Advertisement creativity real-time optimization and delivery control method and system based on artificial intelligence, electronic equipment and computer readable medium

Abstract

The invention discloses an advertisement creative real-time optimization and putting control method and system based on artificial intelligence, electronic equipment and a computer readable medium, and the method achieves the precise matching, effect prediction and dynamic regulation and control of the advertisement creative through an intelligent technology, thereby improving the efficiency and effect of advertisement putting. The core process of the method comprises the following steps: firstly, carrying out fusion and joint modeling on multi-source features such as user portraits, advertisement creative contents and real-time context environments so as to realize high-matching-degree real-time optimization of creativity, users and scenes; and secondly, a deep cross network (DCN) model is adopted to capture a complex high-order interaction relationship among the user, the creativity and the environment, so that key indexes such as a click rate (CTR) and a conversion rate (CVR) are predicted with high precision. Besides, the system establishes a real-time monitoring mechanism based on a sliding time window and a change slope, creative fatigue is actively identified through the change trend of data such as exposure, clicking and conversion, and intervention and adjustment are carried out in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet advertising, and in particular to an advertisement creative real-time optimization and delivery control method and system based on artificial intelligence, an electronic device and a computer readable medium. BACKGROUND

[0002] Current Internet advertising delivery systems generally use DSP (demand side platform) for programmatic delivery. Advertisement creative selection usually relies on historical statistical data or fixed rules for rotation or weighted display. With the high fragmentation of user behavior and the complex media environment, traditional rule-based or shallow machine learning model methods have been difficult to meet the real-time and precision requirements Disadvantages / deficiencies of prior art: Low creative matching accuracy and lack of context awareness: existing systems are difficult to deeply integrate and model multi-source heterogeneous information such as user profile data, creative content features, and real-time delivery environment data. Usually only based on user basic attributes or historical delivery effects for matching, lacking comprehensive use of real-time context information such as user immediate interest, creative multi-modal features, and geographic location, time, terminal type, etc., resulting in difficulty in matching the pushed creative with the current user scenario.

[0003] Limited effect prediction accuracy: existing technologies mostly use logistic regression (LR) or shallow machine learning models, which have limited ability to describe the complex non-linear relationships and high-order feature intersections between user attributes, creative features, and delivery environment, making it difficult to effectively capture the deep combination relationships behind the advertisement effect, resulting in difficulty in meeting the fine-grained delivery requirements of key indicators such as click-through rate (CTR) and conversion rate (CVR).

[0004] Creative fatigue recognition lag: users are prone to aesthetic fatigue after multiple exposures to the same advertisement creative, resulting in rapid decay of advertisement effect. However, the recognition of fatigue in existing systems generally relies on simple exposure thresholds or long-period average effect changes, lacks real-time time series analysis-based perception, and is difficult to predict fatigue inflection points and actively intervene in a timely manner.

[0005] Poor model adaptability: existing models generally rely on offline periodic training, have slow response to sudden traffic changes and new creative online, and lack online learning and incremental update mechanisms, resulting in insufficient real-time and adaptive ability of delivery decisions. SUMMARY

[0006] According to a first aspect of an embodiment of the present application, a real-time advertisement creative optimization and delivery control method based on artificial intelligence is provided, comprising the following steps: When the advertisement creative material is uploaded or updated, the pre-trained visual model and the text model are called asynchronously to extract the visual feature vector and the text feature vector of the creative material respectively, and the two are fused to generate a unified creative content feature vector , stored in the feature platform In response to the received real-time advertisement request, the corresponding and the user portrait features of the corresponding user are queried and obtained from the feature platform , and the real-time obtained delivery environment context features are combined to construct an initial feature vector group for prediction The initial feature vector group is input into the deep cross network DCN model; the core cross layer of the deep cross network DCN model performs high-order nonlinear cross coding between 、 and to generate a joint feature vector ; the deep side layer of the deep cross network DCN model synchronously predicts the click rate and the conversion rate of each candidate creative through multi-task learning based on ; According to the predicted click rate and conversion rate, the comprehensive score S of each candidate creative is calculated, the creatives are sorted according to the comprehensive score S, and the creatives with high ranking are selected for real-time delivery During delivery, based on the sliding time window, the change trend of the effect index of each delivered creative with time is counted, when the change rate Δ of the effect index is lower than the preset negative threshold λ, it is determined that the creative enters the fatigue state, and the weight reduction or switching operation of the creative is automatically executed Based on the online learning mechanism, the exposure, click and conversion data fed back in real time after the advertisement is delivered are collected as incremental training samples, and the parameters of the deep cross network DCN model are updated incrementally using an incremental optimization algorithm to realize the adaptive optimization of the model.

