curve more continuous, robust, and trend interpretable. By performing structured compression on the smoothed curve, a compact period recommendation weight expression is obtained, which not only significantly reduces the transmission and storage burden but also facilitates its rapid deployment and real-time calculation on resource-constrained devices such as mobile terminals and edge computing terminals. Under complex conditions such as multi-device heterogeneity and obvious differences in user period behaviors faced by the platform side, this process generates a dynamic and personalized period weight model by integrating device responses and user active behaviors, enabling the recommendation system to have stronger environmental adaptability. The generated "period recommendation weight" serves as an important input benchmark for the recommendation scheduling system, which can significantly improve the timing accuracy and user acceptance of content pushing without increasing the additional complexity of the system.

[0047] Preferably, step S4 includes the following steps:

[0048] Step S41: Perform bandwidth sampling on the network interface to obtain bandwidth sampling data; perform bandwidth fluctuation analysis on the bandwidth sampling data to generate the real-time network bandwidth fluctuation value;

[0049] Step S42: Construct a dynamic recommendation scoring model through the device compatibility index and the period recommendation weight;

[0050] Step S43: Based on the real-time network bandwidth fluctuation value, use the dynamic recommendation scoring model to execute a lightweight optimization strategy on the cross-platform content feature library to generate a lightweight recommended content pool, where the lightweight optimization strategy includes image compression, video bitrate reduction, and text first loading;

[0051] Step S44: Perform distributed cross-platform content recommendation on the lightweight recommended content pool to execute a multi-platform media content optimization recommendation task.

[0052] By performing real-time sampling on the network bandwidth and conducting fluctuation analysis, the system of the present invention can dynamically capture the changes in network bandwidth, thereby providing a more accurate bandwidth adaptation model for recommended content and ensuring a smooth and fast recommendation experience in different network environments. By constructing a dynamic recommendation scoring model based on the device compatibility index and the time period recommendation weight, the system can adjust the recommendation strategy in real time, combine the network status and device performance, and intelligently optimize content distribution to achieve a more accurate cross-platform content adaptation and push effect. Lightweight optimization strategies (such as image compression, video bitrate reduction, and text first loading) are deeply optimized for environments with large bandwidth fluctuations, reducing the bandwidth pressure on user devices and improving the speed and stability of content loading. Especially for users or network environments with limited bandwidth resources, this strategy significantly improves the efficiency of the recommendation system and the user experience. By performing lightweight optimization on content based on the dynamic scoring model and conducting distributed cross-platform recommendation on the optimized content pool, the system can efficiently distribute optimized content between different devices and platforms, ensuring that users can enjoy accurate and fast recommendation services regardless of the platform they are on. The generation of the lightweight content pool and the execution of the cross-platform recommendation strategy effectively reduce the bandwidth requirements and computing resource consumption of content, enabling the recommendation system to maintain low latency and high stability even under high concurrency, greatly improving resource utilization, especially in resource-constrained devices and network conditions. Through the optimized recommended content pool, users can obtain faster and more accurate personalized recommendations, and even in a poor network environment, the system can still provide a smooth experience through lightweight content and intelligent loading strategies, thereby enhancing user satisfaction and user stickiness of the platform. The system can provide real-time feedback and adapt to changes in the network environment, dynamically adjusting the priority, display method, and loading strategy of recommended content, thus ensuring that in different usage scenarios, the system can flexibly respond to changes in real-time bandwidth fluctuations and device performance, improving the intelligence level of recommendations.

[0053] Preferably, step S42 includes the following steps:

[0054] Step S421: Perform weighted fusion on the device compatibility index and the time period recommendation weight to generate a platform scenario fusion weight;

[0055] Step S422: Screen the content to be recommended from the preset cross-platform content feature library according to the platform scenario fusion scenario weight and conduct data set division to generate a model training set and a model test set;

[0056] Step S423: Train the model training set through a convolutional neural network algorithm to generate a dynamic recommendation scoring pre-model; use the model test set to perform model optimization iteration on the dynamic recommendation scoring pre-model, thereby generating a dynamic recommendation scoring model.

[0057] Through weighted fusion of the device compatibility index and the time period recommendation weight, the system can comprehensively consider the device performance and the characteristics of the user's active time period, generate a more representative platform scenario fusion weight. This fusion strategy improves the adaptability of the model to different devices and usage scenarios, enabling the recommendation algorithm to perform personalized content recommendation on a more accurate basis. The screening of the content to be recommended based on the platform scenario fusion weight enables the data in the cross-platform content feature library to better conform to the feature distribution of user preferences. On this basis, the dataset is partitioned, and a model training set and a test set are generated, effectively reducing unnecessary interference and data noise in the model training process, improving the quality of the training data, and thus accelerating the learning and optimization of the model. The training set is trained using the convolutional neural network algorithm, and through continuous optimization and iteration, the model can better capture complex feature patterns and hierarchical data relationships. The powerful features of CNN enable the recommendation system to accurately model multiple factors such as user preferences, device adaptability, and time period requirements, further improving the accuracy and robustness of the dynamic recommendation scoring model. Using the model test set for optimization and iteration can not only finely tune the model but also ensure the generalization ability of the model under different datasets. Through this iterative optimization process, the dynamic recommendation scoring model can gradually reduce the overfitting phenomenon and improve its performance in practical applications, enabling the recommendation system to quickly adapt to the changing user needs and external environment. Through the training and optimization of the dynamic recommendation scoring model, the system can automatically adjust the recommendation strategy under different devices, network environments, and time period characteristics to ensure that users can receive the recommended content that best meets their needs. This highly customized recommendation ability not only enhances the user experience but also effectively improves the utilization rate of platform content and user stickiness. The training and optimization process of the model helps to improve the cross-platform consistency of the system and ensure the provision of stable and efficient recommendation services on different platforms or devices. Through refined model design and iterative optimization, the system can still maintain a high recommendation accuracy and response speed when facing different terminals and network environments. Through the platform scenario weight after weighted fusion and dataset partitioning, the system can more accurately handle the content recommendation tasks in different scenarios, realizing efficient data circulation and content screening. Combining the advantages of deep learning, the recommendation process becomes more intelligent and adaptive, further optimizing the personalized presentation method and recommendation strategy of the content.

[0058] In this specification, an AI-based multi-platform media content optimization recommendation system is provided for performing the above-mentioned AI-based multi-platform media content optimization recommendation method. The AI-based multi-platform media content optimization recommendation system includes:

[0059] A data collection module for obtaining user device parameters and user platform behavior data; and confirming user preferences based on the user platform behavior data to obtain user platform preference data;

[0060] An adaptation analysis module, which is used to analyze the hardware performance of the user's device according to the user device parameters, classify it, and generate a device adaptability matrix through the classification results; extract the user active time period from the user platform preference data to obtain user active time period feature data; analyze the content loading time consumption of the device adaptability matrix to generate device time consumption data;

[0061] A recommendation optimization module, which is used to extract format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library; calculate the matching degree between the device adaptability matrix and the format adaptation parameters to generate a device compatibility index; generate a time period recommendation weight by combining the device time consumption data with the user active time period feature data;

[0062] A content recommendation module, which is used to obtain the real-time fluctuation value of the network bandwidth; construct a dynamic recommendation scoring model through the device compatibility index and the time period recommendation weight, and perform lightweight recommendation of graphic and text content on the cross-platform content feature library based on the real-time fluctuation value of the network bandwidth to execute a multi-platform media content optimization recommendation task.

