AI-based landing page structure optimization method and system, and electronic device

Abstract

The application provides an AI-based landing page structure optimization method and system and electronic equipment, and relates to the field of structure optimization. The method comprises the following steps: fusing a convolutional neural network and a gradient boosting tree to obtain an initial landing page structure AI deep learning model; taking historical landing page structure features and associated delivery data in a training sample set as inputs of the initial landing page structure AI deep learning model and taking conversion effect data in the training sample set as outputs of the initial landing page structure AI deep learning model to perform model training on the initial landing page structure AI deep learning model, and optimizing hyperparameters through a cross-validation method in the model training process to obtain an intermediate landing page structure AI deep learning model; regarding a landing page optimization process as a sequential decision problem and taking a conversion effect as a reward signal, and real-time receiving delivery data feedback and updating model parameters of the intermediate landing page structure AI deep learning model to obtain a final landing page structure AI deep learning model.

Description

Technical Field

[0001] This application relates to the field of structural optimization technology, and in particular to an AI-based landing page structural optimization method, system, and electronic device. Background Technology

[0002] Currently, in the context of mobile internet performance advertising, the landing page structure lacks precise matching with the target audience attributes, core selling points of the product, and characteristics of the promotional placement. This results in a low conversion rate (CVR) after users are redirected to the landing page, and the inability to quickly respond to data feedback for dynamic optimization leads to a poor user experience for the landing page structure. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-based method, system, and electronic device for optimizing landing page structure, in order to solve the technical problem of poor user experience in landing page delivery.

[0004] Firstly, this application provides an AI-based landing page structure optimization method, the method comprising: Obtain historical landing page structure features, associated campaign data, and corresponding conversion effect data, and construct a training sample set based on the historical landing page structure features, the associated campaign data, and the corresponding conversion effect data; wherein, the conversion effect data includes at least one of CVR, order rate, and retention rate; A convolutional neural network and a gradient boosting tree are fused to obtain an initial landing page structure AI deep learning model; wherein, the convolutional neural network is used to extract deep visual layout features in the landing page structure, and the gradient boosting tree is used to learn the non-linear correlation between features; The initial landing page structure AI deep learning model is trained by using the historical landing page structure features and the associated delivery data in the training sample set as inputs and the conversion effect data in the training sample set as outputs. The hyperparameters are optimized by cross-validation during the model training process to obtain an intermediate landing page structure AI deep learning model. The landing page optimization process is treated as a sequential decision problem, with conversion results as the reward signal. The model parameters of the intermediate landing page structure AI deep learning model are updated in real time based on the feedback of the campaign data, so that the model can adapt to the optimization needs of different campaign stages and obtain the final landing page structure AI deep learning model. The campaign stages include at least one of the cold start period, the scaling period, and the stabilization period.

[0005] In one possible implementation, using conversion results as a reward signal includes: The conversion potential of landing page structure is quantified using a composite analysis approach combining feature synergistic multiplication, Sigmoid interaction gain, and logarithmic data feedback enhancement, with the following formula. The feature synergistic multiplication reflects the synergistic adaptation effect of core dimensions; the Sigmoid interaction gain smooths the influence weight of the interactive experience using the Sigmoid function; and the logarithmic data feedback enhancement dynamically corrects the conversion results by incorporating historical data feedback gains, ensuring the conversion results align with the actual campaign scenario.

[0006] in, This indicates the conversion potential index of the landing page; The target audience attribute matching score represents the quantification of the degree of fit between the landing page's visual style, content description, and the target audience's profile. Product selling point alignment refers to the accuracy of the alignment between the core content displayed on the landing page and the core price of the product. To promote placement suitability, it means that the landing page loading speed and content length are evaluated in conjunction with the user behavior habits of the placement to assess the suitability of the placement. Represents the natural base in mathematical symbols; The interaction gain coefficient is obtained through training with historical data and is used to adjust the slope of the Sigmoid function and control the impact of the interaction experience on conversion potential. This is the interactive experience value, calculated based on metrics such as landing page loading speed, interactive operation steps, and ease of use of landing page buttons. The historical data feedback gain value represents the calculation result based on historical conversion data of similar landing pages in the same delivery scenario. The historical data feedback gain value is equal to the historical best CVR divided by the current scenario average CVR minus one, and is used to reflect the gain effect of historical high-quality experience on the current landing page. The conversion effect corresponding to the conversion potential of the landing page structure is used as a reward signal.

[0007] In one possible implementation, the Sigmoid interaction gain corresponds to user data of the touch operation interaction object; the method further includes: In response to a touch operation on a target landing page, the target terminal model corresponding to the touch operation is obtained, and the touch operation sliding trajectory, touch operation user habits, operation speed, touch operation contact area, touch operation contact size, and touch operation force are determined based on the touch operation. The user data of the touch operation interaction object is analyzed based on the target terminal model, the touch operation swiping trajectory, the touch operation user habits, the operation speed, the touch operation contact area, the touch operation contact size, and the touch operation force; wherein, the user data includes the user's gender and the user's age; Based on the user data, determine the target related delivery data for the target users corresponding to the user data; Based on multiple target-related delivery data, the comprehensive delivery data associated with the target landing page structure corresponding to the target landing page is determined.

[0008] In one possible implementation, analyzing user data of the touch operation interaction object based on the target terminal model, the touch operation swiping trajectory, the touch operation user habits, the operation speed, the touch operation contact area, the touch operation contact size, and the touch operation force includes: In response to the touch operation on the target landing page, the touch screen conductivity sensitivity of the touch operation is determined based on the touch operation; The user's skin texture and stratum corneum of the touch operation interaction object are determined based on the conductivity sensitivity of the touch screen, and the user's age is analyzed based on the user's skin texture and stratum corneum, the target terminal model, the touch operation sliding trajectory, the touch operation user habits, the operation speed, the touch operation contact area, the touch operation contact size, and the touch operation force.

[0009] In one possible implementation, the landing page structure has multiple grating structures on its display screen, and the different angles of the grating strips correspond to different display contents on the display screen; the method further includes: In response to detecting multiple viewing users corresponding to the page display terminal, the user group type of the viewing user is determined according to the user age and user gender of each of the multiple viewing users, and the angle of the raster bar set for the viewing user is determined according to the relative position of each viewing user relative to the page display terminal; The targeting data and corresponding display content for the viewing user are determined based on the user group type. The raster bar setting angle is controlled for each viewing user based on the display content and the raster bar setting angle for the viewing user. Different viewing users correspond to different raster bar setting angles and different display content.