[0007] Further, the visual model is a convolutional neural network (CNN) ResNet, and the text model is a natural language processing (NLP) BERT model.

[0008] Further, the comprehensive score S includes the predicted click rate, conversion rate and expected revenue ; Among them, is the click rate is the conversion rate is the expected revenue is the value after clicking is the bid of the advertiser

[0009] Further, the effect indicator is the click rate; the change rate Δ is the change slope of the click rate with time t in the sliding time window W, that is, Δ= .

[0010] Further, the condition for determining that the creative enters the fatigue state is that the change rate Δ continuously falls below the negative threshold λ for more than a preset time length.

[0011] Further, the Adagrad optimization algorithm is used to incrementally update the parameters of the DCN model, and the update period is minute-level.

[0012] Further, when the creative is put in real time, the cross-terminal exposure control is also completed in combination with the preset frequency control rule.

[0013] According to the second aspect of the embodiment of the application, an advertisement creative real-time optimization and putting control system based on artificial intelligence is provided, comprising: An asynchronous processing module is configured to asynchronously call the pre-trained visual model and the text model when the advertisement creative material is uploaded or updated, extract the visual feature vector and the text feature vector of the creative material respectively, and fuse the two to generate a unified creative content feature vector , which is stored in a feature platform; A response construction module is configured to, in response to a received real-time advertisement request, query and acquire the and the user portrait features of the corresponding user from the feature platform , and combine the real-time acquired putting environment context features to construct an initial feature vector group for prediction; A joint prediction module is configured to input the initial feature vector group into a deep cross network DCN model; the core cross layer of the deep cross network DCN model performs high-order nonlinear cross coding between 、 and to generate a joint feature vector ; the deep side layer of the deep cross network DCN model is based on , and synchronously predicts the click rate and the conversion rate of each candidate creative through multi-task learning; A calculation putting module is configured to calculate the comprehensive score S of each candidate creative according to the predicted click rate and conversion rate, sort the candidate creative according to the comprehensive score S, and select the creative with a high ranking for real-time putting; A monitoring intervention module is configured to, in the putting process, based on a sliding time window, statistically determine the change trend of the effect indicator of each put creative with time, determine that the creative enters the fatigue state when the change rate Δ of the effect indicator is lower than a preset negative threshold λ, and automatically perform the weight reduction or switching operation on the creative; The learning updating module is configured to collect exposure, click and conversion data of real-time feedback after the advertisement is launched as incremental training samples based on an online learning mechanism, and update parameters of a deep cross network (DCN) model by using an incremental optimization algorithm to realize adaptive optimization of the model.

[0014] According to a third aspect of the embodiments of the present application, an electronic device is provided, comprising a memory, a processor and a computer program, the computer program is stored in the memory, and the processor runs the computer program to execute the method for real-time optimization and control of advertisement creatives based on artificial intelligence according to the first aspect.

[0015] According to a fourth aspect of the embodiments of the present application, a computer readable medium having non-volatile program codes executable by a processor is provided, the program codes cause the processor to run the method for real-time optimization and control of advertisement creatives based on artificial intelligence according to the first aspect.

[0016] The method for real-time optimization and control of advertisement creatives based on artificial intelligence according to the embodiments of the present application has the following beneficial effects: High-precision joint matching: by unified modeling of multi-source data and high-order feature cross implemented by DCN, high-precision joint matching of "user-creative-scene" is realized, which significantly improves the matching degree of advertisement creatives and user interests and the scene of launching, thereby effectively improving the click rate and conversion rate of the advertisement.

