Advertisement display method and device, equipment and medium

Abstract

This application relates to an advertising display method, apparatus, device, and medium. The method includes: responding to an advertising display event; constructing an ad placement pool containing multiple candidate ad placements based on the layout structure information of the target page corresponding to the event; obtaining an ad data package of an ad to be displayed, the ad data package including an ad performance value and ad rendering data; calculating a comprehensive score for each candidate ad placement based on the ad performance value, attribute information of each candidate ad placement in the ad placement pool, and contextual feature information of the target page; and displaying the ad to be displayed in the candidate ad placement with the highest comprehensive score based on the ad rendering data. This application calculates the comprehensive score of the target page and intelligently adapts the ad placement with the highest comprehensive score for ad placement, thereby improving the accuracy of ad placement.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an advertising display method, apparatus, device, and medium. Background Technology

[0002] With the booming development of internet digital advertising, balancing the accuracy of ad placement with user experience has become a key challenge for platform operations. Current advertising practices suffer from significant technical limitations in ad placement selection, impacting both the maximization of advertising effectiveness and the optimization of user experience.

[0003] In existing technologies, ad placements are typically static and predefined. Platforms pre-allocate ad slots in fixed locations on the page, such as the top banner, sidebar, and fixed slots in the feed. When an ad needs to be displayed, it is selected from the ad library based on one or a few business metrics, such as click-through rate and bid, and then mechanically placed into a pre-defined fixed ad slot. However, static ad placements cannot fully utilize the dynamic layout and blank areas of the page, resulting in low resource utilization and an inability to flexibly adapt to the specific structure and content of each page load. Summary of the Invention

[0004] The primary objective of this application is to solve at least one of the aforementioned problems by providing an advertising display method, apparatus, device, or medium.

[0005] To achieve the various objectives of this application, the following technical solution is adopted: An advertising display method provided for one of the purposes of this application includes the following steps: In response to an ad display event, an ad pool containing multiple candidate ad slots is constructed based on the layout structure information of the target page corresponding to the event. Obtain the advertising data package of the advertisement to be delivered, wherein the advertising data package includes the advertising effectiveness value and the advertising rendering data; Based on the advertising effectiveness score, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page, a comprehensive score for each candidate ad slot is calculated. The advertisement to be delivered will be displayed in the candidate advertisement position with the highest comprehensive score based on the advertisement rendering data.

[0006] An advertising display device proposed for an advertising display method suited to one of the purposes of this application includes: The ad placement pool construction module is configured to respond to ad display events and, based on the layout structure information of the target page corresponding to the event, construct an ad placement pool containing multiple candidate ad placements. The data packet acquisition module is configured to acquire the advertising data packet of the advertisement to be delivered, wherein the advertising data packet includes the advertising effectiveness value and the advertising rendering data; The comprehensive score calculation module is configured to calculate the comprehensive score of each candidate ad slot based on the ad performance value, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page. The ad rendering module is configured to display the ad to be delivered in the candidate ad slot with the highest comprehensive score based on the ad rendering data.

[0007] In another aspect, a computer device provided for one of the purposes of this application includes a central processing unit and a memory, the central processing unit being used to invoke and run a computer program stored in the memory to perform the steps of the advertising display method described in this application.

[0008] In another aspect, a computer-readable storage medium is provided to suit another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the described advertising display method, which, when invoked by a computer, performs the steps included in the corresponding method.

[0009] Compared with existing technologies, the advantages of this application are as follows: This application dynamically analyzes the layout structure information of the target page in response to ad display events, and intelligently constructs an ad pool containing multiple candidate ad positions based on this. This solution is no longer limited to fixed positions predefined on the page, but can discover potential display areas in the page layout in real time, including dynamic blank spaces such as paragraph gaps and content separation areas, thereby greatly improving the utilization rate of ad space resources and enabling ad display to flexibly adapt to the specific content and structure of each page load, realizing the transformation from static ad positions to dynamic intelligent ad positions.

[0010] Furthermore, by incorporating ad performance metrics, candidate ad placement attribute information, and contextual features of the target page, a comprehensive score is calculated for each candidate ad placement, balancing commercial benefits and user experience. This comprehensive score not only considers the ad's own economic benefits but also incorporates contextual factors such as the ad placement's specific attributes and the current page's contextual features. This multi-dimensional evaluation mechanism ensures that high-value ads are placed in optimal positions with good visual appeal, low interference, and high integration with page content. This enhances ad click-through conversion potential while effectively avoiding issues of ad incompatibility with the page and disruption of the user browsing experience, achieving synergistic optimization of commercial value and user experience. Attached Figure Description

[0011] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic flowchart of a typical embodiment of the advertising display method of this application; Figure 2 This is a schematic block diagram of the advertising display device of this application; Figure 3 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0012] This application discloses an advertising display method that can be programmed into a computer program product and deployed on a content distribution server, advertising server, or client browser / application to achieve its functionality. For example, in an exemplary application scenario of this application, the method can be deployed in the backend advertising system of various content platforms such as news, social media, and e-commerce, enabling the platform to achieve intelligent and contextualized advertising display functions. When a user browses platform content, the method can dynamically evaluate the current page environment and the advertisement to be displayed, automatically select the best advertisement display position, and perform adaptive rendering to improve advertising effectiveness and user experience.

[0013] In a typical embodiment of this application, in response to an ad display event, the document object model (DOM) or layout structure of the current target page is first parsed to dynamically identify and construct a candidate ad space pool containing multiple potential display areas. The ad display event can be triggered in the following ways: Specifically, in response to user browsing behavior on the content page, including swiping and hovering, based on various real-time interaction characteristics such as the corresponding browsing behavior, it is determined whether preset ad triggering conditions are met. If met, the ad display event is further triggered. The ad triggering conditions are preset in advance by those skilled in the art. Simultaneously, the ad data package of the ad to be displayed is obtained. This data package contains the ad effectiveness value calculated from historical data, as well as the ad rendering data corresponding to the original rendering materials such as images, text, and videos of the ad.

[0014] After determining the ad placement pool and obtaining the ad data package, based on the ad performance value of the ad to be placed, the attribute information of each candidate ad placement in the ad placement pool, and the contextual features of the current page, a comprehensive scoring model is used to calculate a comprehensive score for each candidate ad placement that balances commercial benefits and user experience. Finally, the candidate ad placement with the highest comprehensive score is selected as the display position for this ad.

[0015] In one embodiment, after determining the candidate ad placement with the highest overall score, the most suitable target style template is intelligently matched or dynamically generated from a preset style template library based on the specific compatibility style requirements (such as size, shape, and color specifications) of the best ad placement and the ad rendering data carried by the ad. Based on this template and the ad rendering data, an ad component matching the page style is generated and accurately added to the corresponding position on the target page for display.

