Advertisement creativity dynamic generation and adaptation method, device, equipment and medium

Abstract

The invention discloses an advertisement creativity dynamic generation and adaptation method and device, equipment and a medium. The method comprises the following steps: constructing a media characteristic knowledge graph for storing context characteristics, content format specifications and user portraits of different media platforms; carrying out multi-modal semantic analysis on the input initial creative material and extracting semantic elements; inputting the knowledge graph and the semantic elements into a generative adaptive model for processing, and outputting adaptive creative content which is matched with a target media platform in context and format and retains core semantics; putting the adaptive creative content on the target media platform and collecting deep behavior data generated by interaction between the user and the adaptive creative content; and performing iterative optimization on the generative adaptive model based on the depth behavior data, and reversely updating the media characteristic knowledge graph according to the optimization. According to the invention, deep and real-time intercommunication between creative design and media characteristics is realized, and the propagation effectiveness and brand consistency of advertisement creativity are improved.

Description

Technical Field

[0001] This invention relates to the technical field of digital marketing, and in particular to a method, apparatus, device, and medium for dynamically generating and adapting advertising creatives. Background Technology

[0002] In the current advertising industry, creative design and advertising media placement are usually separate processes. The creative team completes the initial creative draft based on common requirements, while the media team then manually adjusts the creative (such as mechanically converting the size and format according to preset rules) to adapt it to different media platforms (such as short videos, news feeds, outdoor billboards, etc.).

[0003] However, this linear collaboration model leads to a disconnect between creative design and media characteristics. Post-adaptation requires significant manual intervention, resulting in low efficiency and high costs. For example, this model fails to understand the deep connection between the semantic information of creative design content (such as core selling points, brand elements, and emotional tone) and the contextual characteristics of media placement (such as user intent, scene atmosphere, and interaction patterns), leading to stiff and unengaging content after adaptation. Furthermore, performance data cannot be fed back to the creative team in real time or only superficial metrics such as click-through rates are provided. This fails to diagnose the root causes of the mismatch between the content and the media context, resulting in long creative optimization cycles, poor multimedia synergy, and negatively impacting advertising effectiveness and brand consistency. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, device, and medium for dynamic generation and adaptation of advertising creative ideas, aiming to solve the existing problem of how to achieve real-time interoperability and automated adaptation between creative design and media characteristics, so as to improve the effectiveness and consistency of the dissemination of creative design content in different media environments.

[0005] In a first aspect, embodiments of the present invention provide a method for dynamically generating and adapting advertising creatives, including: Construct a dynamically updatable media feature knowledge graph, which is used to store contextual features, content format specifications, and user profiles of different media platforms; The initial creative materials are subjected to multimodal semantic analysis and semantic elements are extracted, including core selling points, visual focus and emotional tags; The media feature knowledge graph and the semantic elements are input into the generative adaptation model for processing, and the output is adapted creative content that matches the target media platform in terms of context and format while retaining the core semantics. The adapted creative content is delivered to the target media platform, and in-depth behavioral data generated by user interaction with the adapted creative content is collected. The deep behavioral data is input into a policy network-based reinforcement learning optimization module for processing to update the internal parameters of the generative adaptation model, and the media feature knowledge graph is updated in reverse based on the optimization results.

[0006] Secondly, embodiments of the present invention provide an advertising creative dynamic generation and adaptation device, comprising: The graph construction unit is used to construct a dynamically updatable media feature knowledge graph, which is used to store the contextual features, content format specifications and user profiles of different media platforms. The element extraction unit is used to perform multimodal semantic parsing on the input initial creative materials and extract semantic elements, wherein the semantic elements include core selling points, visual focus and emotional tags; The content output unit is used to input the media feature knowledge graph and the semantic elements into the generative adaptation model for processing, and output adapted creative content that matches the target media platform in terms of context and format while retaining the core semantics. The behavior collection unit is used to deliver the adapted creative content to the target media platform and collect in-depth behavioral data generated by the user's interaction with the adapted creative content. The model optimization unit is used to input the deep behavioral data into the policy network-based reinforcement learning optimization module for processing, so as to update the internal parameters of the generative adaptation model, and to update the media feature knowledge graph in reverse according to the optimization results.