[0010] In one possible implementation, it also includes: In response to detecting a position movement event of a first viewing user among the plurality of viewing users, a first relative position of the first viewing user's movement position relative to the page display terminal is determined based on the position movement event; Based on the first relative orientation transformation, the angle of the first raster strip for the first viewing user is set so that the angle of the first raster strip corresponds to and follows the movement position of the first viewing user.

[0011] In one possible implementation, it also includes: In response to the detection that the second relative position of the second viewing user among the plurality of viewing users relative to the page display terminal has not changed within a specified time period, it is determined that the second viewing user is standing and watching the second projected display content; The angle of the second grating strip corresponding to the second relative position is controlled to move to display the purchase intention data corresponding to the second display content, so as to guide the second viewing user to view the purchase intention data by moving the angle of the second grating strip; wherein, the purchase intention data includes purchase intention QR code content; Based on the second relative orientation, project the corresponding purchase intention QR code content below the location of the second viewing user to improve the conversion effect data after the second projected display content is displayed.

[0012] Secondly, this application provides an AI-based landing page structure optimization system, including: A construction module is used to acquire historical landing page structure features, associated delivery data and corresponding conversion effect data, and to construct a training sample set based on the historical landing page structure features, the associated delivery data and the corresponding conversion effect data; wherein, the conversion effect data includes at least one of CVR, order rate and retention rate; The fusion module is used to fuse the convolutional neural network and the gradient boosting tree to obtain an initial landing page structure AI deep learning model; wherein, the convolutional neural network is used to extract deep visual layout features in the landing page structure, and the gradient boosting tree is used to learn the non-linear correlation between features; The training module is used to train the initial landing page structure AI deep learning model by taking the historical landing page structure features and the associated delivery data in the training sample set as inputs to the initial landing page structure AI deep learning model and taking the conversion effect data in the training sample set as outputs to the initial landing page structure AI deep learning model. During the model training process, the hyperparameters are optimized by cross-validation to obtain an intermediate landing page structure AI deep learning model. The update module is used to treat the landing page optimization process as a sequential decision problem and use the conversion effect as a reward signal. It receives the feedback of the campaign data in real time and updates the model parameters of the intermediate landing page structure AI deep learning model so that the model can adapt to the optimization needs of different campaign stages and obtain the final landing page structure AI deep learning model. The campaign stages include at least one of the cold start period, the scaling period, and the stabilization period.

[0013] Thirdly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method described in the first aspect above.

[0014] Fourthly, this application also provides a computer-readable storage medium storing computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method described in the first aspect above.

[0015] This application brings the following beneficial effects: This application provides an AI-based landing page structure optimization method, system, and electronic device. It can acquire historical landing page structure features, associated advertising data, and corresponding conversion effect data, and construct a training sample set based on the historical landing page structure features, the associated advertising data, and the corresponding conversion effect data. The conversion effect data includes at least one of CVR, order rate, and retention rate. A convolutional neural network and a gradient boosting tree are fused to obtain an initial landing page structure AI deep learning model. The convolutional neural network is used to extract deep visual layout features from the landing page structure, and the gradient boosting tree is used to learn the non-linear relationships between features. The historical landing page structure features and the associated advertising data in the training sample set are used as input to the initial landing page structure AI deep learning model, and the conversion effect data in the training sample set is used as the input to the model. The output of the initial landing page structure AI deep learning model is used to train the initial landing page structure AI deep learning model. During the model training process, the hyperparameters are optimized using cross-validation to obtain an intermediate landing page structure AI deep learning model. The landing page optimization process is treated as a sequential decision problem, and the conversion effect is used as a reward signal. The model parameters of the intermediate landing page structure AI deep learning model are updated in real time with the feedback of the campaign data, so that the model can adapt to the optimization needs of different campaign stages to obtain the final landing page structure AI deep learning model. The campaign stages include at least one of the cold start period, the scaling period, and the stabilization period. In this solution, the conversion potential is accurately quantified through a composite logic formula, achieving deep adaptation between the landing page structure and the campaign scenario. Compared with traditional optimization methods, this significantly improves the landing page conversion rate (CVR) and solves the technical problem of low campaign experience of the landing page structure.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the AI-based landing page structure optimization method provided in this application embodiment; Figure 2 Another flowchart illustrating the AI-based landing page structure optimization method provided in this application embodiment; Figure 3 A schematic diagram of an AI-based landing page structure optimization system provided in this application embodiment; Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this application, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0021] Currently, the iteration of landing pages for different industries and product categories relies on human experience, resulting in a poor user experience for landing page structures. Therefore, this application provides an AI-based landing page structure optimization method, system, and electronic device, which can solve the technical problem of poor user experience for landing page structures.

[0022] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart illustrating an AI-based landing page structure optimization method provided in an embodiment of this application. Figure 1 As shown, the method includes: Step S110: Obtain historical landing page structure features, associated delivery data and corresponding conversion effect data, and construct a training sample set based on historical landing page structure features, associated delivery data and corresponding conversion effect data.

[0024] The conversion performance data includes at least one of CVR, order rate, and retention rate. In this step, a training sample set is constructed based on historical campaign data. The sample set includes landing page structure features, related campaign data, and corresponding conversion performance data (CVR, order rate, retention rate, etc.). The sample data is then normalized to eliminate the differences in the dimensions of the data.

[0025] Step S120: The convolutional neural network and gradient boosting tree are fused to obtain the initial landing page structure AI deep learning model.

[0026] Among them, the convolutional neural network is used to extract deep visual layout features in the landing page structure, and the gradient boosting tree is used to learn the non-linear relationship between features.

[0027] As an optional implementation, the AI ​​optimization model training module employs a deep learning model (integrating convolutional neural networks and gradient boosting trees), using landing page structural features and related ad delivery data as input, and conversion performance data as the output target for model training. Specifically, the convolutional neural network is used to extract deep visual layout features from the landing page structure, while the gradient boosting tree is used to learn the non-linear relationships between features.