[0016] The advertising display method described in this application can be widely applied to various digital media platforms, enabling refined and intelligent ad placement. This method helps platforms deeply mine the advertising value of page traffic without excessively interfering with users. Through dynamic placement and intelligent rendering, it simultaneously achieves the dual goals of improving ad click-through rates and optimizing the user browsing experience, overcoming the shortcomings of traditional fixed-position ads in terms of resource utilization and contextual adaptability.

[0017] Please see Figure 1 The advertising display method of this application, in its typical embodiment, includes the following steps: Step S3100: Respond to the ad display event and construct an ad pool containing multiple candidate ad slots based on the layout structure information of the target page corresponding to the event; Ad display events can be triggered in the following ways. In one embodiment, the system continuously monitors user interactions while browsing content pages (such as news articles, social media feeds, e-commerce product detail pages, etc.), collecting corresponding real-time interaction features. These features include, but are not limited to, user scrolling speed, click behavior, mouse hover position, page dwell time, and semantic tags extracted from the current page through content analysis (e.g., "technology news," "beauty tutorials," "outdoor gear," etc.). A built-in preset ad triggering condition judgment model is used to determine whether the preset ad triggering conditions are met. For example, when the model detects that the user's dwell time on the current page exceeds a preset threshold (e.g., 10 seconds), and the scrolling behavior indicates that the user has entered a deep reading state, and the page content tags are highly relevant to a certain ad category, the ad triggering condition is determined to be met. When the condition is met, the page currently being viewed by the user is identified as the target page, and an ad display event is triggered based on this target page.

[0018] After an ad display event is triggered, this step responds to the event by parsing the layout structure information of the target page. This layout structure information refers to the visual and logical structure represented by the page's Document Object Model (DOM) tree or rendering tree. In one embodiment, a page layout parsing engine is invoked. This engine first obtains the complete DOM tree of the target page, and then calculates the CSS box model properties of each user interface component to accurately determine its absolute position coordinates, size, z-index, and visual visibility within the viewport. In addition to these basic geometric properties, the engine also analyzes the semantic role of the component (such as whether it is a content container, navigation bar, sidebar, image gallery, comment section, etc.) and its content density.

[0019] Based on the detailed layout structure information obtained from the parsing, an ad placement pool is dynamically constructed to intelligently determine suitable candidate ad placements for embedding ads from the target page. Specific rules may include: identifying visual blank areas or content separation areas on the page, such as natural gaps between paragraphs in an article or between images and text; evaluating the size and position of existing components to determine if it is possible to insert containers that conform to common ad sizes without disrupting the main content layout; or identifying reserved slots designed to insert dynamic content, such as the positions between items in a news feed list. In some embodiments, the rule engine iterates through all parsed components and calculates a suitability score for ad placement based on their position, size, spacing from surrounding content, and semantic role. When the fit score exceeds a preset threshold, the corresponding area is initially marked as a candidate ad placement. For each identified candidate ad placement, corresponding attribute information is generated. This attribute information may include the relative position and size range of the candidate ad placement on the target page, a preset basic interference value (measuring the degree of interference that an ad placed here may cause to the user's browsing of the original content), and its compatibility style information, such as its size range, container style, visual blending requirements (such as rounded corners, background color, shadows), and interactive behavior restrictions. Finally, based on the attribute information of each candidate ad placement, a dynamically generated ad placement pool is formed for this ad display event and the target page.

[0020] Step S3200: Obtain the advertising data package of the advertisement to be delivered, wherein the advertising data package includes the advertising effectiveness value and the advertising rendering data; In one embodiment, when an ad display event is triggered and the candidate ad pool for the target page is constructed, ads that highly match the display environment of the target page are matched from a preset ad library. Specifically, an ad request is first sent to the backend server of the ad delivery platform. This request may carry context parameters to support precise matching, including the identifier of the target page, page content tags, user profile features, and summary information of the candidate ad pool, such as the size range and location type of each candidate ad. After receiving the request, the server, based on a preset real-time bidding mechanism or targeted matching algorithm, filters out ads from the ad library that match the current request context. In some embodiments, when multiple candidate ad slots in the ad pool meet the display requirements, multiple ads can be selected for display.

[0021] For the ad to be deployed, its detailed ad data package is further obtained, including ad performance metrics and ad rendering data. When extracting the ad performance metrics, the historical ad data storage layer is first accessed to retrieve the ad's performance data over historical campaign periods, including cross-channel display logs, clickstream data, conversion attribution records, and revenue settlement details. Based on this raw data, basic performance metrics for the ad are calculated, such as revenue per thousand impressions, click-through rate, and cost per conversion, and these are then packaged into the ad data package as ad performance metrics.

[0022] The ad rendering data includes main creative materials, such as image file addresses for different sizes and rules, video stream addresses for multi-bitrate adaptation, and structured data for dynamic creative templates; it also includes copy content components, such as main titles, subtitles, and descriptive text; it includes style specification definitions, such as the advertiser's brand color scheme, font usage guidelines, logo display requirements, and safe area margins, which ensure consistent brand recognition when the ad is adapted to different ad placements; and it includes interactive behavior configurations, such as the target landing page address for click-throughs, supported user interaction modes such as hover preview, swipe switching, form filling, and the injection location of monitoring tracking code, which define the interaction logic and data feedback mechanism between the ad and the user. The acquired ad rendering data and ad performance values ​​are packaged together into an ad data package for subsequent comprehensive score calculations and ad rendering.

[0023] Step S3300: Calculate the comprehensive score of each candidate ad slot based on the ad performance value, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page. Based on the ad performance score and exposure potential coefficient, a corresponding revenue score is calculated. The exposure potential coefficient is determined based on the size range and relative position information contained in the candidate ad placement's attribute information. Simultaneously, based on the basic interference value and scene interference correction factor contained in the candidate ad placement's attribute information, an experience score is calculated. The scene interference correction factor is determined based on the page's visual features, real-time interaction features, and scene sensitivity contained in the target page's contextual features. Based on preset revenue weight coefficients and experience weight coefficients, and combining the revenue score and experience score, a comprehensive score for each candidate ad placement is obtained.

[0024] Specifically, when calculating the revenue score, the first step is to calculate the salience score of the candidate ad placements in the page's visual hierarchy based on their relative position information. For example, ad placements located in the center of the first screen receive the highest position weight, which decreases non-linearly as the page scrolling depth increases or the degree of deviation from the center of the view increases. Simultaneously, the information carrying capacity and visual impact scores are calculated based on the size range of the ad placements. Larger display areas mean greater freedom in creative expression and stronger user capture capabilities. However, a size adaptation correction factor is introduced. When there is a significant deviation between the ad placement size and the standard proportion of the ad creative, the exposure potential is reduced to reflect potential issues of creative cropping or wasted white space. Furthermore, the timeliness weight of the exposure potential is dynamically adjusted by incorporating real-time user interaction characteristics from the contextual feature information. For example, when a user is detected scrolling rapidly, ad placements located at the front of the page and with longer dwell times receive higher timeliness bonuses, while bottom ad placements that require longer scrolling times to reach face a discount in exposure probability. The ad performance value is then fused with the multi-dimensionally corrected exposure potential coefficient, using either a weighted product or exponential smoothing method to generate the final revenue score.