[0007] Thirdly, embodiments of the present invention provide a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the advertising creative dynamic generation and adaptation method described in the first aspect.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, causes the processor to perform the advertising creative dynamic generation and adaptation method described in the first aspect.

[0009] The beneficial effects of this invention are as follows: through dynamic media characteristic knowledge graphs and multimodal parsing, deep and real-time communication between creative design and media characteristics is achieved; through generative models for dual semantic and format adaptation, the automation level and content quality of cross-media creative production are improved; through reinforcement learning closed loop based on deep behavioral data, the system can continuously optimize itself, not only improving the effect of a single campaign, but also enabling the entire advertising campaign strategy to evolve over time, thereby significantly improving the overall effectiveness of advertising creative dissemination and brand consistency. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating the advertising creative dynamic generation and adaptation method provided in this embodiment of the invention; Figure 2 This is a schematic diagram of a sub-process of step S101 provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a sub-process of step S102 provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a sub-process of step S103 provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a sub-process of step S104 provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a sub-process of step S105 provided in an embodiment of the present invention; Figure 7 A schematic block diagram of an advertising creative dynamic generation and adaptation device provided in an embodiment of the present invention; Figure 8 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] Please see Figure 1 , Figure 1 This is a flowchart illustrating the method for dynamically generating and adapting advertising creatives according to an embodiment of the present invention.

[0017] like Figure 1 As shown, the method includes steps S101-S105.

[0018] S101. Construct a dynamically updatable media feature knowledge graph, which is used to store the contextual features, content format specifications, and user profiles of different media platforms. Step S101 aims to establish a dynamically updated media feature knowledge graph, which serves as the core database of the system. This knowledge graph continuously stores and updates contextual features (reflecting the content style of the platform), content format specifications (clarifying technical parameter requirements), and user profiles (describing audience characteristics) from different media platforms.

[0019] S102. Perform multimodal semantic analysis on the input initial creative materials and extract semantic elements, including core selling points, visual focus and emotional tags; Step S102 aims to perform in-depth multimodal semantic analysis on the input initial creative materials to extract semantic elements including core selling points, visual focus and emotional tags. These semantic elements are the basis for subsequent creative adaptation and generation, and provide key information support for creative adaptation and generation across multiple media forms such as text, images and videos.

[0020] S103. Input the media feature knowledge graph and semantic elements into the generative adaptation model for processing, and output adapted creative content that matches the target media platform in terms of context and format while retaining the core semantics. Step S103 aims to input the aforementioned semantic elements and the target media platform data from the media feature knowledge graph into the generative adaptation model. The generative adaptation model then performs an intelligent generation process that integrates semantic understanding and rule adherence, ultimately outputting adapted content that is highly matched to the target media platform in terms of context and physical format, while retaining the core semantics (i.e., the core information of the original creative idea).

[0021] S104. Deploy appropriate creative content on target media platforms and collect in-depth behavioral data generated by user interactions with the appropriate creative content; Step S104 aims to deliver the generated adapted creative content to the target media platform. During this process, the system will record every interaction between the user and the adapted creative content in detail, including but not limited to clicks, browsing time, interactive comments, and other behavioral data that reflect the user's real participation or deep engagement. This data will provide an important basis for subsequent creative optimization.

[0022] S105. Input the deep behavioral data into the policy network-based reinforcement learning optimization module for processing, so as to update the internal parameters of the generative adaptation model, and update the media feature knowledge graph in reverse according to the optimization results.

[0023] Step S105 aims to leverage the collected deep behavioral data to deeply optimize the generative adaptation model through a policy network-based reinforcement learning optimization module. This module intelligently adjusts model parameters based on user deep behavioral data, enabling the generative adaptation model to more accurately match user preferences in subsequent creative generation. Simultaneously, the system updates the media feature knowledge graph with the new knowledge gained during optimization, ensuring that the contextual features, content format specifications, and user profiles in the knowledge graph remain synchronized with actual user behavior and media platform changes. This bidirectional update mechanism enhances the adaptability and accuracy of the generative adaptation model, allowing the entire creative generation and delivery system to continuously evolve, forming a semantic-level continuous optimization loop.