[0028] Step S130: The initial landing page structure AI deep learning model is trained by using the historical landing page structure features and related delivery data in the training sample set as inputs and the conversion effect data in the training sample set as outputs. During the training process, the hyperparameters are optimized by cross-validation to obtain the intermediate landing page structure AI deep learning model.

[0029] It should be noted that during model training, hyperparameters are optimized using cross-validation to improve the model's generalization ability.

[0030] In one alternative implementation, the Sigmoid interaction gain corresponds to user data of the touch operation interaction object; such as... Figure 2 As shown, the method may further include the following steps: Step S210: In response to a touch operation on the target landing page, obtain the target terminal model corresponding to the touch operation, and determine the touch operation swiping trajectory, touch operation user habits, operation speed, touch operation contact area, touch operation contact size, and touch operation force based on the touch operation. Step S220: Analyze the user data of the touch operation interaction object based on the target terminal model, touch operation swiping trajectory, touch operation user habits, operation speed, touch operation contact area, touch operation contact size, and touch operation force; wherein, the user data includes user gender and user age; Step S230: Determine the target related delivery data for the target users corresponding to the user data based on the user data; Step S240: Determine the comprehensive delivery data associated with the target landing page structure corresponding to the target landing page based on multiple target-related delivery data.

[0031] In existing technologies, traditional user profiling typically relies on user-filled registration information, cookie tracking, or third-party data purchases, which faces privacy compliance risks (such as GDPR and personal information protection laws) and low user cooperation.

[0032] This proposed solution utilizes the natural physical characteristics of touch input (fingerprint contact area, pressure, swipe inertia, and other biomechanical features that differ in gender and age) generated when users browse landing pages to calculate the user's gender and age in real time without the user's awareness or additional authorization. This avoids the compliance risks of directly collecting sensitive personal information and solves the problem of not being able to identify cold-start users (users who are not logged in).

[0033] Furthermore, regarding the improvement of conversion rate CTR / CVR, there are significant statistical differences in finger physiology (affecting contact area) and operating habits (affecting speed, trajectory, and force) among users of different ages and genders. For example, young men may operate faster and with greater force; older users may have a larger contact area, slower speed, and a smoother trajectory.

[0034] This application's solution utilizes multi-dimensional cross-analysis of touch parameters to more accurately target specific demographics than relying solely on device model identification. Real-time matching of ad content with derived user attributes (e.g., displaying beauty products to users identified as "young women" and cars to "middle-aged men") significantly improves ad relevance, directly boosting click-through rates and conversion rates.

[0035] This technical solution infers the user's macro attributes (gender, age) by deeply mining the user's micro-touch behavior characteristics (such as force, contact area, swipe trajectory, speed, etc.) and combining them with the terminal model, thereby achieving precise matching of advertisements.

[0036] This application's solution achieves high-precision user profiling and personalized ad delivery based on "implicit biometric behavioral characteristics" without requiring users to actively register or authorize personal information. It transforms "touch behavior data" into "business insight data." Its primary technical benefit is that, while ensuring user privacy and compliance, it significantly improves the accuracy of ad identification and conversion efficiency for users who are not logged in or are anonymous, solving the critical problem of "not knowing who is in front of the screen" in mobile advertising scenarios.

[0037] As an optional implementation, a multi-dimensional landing page structure feature system is constructed through a landing page structure feature extraction module. This system includes content layout features (such as the display location of core selling points, the placement area of ​​call-to-action buttons, and the hierarchy of product information layout), interaction design features (such as loading speed, form filling complexity, and page navigation path), and visual presentation features (such as background color, font type, and the proportion of images / videos). These feature parameters are automatically extracted from the landing page using webpage parsing technology, forming a structured feature dataset. Furthermore, by combining advertising-related data, including target audience attribute data (gender, age, region, interest tags, etc.), product core selling point data (functional features, price advantages, welfare policies, etc.), and promotional placement data (such as user behavior data for placements like WeChat official accounts and mini-programs), a feature association mapping library is established.

[0038] In one possible implementation, analyzing user data of the touch interaction object based on the target terminal model, touch operation swipe trajectory, touch operation user habits, operation speed, touch operation contact area, touch operation contact size, and touch operation force can specifically include the following steps: In response to touch operations targeting the landing page, the system determines the touch screen's conductivity sensitivity based on the touch operation; it then determines the user's skin texture and stratum corneum based on the touch screen's conductivity sensitivity, and analyzes the user's age based on the user's skin texture, stratum corneum, target terminal model, touch operation swipe trajectory, touch operation user habits, operation speed, touch operation contact area, touch operation contact size, and touch operation force.

[0039] By leveraging the correlation between human bioelectric properties and skin physiological structure, a highly robust age estimation model based on "physiological-behavioral dual features" was constructed, which significantly improved the accuracy of user age group identification (especially across generations) in scenarios lacking explicit identity information.

[0040] In existing technologies, age determination based solely on behavioral data such as swipe trajectory, speed, and force is easily influenced by the user's current state (e.g., fatigue, mood), imitation of operating habits, or specific scenarios (e.g., operating while walking), leading to misjudgments. For example, a young person operating slowly might be misjudged as an elderly person.

[0041] This application's solution overcomes the limitations of single-behavioral analysis by introducing a physiological fingerprint dimension. It uses the touchscreen's conductivity sensitivity to infer the thickness / state of the skin's stratum corneum, introducing physiological characteristics that are difficult to fake. It's important to note that capacitive screens rely on human body current sensing. Significant physiological differences exist in the thickness of the skin's stratum corneum, water content, sebum secretion, and dermal conductivity among different age groups (e.g., children have thinner stratum corneums and higher water content, resulting in higher conductivity sensitivity; older people have dry skin, thicker and harder stratum corneums, leading to relatively lower conductivity sensitivity and greater signal attenuation). By using this "physiological hard indicator" as the core weight for age determination, the application effectively corrects for biases caused by relying solely on behavioral habits, making age estimation closer to the user's true physiological age.

[0042] Moreover, this solution eliminates the need for additional biosensors (such as fingerprint scanners or heart rate monitors), directly reusing existing capacitive touchscreen hardware and mining underlying electrical signal data (conductivity sensitivity) through algorithms. It acquires biometrically valuable feature data at low cost without the user's awareness or the need for additional authorization of biometric information (such as facial or fingerprint images). This reduces hardware costs, avoids compliance risks associated with collecting highly sensitive biometric image data, achieves privacy-friendly biometric analysis, and enables contactless, seamless biometric data collection.