[0025] When calculating the experience score, the baseline interference value is first retrieved from the candidate ad placement attribute information. This baseline interference value is a pre-set value for each candidate ad placement in its attribute information, representing the potential for interference with user attention and workflow based on general page layout and design principles. For example, an insert ad that is forcibly inserted between two paragraphs of text, interrupting continuous reading, will have a higher baseline interference value; while a rectangular ad located at the bottom of the sidebar, separated from the main content flow, will have a relatively lower baseline interference value. This value is pre-calculated or assigned based on factors such as placement type and proximity to core content when constructing the ad placement pool.

[0026] Subsequently, a scene interference correction factor is introduced. This factor is calculated comprehensively based on multi-dimensional data from contextual feature information, including: style consistency index from page visual features (when the default style of the ad placement significantly conflicts with the overall visual style of the page, the scene interference correction factor is increased to reflect the additional visual disjointedness); user focus index from real-time interaction features (when it is detected that the user is in a deep reading or highly focused operation state, the correction factor is increased to reflect the user's extremely low tolerance for interference at this time); and scene sensitivity level (high-sensitivity scenes, such as tool office pages or pages playing videos, will trigger an exponential amplification of the correction factor, while low-sensitivity scenes, such as social information feed pages, allow the correction factor to be moderately reduced). The product of the base interference value and the scene interference correction factor constitutes the comprehensive interference evaluation of the ad placement, i.e., the experience score.

[0027] Finally, the final comprehensive score is calculated by using preset revenue weighting coefficients and experience weighting coefficients, along with the aforementioned revenue score and experience score. The weighting coefficients can be dynamically adjusted based on platform operation strategies, advertiser preferences, or user group characteristics. The comprehensive score can be calculated using a linear weighted form, i.e., the comprehensive score equals the revenue score multiplied by the revenue weighting coefficient plus the experience score multiplied by the experience weighting coefficient.

[0028] Step S3400: Display the advertisement to be delivered in the candidate advertisement position with the highest comprehensive score based on the advertisement rendering data.

[0029] After calculating the comprehensive score of each candidate ad slot on the target page, the ad to be placed is displayed in the candidate ad slot with the highest comprehensive score. In one embodiment, based on the ad rendering data and the compatibility style information contained in the attribute information of the candidate ad slot, a suitable style template is matched from a preset style template library to visually integrate the ad with the target page. Specifically, firstly, the compatibility style information in the ad slot attribute information is parsed to extract its precise size value, shape ratio, visual style tendencies such as rounded corner radius, shadow effect, background color suggestions, and interactive behavior restrictions such as whether autoplay is supported and whether floating expansion is allowed. At the same time, the material characteristics in the ad rendering data are analyzed, including the original aspect ratio of the image, the content length and cover frame of the video, the length and language encoding of the text, and the complexity of dynamic elements. Based on the above compatibility style information and the input of the ad rendering data, the preset style template library is traversed. This library stores a variety of preset ad container templates, each template defining a specific layout structure, CSS style rules, and JavaScript interaction logic. For specific matching steps, please refer to the subsequent detailed implementation methods, which will not be elaborated here.

[0030] After selecting the target style template, the corresponding ad component is generated, and precise DOM insertion and ad display are performed. A placeholder insertion strategy can be used, where an empty container element with specific identifier attributes is pre-set at the target position of the candidate ad slot during the initial page rendering or ad pool construction phase. In this case, the generated ad component can simply be appended to this container as a child node. If no placeholder slot is pre-set, the precise coordinates of the target insertion point in the existing DOM structure are calculated, and standard DOM manipulation methods are used to insert the ad component into the specified position, immediately triggering a reflow and repaint to ensure synchronized visual updates.

[0031] As can be seen from the above embodiments, this application has achieved fundamental technical progress compared to traditional methods, including but not limited to: This application dynamically analyzes the layout structure information of the target page in response to ad display events, and intelligently constructs an ad pool containing multiple candidate ad positions based on this. This solution is no longer limited to fixed positions predefined on the page, but can discover potential display areas in the page layout in real time, including dynamic blank spaces such as paragraph gaps and content separation areas, thereby greatly improving the utilization rate of ad space resources and enabling ad display to flexibly adapt to the specific content and structure of each page load, realizing the transformation from static ad positions to dynamic intelligent ad positions.

[0032] Furthermore, by incorporating ad performance metrics, candidate ad placement attribute information, and contextual features of the target page, a comprehensive score is calculated for each candidate ad placement, balancing commercial benefits and user experience. This comprehensive score not only considers the ad's own economic benefits but also incorporates contextual factors such as the ad placement's specific attributes and the current page's contextual features. This multi-dimensional evaluation mechanism ensures that high-value ads are placed in optimal positions with good visual appeal, low interference, and high integration with page content. This enhances ad click-through conversion potential while effectively avoiding issues of ad incompatibility with the page and disruption of the user browsing experience, achieving synergistic optimization of commercial value and user experience.

[0033] Based on any embodiment of the method in this application, before constructing an ad placement pool containing multiple candidate ad placements based on the layout structure information of the target page corresponding to the ad display event, the method includes: Step S2100: Obtain the real-time interaction features corresponding to the current user page in real time. The real-time interaction features include user behavior features, current page content tags, and page dwell time. In one embodiment, a lightweight monitoring script or SDK is deployed on the user's end (such as a web browser or mobile application client). This monitoring method operates by listening to front-end events and periodically reporting data, capturing multi-dimensional user behavior characteristics without significantly impacting page performance and user experience. The collection of user behavior characteristics not only includes basic page-level events (such as onload and onunload), but also focuses on capturing micro-interactions of users within the page content area, including: scrolling behavior, by listening to corresponding events and calculating their speed, acceleration, scrolling direction, and dwell point in the vertical depth of the page, it can be determined whether the user is quickly scrolling to find information, slowly browsing and reading carefully, or repeatedly scrolling up and down in a specific area (such as the middle of the page or the top of the comment section); pointer (mouse / touch) behavior, by tracking corresponding events to obtain the cursor's movement trajectory, the duration of hovering over specific page elements (such as titles, images, and buttons), and the type and location of the target element of the click event, these can infer the user's visual focus and operational intent; keyboard interaction, by listening to corresponding events in the input box, it can be determined whether the user is in an active input state (such as filling out a form or posting a comment), which indicates that the user is performing a high-focus task and is more sensitive to ad insertion. In addition, it also includes events such as viewport changes and page visibility, used to determine whether the user has switched the page to the background.

