AI-based cross-platform advertising material lifecycle management and optimization method and system

Abstract

The present invention discloses an AI-based cross-platform advertising material lifecycle management and optimization method and system, the method comprising: collecting raw data and metadata to construct a heterogeneous data graph; performing layered graph convolution to extract cross-modal features; splicing and dimensionality alignment of the cross-modal features and performing feature dimensionality reduction to obtain unified features; constructing gene individuals based on the unified features, generating cross-platform materials through a multi-objective constrained co-evolutionary algorithm; generating material push strategies based on cross-platform material mapping, optimizing budget allocation and push strategies by constructing a layered game framework, and combining federal constraint synchronization to obtain a multi-platform collaboratively optimized push strategy set; modeling material decay laws and calculating material survival functions to provide real-time warnings of material decay lifecycles and trigger adjustments to material push strategies. The present invention solves technical problems such as insufficient dynamic strategy collaborative optimization and lagging prediction of material lifecycle decay in cross-platform advertising material management.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an AI-based cross-platform advertising material lifecycle management and optimization method and system. Background Art

[0002] As the advertising industry continues to digitize, the management and optimization of advertising creatives have become increasingly crucial for advertisers to improve advertising effectiveness and reduce costs. However, existing methods for managing the lifecycle of advertising creatives have the following problems:

[0003] Cross-platform data fragmentation: Advertising creative delivery data on multiple platforms (such as social media, search engines, and short video platforms) cannot be analyzed uniformly, resulting in delayed strategy adjustments; high manual dependence: Creative generation relies on designer experience, and delivery strategies rely on manual parameter adjustment, which is inefficient and difficult to scale; Insufficient dynamic adaptability: Traditional A / B testing cycles are long, unable to respond to market changes in real time, and lack automated optimization mechanisms based on user behavior feedback.

[0004] The Chinese patent with publication number CN117391776A discloses an advertising delivery method, apparatus, device and medium. The method includes: selecting corresponding advertising materials according to the type of advertisement to be delivered, wherein the advertising materials are stored in an advertising material library constructed based on a historical advertising data set; determining the element components and promotional copy of the advertisement to be delivered based on the advertising materials; generating a target advertising page and a target delivery plan corresponding to the advertisement to be delivered based on the element components of the advertisement to be delivered and the promotional copy of the advertisement to be delivered; delivering the target advertising page according to the target delivery plan, and recording the delivery log of the advertisement to be delivered, so as to optimize the target advertising page and the target delivery plan based on the delivery log. This invention lacks cross-platform dynamic collaborative optimization, cannot achieve global optimization of material push strategy, and may cause imbalance in cross-platform resource competition; and this invention relies on post-optimization of historical logs, which makes it difficult to trigger material iteration and strategy adjustment in a timely manner, and there is a delivery lag. Summary of the Invention

[0005] The purpose of the present invention is to provide an AI-based cross-platform advertising material lifecycle management and optimization method and system to solve the technical problems of difficulty in integrating multi-source heterogeneous data, insufficient dynamic strategy collaborative optimization, and lagging prediction of material lifecycle decline in cross-platform advertising material management.

[0006] The technical solutions of the present invention are as follows:

[0007] In one aspect, the present invention provides an AI-based cross-platform advertising material lifecycle management and optimization method, comprising the following steps:

[0008] Raw data and metadata are collected from different platforms, and a heterogeneous data graph is constructed based on the raw data and metadata. The node types in the heterogeneous data graph are defined as user nodes, material nodes, and context nodes, and the edge types are defined as click relationship edges, browsing relationship edges, modality association edges, and user-context edges.

[0009] Based on the heterogeneous data graph structure, hierarchical graph convolution is performed on each node according to the corresponding relationship type to extract cross-modal features; cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features, and the high-dimensional features are subjected to feature dimensionality reduction through the attention mechanism to obtain unified features after dimensionality reduction.

[0010] Core semantic items are constructed based on unified features, and genetic individuals composed of core semantic items, physical constraint items, and creative expression items are constructed. Cross-platform materials are generated through a multi-objective constraint co-evolution algorithm.

[0011] A material push strategy is generated based on the generated cross-platform material mapping. Budget allocation and push strategy are optimized by constructing a hierarchical game framework including offline Nash equilibrium pre-calculation and online No-regret learning. Combined with federated constraint synchronization, a set of push strategies for multi-platform collaborative optimization is obtained.

[0012] By modeling the material decay rules and calculating the material survival function, real-time warnings of material decay life cycle can be provided and adjustments to the material push strategy can be triggered.