[0063] The beneficial effects of the present invention are as follows: The data collection module can accurately analyze user preferences by obtaining user device parameters and behavior data, and generate user platform preference data. This accurate capture of preferences provides a refined basis for personalized recommendations for the subsequent recommendation optimization module, ensuring that each user can receive content recommendations that meet their needs. The adaptation analysis module can generate a device adaptation matrix for each device by analyzing the hardware performance of the user device, and provide different adaptation strategies according to the performance grading of the device. This performance adaptation management helps to dynamically optimize the content loading strategy according to the actual situation of the device, avoiding content loading delays or incompatibility problems caused by low-performance devices. By extracting the characteristic data of the user's active period, the system can provide the best content recommendations for users at different times. For example, high-quality or video content is pushed during the user's active period, and lightweight content is pushed during the low-active period, thereby enhancing the user stickiness and content interaction rate of the platform. The adaptation analysis module generates device time-consuming data by analyzing the loading time of the device adaptation matrix, which enables the system to identify devices with loading delays and optimize the loading path and resource allocation accordingly, improving the content loading efficiency and user experience. The recommendation optimization module generates a device compatibility index by calculating the matching degree between the device adaptation matrix and the format adaptation parameters, further optimizing the adaptability of the recommended content. This compatibility index helps to provide content recommendations at different levels according to the device performance differences, ensuring the consistency and fluency between different devices on different platforms. By combining the characteristic data of the user's active period and the device time-consuming data, the system generates a time period recommendation weight. This strategy can push optimized content at the appropriate time period, improving the user experience while avoiding system overload or degradation of the user interface performance, especially during peak hours. The content recommendation module can adjust the recommendation strategy according to the network status by real-time monitoring of network bandwidth fluctuations and combining with a dynamic recommendation scoring model. This strategy ensures that in the case of large network fluctuations, the recommendation system can intelligently adjust the graphic and text loading method or compression of the content, reducing the loading delay or fluency degradation caused by unstable bandwidth. Therefore, the present invention improves the compatibility and response speed of cross-platform media content recommendations by comprehensively considering user preferences, device performance, period activity, content adaptability, and network fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a schematic diagram of the step flow of a multi-platform media content optimization recommendation method based on AI;

[0065] Figure 2 is Figure 1 a detailed implementation step flow diagram of step S2 in

[0066] Figure 3 is Figure 1 a detailed implementation step flow diagram of step S4 in

[0067] The realization, functional features, and advantages of the objectives of the present invention will be further described in conjunction with the embodiments and with reference to the accompanying drawings. Specific Embodiments

[0068] The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0069] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0070] It should be understood that although terms such as "first" and "second" may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed associated items.

[0071] To achieve the above object, please refer to Figures 1 to 3 , a multi-platform media content optimization and recommendation method based on AI, the method includes the following steps:

[0072] Step S1: Obtain user device parameters and user platform behavior data; confirm user preferences based on the user platform behavior data to obtain user platform preference data;

[0073] Step S2: Analyze the hardware performance of the user's device according to the user device parameters and classify it, and generate a device adaptability matrix through the classification results; extract the user's active time period from the user platform preference data to obtain user active time period feature data; analyze the content loading time consumption of the device adaptability matrix to generate device time consumption data;

[0074] Step S3: Extract the format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library; calculate the matching degree between the device adaptability matrix and the format adaptation parameters to generate a device compatibility index; generate a time period recommendation weight by combining device time-consuming data with user active period feature data;

[0075] Step S4: Obtain the real-time fluctuation value of the network bandwidth; construct a dynamic recommendation scoring model through the device compatibility index and the time period recommendation weight, and perform lightweight recommendation of graphic and text content on the cross-platform content feature library based on the real-time fluctuation value of the network bandwidth to execute the multi-platform media content optimization recommendation operation.

[0076] The present invention accurately depicts the content preferences of users at different times and on different devices by collecting user device parameters and platform behavior data and combining user active period features, and realizes the refined construction of personalized recommendation strategies. By grading the device hardware performance and generating a device adaptability matrix, and combining the content format adaptation parameters, the device compatibility index is calculated, so as to ensure the display and operation effects of the content on different devices and avoid problems such as loading failure or card lag. The time period recommendation weight jointly constructed by the user active period features and the device time-consuming data enables the recommendation system to push the most matching content during the time period when the user is most likely to be active, improving the user click-through rate and satisfaction. Combining the real-time fluctuation data of the network bandwidth, a real-time scoring mechanism is introduced into the recommendation model to realize the lightweight processing of graphic and text content, and still ensure the content loading and reading experience in a low-bandwidth environment, significantly optimizing the recommendation effect in a weak network environment. This method organically integrates device performance, user behavior, content features and network environment, provides a general and extensible optimization recommendation framework for cross-platform content distribution systems, and can be applied to various terminal forms such as APPs, applets, and web pages. Therefore, the present invention improves the compatibility and response speed of cross-platform media content recommendation by comprehensively considering user preferences, device performance, time period activity, content adaptability and network fluctuations.

[0077] In the embodiment of the present invention, with reference to Figure 1 As shown, it is a schematic diagram of the step flow of a multi-platform media content optimization recommendation method based on AI according to the present invention. In this example, the multi-platform media content optimization recommendation method based on AI includes the following steps:

[0078] Step S1: Obtain user device parameters and user platform behavior data; confirm user preferences based on the user platform behavior data to obtain user platform preference data;

[0079] In the embodiments of the present invention, parameters of a user device are collected through a device management tool (such as Google Analytics or Firebase), including hardware configurations (such as processor type, memory capacity, storage space, etc.), software environments (such as operating system version, browser type), and network status (Wi-Fi or mobile data), etc. Assume that the memory capacity of the user device is between 2GB and 8GB, and the storage space is between 32GB and 256GB. Through embedded event tracking codes (such as Mixpanel or Kissmetrics), behavioral data of users on the platform is collected in real time, such as browsing time of each page, click frequency, search records, purchase behavior, etc. For example, a certain user conducts an average of 20 searches per week, and the stay time for each time is 3 minutes. The collected behavioral data is stored in a big data platform (such as Amazon Redshift or Google BigQuery), and is cleaned and standardized to remove invalid data. Next, machine learning algorithms, such as collaborative filtering and K-means clustering, are used to analyze the behavioral data of users and construct a user preference model. For example, through the collaborative filtering algorithm, it can be found that a certain user likes product categories of electronic products, while another user's preferences focus on fashion and beauty products. Based on this data, a preference dataset for each user can be generated, including the product categories they frequently browse, the content they are interested in, etc. These data will be stored in a database (such as MongoDB), and the real-time nature of the preference data is ensured through a real-time update mechanism (such as Apache Kafka or Redis). Finally, these generated user preference data are input into a personalized recommendation system to recommend content, products, or services that match the users' interests.