[0043] Furthermore, by combining dual verification based on stratum corneum (physiological) and user habits (behavioral), the system can more precisely differentiate adjacent age groups (e.g., "20-25 years old" and "45-50 years old"), and even identify subgroups with specific skin characteristics (e.g., young people with premature skin aging or elderly people with excellent skin condition). Based on more accurate age and skin type inferences, the advertising system can not only match products that "match age," but also further match products related to "skin type needs" (e.g., recommending gentle skincare products to users inferred to have "thin stratum corneum / sensitive skin," regardless of their inferred age), thereby upgrading from "demographic targeting" to "physiological characteristic targeting" and significantly improving conversion rates. Therefore, it enhances the granularity of user profiles and the accuracy of ad matching.

[0044] This solution transforms the "signal noise" (differences in conductivity sensitivity) of touchscreens into "biosignals" (skin texture and age characteristics). Its most significant technical achievement is the establishment of a low-cost, highly privacy-secure, and interference-resistant user age and physiological estimation mechanism, solving the key challenge of "missing physiological attributes" in accurate profiling of anonymous mobile users.

[0045] Step S140: Treat the landing page optimization process as a sequential decision problem and use the conversion effect as a reward signal. Receive real-time feedback on the campaign data and update the model parameters of the intermediate landing page structure AI deep learning model so that the model can adapt to the optimization needs of different campaign stages, thus obtaining the final landing page structure AI deep learning model.

[0046] The deployment phase includes at least one of the following: cold start period, initial growth period, and stabilization period.

[0047] By introducing a reinforcement learning mechanism, the landing page optimization process is treated as a sequential decision problem. Improved conversion rate is used as a reward signal. The model receives real-time feedback from campaign data and updates model parameters, enabling the model to adapt to the optimization needs of different campaign stages (cold start, initial growth, and stable period).

[0048] As one possible implementation, using conversion results as a reward signal, the above-mentioned method may specifically include the following steps: The conversion potential of landing page structure is quantified using a composite analysis approach combining feature synergistic multiplication, Sigmoid interaction gain, and logarithmic data feedback enhancement, with the following formula. Feature synergistic multiplication reflects the synergistic adaptation effect of core dimensions; Sigmoid interaction gain smooths the impact weight of the interactive experience using the Sigmoid function; and logarithmic data feedback enhancement dynamically corrects the conversion results by incorporating historical data feedback gains, ensuring the conversion results align with the actual campaign scenario.

[0049] in, This indicates the conversion potential index of the landing page; The target audience attribute matching score represents the quantification of the degree of fit between the landing page's visual style, content description, and the target audience's profile. Product selling point alignment refers to the accuracy of the alignment between the core content displayed on the landing page and the core price of the product. To promote placement suitability, it means that the landing page loading speed and content length are evaluated in conjunction with the user behavior habits of the placement to assess the suitability of the placement. Represents the natural base in mathematical symbols; The interaction gain coefficient is obtained through training with historical data and is used to adjust the slope of the Sigmoid function and control the impact of the interaction experience on conversion potential. This is the interactive experience value, calculated based on metrics such as landing page loading speed, interactive operation steps, and ease of use of landing page buttons. The historical data feedback gain value represents the calculation result based on historical conversion data of similar landing pages in the same delivery scenario. The historical data feedback gain value is equal to the historical best CVR divided by the current scenario average CVR minus one, and is used to reflect the gain effect of historical high-quality experience on the current landing page. The conversion effect corresponding to the conversion potential of the landing page structure is used as the reward signal.

[0050] For example, the intelligent optimization solution generation module determines the landing page conversion potential index (P). The formula for calculating the landing page conversion potential index is based on the following principle: through a composite calculation logic of "feature synergistic product + Sigmoid interaction gain + data feedback logarithmic enhancement", the conversion potential of the landing page structure is quantified. This not only reflects the synergistic adaptation effect of the core dimensions, but also smooths the influence weight of the interactive experience through the Sigmoid function. Finally, it combines historical data feedback gain to achieve dynamic correction, ensuring that the results are more in line with the actual deployment scenario.

[0051] in, The target audience attribute matching score (1-10) is quantified based on the degree of fit between the landing page's visual style, content description, and the target audience profile (age, gender, interests, etc.). The higher the matching score, the larger the score. The product selling point matching score (1-10) measures the accuracy of the correspondence between the core content displayed on the landing page and the core selling points of the product (functions, price, benefits, etc.). The more prominent the core selling points are and the more accurate the correspondence, the higher the score. The ad placement suitability score (1-10) is determined by considering user behavior habits (e.g., users on WeChat Moments prefer short and fast-paced content, while users on Youlianghui have a higher tolerance for content). The score assesses the suitability of the landing page loading speed, content length, and other factors with the ad placement. The better the suitability, the higher the score. The interaction design experience score (0-5) is calculated based on indicators such as landing page loading speed, form filling steps, and ease of clicking call-to-action buttons. The better the experience, the higher the score. The interaction gain coefficient (a positive number) is obtained through training with historical data and is used to adjust the slope of the Sigmoid function, thereby controlling the impact of the interaction experience on conversion potential. The historical data feedback gain value (≥0) is calculated based on historical conversion data of similar landing pages under the same advertising scenario. The formula is as follows: This reflects the positive impact of past successful experiences on the current landing page.

[0052] For the above optimization scheme generation logic, for example, input the target audience attributes, core selling points of the product, and promotional placement information for the current ad campaign, and calculate the current landing page conversion potential index (P) using the above formula. Set a conversion potential threshold (P0, determined based on industry averages and client KPI targets): When P < P0 × 0.8, it is determined to be a high optimization priority. The system optimizes all dimensions at the same time and generates a "full-dimensional reconstruction plan", including visual style reshaping, core selling points being placed at the top, layout adaptation adjustment, and simplified interaction process. When P0×0.8≤P<P0, it is determined to be of medium optimization priority, and the system identifies it. The two dimensions with the lowest median values ​​are used to generate "key dimension optimization plans", such as adjusting the content description to improve audience matching and optimizing the logic of selling points display to improve relevance. When P≥P0, it is determined to be a low optimization priority, and only applies to... and The corresponding details are fine-tuned to generate a "detailed iteration plan", such as optimizing the button click hotspot and adjusting the page loading compression ratio.