[0013] Preferably, based on the heterogeneous data graph structure, a hierarchical graph convolution is performed on each node corresponding to the relationship type, and the cross-modal feature is extracted and expressed as:

[0014]

[0015] Where, , , is a set of predefined relationship types; For nodes In the Features of layer graph convolution, express dimensional real vector space; is a nonlinear activation function; For nodes In relationship The set of neighbor nodes under ; For the relationship In the The learnable weight matrix of the layer graph convolution is obtained through back-propagation training; For nodes In the Features of layer graph convolution; , is the predefined number of graph convolution layers.

[0016] Preferably, feature dimensionality reduction is performed on high-dimensional features through the attention mechanism as follows:

[0017] Perform global feature extraction on the high-dimensional features of all nodes after layered graph convolution in the heterogeneous data graph, expressed as:

[0018]

[0019] Where, is the global average feature, express dimensional real vector space; is the total number of nodes in the heterogeneous data graph; For nodes high-dimensional features.

[0020] Calculate the attention weight of each node:

[0021]

[0022] Where, is the attention weight, indicating the importance of each target dimension; is the normalization function; is a learnable matrix, mapping the concatenated features to dimension, ; For splicing operation.

[0023] Based on the concatenated high-dimensional features and the global average features, the obtained attention weights are weighted summed to obtain the unified features after dimensionality reduction.

[0024] Preferably, constructing a gene individual consisting of core semantic items, physical constraint items, and creative expression items, and generating cross-platform materials through a multi-objective constraint co-evolution algorithm is as follows:

[0025] S1: Construct core semantic items based on unified features, construct physical constraint items by sampling from the multi-platform constraint rule library, and construct creative expression items by probabilistic sampling based on the historical material library and the predefined style probability matrix. The core semantic items, physical constraint items, and creative expression items are encoded to generate gene individuals to form the main population for global search, and TOPP gene individuals are selected from the main population to form a local population for single-platform search.

[0026] S2: During single-generation evolution, a multi-objective fitness function is calculated for each gene individual in the main population and the local population. Elite-retention selection, crossover recombination, and directed mutation operations are used for evolution. The weights of each objective in the multi-objective fitness function are adjusted through an adaptive optimization strategy to generate a new generation of main population and local population.

[0027] S3: Repeat step S2 and perform cross-population collaborative operations to optimize the local population. After the stopping conditions of reaching the maximum number of iterations or the convergence of the multi-objective fitness function are met, the multi-objective constrained collaborative evolution algorithm is terminated, and the gene individual set with the highest fitness of the main population and the local population is output to generate cross-platform materials.

[0028] Preferably, the multi-objective fitness function calculation is expressed as:

[0029]

[0030] Where, is the multi-objective fitness function; For genetic individuals; For genetic individuals on the platform The parameter vector of is the total number of platforms; For the platform Semantic standard vector of ; For the platform The semantic weight of is the semantic matching degree; For the The satisfaction of the item constraints; is the total number of constraints; To constrain the penalty coefficient, it is adaptively adjusted through each generation of evolution; score creative diversity; is the diversity bonus coefficient.

[0031] Preferably, a material push strategy is generated based on the generated cross-platform material mapping. Budget allocation and push strategy are optimized by constructing a hierarchical game framework including offline Nash equilibrium pre-calculation and online No-regret learning. Combined with federated constraint synchronization, a multi-platform collaboratively optimized push strategy set is obtained, specifically:

[0032] Each platform is modeled as an intelligent agent. The strategy of the intelligent agent includes: the budget allocation ratio of the platform and the material push strategy generated according to the generated cross-platform material mapping. The initial budget allocation ratio of each platform is determined by offline Nash equilibrium pre-calculation using the Lemke-Howson algorithm.

[0033] Combining real-time ROI and user behavior data, the No-regret learning mechanism optimizes each platform's strategy. Through federated constraints, the budget boundary information of each platform is synchronously exchanged to ensure that the global budget constraint is met. The optimized push strategy set is output. The strategy update rule of the No-regret learning mechanism is as follows:

[0034]

[0035] Where, For the platform Optional policy adjustment actions; For the platform exist Always choose your action probability; For the platform exist Always choose your action probability; To remove the platform All other platforms except For other platforms Joint strategies at all times; For action immediate benefits; It is a dynamic learning rate, which is adjusted according to the freshness of the data; Adjust actions for any candidate policy; For the platform exist Always choose your action probability.

[0036] Preferably, the material decay law is defined as follows:

[0037]

[0038] Where, For the platform exist The intensity of the material click at the moment; is the baseline decay rate; To remove the platform External platforms; is the cross-platform propagation coefficient; Affects the decay rate across platforms; For the moment; For the platform exist The event increment at the moment; Number the external event type; For events Impact strength; For events Indicator function that occurred.

[0039] The survival function of the material is defined as:

[0040]

[0041] Where, is the survival function, representing the platform The material is The probability that the moment has not decayed; is the integration variable.

[0042] A multi-level decay level response mechanism is preset based on the survival function value. The decay level of the material is warned in real time according to the calculated survival function value, and the material push strategy is adjusted according to the adjustment method corresponding to the material decay level triggered by the response mechanism.