[0024] In steps S101-S105 of this embodiment, deep and real-time communication between creative design and media characteristics is achieved through dynamic media characteristic knowledge graphs and multimodal parsing; semantic and format dual adaptation is performed through generative models, improving the automation level and content quality of cross-media creative production; through reinforcement learning closed loop based on deep behavioral data, the system can continuously optimize itself, not only improving the effect of a single campaign, but also enabling the entire advertising campaign strategy to evolve over time, thereby significantly improving the overall effectiveness of advertising creative dissemination and brand consistency.

[0025] In one embodiment, such as Figure 2 As shown, step S101 includes: S201. Obtain content format specification data from the official interfaces of various media platforms and perform structured processing on the content format specification data; S202. Collect content data from the publicly available content of various media platforms, analyze and process the content data, and output the contextual characteristics and user profiles of the media platforms. S203. Link and integrate content format specifications, contextual features, and user profiles to construct a media characteristic knowledge graph in a graphical structure.

[0026] For step S201, official, machine-readable raw specification data can be directly obtained through programmatic interfaces (such as the Official API provided by various media platforms to developers), and these unstructured or semi-structured data can be transformed into unified and regular structured data using data processing tools (such as customized parsing scripts or data pipelines) to obtain visual feature vectors and text feature vectors. For example, the size requirement of the text description "1080x1920 pixels" can be parsed into a data object with explicit numeric fields.

[0027] For step S202, content data can be collected from the public content stream of the media platform (such as information stream, popular video list) and analyzed and processed in parallel. One path uses natural language processing technology to analyze popular content to extract contextual features, and the other path uses clustering analysis technology to process user behavior data to outline user profiles.

[0028] For step S203, the content format specifications, contextual features and user profiles obtained in the aforementioned steps S201-S202 are associated and integrated. Specifically, graph database technology (such as Neo4j) can be used to treat "media platform" as a node, the three types of data as their attributes or associated nodes, and the relationship between them is defined by edges, thereby constructing an interconnected and efficiently queryable media characteristic knowledge graph.

[0029] In steps S201-S203 of this embodiment, the accuracy of the creative adaptation benchmark is ensured by directly and automatically obtaining and structuring content format specifications from the official interface, thus avoiding compliance issues caused by specification errors from the outset. Subsequently, by combining multi-dimensional analysis of publicly available content data, the system can dynamically capture the contextual trends and user group changes of media platforms. Finally, through graph-structured data fusion, the three dimensions of format, context, and user information are integrated, providing a rich, relevant, and easily searchable and reasoned core data foundation for subsequent intelligent generation and adaptation, greatly enhancing the system's depth of understanding of the media ecosystem.

[0030] In one embodiment, step S201 includes: performing semantic analysis and style recognition processing on popular platform content in the content data using natural language processing technology to output the contextual features of the media platform; and performing group segmentation processing on user behavior data in the content data using clustering analysis technology to output user profiles of the media platform.

[0031] This embodiment further refines the specific technical path of content data analysis and processing in step S201.

[0032] Specifically, the natural language processing techniques used for extracting contextual features can include pre-trained language models (such as BERT and RoBERTa) or topic models (such as LDA). Taking the BERT model as an example, its specific working process is as follows: First, it segments and vectorizes the collected popular content from the platform (such as titles, copy, and comments). Then, it uses its deep Transformer architecture to understand the contextual semantics of the text, thereby automatically identifying and summarizing the frequently occurring style tone (such as "humorous" and "professional review"), emotional tendency (such as "positive support" and "rational discussion"), and popular topic categories in the platform's content. These quantified features together constitute the contextual features of the media.