[0017] Deep learning high-precision prediction: a deep cross network (DCN) prediction model is constructed, which has stronger fitting ability for high-order feature combination relationship and can more accurately predict the launching effect to provide high-reliability quantitative basis for launching decision.

[0018] Real-time fatigue active intervention: an creative fatigue recognition algorithm based on sliding time window and time series change rate is introduced to realize real-time recognition and early warning of creative fatigue state, which avoids passive lag judgment based on the traditional exposure number threshold and can complete active intervention before the creative effect has a significant decline.

[0019] Model continuous self-adaptation: through online learning and incremental updating mechanism, the prediction model and the launching control strategy can be continuously self-adapted without stopping launching, which can quickly respond to dynamic changes such as changes in user interests and fluctuations in traffic structure, and significantly improves the real-time performance and environmental adaptability of the system.

[0020] Integrated closed-loop control: the present application forms a complete closed-loop launching control system at the system level, realizes the change from experience-driven to data-intelligent-driven, and improves the automation level and operation efficiency in the large-scale advertisement launching scene.

[0021] It is to be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further explanation of the subject technology claimed. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flow chart of a method for real-time optimization and delivery control of an advertising creative based on artificial intelligence according to an embodiment of the present application.

[0023] Figure 2 A structure diagram of a system for real-time optimization and delivery control of an advertising creative based on artificial intelligence according to an embodiment of the present application.

[0024] Figure 3 A structure diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The preferred embodiments of the present application will be described in detail with reference to the drawings, and the present application will be further explained.

[0026] First, a method for real-time optimization and delivery control of an advertising creative based on artificial intelligence according to an embodiment of the present application will be described, which is applied to Internet advertising and has a wide range of application scenarios. Figure 1

[0027] As shown in Figure 1 , the method for real-time optimization and delivery control of an advertising creative based on artificial intelligence according to an embodiment of the present application comprises the following steps: As shown in Figure 1 , in S1, when the advertising creative material is uploaded or updated, the pre-trained visual model and the text model are called asynchronously to extract the visual feature vector and the text feature vector of the creative material, respectively, and to generate a unified creative content feature vector by fusing the two, which is stored in a high-IOPS feature platform.

[0028] It should be noted that before step S1, a multi-source data acquisition module can be constructed to acquire user behavior data, user portrait data, advertising creative material data, and delivery environment data in real time, and to clean and standardize them. Step S1 is a multi-modal creative content preprocessing step (offline / asynchronous), the visual model is a convolutional neural network (CNN) ResNet, and the text model is a natural language processing (NLP) BERT model. The visual feature vector is generated by the pre-trained convolutional neural network (CNN) ResNet by extracting features from pictures or video frames in the creative material. The text feature vector is generated by the natural language processing (NLP) BERT model by extracting semantic features of the creative copy.​

[0029] like Figure 1 As shown, in S2, in response to the received real-time ad request, the feature platform is queried to obtain the corresponding candidate creative. and corresponding user profile characteristics And combined with the real-time acquired contextual features of the deployment environment Construct an initial set of feature vectors for prediction.

[0030] like Figure 1 As shown, in S3, the initial feature vector set is input into the Deep CrossNetwork (DCN) model; the core cross network layer of the DCN model... , and Perform high-order nonlinear cross-coding between features to generate a joint feature vector. Achieving high-match creative optimization; the deep side layer of the Deep Cross Network (DCN) model is based on... It employs a multi-layer feedforward neural network (MLP) structure and simultaneously predicts the click-through rate of each candidate creative through multi-task learning. and Conversion rate It enables simultaneous prediction of click-through rate and conversion rate, providing a reliable basis for real-time optimization.

[0031] like Figure 1 As shown, in S4, based on the predicted click-through rate and conversion rate, the comprehensive score S of each candidate creative is calculated. Creatives are then ranked according to their comprehensive scores S, and the top-ranked creatives are selected for real-time deployment. The comprehensive score S includes the predicted click-through rate, conversion rate, and expected revenue. ; in, For click-through rate; Conversion rate; For expected returns, The value after clicking; The bid from the advertiser.