[0043] On the other hand, the present invention provides an AI-based cross-platform advertising material lifecycle management and optimization system, including a heterogeneous data graph construction module, a cross-modal feature extraction and dimensionality reduction module, a cross-platform material generation module, a material push strategy generation and optimization module, and a material lifecycle warning and push strategy adjustment module.

[0044] A heterogeneous data graph construction module is used to collect raw data and metadata from different platforms and construct a heterogeneous data graph based on the raw data and metadata. The node types in the heterogeneous data graph are defined as user nodes, material nodes, and context nodes, and the edge types are defined as click relationship edges, browsing relationship edges, modal association edges, and user-context edges.

[0045] The cross-modal feature extraction and dimensionality reduction module is used to perform layered graph convolution of the corresponding relationship type on each node based on the heterogeneous data graph structure to extract cross-modal features; the cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features, and the high-dimensional features are reduced in dimension through the attention mechanism to obtain unified features after dimensionality reduction.

[0046] The cross-platform material generation module is used to construct core semantic items based on unified features, and to construct genetic individuals composed of core semantic items, physical constraint items, and creative expression items, and to generate cross-platform materials through a multi-objective constraint co-evolution algorithm.

[0047] The material push strategy generation and optimization module is used to generate material push strategies based on the generated cross-platform material mapping. It optimizes budget allocation and push strategies by constructing a hierarchical game framework that includes offline Nash equilibrium pre-calculation and online No-regret learning. Combined with federal constraint synchronization, it obtains a set of push strategies for multi-platform collaborative optimization.

[0048] The material life cycle warning and push strategy adjustment module is used to provide real-time warning of material life cycle decline and trigger material push strategy adjustments by modeling material decay laws and calculating material survival functions.

[0049] On the other hand, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the AI-based cross-platform advertising material lifecycle management and optimization method as described in any embodiment of the present invention is implemented.

[0050] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the AI-based cross-platform advertising material lifecycle management and optimization method as described in any embodiment of the present invention.

[0051] Compared with the prior art, the present invention has the following technical effects:

[0052] 1. This invention solves the data silo problem by modeling multi-type relationships among users, materials, and context nodes, integrating cross-platform behaviors, content attributes, and environmental factors. It also improves the ability to characterize complex user-material interactions by extracting cross-modal features and performing dimensionality reduction, thus supporting the generation of precise push strategies.

[0053] 2. This invention combines core semantics (user preferences), physical constraints (platform rules), and creative expression (content form) to design individual genes, and conducts multi-objective co-evolution to generate differentiated materials that conform to the characteristics of multiple platforms, avoiding creative homogeneity caused by single-objective optimization.

[0054] 3. This invention pre-calculates the initial strategy combination through offline Nash equilibrium to ensure the balance of interests among multiple platforms, and dynamically adjusts the strategy based on real-time ROI and user retention data to adapt to market changes.

[0055] 4. By quantifying the cross-platform dissemination effect and the impact of external events, the present invention realizes real-time early warning and precise management of the material life cycle, ensuring that the delivery strategy always matches the actual benefits of the material, thereby avoiding ineffective delivery and waste of resources, and improving the accuracy and effectiveness of the cross-platform material push strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is an overall flow chart of the AI-based cross-platform advertising material lifecycle management and optimization method described in the present invention. DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0058] Example 1

[0059] This embodiment provides an AI-based cross-platform advertising material lifecycle management and optimization method, see Figure 1 As shown, the following steps are included:

[0060] Raw data and metadata are collected from different platforms, and a heterogeneous data graph is constructed based on the raw data and metadata. Node types in this heterogeneous data graph are defined as user nodes, material nodes, and context nodes, and edge types are defined as click-related edges, browse-related edges, modality-related edges, and user-context edges. Specifically, the raw data includes, but is not limited to, user behavior (number of clicks, dwell time, etc.), material content (image pixel matrix, video frame sequence, etc.), and context (such as timestamp and geolocation). Metadata includes, but is not limited to, material ID (unique identifier), platform type, and user tags (such as age, gender, and interests). The user node is a feature vector containing user profile features, the material node is a feature vector containing material metadata, and the context node is a feature vector containing platform environment features. Click-related edges are defined as users clicking on materials, representing user behavioral preferences; browse-related edges are defined as users browsing materials, representing the depth of user interest. Modality-related edges are defined as materials linking video frames, linking materials to content details; and user-context edges are defined as users linking contexts, representing the temporal patterns of user behavior.

[0061] Based on the heterogeneous data graph structure, hierarchical graph convolution is performed on each node according to the corresponding relationship type to extract cross-modal features; cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features, and the high-dimensional features are subjected to feature dimensionality reduction through the attention mechanism to obtain unified features after dimensionality reduction.