[0080] Step S2: Analyze the hardware performance of the user device according to the user device parameters and classify it, and generate a device adaptability matrix through the classification results; extract the user active time period from the user platform preference data to obtain user active time period feature data; analyze the content loading time consumption of the device adaptability matrix to generate device time consumption data;

[0081] In the embodiments of the present invention, hardware parameter collection of user devices is performed through a hardware performance evaluation algorithm and a big data analysis platform, including information such as processor frequency, memory capacity, and storage space. For example, the device's memory is between 4GB and 16GB, and the processor frequency is between 2.0GHz and 3.5GHz. Through hardware performance testing tools (such as Geekbench or CPU-Z), a performance benchmark test is performed on each device to evaluate its processing ability. According to these test results, the devices are divided into different performance levels, such as high-performance devices (with more than 8GB of memory and a processor above 3.0GHz), medium-performance devices (with 6GB of memory and a 2.5GHz processor), and low-performance devices (with 4GB of memory and a processor below 2.0GHz). Based on these device performance levels, a device adaptability matrix is constructed to evaluate the content loading time, stability, etc. of different devices. Then, time series analysis and machine learning techniques (such as K-means clustering) are used to extract active period features from the user's behavior data. By analyzing the user's daily usage duration, access frequency, and time period, the user's active peak time period is determined. For example, a certain user is most active between 7 pm and 10 pm every day. Through the dynamic time warping (DTW) technique, the user's active pattern is further identified to generate feature data of the user's active period, which will help optimize content push and platform services. Finally, a performance monitoring tool (such as Lighthouse or Appium) is used to analyze the content loading time consumption of different devices. For example, on high-performance devices, the web page content is loaded within 2 seconds, while on low-performance devices, it takes 8 seconds. According to the loading time of different devices, a device time consumption data table is generated to help the platform identify performance bottlenecks and optimize them.

[0082] Step S3: Extract format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library; calculate the matching degree between the device adaptability matrix and the format adaptation parameters to generate a device compatibility index; generate a time period recommendation weight by combining the device time consumption data with the user active period feature data;

[0083] In the embodiments of the present invention, by extracting the format adaptation parameters of the content to be recommended from a preset cross-platform content feature library, the system can adjust the format of the recommended content according to the requirements of different platforms, such as video resolution, image size, audio quality, etc. For example, a video needs to be adapted to 720p on a mobile device, while 1080p content is provided on a desktop device. Then, the system uses a matching degree calculation model to generate a device compatibility index based on the matching degree between the device performance and the content format adaptation parameters. The device adaptability matrix records the performance of different devices in loading various types of content. By calculating the matching degree between the maximum content format supported by the device and the format of the recommended content, a compatibility index is generated to measure the adaptability of the device to the recommended content. The device compatibility index generates a comprehensive compatibility score based on the device performance and the supported format. For example, if the device supports 1080p content and the recommended content is 720p, the compatibility index is 100%; if the device only supports 720p, the compatibility index is lower. In addition, by combining the device time-consuming data and the user active period feature data, the system generates a period recommendation weight through a weighted average model. Specifically, a low-performance device takes a longer time to load content during the user active period, so the recommendation weight during this period is lower, while the recommendation weight of a high-performance device is higher. By calculating the comprehensive weighting of the device compatibility index, the device time-consuming index, and the user active period, the system can generate accurate content recommendation weights for different devices and users, thereby optimizing the recommendation strategy and improving the user experience and content loading efficiency.

[0084] Step S4: Obtain the real-time fluctuation value of the network bandwidth; construct a dynamic recommendation scoring model through the device compatibility index and the period recommendation weight, and perform lightweight recommendation of graphic and text content on the cross-platform content feature library based on the real-time fluctuation value of the network bandwidth, so as to execute the multi-platform media content optimization recommendation operation.

[0085] In the embodiments of the present invention, the network bandwidth of the user device is regularly monitored by integrating network monitoring tools (such as NetFlow, Wireshark, or Pingdom). By real-time sampling and recording network traffic data, the system can obtain the real-time fluctuation value of the bandwidth. Whenever the user device connects to the Internet, the system automatically records the real-time changes in network bandwidth, including upload and download speeds, latency times, etc. These data will be updated in real-time and recorded in minutes or seconds. The system will calculate the volatility and stability of the network based on the real-time fluctuation value of the bandwidth. For example, if the bandwidth is stable and high during a certain period, the content loading experience of the user device will be relatively good; while if the bandwidth fluctuates violently or is low, it will lead to loading delays and a decline in the user experience. Using a weighted scoring model and multi-factor decision analysis, combined with the device compatibility index, time period recommendation weight, and network bandwidth fluctuation value, a dynamic recommendation scoring model is constructed. According to the device compatibility index, time period recommendation weight, and network bandwidth fluctuation value, the system constructs a dynamic recommendation scoring model. This model will comprehensively calculate the scores of the recommended content based on factors such as different device performances, active time periods, and network bandwidth conditions. For example, if a user is on a high-performance device and in an active time period, and the network bandwidth is stable, the user's recommendation score will be higher; if the user's network bandwidth is unstable, the recommendation score will be appropriately reduced to avoid affecting the experience due to slow-loading content. The model assigns weights according to different factors: the device compatibility index, time period recommendation weight, and the fluctuation value of the network bandwidth, and these weights will be adjusted over time and user behavior. For example, during periods of low network bandwidth, the system will increase the proportion of lightweight content recommended to ensure that the recommended content can be loaded quickly and provide a better user experience. The calculation formula can be set as: DRS = w1×CI + w2×RW + w3×BW; where DRS is the dynamic recommendation score, CI is the device compatibility index, RW is the time period recommendation weight, BW is the network bandwidth fluctuation value, and w1, w2, w3 are the weight coefficients of the corresponding factors and can be adjusted according to actual needs. Using content lightweighting technology and multi-platform adaptation algorithms, the cross-platform content feature library is recommended and optimized based on the dynamic recommendation scoring model. Through lightweighting technology, such as image compression (JPEG, WebP), video resolution reduction (from 1080p to 720p or lower), audio compression (reducing the bit rate), etc., the system can recommend lightweight content that meets the user device and network conditions according to the device compatibility index and the real-time fluctuation of the network bandwidth. For users with low network bandwidth or poor device performance, the system preferentially recommends compressed images or low-resolution videos to avoid loading delays caused by heavy content. Using multi-platform adaptation algorithms, the system automatically adjusts the format of the recommended content according to the user's current device, platform, and network status. For example, for mobile devices, the system recommends smaller-sized pictures and low-resolution videos, while for desktop devices, it recommends higher-quality content.During the cross-platform recommendation process, the system will screen the cross-platform content feature library according to the dynamic recommendation scoring model, and give priority to recommending content that suits the current network and device environment, ensuring that the platform content can be presented to users in the most appropriate way. Finally, based on the results of the above model, the platform will execute the optimized content recommendation task, push the most suitable content through different platforms, and achieve content loading optimization and user experience improvement. For example, during periods when the user's network bandwidth is low, the platform will give priority to pushing lightweight graphic and text content, while when the network bandwidth is high, the platform will recommend high-quality multimedia content.

[0086] Preferably, step S1 includes the following steps:

[0087] Step S11: Obtain user device parameters and user platform behavior data using network protocols;

[0088] Step S12: Conduct usage log analysis and scenario feature extraction on the user platform behavior data to generate user behavior feature data, where the user behavior feature data includes usage feature data and usage scenario feature data;

[0089] Step S13: Conduct user interaction pattern analysis based on the usage feature data to generate user interaction pattern data;

[0090] Step S14: Conduct behavior clustering analysis on the user interaction pattern data to generate user behavior clustering data; combine the user device parameter data and the user behavior clustering data to conduct personalized preference modeling and generate user preference feature data;

[0091] Step S15: Calculate the preference weights of the user in different usage scenarios for the user preference feature data according to the usage scenario feature data to generate user platform preference data.