[0032] Furthermore, in this embodiment, when the creative content is delivered in real time, cross-platform exposure control is also achieved by combining preset frequency control rules.

[0033] like Figure 1As shown, in S5, during the campaign, the performance metrics of each campaign are statistically analyzed over time based on a sliding time window (e.g., 5 minutes). When the rate of change Δ of the performance metric falls below a preset negative threshold λ, the campaign is determined to be in a fatigue state, and automatic actions are taken to downgrade or switch the campaign, achieving early warning and proactive intervention. The performance metric is click-through rate (CTR); the rate of change Δ is the slope of the CTR change over time t within the sliding time window W, i.e., Δ = Performance metrics can also include ad impressions and conversion rates.

[0034] Furthermore, in this embodiment, the condition for determining that the creative idea has entered a fatigue state is: the rate of change Δ remains below the negative threshold λ for a preset duration. For example: when Continuously below the negative threshold If more than 10 minutes have passed, the system will automatically trigger a demotion policy, reducing its ranking for the next hour. The weighting was reduced by 30%.

[0035] like Figure 1 As shown, in S6, based on the online learning mechanism, real-time feedback data on exposure, clicks, and conversions after ad delivery is collected as incremental training samples. An incremental optimization algorithm is used to incrementally update the parameters of the deep cross-network (DCN) model to achieve adaptive optimization of the model. Specifically, the Adagrad optimization algorithm is used to update the DCN prediction model. The parameters are continuously and frequently updated, with an update cycle of minutes (e.g., every 5 minutes), enabling the model to complete adaptive learning without stopping delivery, thereby achieving dynamic adjustment and optimization of the delivery strategy.

[0036] As described above, the real-time optimization and delivery control method for advertising creatives based on artificial intelligence according to an embodiment of the present invention has the following beneficial effects: High-precision joint matching: By unifying and modeling multi-source data and using DCN to achieve high-order feature cross-matching, high-precision joint matching of "user-creative-scenario" is achieved, which significantly improves the fit between advertising creatives and users' immediate interests and placement scenarios, thereby effectively improving the click-through rate and conversion rate of advertisements.

[0037] High-precision prediction through deep learning: Constructing a deep cross-network (DCN) prediction model has a stronger fitting ability for high-order feature combination relationships, enabling more accurate prediction of campaign performance and providing highly reliable quantitative basis for campaign decisions.

[0038] Real-time proactive fatigue intervention: Introducing a creative fatigue identification algorithm based on sliding time window and time series change rate, enabling real-time identification and early warning of creative fatigue status, avoiding the passive and delayed judgment based on traditional exposure number thresholds, and enabling proactive intervention before the creative effect has significantly deteriorated.

[0039] Continuous Model Adaptation: Through online learning and incremental update mechanisms, the prediction model and delivery control strategy can be continuously and adaptively updated without stopping delivery. This enables rapid response to dynamic changes such as changes in user interests and fluctuations in traffic structure, significantly improving the system's real-time performance and environmental adaptability.

[0040] Integrated closed-loop control: This invention forms a complete closed-loop delivery control system at the system level, realizing the transformation from experience-driven to data intelligence-driven, and improving the level of automation and operational efficiency in large-scale advertising delivery scenarios.

[0041] The above combined with the appendix Figure 1 This paper describes a method for real-time selection and delivery control of advertising creatives based on artificial intelligence, according to an embodiment of the present invention. Furthermore, the present invention can also be applied to a control system for real-time selection and delivery of advertising creatives based on artificial intelligence.