[0062] As a preferred implementation of this embodiment, based on the heterogeneous data graph structure, a layered graph convolution of the corresponding relationship type is performed on each node to extract cross-modal features represented as follows:

[0063]

[0064] Where, , , is a set of predefined relationship types; For nodes In the Features of layer graph convolution, express dimensional real vector space; It is a nonlinear activation function, such as the LeakyReLU function, to enhance the nonlinear expression ability; For nodes In relationship The set of neighbor nodes under ; For the relationship In the The learnable weight matrix of the layer graph convolution is randomly initialized and trained through backpropagation; For nodes In the The features of the layer graph convolution are initially the original features of the nodes and are updated layer by layer through graph convolution; , The number of layers is predefined. There is no limit on the number of layers you can choose. You can set it according to your actual needs. , directly learn neighbor nodes; set , which can capture second-order neighbor nodes and expand the perception domain (such as “the video frame associated with the material clicked by user A”).

[0065] The cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features. Specifically, for different types of nodes, zero padding is performed on the non-participating relationship types to align the feature dimensions. For example, for user nodes, The dimension of relationship type is completed.

[0066] As a preferred implementation of this embodiment, feature dimensionality reduction is performed on high-dimensional features through the attention mechanism as follows:

[0067] Perform global feature extraction on the high-dimensional features of all nodes after layered graph convolution in the heterogeneous data graph, expressed as:

[0068]

[0069] Where, is the global average feature, express dimensional real vector space; is the total number of nodes in the heterogeneous data graph; For the High-dimensional features of nodes.

[0070] Calculate the attention weight of each node:

[0071]

[0072] Where, is the attention weight, indicating the importance of each target dimension; It is a normalization function used to normalize the weights into a probability distribution to ensure that the sum of the weights of each dimension is 1; is a learnable matrix, mapping the concatenated features to dimension, randomly initialized and optimized through training, ; For splicing operation; For nodes The eigenvector of .

[0073] Based on the concatenated high-dimensional features and the global average features, the obtained attention weights are weighted summed to obtain the unified features after dimensionality reduction, specifically:

[0074]

[0075] Where, As the projection matrix, the spliced 8d-dimensional features are mapped to dimension.

[0076] The unified features of the output capture the comprehensive semantics of cross-platform users, materials, and context from the following feature dimensions: user behavior including click patterns (such as the frequency and time distribution of users' clicks on different platforms), dwell time (such as the average dwell time of users on images, texts, videos, and other materials), cross-platform conversion paths, etc.; material content including visual features (such as the HSV value of image color matching, dynamic elements of videos), text keywords, multimodal associations (such as the semantic matching degree between images, texts, and videos), etc.; context features including time context (such as delivery time period, holiday markings), geographic context (such as the compatibility between the user's geographic location and the regional location of the materials), device context (such as the display difference between mobile and PC terminals), etc.

[0077] Core semantic items are constructed based on unified features, and genetic individuals composed of core semantic items, physical constraint items, and creative expression items are constructed. Cross-platform materials are generated through a multi-objective constraint co-evolution algorithm.

[0078] As a preferred implementation of this embodiment, a gene individual consisting of core semantic items, physical constraint items, and creative expression items is constructed, and cross-platform materials are generated through a multi-objective constraint co-evolution algorithm as follows:

[0079] S1: Core semantic items are constructed based on unified features, physical constraints are sampled from a multi-platform constraint rule library, and creative expression items are constructed by probabilistic sampling based on a historical material library and a predefined style probability matrix. Gene individuals are generated from the encoding of the core semantic items, physical constraints, and creative expression items to form a main population for global search. Top 10 gene individuals are selected from the main population to form a local population for single-platform search (by restricting the search space of the local population to only allow parameter combinations on the same platform, improving platform adaptation accuracy). Specifically, the core semantic items directly inherit the unified features and contain abstract semantic information such as user behavior and cross-modal relationships. The physical constraint items are encoded by dynamically loading the technical specification parameters of each platform from the multi-platform rule library. The creative expression items include the encoding of specific creative elements such as copywriting, visuals, and sound effects.

[0080] S2: During single-generation evolution, a multi-objective fitness function is calculated for each individual gene in the main population and local population. Evolution is performed using elite-retention selection, crossover recombination, and directed mutation. The weights of each objective in the multi-objective fitness function are adjusted through an adaptive optimization strategy to generate a new generation of the main population and local population. Furthermore, the elite-retention operation is specifically performed by retaining the individuals with the highest fitness (e.g., set to 20%) and directly advancing them into the next generation. The crossover recombination operation is specifically performed by using a high crossover rate to mix different platform parameters in the main population, while the local population focuses on optimizing the same platform parameters. The directed mutation operation is specifically performed by adjusting the mutation direction based on real-time feedback. Furthermore, the multi-constraint co-evolution algorithm also performs real-time data fusion, including collecting user behavior to receive real-time data streams such as user click heat maps and platform rule changes to update semantic weights, adjust gene mutation directions, and dynamically update the constraint library.