[0092] In the embodiments of the present invention, by connecting to the network protocol of the user device, standard protocols (such as HTTP header information, WebSocket, etc.) are used to obtain device parameter information. Device parameters include device model, operating system version, CPU, memory, screen resolution, network status (such as bandwidth and latency), etc. The basic information of the device is obtained using the User-Agent and HTTP request headers. The operation behavior data of the user on the platform is collected through JavaScript scripts or API interfaces. The behavior data includes, but is not limited to, activities such as clicks, browsing, searching, video viewing, product browsing, etc. The collected data is stored in the database in real time and marked based on timestamps. Log analysis frameworks (such as ELK Stack, Apache Flume) are used to analyze the user's behavior logs to extract key behavior features, such as page access frequency, click time, click type, user stay duration, etc. The specific scenarios in the behavior data are classified and feature extracted. For example, the behavior characteristics of users in the shopping scenario (browsing products, adding to cart, checking out, etc.) are significantly different from those in the social scenario (liking, commenting, sharing). These behavior characteristics are classified through machine learning algorithms (such as clustering analysis, feature selection) to obtain behavior data for different usage scenarios. Based on the extracted scenario features and log data, user behavior feature data including usage feature data (such as active period, access frequency, click volume, etc.) and usage scenario feature data (such as shopping scenario, entertainment scenario, social scenario) is generated. Clustering algorithms (such as K-Means, DBSCAN, etc.) are used to classify the user's usage feature data to identify the user's interaction patterns. For example, some users tend to quickly browse and skip content, while others are more inclined to deeply participate and stay on the page for a long time. The interaction behavior of each user is converted into interaction pattern data, such as "quick browsing type", "deep participation type", "intermittent access type", etc., and each user is marked with their main interaction pattern. The clustering analysis method is used to process the user interaction pattern data, and clustering is performed based on data such as the user's interaction frequency, stay duration, interaction type, etc. For example, users are divided into "active users", "occasional users", and "silent users", etc. According to the results of the clustering analysis, user behavior clustering data is generated, different user behavior patterns are mapped to specific clustering categories, and each category is given a label, such as "high activity users", "low activity users", etc. The device parameters of the user are combined with the behavior clustering data, and collaborative filtering or deep learning recommendation models (such as neural collaborative filtering) are used for preference modeling. The model provides personalized content recommendations for users based on their behavior patterns on different devices and in different scenarios.After completing the personalized preference modeling, the system generates preference feature data for each user, including information such as the content type, usage time, and device selection preferred by the user. These preference feature data can accurately reflect the personalized needs of the user and provide a basis for subsequent recommendations. According to the behavior and needs of the user in different scenarios, the system combines the scenario feature data with the preference feature data and calculates the preference weights of the user in different usage scenarios through weighted algorithms (such as weighted average method, TF-IDF model). For example, in a social scenario, the user prefers content for easy interaction, while in a work scenario, the user prefers professional and information-intensive content. Through weight calculation, the system generates user platform preference data for different scenarios. The preference data for each scenario will include the user's preference for specific content types, interaction frequency, usage habits, etc., to help the platform make more accurate content recommendations.

[0093] Preferably, step S15 includes the following steps:

[0094] Step S151: Perform high-dimensional vector mapping transformation on the user preference feature data to generate user multi-dimensional preference vector data; perform time-series dynamic clustering on the usage scenario feature data to generate a scenario dynamic feature matrix;

[0095] Step S152: Perform cross-scenario preference projection based on the user multi-dimensional preference vector data and the scenario dynamic feature matrix to generate a user scenario projection parameter set;

[0096] Step S153: Use the user scenario projection parameter set to perform adaptive preference weight allocation calculation to generate a user scenario adaptive weight matrix;

[0097] Step S154: Perform deep feature fusion on the user scenario projection parameter set according to the user scenario adaptive weight matrix and perform multi-level normalization processing to calculate the final preference weights of the user in different usage scenarios and generate user platform preference data.

[0098] In the embodiments of the present invention, high-dimensional vector mapping is performed on user preference feature data by using a deep learning model (such as an autoencoder or Word2Vec). This method maps multi-dimensional information such as user behavior data and preference data into a high-dimensional vector of a fixed length. Through deep learning frameworks such as TensorFlow or PyTorch, a neural network model is designed and trained to obtain multi-dimensional preference vector data of users. Time series clustering analysis is performed on the usage scenario feature data, and time series clustering algorithms (such as K-means, DBSCAN, dynamic time warping (DTW), etc.) are used to perform dynamic clustering on the usage scenario data in different time periods. Through time series clustering, the behavior characteristics of users in different time periods are aggregated to capture the dynamic changes of scenarios in different time periods. The user preferences (such as viewing duration, click volume, usage frequency, etc.) are converted into high-dimensional vectors and mapped into a continuous vector space by using a neural network model. Time series dynamic clustering is performed on the usage scenario data, and the clustering algorithm is used to extract the scenario features in each time period according to the changes in the time dimension. Through the generation of the time series dynamic feature matrix, the usage patterns and preference changes of users in different time periods can be analyzed. Based on the multi-dimensional preference vector data of users and the scenario dynamic feature matrix, matrix factorization algorithms or projection algorithms are used to map these data into a low-dimensional space for fusion. This operation aims to discover the preference relationships of users in different scenarios. Principal component analysis (PCA) or non-linear dimensionality reduction (t-SNE) is used to project the user preferences and scenario feature data across scenarios to obtain a set of user scenario projection parameters. Based on the generated set of user scenario projection parameters, an adaptive weighting algorithm is used to assign weights to the preferences of users in different scenarios, which can be achieved through weighted regression, reinforcement learning strategies, or neural network weighted learning. The system will automatically adjust the preference weights in each scenario according to the behavior patterns and scenario features of users in different scenarios. For example, the preferences of users in the entertainment scenario have a higher weight, while the preferences in the work scenario are lower. Using the weighting algorithm, the preference weights in each scenario are calculated and a user scenario adaptive weight matrix is formed. This matrix represents the weight distribution of users in each usage scenario and reflects the influence of each scenario on user behavior. Based on the set of user scenario projection parameters and the user scenario adaptive weight matrix, feature fusion is performed through a deep neural network (such as a multi-layer perceptron (MLP)). The neural network will deeply mine the behavior patterns of users according to the feature data and weights of different scenarios. In the process of deep feature fusion, techniques such as Batch Normalization or Layer Normalization are applied to normalize the fused features to reduce data bias and improve the recommendation effect.After deep fusion and normalization, the user's preference weights in different usage scenarios are finally calculated. These weights represent the user's demand intensity in a specific scenario and can provide a more accurate basis for subsequent content recommendations. Finally, based on the calculated user scenario preference weights, the system generates user platform preference data to provide input for the personalized recommendation algorithm. These data reflect the user's different preferences in various usage scenarios, enabling the platform to optimize recommended content in a targeted manner.