[0034] Simultaneously, it is necessary to obtain the current page content tags. In one embodiment, this is injected by the server during page rendering, or obtained by the client through analysis after the page has loaded. The current page content tags are a structured description of the page's theme and key entities. For example, content publishing platforms manually or automatically assign a series of tags to each article or page in the background, such as artificial intelligence, financial markets, and spring fashion. These tags are output as structured data in the corresponding tags on the page, which the monitoring script can directly read. In another embodiment, after the page loads, the script can extract the page title and main body content (through identification...) <article>The system extracts the areas with the highest text density using tags or heuristic rules, and then uses lightweight natural language processing models or keyword extraction algorithms to calculate the core keywords, entities (such as names of people, places, and products), and sentiment of the page in real time.

[0035] Simply calculating the time interval from page load (onload) to before unload (before unload) is not precise because it includes inactive time such as switching browser tabs and minimizing windows. Therefore, in this specific implementation, page dwell time is calculated as active dwell time. Specifically, the timer only accumulates data when the page is visible and there has been detectable user interaction (scrolling, clicking, moving) within a recent preset time period (e.g., within the past 3 seconds). When the page is hidden or the user has no activity for an extended period, the timer pauses. The resulting active dwell time reflects the user's actual engagement with the page content.

[0036] Ultimately, the monitoring script encapsulates these collected real-time interaction features into data packets at a certain frequency and sends them to the backend server via asynchronous requests. The server receives and updates this data in real time.

[0037] Step S2200: Based on the real-time interaction features, determine whether the preset advertising triggering conditions are met. In one embodiment, an ad trigger condition library is maintained, containing several ad trigger conditions defined by those skilled in the art. These ad trigger conditions are not always simple thresholds of a single dimension; they can also be composite logical expressions combining multiple real-time interaction features with different weights. A rule engine-based condition-action paradigm can be adopted, where each ad trigger condition includes a condition judgment logic part. For example, an ad trigger condition might be defined as: (page content tag contains "travel guide") AND (active dwell time >= 15 seconds) AND (average scroll speed in the last 5 seconds < 50 pixels / second). The rule engine continuously matches the real-time input feature values ​​(i.e., real-time interaction features) with each ad trigger condition in the ad trigger condition library to determine whether the preset ad trigger condition is met.

[0038] Step S2300: When the conditions are met, the current user page is used as the target page to trigger the advertisement display event.

[0039] When the preset ad triggering conditions are met at the current moment, the context state of the browser page (or in-app page view) that the user is currently interacting with is locked. The page state at the moment of triggering is fixed as an immutable snapshot to prevent the judgment from becoming invalid due to changes in page content or user interaction state during subsequent processing. This locked page is designated as the target page for this ad delivery, and an ad display event is triggered based on this target page. This ad display event is a structured instruction object used to start the subsequent complete ad display method.

[0040] Through the above embodiments, by constructing real-time interactive features and rule-based advertising triggering conditions, the timing of ad display is transformed from traditional time-driven or fixed-location-driven to data-driven decision-making that deeply depends on the user's real-time intent, content context, and interaction state. Based on this, it is possible to accurately identify acceptable moments when users have a high level of engagement and interest in the current page content, effectively filtering out scenarios where attention is scattered, operations are frequent, or it is not suitable for ad insertion, thereby improving the relevance of ad exposure and user acceptance from the source.

[0041] Based on any embodiment of the method in this application, before obtaining the advertising data packet of the advertisement to be delivered, the advertising data packet includes the following: Step S1100: Obtain the historical data of the advertisement to be placed, including historical impressions, historical clicks, historical conversion data and historical revenue data; Before this step, a candidate list of advertisements to be delivered is determined. For each advertisement in the list, an asynchronous data query request is initiated based on its advertisement identifier. This request can be sent to a built-in advertising history data service or an online analytical processing (OLAP) database. Then, based on the advertisement identifier, its delivery history data is extracted from a massive advertising log stream. This delivery history data is stored in time-sharded log tables (e.g., by day, by hour) and indexed by the advertisement identifier. In one embodiment, aggregated data from a specific past time window (e.g., the past 7 days, 30 days) is obtained to ensure data timeliness and avoid misleading current value assessments due to excessively distant historical data.

[0042] The query returns historical data including: historical impressions, which is the total number of times the ad was successfully loaded and displayed to users within a specified time window, serving as a fundamental metric for measuring ad exposure. Invalid impressions, such as those placed outside the visible area or with extremely short exposure times (less than 1 second), are excluded during recording; historical clicks, which is the total number of valid clicks (such as clicking the ad creative or call-to-action button) made by users during their interaction with the ad, reflecting the ad's ability to attract users' initial interest; historical conversion data, the definition of which varies depending on the ad's objectives and may include specific actions completed by users on the landing page after clicking the ad, such as form submission, app download, product purchase, account registration, adding to cart, or making a contact call. Conversion data can be obtained by transmitting monitoring code deployed by the advertiser and attributing it against ad clicks; and historical revenue data, which quantifies the direct economic benefits generated by the ad. Its calculation method depends on the ad's billing model. For pay-per-click (PPC) ads, revenue data can be the sum of the advertiser's bids (or the platform's actual revenue) for all clicks generated by the ad within a specified time window. For pay-per-impression (PPC) or pay-per-conversion (PCC) ads, the revenue data is the sum of impression revenue or conversion revenue, respectively.

[0043] In addition to these four core fields, some implementations also include more granular dimensional information in the historical data for subsequent finer performance optimization. For example: historical click-through rate (historical clicks / historical impressions), which is a normalized metric used to measure the relative attractiveness of ad creatives; historical conversion rate (historical conversions / historical clicks), which measures the conversion efficiency from interest to action; historical average cost per click or historical revenue per thousand impressions, which directly reflects its economic efficiency; and segmented performance data broken down by time period (e.g., by hour, by weekday / weekend), by user profile tags, by media attributes, etc.

[0044] Step S1200: Based on the historical display counts and historical revenue data, calculate the basic performance value of the advertisement to be deployed; In one embodiment, the average revenue per thousand historical impressions can be directly calculated as the base performance value of the advertisement to be placed using a preset mathematical expression. The mathematical expression is: Base Performance Value = (Historical Revenue Data / Historical Impressions) * 1000. This calculation has a clear business meaning: if an advertisement has received 1 million impressions in the past and generated a total revenue of 5,000 yuan, its base performance value is (5000 / 1000000) * 1000 = 5. That is, based on historical performance, the advertisement is expected to generate an average revenue of 5 yuan per thousand impressions. In the above specific implementation, when the historical impression count is zero or extremely low (e.g., less than a minimum statistical sample threshold, such as 100 times), direct division will lead to unstable or infinitely large values. Therefore, a preset default base performance value is used. This default value can be the average historical revenue per thousand impressions for the product category to which the advertisement belongs (e.g., "automobiles," "beauty"), or it can be the value of the advertiser's initial bid after a certain discount.