[0081] S3: Repeat step S2 and perform cross-population coordination to optimize the local population. The multi-objective constrained co-evolutionary algorithm terminates when the stopping condition of reaching the maximum number of iterations or convergence of the multi-objective fitness function is met. The sets of genetic individuals with the highest fitness in the main population and the local population are output, generating cross-platform material. Specifically, the cross-population coordination involves performing elite migration every X generations (e.g., 5 generations), injecting the best genetic individuals (e.g., the top 5%) from the main population into the local population to replace the same number of the worst genetic individuals in the local population. The main population output is the set of genetic individuals with the highest global fitness (e.g., the top 10%), covering different platform combinations. The local population output is the set of optimal candidate genetic individuals (e.g., the top 5) retained for each platform.

[0082] As a preferred implementation of this embodiment, the multi-objective fitness function is evaluated by semantic matching, platform constraint satisfaction (checking whether parameters such as resolution and duration meet the requirements of each platform), and creative diversity score (comparing the degree of difference between the current gene and historical materials), and is expressed as:

[0083]

[0084] Where, is the multi-objective fitness function; For genetic individuals; For genetic individuals on the platform The parameter vector of is the total number of platforms; For the platform The semantic standard vector is the unified feature obtained by dimensionality reduction; For the platform The semantic weight of is the semantic matching degree, which is generally calculated by cosine similarity; For the The satisfaction of the item constraints; is the total number of constraints; To constrain the penalty coefficient, the adaptive optimization strategy is adjusted through each generation of evolution; score creative diversity; is the diversity bonus coefficient.

[0085] The semantic weight The calculation is as follows:

[0086]

[0087] The constraint penalty coefficient Adaptive optimization strategy adjustment through each generation of evolution:

[0088]

[0089] Where, For the Generation constraint penalty coefficient; for Generation constraint penalty coefficient; is the total number of genetic individuals in the current main / local population.

[0090] The degree of satisfaction The calculation is as follows:

[0091]

[0092] Where, is the parameter deviation; These are the hard limit thresholds for material parameters on each platform (such as duration constraint thresholds and resolution constraint thresholds).

[0093] Furthermore, the parameter deviation The calculation is as follows:

[0094]

[0095] Where, is the total number of parameters of individual genes, ; For parameters Importance weights can be preset in the platform rule library and dynamically adjusted based on real-time user click data; For the genetic individual parameters; For the platform For parameters The hard limit threshold.

[0096] The diversity bonus coefficient The calculation is as follows:

[0097]

[0098] Where, is the initial value of the diversity reward coefficient; a repeated batch is counted as one repeated batch each time the similarity of consecutive preset generations of creative ideas is detected to be greater than the set percentage.

[0099] The Creative Diversity Score The calculation is as follows:

[0100]

[0101] A material push strategy is generated based on the generated cross-platform material mapping. Budget allocation and push strategy are optimized by constructing a hierarchical game framework including offline Nash equilibrium pre-calculation and online No-regret learning. Combined with federated constraint synchronization, a set of push strategies for multi-platform collaborative optimization is obtained.

[0102] As a preferred implementation of this embodiment, the steps are specifically as follows:

[0103] Each platform is modeled as an agent, and the agent's strategy includes: platform Budget allocation ratio And generate a material push strategy based on the generated cross-platform material mapping, and use the Lemke-Howson algorithm to perform offline Nash equilibrium pre-calculation to determine the initial budget allocation ratio of each platform.

[0104] Furthermore, the Lemke-Howson algorithm is used to pre-calculate the offline Nash equilibrium to determine the initial budget allocation ratio of each platform:

[0105] Define the global optimization objective as the profit function :

[0106]

[0107] Where, is the total number of platforms, ; For the platform ROI benefits, It is a fairness adjustment factor used to balance total revenue and fairness among platforms; the Gini coefficient is used to measure the inequality of platform revenue.

[0108] The initial continuous budget allocation ratio in the agent strategy is discretized into a finite action set, and the corresponding benefit function is calculated for each strategy combination to construct a payment matrix, where the elements of the payment matrix are the benefit function values of the corresponding strategy combination.

[0109] Through complementary pivot iteration, find satisfaction The equilibrium point is that the editing utility of each platform strategy is zero, and the overall benefit value cannot be improved by slightly adjusting the budget allocation. The initial strategy combination of each platform is output.