[0099] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0100] Step S21: Evaluate the computing power of the user equipment parameters to generate user equipment performance evaluation data; perform a hierarchical clustering analysis of the hardware computing power of the user equipment parameters based on the user equipment performance evaluation data to generate a device performance grading label set;

[0101] Step S22: extracting the device driver compatibility and rendering capability of the user device parameters, and constructing an adaptability matrix for the device performance grading label set to generate a device adaptability matrix;

[0102] Step S23: performing time series pattern recognition on the user platform preference data, extracting the user interaction peak features, and generating user active period trend data; performing time series distribution calculation on the user active period trend data, and generating user active period feature data;

[0103] Step S24: simulate the content loading path of the device adaptability matrix, calculate the loading delay of different devices in different scenarios, and generate device time consumption data.

[0104] In the embodiments of the present invention, a performance analysis tool (such as Geekbench, PassMark, or Android Profiler, etc.) is used to perform performance tests on the user device to evaluate the hardware capabilities of the device, such as CPU performance, GPU performance, memory bandwidth, storage speed, etc. Based on these test results, user device performance evaluation data is generated, which contains various performance indicators of the device and its performance in specific tasks. A clustering algorithm (such as K-means, DBSCAN) is used to analyze the user device performance evaluation data, and the devices are hierarchically classified according to their performance. For example, clustering can be performed based on multi-dimensional parameters such as CPU frequency, GPU capabilities, RAM capacity, etc., to generate a hierarchical label set of device performance (such as high, medium, low levels). Obtain the hardware performance data of the user device, including multiple performance indicators such as CPU, GPU, and memory. Use a clustering analysis algorithm (such as K-means) to classify the devices, and group devices with different performances into the same category (such as high-performance devices, medium-low performance devices). Assign corresponding performance grading labels to each device, such as high-end devices, mid-range devices, and low-end devices. Extract the driver version and rendering capabilities of the device, such as OpenGL version, DirectX support, GPU driver compatibility, etc., to evaluate whether the device supports certain specific hardware acceleration and graphics rendering requirements. Use system APIs or hardware diagnostic tools (such as OpenGL Extensions Viewer or DirectX Diagnostic Tool) to extract these device characteristics. Combine the device performance grading label set with data such as rendering capabilities and driver compatibility, and construct an adaptability matrix of the device through matrix analysis. The adaptability matrix will show the rendering compatibility and driver support of different devices in different scenarios, so as to determine whether the device can run specific content smoothly. Extract relevant information from the device driver compatibility and rendering capabilities, and analyze whether each device meets specific hardware rendering requirements. According to the performance grading label of the device, combined with its driver compatibility and rendering capabilities, generate an adaptability matrix, indicating the adaptability performance of each device in a specific content scenario. Use a time series analysis algorithm (such as dynamic time warping (DTW) or long short-term memory network (LSTM)) to analyze the user's platform preference data and identify the user's behavior patterns on the platform. Through the identification of time series patterns, extract the interactive peak features in the user's behavior, such as peak access times, active times, etc., to form user active time trend data. Use time series analysis tools (such as ARIMA, Holt-Winters, etc.) to perform distribution calculations on the user active time trend data, and generate features such as active frequency and active intensity within a time period. According to the analysis results of the active time period, generate user active time period feature data, reflecting the user's active periodicity and time period preference. Use a time series analysis algorithm to perform pattern recognition on the user's preference data and extract interactive peak features.Through time series distribution calculation, based on the user's active time period data, calculate the active period characteristic data to further understand the user's active pattern at different times. Through load testing tools (such as WebPageTest, Lighthouse, or LoadRunner), simulate the content loading path under different devices and scenarios, and evaluate the latency, bandwidth utilization, page rendering time, etc. of the user device during the loading process. Use simulation modeling (such as Monte Carlo simulation) to simulate the device loading process, considering factors such as different network conditions, hardware performance, and application scenarios, and calculate the loading latency. Based on the loading path simulation, collect the loading latency data of each device under different scenarios to form device time-consuming data, and perform weighted calculation according to the hardware performance and network conditions of the device to obtain the specific loading time of each device. Use the content loading path simulation tool to simulate the loading latency on different devices, considering factors such as device performance and network conditions. Calculate the loading latency and generate device time-consuming data to provide a basis for subsequent content loading optimization.

[0105] Preferably, step S24 includes the following steps:

[0106] Step S241: Extract the device characteristics of the device adaptability matrix to generate device performance data; perform network condition matching on the device performance data to generate network status data;

[0107] Step S242: Analyze the storage medium performance of the network status data, and evaluate the system overhead of the device adaptability matrix based on the storage medium performance to generate system load data;

[0108] Step S243: Develop a content chunking strategy for the system load data to generate chunking scheme data; perform priority sorting on the chunking scheme data to generate loading sequence data;

[0109] Step S244: Analyze the dependency relationship of the loading sequence data to generate dependency graph data; optimize the loading path of the dependency graph data to generate path scheme data;

[0110] Step S245: Perform multi-dimensional loading latency calculation on the path scheme data to obtain device time-consuming data, where the multi-dimensional loading latency calculation includes transmission latency, storage read / write latency, decoding and rendering latency, and system scheduling latency.

[0111] In the embodiments of the present invention, hardware characteristics of each device, such as CPU performance, memory capacity, GPU rendering ability, etc., are extracted from the device adaptability matrix to form device performance data. The device performance data includes the processing ability of the device in specific tasks, such as graphics rendering, data calculation speed, etc. Network monitoring tools (such as iPerf, Wireshark, etc.) are used to obtain information such as the network bandwidth, latency, packet loss rate, etc. of the device. The network condition data is matched with the device performance data to generate network status data, indicating the performance of different devices under different network conditions. Specifically, the hardware characteristics of the device are extracted from the device adaptability matrix. The network status data of the device is obtained, including information such as network bandwidth and latency. The device performance data is matched with the network status data to evaluate the performance of the device in the actual network environment. The storage performance of the device is analyzed through IO performance testing tools (such as CrystalDiskMark, HD Tune), and the read and write speed, latency, random access performance, etc. are evaluated. The performance of the storage medium will affect operations such as content loading and cache reading, so these factors need to be considered for their impact on the device. Performance analysis tools (such as Intel VTune, Windows Performance Monitor) are used to evaluate the system overhead of the device. Based on the storage medium performance and the hardware performance of the device, the overall performance of the device is evaluated to generate system load data for optimizing the subsequent content loading path. Specifically, storage performance analysis tools are used to test the storage read and write speed, random access ability, etc. of the device. According to the storage medium performance and hardware configuration, a system load evaluation tool is used to calculate the overall system load data of the device and record the bottlenecks. According to the system load data, a reasonable content chunking strategy is formulated. The chunking strategy should reasonably divide and load the content according to the storage performance, computing power, and network conditions of the device. A dynamic chunking algorithm (such as the divide-and-conquer algorithm) is used to chunk the content so that each chunk can be loaded at the most suitable time. After the content is chunked, a priority sorting algorithm (such as time-based sorting, resource requirement-based sorting) is used to sort the chunks to generate loading sequence data, indicating the loading priority order of different content chunks on different devices. Specifically, a content chunking strategy is formulated according to the system load data, and the content is divided into multiple chunks. The chunked content is sorted by priority to generate loading sequence data to ensure that the device first loads the most important or most needed content. Graph theory algorithms (such as topological sorting, DAG graph analysis) are used to analyze the dependency relationships between different content chunks to determine which content chunks need to be loaded first and which can be loaded in parallel. A dependency graph data is constructed to represent the loading dependencies of the content chunks in the form of a graph. Based on the dependency relationships, a path optimization algorithm (such as Dijkstra's algorithm) is used to optimize the content chunk loading path to reduce latency and resource consumption, generating an optimized path scheme data, that is, the optimal loading order in the case of multi-task parallel loading.Calculate the transmission delay between the computing content from the network server to the device, considering factors such as network bandwidth and latency. Calculate the latency of data reading from the storage medium to the memory according to the device storage performance. Evaluate the latency of the device decoding content and rendering graphics, which is especially crucial during the loading process of image and video content. Calculate the time latency of how the operating system schedules each process when the device is processing multiple tasks concurrently. Perform a weighted sum of the latencies in each dimension to obtain the total latency time of each device when loading content, thereby generating device time-consuming data.