[0033] Specifically, clustering analysis techniques used to generate user profiles can include algorithms such as K-means and DBSCAN. Taking K-means as an example, its specific working process is as follows: feature extraction and standardization are performed on user behavior data (such as behavioral vectors such as likes, comments, shares, and viewing completion rates). Then, the number of clusters K is preset or dynamically determined according to the data distribution characteristics. Through iterative calculation, users with similar behavioral patterns are grouped into the same cluster. Each cluster represents a user group with significant commonalities, thereby outputting structured user profile labels, such as "price-sensitive user", "deep ingredient enthusiast", and "trend follower".

[0034] Based on this, this embodiment utilizes natural language processing models such as BERT to accurately and automatically extract abstract yet crucial media contextual features from massive amounts of text, ensuring that creative content seamlessly aligns with the platform's atmosphere in terms of style and tone. Simultaneously, by employing clustering algorithms such as K-means to accurately segment user behavior, it can create refined user profiles, providing solid data support for personalized creative targeting. The combination of these two aspects means that the media characteristic knowledge graph not only includes hard format specifications but also soft style and audience insights, effectively improving the accuracy and appeal of subsequent creative adaptation.

[0035] In one embodiment, such as Figure 3 As shown, step S102 includes: S301. Perform multimodal feature extraction on the initial creative materials and output visual feature vectors and text feature vectors; S302. Input the visual feature vector and text feature vector into the semantic understanding model for processing, and output structured data including core selling points, visual focus and sentiment tags; S303. The structured data is fused and normalized to obtain semantic elements.

[0036] In step S301, a convolutional neural network (CNN, such as ResNet, VGG) or its variants can be used to obtain visual feature vectors. Specifically, the image or video frame in the initial creative material is input into the convolutional neural network, and through multi-layer convolution and pooling operations, hierarchical features from edges and textures to objects and scenes are gradually extracted, and finally a high-dimensional visual feature vector that can represent visual content is output.

[0037] In step S301, a pre-trained language model such as BERT or GPT can be used to generate text feature vectors. Specifically, the text information in the initial creative material is input into the pre-trained language model, and the deep semantic relationships between words are captured using self-attention mechanisms, and finally the text feature vectors are output.

[0038] In step S302, the semantic understanding model can use a multimodal Transformer network. Its working process is to use an attention mechanism to enable bidirectional and in-depth interaction and alignment between visual feature vectors and text feature vectors, thereby comprehensively understanding the entire creative material and finally outputting structured data to accurately label the core selling points (such as "12-hour makeup lasting power" in cosmetic creative design), visual focus (such as "lipstick close-up" in cosmetic creative design) and emotional labels (such as "confidence").

[0039] In step S303, the structured data output by the semantic understanding model, including core selling points, visual focus, and emotional tags, is fused and normalized. This can be achieved by weighted averaging and / or concatenation, followed by integration into a unified and regular semantic element through a fully connected layer, so as to facilitate unified processing in subsequent steps.

[0040] Based on this, steps S301-S303 provide rich and accurate underlying data representations for deep semantic understanding. Utilizing a multimodal Transformer network for cross-modal interaction enables the understanding of composite information conveyed by combining text and video, thereby accurately extracting core advertising elements such as key selling points, visual focal points, and emotional tags. Finally, fusion and normalization processes ensure the standardization and consistency of the output data.

[0041] In one embodiment, such as Figure 4 As shown, step S103 includes: S401. Calculate the importance score of each element in the semantic elements relative to the contextual features through a cross-modal alignment network to generate attention weights, and use the attention weights to perform weighted summation on the semantic elements to output a joint semantic vector that represents the deep fusion of semantic elements and contextual features. S402. Decode the joint semantic vector through the content generator, and combine it with the contextual features and content format specifications of the target media platform to output adapted creative content that conforms to the context and format requirements of the target media platform and retains the core semantics.