[0053] In this embodiment of the application, the k value and the quantitative logic of each dimension in the above formula can be dynamically adjusted based on different promotion positions and campaign stages. The model can adapt to different optimization needs in the cold start period, the scaling period, and the stable period. In the cold start period, it helps to quickly accumulate conversion data. In the scaling period and the stable period, it continuously optimizes the backend conversion effect, ensuring the competitiveness of the landing page throughout the entire advertising campaign, thereby adapting to the needs of multiple scenarios.

[0054] As a possible real-time approach, an industry and category landing page structure template library is established. Based on the historical best landing page characteristics of different industries (e-commerce, life services, audio, live streaming, etc.) and different categories (beauty, footwear, fresh food, digital products, etc.), an industry-specific basic template is generated. The model makes personalized adjustments to the basic template based on the current deployment scenario and the formula calculation results, generating multiple candidate optimization solutions.

[0055] In one optional implementation, the effect verification and iteration module employs an A / B testing mechanism. The generated candidate optimization solutions are simultaneously applied to the advertising scenario, and data such as impressions, clicks, and conversions for different solutions are collected in real time. Key metrics such as conversion rate (CVR) and cost per conversion (CPA) are calculated for each solution. Based on the metric comparison results, the optimal landing page structure solution is selected and officially launched. Simultaneously, test data is fed back to the AI ​​optimization model for updates. Values ​​and quantification logic for each dimension; establish an optimization solution archive library, and store the optimal solutions in different scenarios with tags to support rapid reuse in subsequent similar deployment scenarios.

[0056] As an example, the system deployment and monitoring module adopts a distributed deployment architecture, supporting landing page structure parsing, model calculation, and solution generation in high-concurrency scenarios, ensuring that the system response speed meets the real-time optimization needs of ad placement (single optimization solution generation time ≤ 3 minutes). Through the built-in data monitoring module, the operational status of the landing page after its launch is monitored in real time, including page load success rate, user dwell time, conversion funnel data, etc. When core indicators experience abnormal fluctuations (such as a sudden drop in conversion rate exceeding a preset threshold), the model is automatically triggered to recalculate the conversion potential index (P) and generate an emergency adjustment plan.

[0057] In this embodiment, the conversion potential is accurately quantified through a compound logic formula, achieving deep adaptation between the landing page structure and the advertising scenario. Compared with traditional optimization methods, this significantly improves the landing page conversion rate (CVR), with an average increase of over 25% in landing page CVR. Furthermore, the stability of landing page optimization effects across different industries and product categories is improved by 40%.

[0058] Moreover, without relying on the experience and judgment of professional designers and operators, the system can automatically generate targeted optimization solutions through formula-based quantitative analysis, shortening the landing page iteration cycle from the traditional 3-7 days to within 1 day, reducing labor costs by more than 50%, and lowering landing page iteration costs.

[0059] By updating formula parameters and quantification logic through real-time data feedback, a closed-loop mechanism of "data collection - model optimization - solution implementation - effect feedback" is formed, enabling the landing page structure to dynamically adapt to market changes, user preference changes, and product iteration needs, maintaining a high conversion rate for advertising in the long term, and achieving data-driven closed-loop optimization.

[0060] In some embodiments, the display screen of the page display terminal corresponding to the landing page structure is provided with multiple grating structures, and the different grating strip settings angles of the grating structures correspond to different display content on the display screen; the method may further include the following steps: In response to the detection of multiple viewing users corresponding to the page display terminal, the user group of the viewing user is determined according to the user age and gender of each viewing user, and the angle of the raster bar set for the viewing user is determined according to the relative position of each viewing user relative to the page display terminal. The targeting data and corresponding display content for each user are determined based on the user group type. The raster bar angle is then controlled for each user based on the display content and the raster bar angle settings for each user. Different users correspond to different raster bar angle settings and display content.

[0061] The above-mentioned determination of the raster bar setting angle for each viewing user based on the relative position of each viewing user relative to the page display terminal may specifically include the following steps: determining the raster bar setting angle for each viewing user based on the relative position of each viewing user relative to the page display terminal using the following calculation formula:

[0062] in, θi Set the angle / viewpoint angle for the raster strip, for the i-th... i For each user viewing the image, the system needs to set the raster optical opening angle or view offset angle. This angle determines which set of pixels the user can see from their position. It represents the angle from the normal direction of the raster plane (usually directly in front of the screen) to the direction pointing towards the user's eyes.

[0063] This represents the relative horizontal offset, specifically the horizontal distance (lateral offset) of the user's position relative to the center point of the screen. It is calculated from the "user position" captured by a camera or sensor. If the user is in the exact center of the screen, D=0; if the user is on the left, D is negative; and if the user is on the right, D is positive.

[0064] L This is the relative vertical distance / viewing distance, representing the vertical straight-line distance from the user's position to the screen plane of the display terminal (i.e., how far the user is from the screen). It is also calculated from the "user position" data.

[0065] This indicates that the horizontal offset D D Viewing distance L L The ratio is used to calculate the actual viewing angle in degrees using the arctangent function. This is the user's absolute spatial angle relative to the area directly in front of the screen.

[0066] This is the user group type compensation coefficient / display content reference angle, representing a fine-tuning offset of the basic viewing angle based on the user group type (determined by user age and gender). This is the embodiment of "determining the displayed content based on the user group type" in the formula. For example, if children need to view content at a lower position, or if the optimal viewing angle height needs to be adjusted (considering height differences), this coefficient can be used to ensure that specific content is projected precisely at the eye level of a specific group of people.

[0067] In public display scenarios, personalized advertising displays that achieve "space reuse" mean that different users in front of the same screen can see only specific advertising content that is exclusive to their own attributes (age, gender) at the same time and on the same screen, without interfering with each other.

[0068] This application breaks through the limitations of broadcast-style ad delivery, achieving spatial isolation and precise distribution of content on a single screen: In existing technologies, traditional outdoor or public landing page screens are "broadcast-style," with all passersby seeing the same set of advertising materials. This often results in advertising content catering only to the average taste of the general public, failing to differentiate for specific groups (e.g., showing only games to young men while showing health products to elderly women), leading to significant wasted exposure. This application, however, utilizes the directional light-emitting characteristics of a grating structure to optically divide the screen space into multiple independent "visual channels." The system calculates a unique grating angle based on each user's real-time location, making the light emitted from the screen directionally selective. For example, a young person standing on the left sees a game ad, while an elderly person on the right sees a health ad; neither can see the other's content. This physically enables parallel transmission of multiple contents on a single screen, greatly improving the advertising capacity and targeting per unit screen area.