[0045] In another embodiment, a corresponding performance calculation model can be trained, taking historical impression counts and historical revenue data as input, and outputting a corresponding basic performance value after calculation. The model can be trained based on massive historical advertising data, using historical impression counts and historical revenue data as features, and the actual revenue per unit impression (or similar utility metrics) generated subsequently as labels, through supervised learning of regression models (such as linear regression, gradient boosting trees, etc.). This model aims to learn a non-linear mapping from historical performance to stable value prediction, better handling issues such as data sparsity and cold start, and capturing complex feature relationships that are difficult to reflect using traditional ratio calculation methods, thereby outputting a more accurate basic performance value prediction.

[0046] Step S1300: Based on the preset performance value optimization model, obtain the performance value optimization coefficient according to the historical click count and the historical conversion data; The performance evaluation (SEE) tuning model can be set as a rule-based heuristic function or as a trained machine learning model. The model's input consists of historical click counts and historical conversion data obtained in the above steps, within the same time window on which the basic SEE calculation is based. Its core function is to calculate metrics that reflect the quality of ad interaction.

[0047] In one embodiment, the optimization coefficient can be calculated as follows: First, calculate the historical click-through rate (CTR) of the ad = historical clicks / historical impressions. Then, calculate its historical conversion rate (CVR) = historical conversions / historical clicks. Since the benchmark CTR and conversion rate are naturally different for different ad categories (for example, the CTR of game download ads is usually higher than that of brand image ads), directly using the original values ​​may not be accurate. Therefore, benchmark values ​​are introduced for normalization. For example, the average CTR and average conversion rate of all ads under each ad category are determined. At this time, the relative CTR performance of the ad can be calculated as the historical CTR of the ad divided by the average CTR, and the relative conversion rate performance can be calculated as the historical conversion rate of the ad divided by the average conversion rate. Finally, the optimization coefficient K can be defined as K = w1 * relative CTR + w2 * relative conversion rate, where w1 and w2 are preset weights, for example, both can be set to 0.5, indicating that click and conversion quality are equally important; or they can be adjusted according to the platform's optimization goals, such as increasing w2 if more emphasis is placed on conversion. If K > 1, it means that the interaction quality of the advertisement is higher than the average level of the category, and its basic effectiveness value should be adjusted upward; if K < 1, it should be adjusted downward.

[0048] In another embodiment, a machine learning model (such as logistic regression or gradient boosting tree) is used as the tuning model. During training, the model uses features including the historical click-through rate (CTR) and historical conversion rate of the ads. After training, during online inference, the model is input with the historical CTR and historical conversion rate of the ad to be delivered, and it outputs a tuning coefficient.

[0049] Ultimately, whether through rules or models, a specific performance value optimization coefficient (usually a real number greater than 0) will be calculated for the current advertisement to be delivered, which will be used to adjust the basic performance value in the future.

[0050] Step S1400: Based on the performance value optimization coefficient, adjust the basic performance value to generate the advertising performance value.

[0051] Based on the performance value optimization coefficient calculated in the above steps, the corresponding basic performance value is adjusted. In one embodiment, the basic performance value can be directly adjusted by multiplying the performance value optimization coefficient and the basic performance value to obtain the corresponding advertising performance value.

[0052] Through the above embodiments, a basic performance value reflecting the average revenue efficiency of advertising is obtained through basic calculations or model predictions. Then, an optimization coefficient based on historical interaction quality (clicks, conversions) is introduced to correct the basic performance value, and finally, an advertising performance value is generated. This effectively solves the problem that a single revenue indicator cannot comprehensively measure advertising quality, making the evaluation of advertising performance value more scientific, robust and interpretable.

[0053] Based on any embodiment of the method in this application, in response to an ad display event, an ad slot pool containing multiple candidate ad slots is constructed based on the layout structure information of the target page corresponding to the event, including: Step S3110: Analyze the layout structure information of the target page to determine each user interface component and its position and size information; In one embodiment, upon responding to an ad display event, a page layout parsing engine is activated. This engine first obtains and traverses the Document Object Model (DOM) tree of the target page. The DOM tree is the browser's internal representation of the HTML document structure, defining the logical relationships between all elements on the page in a hierarchical node structure. By calling standard APIs provided by the browser or equivalent mobile framework interfaces, the parsing engine can obtain the user interface component nodes present on the page. These nodes correspond to HTML elements, such as... 、 <section> 、 <article> 、 、 、 <button>wait.

[0054] However, simply obtaining a list of DOM nodes is insufficient to determine the layout structure of the target page. The parsing engine needs to calculate the CSS box model properties for each relevant UI component node. This can be achieved by calling the corresponding browser API, aiming to determine the geometric properties of each component in the viewport coordinate system, including: absolute position coordinates, i.e., the pixel values ​​of the component's border box relative to the top-left corner of the current browser viewport: top (top edge), left (left edge), bottom (bottom edge), and right (right edge); size information, i.e., the component's width and height, including content, padding, and border, but excluding margins; hierarchy information, determined by parsing z-index and the DOM stacking context to determine the visual stacking order of elements, used to identify components that may obscure or cover other areas; and visual visibility, filtered out elements that are not actually visible on the screen by checking style properties such as display:none, visibility:hidden, and opacity:0, or whether the component is within the visible area, ensuring that only the user-visible portion of the interface is analyzed.

[0055] Step S3120: Based on the preset ad placement determination rules, determine the candidate ad placements for the target page according to the location and size information; The ad placement determination rules are pre-defined by those skilled in the art, and their design goal is to balance the commercial visibility of the ad with minimal interference to the user's browsing experience. These rules may include size and shape adaptation rules. The rules check whether the width and height of a component (or a potential area composed of multiple adjacent components) conform to a pre-defined set of common ad size specifications, such as full-width banners, rectangular ads, and sidebar square ads. The rules set tolerance ranges for size matching; for example, an area with a width between 728 and 732 pixels and a height between 90 and 95 pixels is considered suitable for a 728x90 banner ad placement. In addition to checking the dimensions of existing components, these rules can also identify visual gaps or content separation areas on the page. For example, by analyzing the vertical spacing between adjacent paragraph containers, if the dimensions (height and width) of this spacing meet the requirements, it can be identified as a potential candidate ad placement for inserting a feed ad.

[0056] The ad placement determination rules may also include location and visibility evaluation rules. These rules are calculated and judged based on the absolute position coordinates of the component. The core of these rules is to evaluate the salience and naturalness of the location within the user's visual flow. For example, the rules may prioritize suitable locations within the first screen area (i.e., the area visible immediately after page load without scrolling), which has the highest initial exposure probability. The rules may also avoid locations too close to the top edge of the page (potentially conflicting with the navigation bar) or completely flush with the bottom edge. Simultaneously, the rules will evaluate the contextual integration of the location.

[0057] After applying the above rules for traversal evaluation, a fit score is calculated for each evaluated location or region. This score integrates the evaluation results from multiple dimensions. Only when the fit score of a region exceeds a preset threshold is it marked as a candidate ad placement.

[0058] Step S3130: Based on the attribute information of the determined candidate ad slots, construct the ad slot pool.