[0110] Combining real-time ROI and user behavior data, the No-regret learning mechanism optimizes each platform's strategy. Through federated constraints, the budget boundary information of each platform is synchronously exchanged to ensure that the global budget constraint is met. The optimized push strategy set is output. The strategy update rule of the No-regret learning mechanism is as follows:

[0111]

[0112] Where, For the platform Optional strategic adjustment actions (including budget allocation adjustment and material push strategy selection); For the platform exist Always choose your action probability; For the platform exist Always choose your action probability; To remove the platform All other platforms except For other platforms Joint strategies at all times; For action immediate benefits; It is a dynamic learning rate, which is adjusted according to the freshness of the data; Adjust actions for any candidate policy; For the platform exist Always choose your action probability.

[0113] Furthermore, the No-regret learning mechanism can be driven to update based on timing (update at fixed time intervals) or events (update is triggered when ROI fluctuation exceeds a threshold).

[0114] The implementation process of the federal constraint synchronization is as follows: Each platform encrypts the local budget ratio ; Collaborative Decryption , verify whether the global budget constraint is met; if it exceeds the limit, compress the budget of each platform proportionally to ensure .

[0115] As a preferred implementation of this embodiment, revenue improvement verification is also performed to ensure that the adjustment of the material push strategy is in line with the overall interests of the intelligent system composed of various platforms. The revenue improvement verification is specifically as follows:

[0116] Calculate the revenue changes of each platform:

[0117]

[0118] Where, For the platform Changes in earnings, For the platform Changing to strategy After the income; For the platform In implementing strategy The income when.

[0119] If exists And other platforms , the policy change is accepted.

[0120] By modeling the material decay law and calculating the material survival function, a real-time warning of the material decay life cycle is provided and the material push strategy adjustment is triggered. As a preferred implementation method of this embodiment, the above steps are specifically as follows:

[0121] The material decay law is defined as follows:

[0122]

[0123] Where, For the platform exist The click intensity of the material at a certain moment can be obtained by fitting the historical click features in the unified features after dimensionality reduction. is the baseline decay rate, which is fitted by the click decay curve without external intervention; To remove the platform External platforms; is the cross-platform propagation coefficient; Affects the decay rate across platforms; For the moment; For the platform exist The event increment at the moment; Number the external event type; For events The impact strength is verified by double difference method, positive value means improvement and negative value means inhibition; For events Indicator function, event that occurred The value is 1 when the function is enabled and 0 when the function is disabled.

[0124] Furthermore, the cross-platform propagation coefficient and the cross-platform impact attenuation rate are estimated and calculated using the EM algorithm: Assuming the current parameters, the number of potential cross-platform impact events is calculated:

[0125]

[0126] Where, is the number of potential cross-platform impact events, indicating the platform exist Always on the platform exist The click contribution ratio at the moment.

[0127] Update the cross-platform propagation coefficient and the cross-platform influence attenuation rate to maximize the log-likelihood function until the parameters converge, and obtain and .

[0128] The survival function of the material is defined as:

[0129]

[0130] Where, is the survival function, representing the platform The material is The probability that the moment has not decayed; is the integral variable, which represents the temporary variable in the integration process.

[0131] A multi-level decay response mechanism is preset based on the survival function value. The decay level of the material is warned in real time according to the calculated survival function value, and the adjustment method corresponding to the decay level of the material is triggered according to the response mechanism to guide the previous step to dynamically adjust the budget allocation and material push strategy to avoid ineffective delivery. The total number of level divisions is not limited here and can be divided according to actual needs. The following is a simple strategy adjustment example: when the survival function value of a certain material reaches the first decay level, the delivery frequency of the material is reduced, and the initial budget allocation ratio of each platform is dynamically adjusted. At the same time, it generates iterative instructions such as "Replace the promotional slogan with "Limited Time Discount"".

[0132] Example 2

[0133] Accordingly, this embodiment provides an AI-based cross-platform advertising material lifecycle management and optimization system, which is used to implement the AI-based cross-platform advertising material lifecycle management and optimization method as described in Example 1 of the present invention, including a heterogeneous data graph construction module, a cross-modal feature extraction and dimensionality reduction module, a cross-platform material generation module, a material push strategy generation and optimization module, and a material lifecycle warning and push strategy adjustment module.

[0134] A heterogeneous data graph construction module is used to collect raw data and metadata from different platforms and construct a heterogeneous data graph based on the raw data and metadata. The node types in the heterogeneous data graph are defined as user nodes, material nodes, and context nodes, and the edge types are defined as click relationship edges, browsing relationship edges, modal association edges, and user-context edges.

[0135] The cross-modal feature extraction and dimensionality reduction module is used to perform layered graph convolution of the corresponding relationship type on each node based on the heterogeneous data graph structure to extract cross-modal features; the cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features, and the high-dimensional features are reduced in dimension through the attention mechanism to obtain unified features after dimensionality reduction.

[0136] The cross-platform material generation module is used to construct core semantic items based on unified features, and to construct genetic individuals composed of core semantic items, physical constraint items, and creative expression items, and to generate cross-platform materials through a multi-objective constraint co-evolution algorithm.