[0069] This application solution not only relies on static user attributes (age, gender) but also incorporates dynamic relative positioning. The system can respond to user movement in real time. When a user moves from the left side of the screen to the right, or when a new user enters the field of view, the raster angle and displayed content will dynamically adjust accordingly. This "human-centric, image-driven" interaction ensures that no matter what angle the user is standing at, as long as they are detected, they will receive the most suitable targeting data for their demographic (age / gender), achieving true contextualized precision marketing and realizing dynamic real-time matching based on both physiological and location-based dimensions.

[0070] Furthermore, for advertisers, this means every exposure is a "effective exposure" because the content is specifically targeted at this demographic, significantly increasing click-through rates and conversion rates. For users, it avoids viewing irrelevant or even offensive ads (e.g., men don't need to see ads for feminine hygiene products), improving the browsing experience. Simultaneously, this process is typically based on macroscopic features captured by cameras (age / gender / location) rather than biometrics (such as facial recognition IDs), achieving precise targeting while relatively reducing the risk of personal privacy leaks, thus improving ad conversion rates while protecting user privacy.

[0071] This technical solution upgrades public screens from passive viewing to intelligent interactive terminals that actively perceive and project content. Its most significant technological advantage lies in using optical principles to construct parallel virtual advertising spaces in physical space, completely solving the problem of public screens being unable to simultaneously display differentiated content to multiple users, thus maximizing the utilization of advertising resources.

[0072] In some embodiments, the method may further include the following steps: in response to detecting a position movement event of a first viewing user among a plurality of viewing users, determining a first relative orientation of the first viewing user's movement position relative to the page display terminal based on the position movement event; and setting an angle for a first raster bar for the first viewing user based on the first relative orientation change, so that the angle of the first raster bar correspondingly follows the movement position of the first viewing user.

[0073] In this application, the solution achieves continuous dynamic tracking of naked-eye 3D / multi-viewpoint display content, ensuring that the user is always in the best viewing area (Sweet Spot) during movement, completely eliminating screen crosstalk, black screen or content jump caused by user displacement, and providing a smooth and uninterrupted personalized visual experience.

[0074] In existing technologies, lenticular stereoscopic displays or multi-viewpoint screens typically have a fixed "viewing cone angle" or "optimal viewing point." Once the user moves out of this narrow physical range, the image seen will experience crosstalk (the left eye sees the image of the right eye, resulting in ghosting / blurring) or will simply become a black screen / invalid content. This restricts the user's freedom of movement, forcing viewers to stand still to view the image.

[0075] In this application, by detecting positional movement events and calculating the new relative orientation in real time, the system dynamically adjusts the angle of the raster strips (or the corresponding pixel mapping). This is equivalent to making the "optimal viewing point" follow the user's movement. Regardless of whether the user moves left, right, or forward and backward, the system can reconstruct the light path in real time, ensuring that the user's eyes always receive the correct light beam. This transforms the originally static "narrow viewing angle" screen into an intelligent display with wide-area dynamic tracking capabilities. It solves the pain point of narrow viewing area and expands the effective viewing range.

[0076] This application emphasizes "following" the moving position, meaning that angle changes are continuous or frequently updated. On traditional static raster screens, users may experience a jumpy experience of "clear -> blurry -> clear" when moving, or abrupt transitions between content from different viewpoints. This solution, through real-time calculation and fine-tuning of the raster angle, makes the switching of displayed content visually smooth and continuous. Users are unaware of the technology's intervention; they only feel that the image is always clearly "locked" in front of them, greatly enhancing visual immersion and the naturalness of interaction. It eliminates visual gaps and achieves a seamless, smooth transition.

[0077] Furthermore, in scenarios with "multiple viewing users" (in the context of the preceding text), everyone is moving. This mechanism can independently process the movement trajectory of each user (such as the "first viewing user"). Even if multiple people are moving around in front of the screen simultaneously, the system can calculate and maintain their individual raster angle for each person. This ensures that in complex, dynamic crowd environments, each user can continuously and stably see their specific content, without losing personalized advertising information due to their own movement or obstruction by others, thus guaranteeing the continuity and effectiveness of advertising delivery and ensuring the stability of independent tracking in multi-user scenarios.

[0078] Therefore, the technical solution of this application upgrades the lenticular display from "static fixed point" to "dynamic follow-up". Its most important technical effect is to break the strict limitation of the lenticular screen on the viewer's position. Through real-time optical path reconstruction, users can enjoy a stable, clear and exclusive naked-eye 3D / personalized visual experience while moving freely, which greatly improves the practicality of public display terminals and user retention rate.

[0079] In some embodiments, the method may further include the following steps: In response to the detection that the second relative position of the second viewing user among multiple viewing users has not changed relative to the page display terminal within a specified time period, it is determined that the second viewing user is standing and watching the second displayed content. The angle of the second grating strip corresponding to the second relative position is controlled to move to display the purchase intention data corresponding to the second display content, so as to guide the second viewing user to view the purchase intention data by moving the angle of the second grating strip; wherein, the purchase intention data includes purchase intention QR code content; Based on the second relative orientation, project the corresponding purchase intention QR code content below the location of the second viewing user to improve the conversion effect data of the second projected content.

[0080] In this embodiment of the application, an immersive closed-loop marketing scenario of "conversion upon stopping" is constructed. Through the dual interactive mechanism of "eye-guided (grating angle movement) + spatial guidance (ground projection)," the static viewing behavior of users is seamlessly transformed into actual QR code scanning and purchase actions, which significantly improves the conversion rate of advertising from "exposure" to "transaction."

[0081] In existing technologies, traditional advertising screens play content in a loop regardless of whether the user is interested or not, or forcibly pop up a QR code when the user is not ready, which can easily cause resentment or be ignored.