[0059] Obtain the candidate ad placements identified in the above steps, and generate a complete, structured attribute information object for each candidate ad placement. The attribute information object includes the following fields: a unique identifier, which assigns a unique ID to the candidate ad placement within the current ad display event; size range and relative position information, which precisely records its geometric attributes such as coordinate range, width, and height within the target page viewport, as well as its relative position in the page's visual flow (e.g., whether it's on the first screen, its offset relative to the viewport center); a base interference value, which is a predefined or calculated scalar value used to quantify the baseline interference that placing the ad there may cause to the user's browsing of the original page content. For example, an insert card located in the center of the main text might be assigned a higher base interference value, while a position at the bottom of the sidebar might be assigned a lower value; and a style compatibility description, which describes potential constraints or suggestions regarding the UI style of the position, such as needing to adapt to rounded corner card styles, background color blending with the surrounding area, and support for automatic video playback.

[0060] Based on the attribute information of each candidate ad slot identified above, a corresponding ad slot pool is constructed.

[0061] Through the above embodiments, the underlying layout structure of the page is analyzed, and potential ad positions that meet the requirements of size, position and experience are dynamically identified through preset intelligent rules. Structured attribute information including geometry, interference degree, style compatibility and other dimensions is generated for each candidate position, and finally a standardized ad position pool is constructed.

[0062] Based on any embodiment of the method in this application, a comprehensive score for each candidate ad slot is calculated according to the advertising effectiveness value, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page, including: Step S3310: Determine the corresponding revenue score based on the advertising effectiveness value and exposure potential coefficient. The exposure potential coefficient is determined based on the size range and relative position information contained in the attribute information of the candidate advertising position. The exposure potential coefficient is a real number greater than 0, calculated based on the attribute information of the candidate ad placement. In one embodiment, it is calculated based on the size range and relative position information in the attribute information. The size (width and height) of the ad placement directly affects its visual impact, information capacity, and probability of being noticed by users. Generally, within a reasonable page layout, the larger the size, the higher the exposure potential. Specifically, a size value mapping table can be preset to pre-set corresponding basic size coefficients for common ad size specifications. For non-standard sizes, an interpolation function is used to calculate the relative value based on its actual area and aspect ratio. For example, the larger the area, the higher the size coefficient; certain aspect ratios (such as horizontal rectangles) are considered better than slender strips, thus receiving a slight bonus. Relative position information mainly refers to the ad placement's position relative to the current browser viewport and the first screen of the page. During calculation, it's determined whether it's on the first screen. Ad placements entirely within the area visible immediately after page load (the first screen) receive a higher position weight. Its position in the page's visual flow is also considered; even if scrolling is required, ad placements near the front of the page (e.g., after the first screen, the second screen) have significantly higher exposure potential than those at the bottom. A weight curve that decreases with page depth can be calculated based on the top border's ordinate value. Relative centering within the viewport also plays a role, as users' attention is more easily focused on the center of the screen. Therefore, the distance between the ad placement's center point and the current viewport center can be calculated; the closer the distance, the higher the visual focus bonus.

[0063] The final exposure potential coefficient is synthesized based on the size coefficient and the position coefficient using a weighted or multiplicative formula. After obtaining the exposure potential coefficient of the current candidate ad placement, the revenue score is calculated. The revenue score aims to predict the expected economic return of this impression. The most straightforward calculation formula is: Revenue Score = Ad Performance Value * Exposure Potential Coefficient. Finally, the revenue score for each candidate ad placement is calculated and used in the subsequent calculation of the overall score.

[0064] Step S3320: Calculate the experience score of the candidate ad slot based on the basic interference value contained in the attribute information of the candidate ad slot and the scene interference correction factor. The scene interference correction factor is determined based on the page visual features, real-time interaction features and scene sensitivity contained in the context feature information. The scene interference correction factor is a dynamic multiplier used to amplify or reduce the base interference value based on the current browsing environment. Its calculation is based on the contextual features of the target page, primarily its visual characteristics. Specifically, it analyzes the overall color distribution, font style, and density and style of graphic elements. For example, if the current page uses a dark mode and a minimalist text layout for serious reading (such as a long blog post or technical document), inserting a brightly colored banner ad with flashing animation would create a strong visual clash and a jarring experience. The scene interference correction factor recognizes this difference in visual style and amplifies the interference effect (i.e., correction factor > 1). Conversely, if the page is a colorful, element-rich product listing page or social media feed, where newly inserted ads blend in visually, the correction factor may be close to or even less than 1, indicating that the interference is reduced or considered acceptable.

[0065] The scene interference correction factor is also determined based on real-time interaction characteristics. Specifically, it analyzes the user's current behavioral state. If the user is scrolling rapidly, their focus on the page content is low, and the perceived interference from inserted ads is weak. Conversely, if the user moves the mouse slowly, stays on a paragraph for a long time, or is reading linearly in reader mode, it indicates that the user is in a state of high focus. In this state, inserted non-content elements will appear abrupt, and the correction factor will significantly amplify the base interference value. Similarly, if the user's cursor is hovering over a button or in an input state, it indicates that they are performing a specific operation task. Inserting ads near the operation path at this time is very likely to cause accidental clicks, and the correction factor will also increase accordingly.

[0066] The scene interference correction factor is also determined based on scene sensitivity. For example, when the current page is determined to be a full-screen video playback page, an online payment confirmation page, or a serious news details page, these scenes have extremely low tolerance for the insertion of irrelevant information, that is, the scene sensitivity is very high. The scene interference correction factor will be significantly increased to strictly limit the insertion of advertisements.

[0067] The specific scene interference correction factor can be determined by training a correction factor calculation model. This model takes page visual features, real-time interaction features, and scene sensitivity as input features, and directly outputs the corresponding scene interference correction factor. Training this model first requires constructing a training sample set, which includes training samples and their supervision labels. The training samples are structured feature vectors of page visual features, real-time interaction features, and scene sensitivity corresponding to historical ad display events. The supervision labels are quantitative indicators based on the actual degree of interference caused to the user experience by the ad display, i.e., the scene interference correction factor. The correction factor calculation model is trained based on this training sample set until it converges, learning the ability to generate corresponding scene interference correction factors based on the input page visual features, real-time interaction features, and scene sensitivity.

[0068] After obtaining the base interference value and the scene interference correction factor, the experience score of the corresponding candidate ad slot is calculated. The experience score can be obtained by multiplying the base interference value by the scene interference correction factor.

[0069] Step S3330: Based on the revenue score and the experience score, and using preset revenue weighting coefficients and experience weighting coefficients, calculate the comprehensive score of the corresponding candidate ad slot.

[0070] Those skilled in the art have pre-set revenue weighting coefficients and experience weighting coefficients. After the above steps calculate the revenue score and experience score, the revenue score is multiplied by the revenue weighting coefficient to obtain a first intermediate value, and the experience score is multiplied by the experience weighting coefficient to obtain a second intermediate value. The first intermediate value and the second intermediate value are added together to obtain the comprehensive score of the corresponding candidate ad slot.