[0042] like Figure 2 As shown, according to a second aspect of the present invention, an artificial intelligence-based real-time optimization and delivery control system for advertising creatives is provided, comprising: The asynchronous processing module 100 is used to asynchronously call the pre-trained visual model and text model when the advertising creative material is uploaded or updated, extract the visual feature vector and text feature vector of the creative material respectively, and fuse the two to generate a unified creative content feature vector. Stored on the feature platform; Response building module 200 is used to respond to received real-time advertising requests by querying the feature platform to obtain the corresponding candidate creative. and corresponding user profile characteristics And combined with the real-time acquired contextual features of the deployment environment Construct an initial set of feature vectors for prediction; The joint prediction module 300 is used to input the initial feature vector set into the Deep Cross Network (DCN) model; the core cross layer of the Deep Cross Network (DCN) model... , and Perform high-order nonlinear cross-coding between features to generate a joint feature vector. The deep side layers of the Deep Cross Network (DCN) model are based on It can simultaneously predict the click-through rate and conversion rate of each candidate creative through multi-task learning; The delivery calculation module 400 is used to calculate the comprehensive score S of each candidate creative based on the predicted click-through rate and conversion rate, sort them according to the comprehensive score S, and select the top-ranked creatives for real-time delivery. The monitoring and intervention module 500 is used to statistically analyze the changing trend of the performance indicators of each deployed creative over time based on a sliding time window during the campaign. When the rate of change of the performance indicator Δ is lower than the preset negative threshold λ, the creative is determined to have entered a fatigue state, and the module automatically performs a downgrading or switching operation on the creative. The learning update module 600 is used to collect real-time feedback data on exposure, clicks, and conversions after ad placement based on an online learning mechanism. This data serves as incremental training samples, and the parameters of the Deep Cross Network (DCN) model are incrementally updated using an incremental optimization algorithm to achieve adaptive optimization of the model.

[0043] The above combined with the appendix Figure 2 This invention describes an artificial intelligence-based real-time optimization and delivery control system for advertising creatives according to an embodiment of the present invention. Furthermore, the invention can also be applied to an electronic device.

[0044] like Figure 3 As shown, according to a third aspect of the present invention, an electronic device is provided, comprising: a memory 1, a processor 2, and a computer program 3, wherein the computer program 3 is stored in the memory 2, and the processor 1 executes the computer program 3 to perform a method for real-time selection and delivery control of advertising creatives based on artificial intelligence, as described in the first aspect.

[0045] The above combined with the appendix Figure 3 This invention describes an artificial intelligence-based real-time optimization and delivery control system for advertising creatives according to an embodiment of the present invention. Furthermore, the invention can also be applied to an electronic device.

[0046] According to a fourth aspect of the present invention, a computer-readable medium having processor-executable non-volatile program code, the program code causing the processor to perform a method for real-time selection and delivery control of advertising creatives based on artificial intelligence, as described in the first aspect.

[0047] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of a computer program from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the ASIC can reside within a device. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. The present invention also provides a program product including execution instructions stored in the readable storage medium. At least one processor of the device can read the execution instructions from the readable storage medium, and the at least one processor executes the execution instructions, causing the device to implement the artificial intelligence-based real-time optimization and delivery control method for advertising creatives provided in the various embodiments described above. In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0048] It should be noted that, in this specification, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0049] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A method for real-time selection and delivery control of advertising creatives based on artificial intelligence, characterized in that, It includes the following steps: When advertising creative materials are uploaded or updated, pre-trained visual and text models are asynchronously invoked to extract visual and text feature vectors of the creative materials, respectively, and then the two are fused to generate a unified creative content feature vector. Stored on the feature platform; In response to a received real-time ad request, the system queries the feature platform to retrieve the corresponding candidate creative. and corresponding user profile characteristics And combined with the real-time acquired contextual features of the deployment environment Construct an initial set of feature vectors for prediction; The initial feature vector set is input into the Deep Cross Network (DCN) model; the core cross layer of the DCN model... , and Perform high-order nonlinear cross-coding between features to generate a joint feature vector. The deep side layers of the Deep Cross Network (DCN) model are based on It can simultaneously predict the click-through rate and conversion rate of each candidate creative through multi-task learning; Based on the predicted click-through rate and conversion rate, calculate the comprehensive score S for each candidate creative, sort them according to the comprehensive score S, and select the top-ranked creatives for real-time delivery. During the campaign, the performance metrics of each campaign are statistically analyzed over time based on a sliding time window. When the rate of change of the performance metrics Δ is lower than the preset negative threshold λ, the campaign is determined to be in a fatigue state, and the campaign is automatically downgraded or switched. Based on the online learning mechanism, real-time feedback data on exposure, clicks, and conversions after ad placement is collected as incremental training samples. The parameters of the Deep Cross Network (DCN) model described in the incremental optimization algorithm are incrementally updated to achieve adaptive optimization of the model.