[0082] This proposed solution uses the logic of "no changes within a specified time" to intelligently identify high-potential users who are genuinely interested in the content and stay engaged. Subsequent actions are only triggered after user interest is confirmed. This "on-demand response" mechanism ensures that marketing actions occur at the moment when users are most receptive, greatly reducing distractions and increasing user acceptance of subsequent purchase information. It accurately captures moments of high intent, achieving a "zero-disruption" push notification effect.

[0083] In this application, the angle of the grating strip is controlled to "move," producing dynamic visual effects on the naked-eye 3D screen (such as a QR code "flying" off the screen or an arrow pointing downwards), directly attracting the user's attention to purchase information.

[0084] By projecting a QR code onto the ground below the user's location, this not only solves the pain points of "not being able to find the code" or "not being able to see it clearly due to glare" when scanning with a handheld phone, but also utilizes the human eye's natural habit of looking down at the phone, forming a natural visual flow and spatial guidance of "screen attraction -> eye movement -> ground landing point -> QR code payment".

[0085] This cross-dimensional visual guidance (from the screen in the air to the ground beneath your feet) creates a strong sense of spatial interaction and ritual, making it easier for users to subconsciously complete the scanning action and significantly shortening the decision-making path. It achieves the construction of a three-dimensional visual guidance path using "dynamic lenticular combined with ground projection".

[0086] This application also addresses the specific relative location of the "second viewing user." This means that the dynamic guidance and ground projection seen by each user who stops at the site are independently generated and precisely matched in location. For example, if user A is standing on the left, the QR code dynamically moves to the left and is projected at A's feet; if user B is standing on the right, it independently serves B. This independence not only avoids confusion in multi-user scenarios but also accurately attributes specific scanning behaviors (conversion data) to specific user locations and the content being displayed at that time, providing high-precision data support for subsequent ROI (Return on Investment) analysis and user profile optimization. It achieves independent conversion tracking and data closure for each user.

[0087] In this application, the passively viewed "display terminal" is upgraded to an actively guided "transaction terminal." Its main technical effect is to construct the shortest and most natural QR code payment path the moment the user's interest is generated through intelligent behavior recognition and multimodal visual guidance (dynamic screen grating + ground projection), thereby maximizing the value of each customer and achieving a combination of brand and performance in the campaign.

[0088] Figure 3 A schematic diagram of an AI-based landing page structure optimization system is provided. (For example...) Figure 3As shown, the AI-based landing page structure optimization system 300 includes: The construction module 301 is used to acquire historical landing page structure features, associated delivery data and corresponding conversion effect data, and construct a training sample set based on the historical landing page structure features, the associated delivery data and the corresponding conversion effect data; wherein, the conversion effect data includes at least one of CVR, order rate and retention rate; The fusion module 302 is used to fuse the convolutional neural network and the gradient boosting tree to obtain an initial landing page structure AI deep learning model; wherein, the convolutional neural network is used to extract deep visual layout features in the landing page structure, and the gradient boosting tree is used to learn the nonlinear correlation between features; The training module 303 is used to train the initial landing page structure AI deep learning model by taking the historical landing page structure features and the associated delivery data in the training sample set as inputs to the initial landing page structure AI deep learning model and taking the conversion effect data in the training sample set as outputs to the initial landing page structure AI deep learning model. During the model training process, the hyperparameters are optimized by cross-validation to obtain an intermediate landing page structure AI deep learning model. The update module 304 is used to treat the landing page optimization process as a sequential decision problem and use the conversion effect as a reward signal. It receives the feedback of the delivery data in real time and updates the model parameters of the intermediate landing page structure AI deep learning model so that the model can adapt to the optimization needs of different delivery stages and obtain the final landing page structure AI deep learning model. The delivery stage includes at least one of the cold start period, the volume increase period, and the stable period.

[0089] The AI-based landing page structure optimization system provided in this application has the same technical features as the AI-based landing page structure optimization method provided in the above embodiments, so it can also solve the same technical problems and achieve the same technical effects.

[0090] An electronic device provided in this application embodiment, such as Figure 4 As shown, the electronic device 400 includes a processor 402 and a memory 401. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method provided in the above embodiments.

[0091] See Figure 4 The electronic device also includes a bus 403 and a communication interface 404. The processor 402, the communication interface 404 and the memory 401 are connected via the bus 403. The processor 402 is used to execute executable modules, such as computer programs, stored in the memory 401.

[0092] The memory 401 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 404 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0093] Bus 403 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0094] The memory 401 is used to store programs. After receiving an execution instruction, the processor 402 executes the program. The method executed by the apparatus defined by the process disclosed in any of the preceding embodiments of this application can be applied to the processor 402 or implemented by the processor 402.

[0095] Processor 402 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 402 or by instructions in software form. The processor 402 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 401, and processor 402 reads the information from memory 401 and, in conjunction with its hardware, completes the steps of the above method.

[0096] Corresponding to the above-described AI-based landing page structure optimization method, this application embodiment also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to perform the steps of the above-described AI-based landing page structure optimization method.

[0097] The AI-based landing page structure optimization system provided in this application embodiment can be specific hardware on a device or software or firmware installed on the device. The device provided in this application embodiment has the same implementation principle and technical effects as the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0098] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0099] For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0100] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0101] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0102] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the AI-based landing page structure optimization method described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, external hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0103] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0104] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. An AI-based method for optimizing landing page structure, characterized in that, The method includes: Obtain historical landing page structure features, associated campaign data, and corresponding conversion effect data, and construct a training sample set based on the historical landing page structure features, the associated campaign data, and the corresponding conversion effect data; wherein, the conversion effect data includes at least one of CVR, order rate, and retention rate; A convolutional neural network and a gradient boosting tree are fused to obtain an initial landing page structure AI deep learning model; wherein, the convolutional neural network is used to extract deep visual layout features in the landing page structure, and the gradient boosting tree is used to learn the non-linear correlation between features; The initial landing page structure AI deep learning model is trained by using the historical landing page structure features and the associated delivery data in the training sample set as inputs and the conversion effect data in the training sample set as outputs. The hyperparameters are optimized by cross-validation during the model training process to obtain an intermediate landing page structure AI deep learning model. The landing page optimization process is treated as a sequential decision problem, with conversion results as the reward signal. The model parameters of the intermediate landing page structure AI deep learning model are updated in real time based on the feedback of the campaign data, so that the model can adapt to the optimization needs of different campaign stages and obtain the final landing page structure AI deep learning model. The campaign stages include at least one of the cold start period, the scaling period, and the stabilization period.