[0071] Through the above embodiments, a revenue score and an experience score are independently calculated for each candidate ad placement. The revenue score is obtained by multiplying the ad effectiveness value and the exposure potential coefficient, quantifying the commercial value; the experience score is obtained by multiplying the basic interference value and the scene interference correction factor, quantifying the experience cost. Subsequently, by introducing preset revenue weighting coefficients and experience weighting coefficients, the two scores, representing commercial goals and user experience goals respectively, are combined into a unified comprehensive score. Finally, all candidate ad placements are ranked according to the comprehensive score, and the candidate ad placement with the highest score is determined as the optimal display position.

[0072] Based on any embodiment of the method in this application, the experience score of the candidate ad slot is calculated according to the basic interference value contained in the attribute information of the candidate ad slot and the scene interference correction factor. The scene interference correction factor, before being determined based on the page visual features, real-time interaction features, and scene sensitivity contained in the context feature information, includes: Step S4100: Based on the page color distribution features, font style features, and graphic element features of the target page, construct the corresponding page visual features; Analyzing the target page can involve taking a screenshot of the entire visible area or obtaining the render tree and style information via the browser API. Specifically, it involves extracting page color distribution features, font style features, and graphic element features. Among these: Extraction of page color distribution features. One approach is to obtain a screenshot of the page and then use color quantization algorithms (such as K-means clustering) to extract the most frequently occurring theme colors on the page, recording the RGB or LAB values ​​of each color and its proportion in pixels. Simultaneously, the overall color contrast, saturation distribution, and the warm / cool tendency of the main color tone are calculated. For example, a cool tone with a bluish tint or a warm tone with a yellowish / red tint. Furthermore, the presence of large areas of solid-color backgrounds on the target page, and the specific values ​​of these background colors, are identified. These color features can be formalized as a set of vectors, such as [Main Color 1_R, Main Color 1_G, Main Color 1_B, Proportion 1, Main Color 2_R, ..., Average Saturation, Average Brightness, Warm / Coolness Index].

[0073] Font style features can be extracted by traversing text nodes in the DOM tree. Specifically, this involves statistically analyzing the most frequent font combinations on the page, the most commonly used body font sizes, the hierarchical relationship between headings and body font sizes, and the main font weights. Furthermore, whether the target page prefers serif or sans-serif fonts is also recorded as a feature.

[0074] Extracting graphic element features specifically requires identifying and analyzing non-textual visual elements on the page, including: the style of image elements, such as realistic photographs, simple vector icons, or hand-drawn illustrations; the style of interface components, such as whether buttons have right angles or rounded corners, the shadow intensity of cards, and the style of dividing lines (solid lines, dashed lines, gradients); and the density and white space of the layout, calculated by comparing the area ratio of visual elements (text blocks, images, components) within the main content area with the surrounding white space. In addition, the analyzer will also detect whether the page contains background patterns, gradients, or dynamic visual elements.

[0075] After extracting the raw data from the three dimensions of color, font, and graphic elements, these data are used to construct a unified and structured visual feature representation of the page. This feature representation can be a high-dimensional feature vector that concatenates all the above-mentioned quantitative indicators; or it can be a more structured JSON object that stores feature subsets of each dimension in categories.

[0076] Step S4200: Generate scene sensitivity based on the page type of the target page and the user's current operation behavior; In one embodiment, the page type of the target page is first identified and classified. This page type identification process is based on multi-dimensional feature fusion analysis. Specifically, by parsing the URL structure features of the target page, the distribution of semantic tags in the DOM tree, and the page metadata information, the major category to which the page belongs is initially determined. For example, when the page URL contains specific path identifiers such as " / news / ", " / article / ", or " / blog / ", and the page contains a large number of paragraph text nodes, heading hierarchy structures, and timestamp elements, it is classified as a content reading page; when the page contains a large number of product list containers, price tags, shopping cart buttons, and order checkout forms, it is identified as an e-commerce transaction page; when the page contains video player components, live stream identifiers, or short message stream lists, it is determined to be a video entertainment page. Furthermore, cross-validation is performed using semantic analysis results of page content tags. For example, core keywords are extracted from the page using a natural language processing model. If the keywords are concentrated in terms like "tutorial," "guide," and "analysis report," the confidence level for classifying it as a content-reading page is strengthened. If the keywords are concentrated in terms like "discount," "flash sale," and "coupon," the confidence level for classifying it as an e-commerce transaction page is strengthened. This page type recognition is not a simple binary or multi-class label output, but rather generates a structured description containing a main type label, confidence score, and feature vectors of relevant subcategories, providing a basic contextual framework for subsequent calculations of scene sensitivity.

[0077] Based on the determined page type, further data collection and analysis of user behavior are conducted to capture the user's real-time interaction state and intent intensity on the page. This behavior collection can be achieved through a behavior tracking module deployed on the client side, which non-intrusively listens to and records the user's micro-interaction events on the page. Specifically, real-time monitoring of user scrolling behavior patterns is performed, including scrolling speed, frequency of scrolling direction changes, and distribution characteristics of scrolling pause points. When the user's scrolling speed is detected to be below a preset threshold and the time spent in a certain paragraph area exceeds the average reading time, the user is determined to be in a deep reading state. When the user is detected to frequently perform up and down scrolling back operations or repeatedly switch between browsing specific content blocks, the user is determined to be in an information comparison or decision-making hesitation state. When the user's cursor trajectory shows a regular horizontal scanning pattern or quickly scrolls vertically across a large amount of content, the user is determined to be in a fast browsing or information filtering state. Simultaneously, the system monitors user mouse hovering behavior, recording the duration of cursor hovering over various page elements, the speed of the cursor's movement before hovering, and subsequent actions after hovering. If the cursor hovers over functional buttons such as "Buy Now," "Submit Order," or "Play Video" with a significantly slower movement speed, it indicates that the user is in a highly focused state with a clear operational intention. If the cursor moves erratically mainly within text paragraphs accompanied by slow scrolling, it indicates that the user is in a content consumption state. Furthermore, the system monitors user keyboard interaction events. When continuous character input, shortcut key operations, or tab switching between form fields is detected, it determines that the user is in an active input or task execution state, at which point their focus on the page content is extremely high, and their tolerance for external interference is correspondingly reduced. For touch device users, by analyzing changes in the contact area, sliding acceleration, and multi-touch patterns of touch gestures, the system identifies the user's grip posture and operational intentions. For example, slow single-finger swiping usually corresponds to detailed reading, while rapid multi-finger pinching corresponds to page zooming or back operations.