[0042] For step S401, the cross-modal alignment network can specifically adopt a Transformer architecture based on an attention mechanism. It can input semantic elements (as Query) and contextual features of the target media platform (as Key and Value) into the cross-modal alignment network. By calculating the dot product similarity between Query and Key and applying the Softmax function, a set of attention weights is obtained. This set of weights quantifies the importance of each semantic element to the current target contextual features. Subsequently, the semantic elements (Value) are weighted and summed using these attention weights, and finally outputs a new joint semantic vector that integrates the original creative semantics and the target media contextual preferences.

[0043] In step S402, the content generator is essentially a decoder, which can be built based on architectures such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), or Diffusion Models. The content generator takes the joint semantic vector obtained in step S401 as conditional input and simultaneously retrieves the specific content format specifications (such as size, duration, and file format) of the target media platform from the media characteristic knowledge graph. Through multi-layer deconvolution or denoising diffusion processes of neural networks, it gradually generates (or reconstructs) entirely new, adapted creative content that conforms to the context and format requirements of the target media platform while retaining its core semantics.

[0044] For example, the joint semantic vector and the visual size constraints in the content format specification can be input into the content generator to output an adapted image after intelligent cropping, reconstruction, or style transfer.

[0045] For example, the joint semantic vector and the copy length constraint in the content format specification can be input into the content generator to output a new spoken copy or copy adaptation that conforms to the target media context.

[0046] For example, the joint semantic vector and the interaction patterns in the contextual features are input into the content generator to output interactive advertising content that embeds platform-specific interactive elements.

[0047] Based on this, steps S401-S402, through a cross-modal attention mechanism, achieve intelligent and dynamic integration of the creative core and the media context, ensuring that the generated adapted creative content is tailored to the target audience rather than a forced fit. Guided by joint semantic vectors and combined with strict formatting specifications, the content generator achieves precise and automated creation from semantic concepts to specific audiovisual content. This collectively ensures that the final output adapted creative content is consistent with the original creative core and conforms to the ecological requirements of the target media platform in terms of presentation, thereby achieving high-quality and efficient cross-media creative production.

[0048] In one embodiment, such as Figure 5 As shown, step S104 includes: S501. Collect raw behavior logs generated by users when interacting with adapted creative content through the data interface and tracking system of the target media platform. S502. Clean and aggregate the raw behavior logs to output a structured deep behavior data set, which includes various data such as video completion rate, interaction touchpoint heatmap, sentiment analysis data of user comments, and social relationship chain information behind sharing behavior.

[0049] For step S501, raw behavior logs can be collected using standard data interfaces provided by the media platform (such as Facebook Graph API, TikTok Analytics API), or raw behavior logs can be collected using custom tracking systems embedded in the adapted creative content (such as using SDKs like Google Analytics and Sensors Data). Both tools can record every user interaction event in real time.

[0050] For the cleaning and aggregation process in step S502, the original logs are first deduplicated and invalid or abnormal records are filtered out. Then, data from different sources are correlated and aligned according to user sessions or creative IDs. Finally, low-order events are aggregated into high-order metrics, resulting in a deep behavioral data set. Each piece of information in the deep behavioral data set has specific meaning. For example, video completion rate directly reflects the content's attractiveness and sustained engagement with users; interaction touchpoint heatmaps reveal the distribution of user attention to various elements within the adapted creative content, guiding visual optimization; sentiment analysis data from user comments provides users' subjective attitudes and emotional feedback towards the adapted creative content; and the social relationship chain information behind sharing behavior can assess the viral potential of the adapted creative content and the potential user groups that can be influenced.

[0051] Based on this, steps S501-S502 utilize multi-channel, high-concurrency data collection tools to ensure the comprehensiveness and real-time nature of user feedback information. After cleaning and aggregation, the messy raw logs are transformed into clean, organized, and directly usable structured data for analysis and modeling. The resulting deep behavioral data set, compared to traditional surface indicators, can reveal the quality of interaction between the adapted creative content and users more profoundly from multiple dimensions such as user attention, emotional resonance, and social dissemination, providing data support for subsequent multi-dimensional effect scoring and model optimization.