[0112] Preferably, the generation of the device compatibility index by matching the calculated device adaptability matrix with the format adaptation parameters in step S3 includes:

[0113] Perform eigen-decomposition on the device adaptability matrix, extract the key performance adaptation factors, and obtain the device adaptation eigenvector set, where the formula for eigen-decomposition is as follows: Where, is the device adaptation eigenvector set, are respectively the singular value decomposition components of the device adaptability matrix; perform normalization on the format adaptation parameters and convert them into a high-dimensional vector representation to obtain the format adaptation vector : Construct the format adaptation parameter matrix : Calculate the matching score between the device adaptation eigenvector and the format adaptation parameter matrix to obtain the adaptation matching score matrix, where the calculation formula is as follows: Calculate the normalized matching score of the adaptation matching score matrix : Extract the computing power vector of the device in the device adaptability matrix Calculate the device adaptation weight distribution: Calculate the weighted matching score: Perform fuzzy clustering analysis on the weighted matching score to calculate the device adaptation stability index : Where, is the index number of the device, is the total number of devices, is the fuzzy membership function: In the formula, is the smoothing parameter, is the adaptation threshold;

[0114] Calculate the device compatibility index through the device adaptation stability index, weighted matching score, and normalized matching score to obtain the device compatibility index The calculation formula for the device compatibility index is as follows: In the formula, is the weight parameter.

[0115] Preferably, the generation of the time period recommendation weight by combining the device time-consuming data with the user active time period in step S3 includes:

[0116] Perform time interval segmentation on the device time-consuming data to generate multi-time period device performance response data;

[0117] Perform density modeling on the user active time period feature data to generate time period active density distribution data;

[0118] Perform time period mapping alignment on the multi-time period device performance response data and the time period active density distribution data to generate time period performance active mapping data;

[0119] Perform weighted normalization on the time period performance active mapping data to generate a basic time period adaptation coefficient; perform reverse fitting of the adjustment coefficient on the basic time period adaptation coefficient in combination with the cross-platform content feature library to generate a corrected time period adaptation weight;

[0120] Perform exponential smoothing and outlier correction on the corrected time period adaptation weight to generate a smoothed time period weight curve;

[0121] Perform structured compression on the smoothed time period weight curve to generate a time period recommendation weight.

[0122] In the embodiments of the present invention, by collecting the performance data of the device at different time periods, which usually includes CPU usage rate, memory usage, network latency, etc. The device time-consuming data is segmented according to a specific time interval (such as hours, half an hour, 10 minutes, etc.) to form multiple time-period data sets. Analyze the device performance within each time interval to generate multi-time-period device response data and record the performance metrics of the device at different time periods. Collect the time-period feature data of user activity, which usually includes user active duration, active frequency, platform access volume, etc. Use statistical methods (such as kernel density estimation or Gaussian mixture model) to perform density modeling on the user active time periods to obtain the distribution model of the active time periods. This model can reflect the peak and trough time periods of user activity within a day. By generating the time-period activity distribution data through density modeling, the active density within each time interval can be obtained. Map and align the device performance response data with the user active density distribution data in terms of time to ensure that each device performance data point can correspond to a user activity value. Through the aligned data, generate a time-period performance-active mapping data set, which reflects the relationship between the device performance and user activity at different time periods. Perform weighted normalization on the time-period performance-active mapping data to ensure that the relationship between device performance and user activity can be compared and analyzed on a unified scale. Through the weighted-normalized data, calculate the basic time-period adaptation coefficient, which reflects the adaptation degree of the device performance and user activity within each time period. According to the cross-platform content feature library (for example, differences in user behavior between platforms, device adaptation differences, etc.), adjust the basic time-period adaptation coefficient. Use the inverse fitting method, combined with cross-platform features, to adjust the time-period adaptation coefficient and generate the corrected time-period adaptation weights, which will take into account the differences in device performance and user active characteristics on different platforms. Perform exponential smoothing on the corrected time-period adaptation weights. Remove the outlier values with large fluctuations through the smoothing algorithm to make the weight curve smoother. Further perform outlier correction on the smoothed time-period weight curve to identify and correct the existing abnormal data points. Compress the smoothed time-period weight curve through data compression algorithms (such as principal component analysis, low-rank matrix factorization, etc.) to reduce redundant information. Finally, generate the structured and compressed time-period recommendation weights, which can be used in the recommendation system to help accurately match the user active time period with the device performance and optimize the recommendation results.

[0123] As an example of the present invention, refer to Figure 3 shown, in this example, step S4 includes:

[0124] Step S41: Perform bandwidth sampling on the network interface to obtain bandwidth sampling data; perform bandwidth fluctuation analysis on the bandwidth sampling data to generate the real-time network bandwidth fluctuation value;

[0125] Step S42: Construct a dynamic recommendation scoring model through the device compatibility index and the time period recommendation weight;

[0126] Step S43: Based on the real-time fluctuation value of the network bandwidth, use the dynamic recommendation scoring model to execute a lightweight optimization strategy for the cross-platform content feature library, and generate a lightweight recommended content pool, where the lightweight optimization strategy includes image compression, video bitrate reduction, and text preloading;

[0127] Step S44: Perform distributed cross-platform content recommendation on the lightweight recommended content pool to execute the multi-platform media content optimization recommendation task.