[0137] The material push strategy generation and optimization module is used to generate material push strategies based on the generated cross-platform material mapping. It optimizes budget allocation and push strategies by constructing a hierarchical game framework that includes offline Nash equilibrium pre-calculation and online No-regret learning. Combined with federal constraint synchronization, it obtains a set of push strategies for multi-platform collaborative optimization.

[0138] Example 3

[0139] This embodiment provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the AI-based cross-platform advertising material lifecycle management and optimization method as described in the first embodiment of the present invention is implemented.

[0140] Example 4

[0141] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for managing and optimizing the lifecycle of cross-platform advertising materials based on AI as described in the first embodiment of the present invention is implemented.

[0142] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c or a and b and c, where a, b, c can be single or multiple.

[0143] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented using a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0144] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0145] In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), magnetic disk or optical disk, and other media that can store program code.

[0146] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structures or equivalent process changes made using the contents of the present invention's description and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An AI-based cross-platform advertising material lifecycle management and optimization method, characterized by: The following steps are involved: Collect raw data and metadata from different platforms and construct a heterogeneous data graph based on the raw data and metadata. The node types in the heterogeneous data graph are defined as user nodes, material nodes, and context nodes, and the edge types are defined as click-related edges, browsing-related edges, modality-related edges, and user-context edges. Based on the heterogeneous data graph structure, layered graph convolution is performed on each node according to the corresponding relationship type to extract cross-modal features; Cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features. The high-dimensional features are then dimensionality reduced through the attention mechanism to obtain unified features after dimensionality reduction. Construct core semantic items based on unified features, and construct gene individuals composed of core semantic items, physical constraint items, and creative expression items, and generate cross-platform materials through a multi-objective constraint co-evolution algorithm; Based on the generated cross-platform material mapping, a material push strategy is generated. Budget allocation and push strategies are optimized by building a hierarchical game framework that includes offline Nash equilibrium pre-calculation and online no-regret learning. Combined with federated constraint synchronization, a set of push strategies for multi-platform collaborative optimization is obtained. By modeling the material decay rules and calculating the material survival function, real-time warnings of material decay life cycle can be provided and adjustments to the material push strategy can be triggered.

2. The AI-based cross-platform advertising material lifecycle management and optimization method according to claim 1, characterized in that: Based on the heterogeneous data graph structure, a layered graph convolution of the corresponding relationship type is performed on each node to extract cross-modal features represented as: Where, , , is a set of predefined relationship types; For nodes In the Features of layer graph convolution, express dimensional real vector space; is a nonlinear activation function; For nodes In relationship The set of neighbor nodes under ; For the relationship In the The learnable weight matrix of the layer graph convolution is obtained through back-propagation training; For nodes In the Features of layer graph convolution; , is the predefined number of graph convolution layers.

3. The AI-based cross-platform advertising material lifecycle management and optimization method according to claim 1, characterized in that: The feature dimensionality reduction of high-dimensional features through the attention mechanism is as follows: Perform global feature extraction on the high-dimensional features of all nodes after layered graph convolution in the heterogeneous data graph, expressed as: Where, is the global average feature, express dimensional real vector space; is the total number of nodes in the heterogeneous data graph; For nodes High-dimensional features; Calculate the attention weight of each node: Where, is the attention weight, indicating the importance of each target dimension; is the normalization function; is a learnable matrix, mapping the concatenated features to dimension, ; For splicing operation; Based on the concatenated high-dimensional features and the global average features, the obtained attention weights are weighted summed to obtain the unified features after dimensionality reduction.

4. The AI-based cross-platform advertising material lifecycle management and optimization method according to claim 1, characterized in that: Construct a gene individual consisting of core semantic items, physical constraint items, and creative expression items, and generate cross-platform materials through a multi-objective constraint co-evolution algorithm. Specifically: S1: Construct core semantic items based on unified features, construct physical constraint items by sampling from a multi-platform constraint rule library, and construct creative expression items by probabilistic sampling based on a historical material library and a predefined style probability matrix. Gene individuals are generated by encoding the core semantic items, physical constraint items, and creative expression items to form a main population for global search. The top P gene individuals are selected from the main population to form a local population for single-platform search. S2: In a single-generation evolution, a multi-objective fitness function is calculated for each gene individual in the main population and the local population. The evolution is carried out using elite-retention selection, crossover recombination, and directed mutation operations. The weights of each objective in the multi-objective fitness function are adjusted through an adaptive optimization strategy to generate a new generation of the main population and local population. S3: Repeat step S2 and perform cross-population collaborative operations to optimize the local population. After the stopping conditions of reaching the maximum number of iterations or the convergence of the multi-objective fitness function are met, the multi-objective constrained collaborative evolution algorithm is terminated, and the gene individual set with the highest fitness of the main population and the local population is output to generate cross-platform materials.