[0052] In one embodiment, such as Figure 6 As shown, step S105 includes: S601. Input deep behavioral data into the effect evaluation model for processing, and output a multi-dimensional effect score for the adapted creative content. S602. Construct a reward function for reinforcement learning based on multi-dimensional performance scoring; S603. Adapt the creative content, the corresponding semantic elements, the corresponding contextual features, and the multi-dimensional effect scores as state data, and input them into the policy network contained in the reinforcement learning optimization module. S604. Perform forward propagation calculations on the state data through the policy network and output the update amount of the internal parameters of the generative adaptation model. S605. Adjust the internal parameters of the generative adaptation model according to the update amount to complete the model optimization.

[0053] For step S601, the performance evaluation model can be an ensemble learning model (such as random forest or gradient boosting tree) or a deep neural network. Taking gradient boosting tree (GBDT) as an example, its process of calculating multi-dimensional performance scores is as follows: the collected deep behavioral data (such as completion rate, like rate, positive comment ratio, number of shares, etc.) are used as feature inputs, and the model outputs a comprehensive and quantitative performance score (such as 0 to 1) for this adaptive creative content delivery through the complex non-linear relationships it has learned.

[0054] For the reward function construction in step S602, the multi-dimensional effect score obtained in step S601 can be directly mapped to the reward value, or a baseline can be set, and positive rewards can be given for the part that exceeds the baseline.

[0055] For steps S603-S605, the core of the reinforcement learning optimization module is the policy network, which can be an Actor network in a Deep Deterministic Policy Gradient (DDPG) or Proximal Policy Optimization (PPO) algorithm. The optimization process of the policy network is as follows: state data containing past actions (i.e., the generated adapted creative content and its underlying semantic elements and contextual features) and the consequences of those actions (multi-dimensional effect scores) are input into the policy network; the policy network calculates through forward propagation and outputs an update direction and magnitude (i.e., update amount) for the internal parameters of the generative adaptation model; the system adjusts the parameters of the generative adaptation model accordingly, so that when the model faces similar semantics and contexts in the future, it can generate creative content that has historically received high rewards (i.e., high effect scores) with a higher probability.

[0056] Based on this, in steps S601-S605, the multi-dimensional deep behavioral data is transformed into an authoritative, optimizable single objective through the effect evaluation model, providing a clear optimization direction for reinforcement learning. The policy network, by analyzing complete state data of actions and outcomes, can learn complex creative strategies, making the generative adaptation model no longer a static tool, but an intelligent agent capable of continuously learning from real market feedback and constantly evolving its creative taste and delivery strategies. Ultimately, this achieves autonomous iteration and continuous growth in the effectiveness of adapted creative content delivery, transforming advertising optimization from being driven by human experience to a highly efficient closed loop driven by data and algorithms.

[0057] This invention also provides an apparatus for dynamically generating and adapting advertising creatives, which is used to execute any of the aforementioned methods for dynamically generating and adapting advertising creatives. Specifically, please refer to... Figure 7 , Figure 7 This is a schematic block diagram of the advertising creative dynamic generation and adaptation device provided in the embodiments of the present invention.

[0058] like Figure 7 As shown, the advertising creative dynamic generation and adaptation device 700 includes: a graph construction unit 701, an element extraction unit 702, a content output unit 703, a behavior collection unit 704, and a model optimization unit 705.

[0059] The graph construction unit 701 is used to construct a dynamically updatable media feature knowledge graph. The media feature knowledge graph is used to store the contextual features, content format specifications and user profiles of different media platforms. The element extraction unit 702 is used to perform multimodal semantic analysis on the input initial creative material and extract semantic elements, including core selling points, visual focus and emotional tags; Content output unit 703 is used to input media feature knowledge graph and semantic elements into generative adaptation model for processing, and output adapted creative content that matches the target media platform in terms of context and format and retains the core semantics. The behavior collection unit 704 is used to deliver adapted creative content on the target media platform and collect in-depth behavioral data generated by users' interaction with the adapted creative content. The model optimization unit 705 is used to input deep behavioral data into the policy network-based reinforcement learning optimization module for processing, so as to update the internal parameters of the generative adaptation model, and update the media feature knowledge graph in reverse according to the optimization results.