[0128] In the embodiments of the present invention, bandwidth sampling is periodically performed on the device side (such as client applications, smart devices, etc.). The sampling method can be based on network performance metrics such as network throughput and response time, and record bandwidth data at fixed time intervals (for example, sampling once per second). The obtained data includes network upload / download speed, latency, packet loss rate, etc. The sliding window method is used to perform real-time fluctuation analysis on the bandwidth sampling data to detect the change trend of the bandwidth. Through time-domain analysis methods (such as standard deviation calculation, peak analysis, etc.), real-time fluctuation values of the network bandwidth are generated. These fluctuation values reflect the fluctuation amplitude of the current network state and can be used to evaluate the stability of the bandwidth. Calculate the device compatibility index according to the hardware configuration of the device (such as processor capacity, memory capacity, display resolution, etc.). The device compatibility index represents the processing capacity and compatibility of the device in different network environments. The higher the index, the better the device can support high-quality multimedia content. Use the time period recommendation weight (the time period adaptation weight generated by step S3) to reflect the activity level of users and the device compatibility at different time periods. By combining the device compatibility index and the time period recommendation weight, a dynamic recommendation scoring model is constructed. The scoring model can optimize content recommendations according to the current device status and time period. According to the bandwidth fluctuation value and device performance, formulate lightweight optimization strategies for different network conditions and device statuses. The lightweight optimization strategies include: dynamically adjusting the resolution and compression ratio of pictures according to the real-time fluctuation value of the bandwidth. When the bandwidth is low, automatically compress the picture quality to reduce the bandwidth burden. Dynamically adjust the video bitrate according to network fluctuations. When the network bandwidth fluctuates greatly, reduce the video resolution or bitrate to ensure smooth playback. In the case of limited bandwidth, give priority to loading text content and delay loading image and video content, thereby improving the page loading efficiency. Analyze and optimize the content in the cross-platform content feature library, execute lightweight optimization strategies based on the real-time fluctuation value of the network bandwidth and device compatibility, and generate a lightweight content pool to ensure that the recommended content can be presented in the most optimized way in different devices and network environments. Based on the lightweight recommended content pool, distribute content to multiple platforms (such as mobile devices, PC terminals, smart TVs, etc.) through a distributed content recommendation system. The system will adjust the recommended content according to dynamic factors such as the device compatibility index, user time period activity characteristics, and network bandwidth fluctuations to ensure that each platform can receive content suitable for the current conditions. Perform content format adaptation and distribution strategy adjustment between different platforms. For example, mobile devices give priority to receiving compressed pictures and videos, while PC terminals can receive high-definition content. Achieve cross-platform synchronous optimization to ensure that the recommended content on different devices maintains an efficient and smooth experience in different network environments.

[0129] Preferably, step S42 includes the following steps:

[0130] Step S421: Perform weighted fusion on the device compatibility index and the time period recommendation weight to generate a platform scenario fusion weight;

[0131] Step S422: Screen the content to be recommended in the preset cross-platform content feature library according to the platform scenario fusion scenario weight and perform data set division to generate a model training set and a model test set;

[0132] Step S423: Train the model on the model training set through the convolutional neural network algorithm to generate a dynamic recommendation score pre-model; use the model test set to perform model optimization iteration on the dynamic recommendation score pre-model, thereby generating a dynamic recommendation score model.

[0133] In the embodiments of the present invention, the device compatibility index is calculated according to the hardware configuration of the device (such as processor capacity, memory, screen resolution, etc.). The higher the device compatibility index, the stronger the support ability of the device for different content types. This index can be comprehensively evaluated based on the computing power, storage capacity, network ability, etc. of the device. According to the time period when the user is active, combined with the time period recommendation weight to reflect the device adaptation ability in different time periods. The time period recommendation weight usually comes from the recommendation weight calculation in step S3, indicating the adaptability between the user's activity degree and the device performance in different time periods. The device compatibility index and the time period recommendation weight are weighted and fused, combining the information of both to generate the platform scenario fusion weight. The fusion method can use weighted average, matrix operation or weighted function: W_fusion = α ⋅ W_compatible + β ⋅ W_period; where W_fusion is the platform scenario fusion weight, and α and β are weight coefficients, respectively representing the contribution degrees of device compatibility and time period recommendation weight to the platform scenario fusion weight. According to the generated platform scenario fusion weight, the preset cross-platform content feature library is screened. The platform scenario fusion weight will select the content type suitable for the current environment according to the influence of device compatibility and user active time period. Compressed images, videos with reduced bit rate, and text content preferentially loaded will be screened out for low-compatibility devices or time periods with unstable network bandwidth. According to the screened content, the data set is divided. Usually, the screened content is divided into two parts: Model training set: Used for the training process of the model, containing data for learning and optimizing recommendation rules. Model test set: Used for testing and validating the model to ensure the generalization ability of the recommendation algorithm. The division method is generally carried out according to the ratio of 70%:30% for the training set and the test set. The model training set is trained using a convolutional neural network (CNN). The application of CNN in the recommendation system is usually used to extract features in high-dimensional data. Especially when the input data is an image or text, CNN can automatically learn the latent features of the content. For cross-platform content recommendation, the multi-dimensional features of the content (such as image content, video content, user interaction, etc.) can be converted into a format that CNN can process, and feature learning is carried out through convolutional layers, pooling layers, etc. After being trained by CNN, a preliminary dynamic recommendation scoring pre-model is generated. This model can generate the score of the recommended content according to information such as device compatibility, time period recommendation weight, and content features input. The recommended scoring pre-model obtained by preliminary training is optimized using the model test set. The model optimization adjusts the parameters of the model by comparing the gap between the actual result and the expected result. For example, gradient descent is carried out through the backpropagation algorithm to adjust the weights and biases of the neural network. In the process of iterative optimization, methods such as cross-validation and early stopping can be adopted to prevent overfitting, and the recommendation accuracy is improved by continuously adjusting the model parameters. After iterative optimization, the finally generated dynamic recommendation scoring model can dynamically recommend the most suitable content for users according to factors such as device compatibility, network bandwidth, and user active time period.

[0134] Therefore, in all respects, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be embraced within the present invention.

[0135] The above description is only a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will conform to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-platform media content optimization recommendation method based on AI, characterized in that, It includes the following steps: Step S1: Obtain user device parameters and user platform behavior data; Based on the user platform behavior data, confirm the user preferences to obtain user platform preference data; Step S2: Analyze the user's device hardware performance according to the user device parameters and classify it, and generate a device adaptability matrix through the classification results; Extract the user's active time period from the user platform preference data to obtain user active time period characteristic data; Analyze the content loading time consumption of the device adaptability matrix to generate device time consumption data; Step S3: Extract format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library; Calculate the matching degree between the device adaptability matrix and the format adaptation parameters to generate a device compatibility index; generate a time period recommendation weight by combining the device time consumption data with the user active time period characteristic data; Step S4: Obtain the real-time fluctuation value of the network bandwidth; construct a dynamic recommendation scoring model through the device compatibility index and the time period recommendation weight, and perform lightweight recommendation of graphic and text content on the cross-platform content feature library based on the real-time fluctuation value of the network bandwidth to execute a multi-platform media content optimization recommendation operation.

2. The AI-based multi-platform media content optimization and recommendation method according to claim 1, wherein Step S1 includes the following steps: Step S11: Use the network protocol to obtain user device parameters and user platform behavior data; Step S12: Perform usage log analysis and scenario feature extraction on the user platform behavior data to generate user behavior characteristic data, where the user behavior characteristic data includes usage characteristic data and usage scenario characteristic data; Step S13: Perform user interaction mode analysis based on the usage characteristic data to generate user interaction mode data; Step S14: Perform behavior clustering analysis on the user interaction mode data to generate user behavior clustering data; combine the user device parameter data and the user behavior clustering data to perform personalized preference modeling to generate user preference characteristic data; Step S15: Calculate the preference weights of the user in different usage scenarios according to the usage scenario characteristic data for the user preference characteristic data to generate user platform preference data.