2. The method according to claim 1, characterized in that, The use of conversion results as a reward signal includes: The conversion potential of landing page structure is quantified using a composite analysis approach combining feature synergistic multiplication, Sigmoid interaction gain, and logarithmic data feedback enhancement, with the following formula. The feature synergistic multiplication reflects the synergistic adaptation effect of core dimensions; the Sigmoid interaction gain smooths the influence weight of the interactive experience using the Sigmoid function; and the logarithmic data feedback enhancement dynamically corrects the conversion results by incorporating historical data feedback gains, ensuring the conversion results align with the actual campaign scenario. in, This indicates the conversion potential index of the landing page; The target audience attribute matching score represents the quantification of the degree of fit between the landing page's visual style, content description, and the target audience's profile. Product selling point alignment refers to the accuracy of the alignment between the core content displayed on the landing page and the core price of the product. To promote placement suitability, it means that the landing page loading speed and content length are evaluated in conjunction with the user behavior habits of the placement to assess the suitability of the placement. Represents the natural base in mathematical symbols; The interaction gain coefficient is obtained through training with historical data and is used to adjust the slope of the Sigmoid function and control the impact of the interaction experience on conversion potential. This is the interactive experience value, calculated based on metrics such as landing page loading speed, interactive operation steps, and ease of use of landing page buttons. The historical data feedback gain value represents the calculation result based on historical conversion data of similar landing pages in the same delivery scenario. The historical data feedback gain value is equal to the historical best CVR divided by the current scenario average CVR minus one, and is used to reflect the gain effect of historical high-quality experience on the current landing page. The conversion effect corresponding to the conversion potential of the landing page structure is used as a reward signal.

3. The method according to claim 2, characterized in that, The Sigmoid interaction gain corresponds to user data of the touch operation interaction object; the method further includes: In response to a touch operation on a target landing page, the target terminal model corresponding to the touch operation is obtained, and the touch operation sliding trajectory, touch operation user habits, operation speed, touch operation contact area, touch operation contact size, and touch operation force are determined based on the touch operation. The user data of the touch operation interaction object is analyzed based on the target terminal model, the touch operation swiping trajectory, the touch operation user habits, the operation speed, the touch operation contact area, the touch operation contact size, and the touch operation force; wherein, the user data includes the user's gender and the user's age; Based on the user data, determine the target related delivery data for the target users corresponding to the user data; Based on multiple target-related delivery data, the comprehensive delivery data associated with the target landing page structure corresponding to the target landing page is determined.

4. The method according to claim 3, characterized in that, The analysis of user data of the touch operation interaction object based on the target terminal model, the touch operation swiping trajectory, the touch operation user habits, the operation speed, the touch operation contact area, the touch operation contact size, and the touch operation force includes: In response to the touch operation on the target landing page, the touch screen conductivity sensitivity of the touch operation is determined based on the touch operation; The user's skin texture and stratum corneum of the touch operation interaction object are determined based on the conductivity sensitivity of the touch screen, and the user's age is analyzed based on the user's skin texture and stratum corneum, the target terminal model, the touch operation sliding trajectory, the touch operation user habits, the operation speed, the touch operation contact area, the touch operation contact size, and the touch operation force.

5. The method according to claim 1, characterized in that, The landing page structure has multiple grating structures on its display screen, and the different angles of the grating strips correspond to different display contents on the display screen; the method further includes: In response to detecting multiple viewing users corresponding to the page display terminal, the user group type of the viewing user is determined according to the user age and user gender of each of the multiple viewing users, and the angle of the raster bar set for the viewing user is determined according to the relative position of each viewing user relative to the page display terminal; The targeting data and corresponding display content for the viewing user are determined based on the user group type. The raster bar setting angle is controlled for each viewing user based on the display content and the raster bar setting angle for the viewing user. Different viewing users correspond to different raster bar setting angles and different display content.

6. The method according to claim 5, characterized in that, Also includes: In response to detecting a position movement event of a first viewing user among the plurality of viewing users, a first relative position of the first viewing user's movement position relative to the page display terminal is determined based on the position movement event; Based on the first relative orientation transformation, the angle of the first raster strip for the first viewing user is set so that the angle of the first raster strip corresponds to and follows the movement position of the first viewing user.

7. The method according to claim 5, characterized in that, Also includes: In response to the detection that the second relative position of the second viewing user among the plurality of viewing users relative to the page display terminal has not changed within a specified time period, it is determined that the second viewing user is standing and watching the second projected display content; The angle of the second grating strip corresponding to the second relative position is controlled to move to display the purchase intention data corresponding to the second display content, so as to guide the second viewing user to view the purchase intention data by moving the angle of the second grating strip; wherein, the purchase intention data includes purchase intention QR code content; Based on the second relative orientation, project the corresponding purchase intention QR code content below the location of the second viewing user to improve the conversion effect data after the second projected display content is displayed.

8. An AI-based landing page structure optimization system, characterized in that, include: A construction module is used to acquire historical landing page structure features, associated delivery data and corresponding conversion effect data, and to construct a training sample set based on the historical landing page structure features, the associated delivery data and the corresponding conversion effect data; wherein, the conversion effect data includes at least one of CVR, order rate and retention rate; The fusion module is used to fuse the convolutional neural network and the gradient boosting tree to obtain an initial landing page structure AI deep learning model; wherein, the convolutional neural network is used to extract deep visual layout features in the landing page structure, and the gradient boosting tree is used to learn the non-linear correlation between features; The training module is used to train the initial landing page structure AI deep learning model by taking the historical landing page structure features and the associated delivery data in the training sample set as inputs to the initial landing page structure AI deep learning model and taking the conversion effect data in the training sample set as outputs to the initial landing page structure AI deep learning model. During the model training process, the hyperparameters are optimized by cross-validation to obtain an intermediate landing page structure AI deep learning model. The update module is used to treat the landing page optimization process as a sequential decision problem and use the conversion effect as a reward signal. It receives the feedback of the campaign data in real time and updates the model parameters of the intermediate landing page structure AI deep learning model so that the model can adapt to the optimization needs of different campaign stages and obtain the final landing page structure AI deep learning model. The campaign stages include at least one of the cold start period, the scaling period, and the stabilization period.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

Images