[0060] This device achieves deep and real-time communication between creative design and media characteristics through a dynamic media characteristic knowledge graph and multimodal parsing; it improves the automation and content quality of cross-media creative production through generative models for semantic and format adaptation; and through a reinforcement learning loop based on deep behavioral data, the system can continuously optimize itself, not only improving the effect of a single campaign but also enabling the entire advertising strategy to evolve over time, thereby significantly improving the overall effectiveness of advertising creative communication and brand consistency.

[0061] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0062] The aforementioned advertising creative dynamic generation and adaptation device can be implemented in the form of a computer program, which can, for example, Figure 8 It runs on the computer device shown.

[0063] Please see Figure 8 , Figure 8 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. The computer device 800 is a server, which can be a standalone server or a server cluster composed of multiple servers.

[0064] See Figure 8 The computer device 800 includes a processor 802, a memory, and a network interface 805 connected via a system bus 801. The memory may include a non-volatile storage medium 803 and internal memory 804.

[0065] The non-volatile storage medium 803 may store an operating system 8031 ​​and a computer program 8032. When the computer program 8032 is executed, it enables the processor 802 to perform a method for dynamically generating and adapting advertising creatives.

[0066] The processor 802 provides computing and control capabilities to support the operation of the entire computer device 800.

[0067] The internal memory 804 provides an environment for the operation of the computer program 8032 in the non-volatile storage medium 803. When the computer program 8032 is executed by the processor 802, the processor 802 can execute the method of dynamically generating and adapting advertising creatives.

[0068] This network interface 805 is used for network communication, such as providing data transmission. Those skilled in the art will understand that... Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device 800 to which the present invention is applied. The specific computer device 800 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0069] Those skilled in the art will understand that Figure 8 The embodiments of the computer device shown do not constitute a limitation on the specific configuration of the computer device. In other embodiments, the computer device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. For example, in some embodiments, the computer device may include only memory and a processor. In such embodiments, the structure and function of the memory and processor are different from those shown. Figure 8 The embodiments shown are consistent and will not be described again here.

[0070] It should be understood that, in this embodiment of the invention, the processor 802 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0071] In another embodiment of the invention, a computer-readable storage medium is provided. This computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, wherein when executed by a processor, the computer program implements the advertising creative dynamic generation and adaptation method of the embodiments of the present invention.

[0072] The storage medium is a physical, non-transient storage medium, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk, or any other physical storage medium capable of storing program code.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0074] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for dynamically generating and adapting advertising creatives, characterized in that, include: Construct a dynamically updatable media feature knowledge graph, which is used to store contextual features, content format specifications, and user profiles of different media platforms; The initial creative materials are subjected to multimodal semantic analysis and semantic elements are extracted, including core selling points, visual focus and emotional tags; The media feature knowledge graph and the semantic elements are input into the generative adaptation model for processing, and the output is adapted creative content that matches the target media platform in terms of context and format while retaining the core semantics. The adapted creative content is delivered to the target media platform, and in-depth behavioral data generated by user interaction with the adapted creative content is collected. The deep behavioral data is input into a policy network-based reinforcement learning optimization module for processing to update the internal parameters of the generative adaptation model, and the media feature knowledge graph is updated in reverse based on the optimization results.

2. The method for dynamically generating and adapting advertising creatives according to claim 1, characterized in that, The construction of a dynamically updatable media characteristic knowledge graph, which stores at least the contextual features, content format specifications, and user profiles of different media platforms, includes: Content format specification data is obtained from the official interfaces of various media platforms, and the content format specification data is processed in a structured manner. Content data is collected from publicly available content on various media platforms, and the content data is analyzed and processed to output the contextual features and user profiles of the media platforms. By associating and integrating the content format specifications, contextual features, and user profiles, a media characteristic knowledge graph in the form of a graph structure is constructed.