3. The AI-based multi-platform media content optimization and recommendation method according to claim 1, wherein, Step S15 includes the following steps: Step S151: Perform high-dimensional vector mapping transformation on the user preference characteristic data to generate user multi-dimensional preference vector data; perform time-series dynamic clustering on the usage scenario characteristic data to generate a scenario dynamic feature matrix; Step S152: Perform cross-scenario preference projection based on the user multi-dimensional preference vector data and the scenario dynamic feature matrix to generate a user scenario projection parameter set; Step S153: Use the user scenario projection parameter set to perform adaptive preference weight distribution calculation to generate a user scenario adaptive weight matrix; Step S154: Perform deep feature fusion on the user scenario projection parameter set according to the user scenario adaptive weight matrix, and perform multi-level normalization processing to calculate the final preference weights of the user in different usage scenarios to generate user platform preference data.

4. The AI-based multi-platform media content optimization and recommendation method according to claim 1, wherein Step S2 includes the following steps: Step S21: Evaluate the device computing power of the user device parameters to generate user device performance evaluation data; perform hardware computing ability hierarchical clustering analysis on the user device parameters through the user device performance evaluation data to generate a device performance classification label set; Step S22: Extract the device driver compatibility and rendering capabilities of the user device parameters, construct an adaptation matrix for the device performance grading label set, and generate a device adaptation matrix; Step S23: Perform temporal pattern recognition on the user platform preference data, extract the user interaction peak features, and generate user active period trend data; perform time series distribution calculation on the user active period trend data to generate user active period feature data; Step S24: Simulate the content loading path of the device adaptation matrix, calculate the loading delay of different devices in different scenarios, and generate device time-consuming data.

5. The AI-based multi-platform media content optimization and recommendation method according to claim 4, wherein Step S24 includes the following steps: Step S241: Extract the device features of the device adaptation matrix to generate device performance data; perform network condition matching on the device performance data to generate network status data; Step S242: Analyze the storage medium performance of the network status data, and perform system overhead evaluation on the device adaptation matrix based on the storage medium performance to generate system load data; Step S243: Develop a content chunking strategy for the system load data to generate chunking scheme data; perform priority sorting on the chunking scheme data to generate a loading sequence data; Step S244: Analyze the dependency relationship of the loading sequence data to generate dependency graph data; optimize the loading path of the dependency graph data to generate path scheme data; Step S245: Perform multi-dimensional loading delay calculation on the path scheme data to obtain device time-consuming data, where the multi-dimensional loading delay calculation includes transmission delay, storage read / write delay, decoding and rendering delay, and system scheduling delay.

6. The AI-based multi-platform media content optimization recommendation method according to claim 1, wherein, The calculation of the device compatibility index by calculating the matching degree between the device adaptation matrix and the format adaptation parameters in Step S3 includes: Perform eigenvalue decomposition on the device adaptability matrix, extract the key performance adaptation factors, and obtain the device adaptation feature vector set. The formula for eigenvalue decomposition is as follows: Among them, is the device adaptation feature vector set, are the singular value decomposition components of the device adaptability matrix respectively; normalize the format adaptation parameter and convert it into a high-dimensional vector representation to obtain the format adaptation vector : Construct the format adaptation parameter matrix : Calculate the matching degree score between the device adaptation feature vector and the format adaptation parameter matrix to obtain the adaptation matching degree score matrix. The calculation formula is as follows: Calculate the normalized matching degree score of the adaptation matching degree score matrix : Extract the computing power vector of the device in the device adaptability matrix and calculate the device adaptation weight distribution: Calculate the weighted matching score: Perform fuzzy clustering analysis on the weighted matching score and calculate the device adaptation stability index : Among them, is the index number of the device, is the total number of devices, is the fuzzy membership function: In the formula, is the smoothing parameter, is the adaptation threshold; calculate the device compatibility index through the device adaptation stability index, weighted matching score, and normalized matching degree score to obtain the device compatibility index The calculation formula of the device compatibility index is as follows: In the formula, is the weight parameter.

7. The AI-based multi-platform media content optimization and recommendation method according to claim 1, wherein The generation of the time period recommendation weight by combining the device time-consuming data with the user active period in Step S3 includes: Perform time interval segmentation on the device time-consuming data to generate multi-period device performance response data; Perform density modeling on the user active period feature data to generate period active density distribution data; Perform period mapping alignment on the multi-period device performance response data and the period active density distribution data to generate period performance active mapping data; Perform weighted normalization on the period performance active mapping data to generate a basic period adaptation coefficient; perform reverse fitting of the adjustment coefficient on the basic period adaptation coefficient in combination with the cross-platform content feature library to generate a corrected period adaptation weight; Perform exponential smoothing and outlier correction on the corrected period adaptation weight to generate a smoothed period weight curve; Perform structured compression on the smoothed period weight curve to generate a period recommendation weight.

8. The AI-based multi-platform media content optimization and recommendation method according to claim 1, characterized in that Step S4 includes the following steps: Step S41: Perform bandwidth sampling on the network interface to obtain bandwidth sampling data; perform bandwidth fluctuation analysis on the bandwidth sampling data to generate a real-time network bandwidth fluctuation value; Step S42: Construct a dynamic recommendation scoring model through the device compatibility index and the period recommendation weight; Step S43: Based on the real-time fluctuation value of the network bandwidth, execute a lightweight optimization strategy on the cross-platform content feature library using a dynamic recommendation scoring model to generate a lightweight recommended content pool, where the lightweight optimization strategy includes image compression, video bitrate reduction, and text preloading; Step S44: Perform distributed cross-platform content recommendation on the lightweight recommended content pool to execute a multi-platform media content optimization recommendation job.

9. The AI-based multi-platform media content optimization and recommendation method according to claim 8, wherein Step S42 includes the following steps: Step S421: Perform weighted fusion on the device compatibility index and the time period recommendation weight to generate a platform scenario fusion weight; Step S422: Screen the content to be recommended in the preset cross-platform content feature library according to the platform scenario fusion scenario weight and perform data set division to generate a model training set and a model test set; Step S423: Train the model training set through a convolutional neural network algorithm to generate a dynamic recommendation scoring pre-model; use the model test set to perform model optimization iteration on the dynamic recommendation scoring pre-model to generate a dynamic recommendation scoring model.

10. An AI-based multi-platform media content optimization and recommendation system, characterized in that, For executing the AI-based multi-platform media content optimization recommendation method as described in claim 1, the AI-based multi-platform media content optimization recommendation system includes: A data collection module, configured to obtain user device parameters and user platform behavior data; confirm user preferences based on the user platform behavior data to obtain user platform preference data; An adaptation analysis module, configured to analyze the hardware performance of the user's device according to the user device parameters and classify it, and generate a device adaptability matrix through the classification result; extract the user's active time period from the user platform preference data to obtain user active time period feature data; perform content loading time-consuming analysis on the device adaptability matrix to generate device time-consuming data; A recommendation optimization module, configured to extract format adaptation parameters of the content to be recommended based on a preset cross-platform content feature library; calculate the matching degree between the device adaptability matrix and the format adaptation parameters to generate a device compatibility index; generate a time period recommendation weight by combining the device time-consuming data with the user active time period feature data; A content recommendation module, configured to obtain the real-time fluctuation value of the network bandwidth; construct a dynamic recommendation scoring model through the device compatibility index and the time period recommendation weight, and perform lightweight graphic content recommendation on the cross-platform content feature library based on the real-time fluctuation value of the network bandwidth to execute a multi-platform media content optimization recommendation job.

Images