5. The AI-based cross-platform advertising material lifecycle management and optimization method according to claim 4, characterized in that: The multi-objective fitness function calculation is expressed as: Where, is the multi-objective fitness function; For genetic individuals; For genetic individuals on the platform parameter vector of ; is the total number of platforms; For the platform Semantic standard vector of ; For the platform The semantic weight of is the semantic matching degree; For the The satisfaction of the item constraints; is the total number of constraints; To constrain the penalty coefficient, it is adaptively adjusted through each generation of evolution; score creative diversity; is the diversity bonus coefficient.

6. The AI-based cross-platform advertising material lifecycle management and optimization method according to claim 1, characterized in that: Based on the generated cross-platform material mapping, a material push strategy is generated. By building a hierarchical game framework that includes offline Nash equilibrium pre-calculation and online No-regret learning to optimize budget allocation and push strategies, combined with federated constraint synchronization, the push strategy set for multi-platform collaborative optimization is obtained as follows: Each platform is modeled as an intelligent agent. The agent's strategy includes: budget allocation ratio for each platform and a material push strategy generated based on the generated cross-platform material mapping. The initial budget allocation ratio for each platform is determined by offline Nash equilibrium pre-calculation using the Lemke-Howson algorithm. Combining real-time ROI and user behavior data, the No-regret learning mechanism optimizes each platform's strategy. Through federated constraints, the budget boundary information of each platform is synchronously exchanged to ensure that the global budget constraint is met. The optimized push strategy set is output. The strategy update rule of the No-regret learning mechanism is as follows: Where, For the platform Optional policy adjustment actions; For the platform exist Always choose your action probability; For the platform exist Always choose your action probability; To remove the platform All other platforms except For other platforms Joint strategies at all times; For action immediate benefits; It is a dynamic learning rate, which is adjusted according to the freshness of the data; Adjust actions for any candidate policy; For the platform exist Always choose your action probability.

7. The AI-based cross-platform advertising material lifecycle management and optimization method according to claim 1, characterized in that: The material decay law is defined as follows: Where, For the platform exist The intensity of the material click at the moment; is the baseline decay rate; To remove the platform External platforms; is the cross-platform propagation coefficient; Affects the decay rate across platforms; For the moment; For the platform exist The event increment at the moment; Number the external event type; For events Impact strength; For events The indicator function that occurs; The survival function of the material is defined as: Where, is the survival function, representing the platform The material is The probability that the moment has not decayed; is the integral variable; A multi-level decay level response mechanism is preset based on the survival function value. The decay level of the material is warned in real time according to the calculated survival function value, and the material push strategy is adjusted according to the adjustment method corresponding to the material decay level triggered by the response mechanism.

8. An AI-based cross-platform advertising material lifecycle management and optimization system, characterized by: The system is used to implement the AI-based cross-platform advertising material lifecycle management and optimization method according to any one of claims 1 to 7, comprising a heterogeneous data graph construction module, a cross-modal feature extraction and dimensionality reduction module, a cross-platform material generation module, a material push strategy generation and optimization module, and a material lifecycle warning and push strategy adjustment module; A heterogeneous data graph construction module is used to collect raw data and metadata from different platforms and construct a heterogeneous data graph based on the raw data and metadata. The node types in the heterogeneous data graph are defined as user nodes, material nodes, and context nodes, and the edge types are defined as click-relationship edges, browsing-relationship edges, modality-related edges, and user-context edges. Cross-modal feature extraction and dimensionality reduction module, which is used to extract cross-modal features by performing layered graph convolution on each node based on the corresponding relationship type based on the heterogeneous data graph structure; Cross-modal features of different relationship types are spliced and dimensionally aligned to obtain high-dimensional features. The high-dimensional features are then dimensionality reduced through the attention mechanism to obtain unified features after dimensionality reduction. A cross-platform material generation module is used to construct core semantic items based on unified features, and to construct gene individuals composed of core semantic items, physical constraint items, and creative expression items. Cross-platform materials are generated through a multi-objective constraint co-evolution algorithm. The material push strategy generation and optimization module is used to generate material push strategies based on the generated cross-platform material mapping. It optimizes budget allocation and push strategies by building a layered game framework that includes offline Nash equilibrium pre-calculation and online no-regret learning. Combined with federated constraint synchronization, it obtains a set of push strategies for multi-platform collaborative optimization. The material life cycle warning and push strategy adjustment module is used to provide real-time warning of material life cycle decline and trigger material push strategy adjustments by modeling material decay laws and calculating material survival functions.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the AI-based cross-platform advertising material lifecycle management and optimization method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the AI-based cross-platform advertising material lifecycle management and optimization method according to any one of claims 1 to 7 is implemented.

Images