3. The method for dynamically generating and adapting advertising creatives according to claim 2, characterized in that, The process of collecting content data from publicly available content on various media platforms, analyzing and processing the content data, and outputting the contextual features and user profiles of the media platforms includes: Natural language processing technology is used to perform semantic analysis and style recognition on popular content from the platform in the content data, and the contextual features of the media platform are output. Cluster analysis is used to segment user behavior data in content data and output user profiles for the media platform.

4. The method for dynamically generating and adapting advertising creatives according to claim 1, characterized in that, The process involves performing multimodal semantic analysis on the input initial creative materials and extracting semantic elements, wherein the semantic elements include core selling points, visual focal points, and emotional tags, including: The initial creative materials are subjected to multimodal feature extraction processing to output visual feature vectors and text feature vectors; The visual feature vectors and text feature vectors are input into a semantic understanding model for processing, and the output includes structured data including core selling points, visual focus, and sentiment tags. The structured data is fused and normalized to obtain semantic elements.

5. The method for dynamically generating and adapting advertising creatives according to claim 1, characterized in that, The process of inputting the media feature knowledge graph and the semantic elements into a generative adaptation model for processing, and outputting adapted creative content that matches the target media platform in terms of context and format while retaining the core semantics, includes: An attention weight is generated by calculating the importance score of each element in the semantic element relative to the contextual feature through a cross-modal alignment network. The attention weight is then used to perform a weighted summation on the semantic element to output a joint semantic vector that represents the deep fusion of the semantic element and the contextual feature. The content generator decodes the joint semantic vector and combines it with the contextual features and content format specifications of the target media platform to output adapted creative content that conforms to the context and format requirements of the target media platform and retains the core semantics.

6. The method for dynamically generating and adapting advertising creatives according to claim 1, characterized in that, The process of delivering the adapted creative content to the target media platform and collecting in-depth behavioral data generated by user interaction with the adapted creative content includes: Through the data interface and event tracking system of the target media platform, collect raw behavior logs generated by users when interacting with the adapted creative content; The original behavior logs are cleaned and aggregated to output a structured deep behavior data set, which includes various data such as video completion rate, interaction touchpoint heatmap, sentiment analysis data of user comments, and social relationship chain information behind sharing behavior.

7. The method for dynamically generating and adapting advertising creatives according to claim 1, characterized in that, The step of inputting the deep behavioral data into a policy network-based reinforcement learning optimization module for processing to update the internal parameters of the generative adaptation model, and then updating the media feature knowledge graph in reverse based on the optimization results, includes: The deep behavioral data is input into the effect evaluation model for processing, and a multi-dimensional effect score is output for the adapted creative content. A reward function for reinforcement learning is constructed based on the aforementioned multi-dimensional performance score. The adapted creative content, its corresponding semantic elements, its corresponding contextual features, and its multi-dimensional effect scores are used together as state data and input into the policy network contained in the reinforcement learning optimization module. The policy network is used to perform forward propagation calculations on the state data, and the update amount of the internal parameters of the generative adaptation model is output. The internal parameters of the generative adaptation model are adjusted according to the update amount, thereby completing the model optimization.

8. A device for dynamically generating and adapting advertising creatives, characterized in that, include: The graph construction unit is used to construct a dynamically updatable media feature knowledge graph, which is used to store the contextual features, content format specifications and user profiles of different media platforms. The element extraction unit is used to perform multimodal semantic parsing on the input initial creative materials and extract semantic elements, wherein the semantic elements include core selling points, visual focus and emotional tags; The content output unit is used to input the media feature knowledge graph and the semantic elements into the generative adaptation model for processing, and output adapted creative content that matches the target media platform in terms of context and format while retaining the core semantics. The behavior collection unit is used to deliver the adapted creative content to the target media platform and collect in-depth behavioral data generated by the user's interaction with the adapted creative content. The model optimization unit is used to input the deep behavioral data into the policy network-based reinforcement learning optimization module for processing, so as to update the internal parameters of the generative adaptation model, and to update the media feature knowledge graph in reverse according to the optimization results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for dynamically generating and adapting advertising creatives as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the advertising creative dynamic generation and adaptation method as described in any one of claims 1 to 7.

Images