2. The method for real-time optimization and delivery control of advertising creatives based on artificial intelligence as described in claim 1, characterized in that, The visual model is a convolutional neural network (CNN) ResNet, and the text model is a natural language processing (NLP) BERT model.

3. The method for real-time optimization and delivery control of advertising creatives based on artificial intelligence as described in claim 1, characterized in that, The overall score S includes the predicted click-through rate, conversion rate, and expected revenue; ； in, For click-through rate; Conversion rate; For expected returns, The value after clicking; The bid from the advertiser.

4. The method for real-time optimization and delivery control of advertising creatives based on artificial intelligence as described in claim 1, characterized in that, The performance metric is click-through rate (CTR); the rate of change Δ is the slope of the CTR change with time t within the sliding time window W, i.e., Δ = .

5. The method for real-time optimization and delivery control of advertising creatives based on artificial intelligence as described in claim 1, characterized in that, The condition for determining that the creative idea has entered a state of fatigue is: the change rate Δ remains below the negative threshold λ for a period of time exceeding a preset duration.

6. The method for real-time optimization and delivery control of advertising creatives based on artificial intelligence as described in claim 1, characterized in that, The Adagrad optimization algorithm is used to incrementally update the parameters of the DCN model, with an update cycle of minutes.

7. The method for real-time optimization and delivery control of advertising creatives based on artificial intelligence as described in claim 1, characterized in that, When creative content is deployed in real time, cross-platform exposure control is also achieved by combining preset frequency control rules.

8. A real-time optimization and delivery control system for advertising creatives based on artificial intelligence, characterized in that, Include: The asynchronous processing module is used to asynchronously call pre-trained visual and text models when advertising creative materials are uploaded or updated. These models extract the visual and text feature vectors of the creative materials respectively, and then fuse them to generate a unified creative content feature vector. Stored on the feature platform; The response building module is used to respond to received real-time advertising requests by querying the feature platform to obtain the corresponding candidate creative. and corresponding user profile characteristics And combined with the real-time acquired contextual features of the deployment environment Construct an initial set of feature vectors for prediction; The joint prediction module is used to input the initial feature vector set into the Deep Cross Network (DCN) model; the core cross layer of the DCN model... , and Perform high-order nonlinear cross-coding between features to generate a joint feature vector. The deep side layers of the Deep Cross Network (DCN) model are based on It can simultaneously predict the click-through rate and conversion rate of each candidate creative through multi-task learning; The delivery calculation module is used to calculate the comprehensive score S of each candidate creative based on the predicted click-through rate and conversion rate, sort them according to the comprehensive score S, and select the top-ranked creatives for real-time delivery. The monitoring and intervention module is used to statistically analyze the changing trend of the performance indicators of each deployed creative over time based on a sliding time window during the campaign. When the rate of change Δ of the performance indicator is lower than the preset negative threshold λ, the creative is determined to be in a fatigue state, and the module automatically performs a downgrading or switching operation on the creative. The learning update module is used to collect real-time feedback data on exposure, clicks, and conversions after ad placement based on an online learning mechanism. This data serves as incremental training samples, and the parameters of the Deep Cross Network (DCN) model described in the incremental optimization algorithm are incrementally updated to achieve adaptive optimization of the model.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program, wherein the computer program is stored in the memory, and the processor executes the computer program to perform the real-time optimization and delivery control method for advertising creatives based on any one of claims 1 to 7.

10. A computer-readable medium having processor-executable non-volatile program code, characterized in that, The program code causes the processor to execute the real-time optimization and delivery control method for advertising creatives based on any one of claims 1-7.

Images