[0078] Based on the comprehensive analysis of the page type identification results and the user's current operational behavior, a scene sensitivity calculation model is constructed. This model aims to quantify the tolerance threshold and suitability of the current scene for ad insertion. The generation logic of scene sensitivity follows the calculation paradigm of scene type baseline sensitivity and dynamic correction of operational behavior. First, a basic sensitivity baseline value is preset for different page types. This baseline value reflects the inherent tolerance of that type of page to ad interference under normal conditions. For example, for content-reading pages, since their core user experience relies on an immersive linear reading flow, any interrupting element can cause significant cognitive load and disrupt the experience, thus a high base sensitivity value is assigned. For e-commerce transaction pages, users typically have clear shopping goals and product browsing expectations, and are relatively more accepting of product recommendation ads, but less tolerant of irrelevant ads, so a moderately high base sensitivity value is assigned, and a product relevance adjustment factor is introduced. For video entertainment pages, users are extremely sensitive to visual interference while watching video content, but their tolerance for ads increases during video playback breaks or list browsing, so the base sensitivity value is dynamically adjusted according to the video playback status, with extremely high sensitivity during playback and moderate sensitivity during pauses or list pages.

[0079] Subsequently, the base sensitivity value is dynamically adjusted based on the user's current operational behavior to generate the final scene sensitivity value. The adjustment logic is based on the user's current state of interference tolerance assessment: when the user is detected to be in a deep reading state, the adjustment coefficient is further increased on the base sensitivity value of the content reading page, because at this time the user's cognitive resources are highly concentrated on text comprehension, and the insertion of advertisements can easily cause reading interruption and distraction; when the user is detected to be in a fast browsing state, the sensitivity value is appropriately decreased, because at this time the user's attention is in a divergent scanning mode, and the sensitivity to page elements is relatively low, and moderate advertising display is unlikely to cause strong negative perception; when the user is detected to be in a highly focused state with a clear operational intention, such as hovering over the purchase button, the sensitivity value is drastically increased, because interrupting the user's operation process at this time may lead to direct conversion loss and user frustration; when the user is detected to be in an active input state, the sensitivity value is adjusted to the highest level, forcibly prohibiting any form of advertisement insertion or pop-up interference to avoid input errors and task failure. In addition, correction factors based on time and historical behavior are introduced. For example, when it is detected that the user's cumulative time spent on the page has exceeded the normal browsing time and no obvious intention to leave has been shown, the sensitivity value is appropriately lowered to allow timely ad display. When it is detected that the user has frequently encountered similar ad displays in a short period of time and shows a quick closing or ignoring behavior, the sensitivity value is increased to avoid user fatigue and aversion caused by overexposure.

[0080] Ultimately, the scenario sensitivity output is a normalized numerical value or a hierarchical label. This value not only reflects the overall tolerance of the current page for ad insertion but also implies constraints and suggestions regarding ad display format, timing, and content. For example, high-sensitivity scenarios are advised to use non-intrusive edge display or delayed loading strategies, medium-sensitivity scenarios allow for moderate information flow insertion, and low-sensitivity scenarios can consider more prominent display positions.

[0081] Step S4300: Construct the context feature information of the target page by combining the page visual features, the real-time interaction features corresponding to the current user page, and the scene sensitivity.

[0082] The obtained page visual features, real-time interaction features, and scene sensitivity are combined to form the contextual feature information of the target page. This constructed contextual feature information is then serialized, encoded, and compressed for storage, generating a standardized contextual feature data packet. This data packet employs an efficient binary encoding format or a compact JSON structure to ensure space efficiency and parsing speed in network transmission and memory storage. This contextual feature information provides crucial data for subsequent calculation of the experience score for candidate ad placements.

[0083] Through the above embodiments, the final contextual feature information is constructed by quantitatively analyzing the page's visual style, accurately tracking real-time user behavior, and comprehensively determining the scene sensitivity based on page type and operational intent. The construction of this contextual feature information quantifies the target page into standardized feature data, providing a data foundation for subsequent evaluation of the user experience interference caused by advertisements in different locations and at different times.

[0084] Based on any embodiment of the method in this application, displaying the advertisement to be delivered in the candidate advertisement position with the highest comprehensive score based on the advertisement rendering data includes: Step S3410: Based on the attribute information of the candidate ad slot with the highest comprehensive score, determine the compatible style information of the candidate ad slot; The attribute information object of the candidate ad slot with the highest comprehensive score is obtained and parsed. This object is constructed in the embodiment of building the ad slot pool based on the determined attribute information of the candidate ad slots in the above steps. In this step, compatibility style information is explicitly generated. For specific compatibility style information, please refer to the specific implementation method above, which will not be repeated here. This step determines the compatibility style information of the candidate ad slot with the highest comprehensive score. The compatibility style information may include its size range, container style, visual integration requirements (such as rounded corners, background color, shadow), and interactive behavior restrictions, to ensure that the generated ad components can seamlessly integrate into the page environment of the target location.

[0085] Specifically, compatibility style information can specify that the outer frame size of the final generated ad component must fall within a preset size range. Furthermore, if the ad placement attributes indicate that it is located within a container with specific inner or outer margins, these spacing values ​​will also be incorporated into the style compatibility requirements to ensure that the ad maintains an appropriate distance from surrounding content. Secondly, compatibility style information may also include visual style integration requirements, referencing information in the ad placement attributes regarding the type of the page area it belongs to and surrounding style references. For example, if a candidate ad placement is identified as embedded in a card-style information flow, the compatibility style information will suggest or require the generated ad component to also adopt a similar card style, including but not limited to: having the same corner radius, using the same or coordinated shadow effects, and matching the background color with the card background color. If the candidate ad placement is located in a sidebar, and the sidebar as a whole has a light gray background and thin borders, the compatibility style information may suggest that the ad component also use a transparent or light-colored background.

[0086] Step S3420: Based on the compatibility style information and combined with the advertising rendering data, match the target style template from the preset style template library; The target style templates are preset by those skilled in the art. Each target style template has a corresponding template description file, which defines key attributes such as the basic size range supported by the template, the preset aspect ratio, visual style tags, supported creative types, interactivity, and configurable style variables.

[0087] Specifically, the first step is to conduct an initial screening based on compatibility style information. This involves filtering the template library according to the hard constraints in the compatibility style information, including: size filtering, which retains only templates that support (or can be adapted through simple stretching / scaling) the specified width and height of the target ad slot; type filtering, which excludes templates containing autoplaying videos or complex animations if the compatibility style information explicitly requires only static images or prohibits autoplaying videos; and style pre-screening, which prioritizes templates that explicitly contain the corresponding tags in the visual style labels if the compatibility style information suggests a rounded card style or the same background color as the sidebar.

[0088] After initial screening, a short list of short template candidates that meet the criteria is obtained. Then, a fine-matching stage based on ad rendering data is performed, where the ad rendering data is used to calculate a corresponding fit score for each template in the short list. The calculation analyzes the aspect ratio of the creative materials, the length of the title and body text provided in the ad rendering data. For image and text ads, it checks whether the template has suitable areas to separately accommodate the image and text, and whether the text length causes layout overflow or excessive white space. Furthermore, if specific interaction monitoring points or behaviors are defined in the ad rendering data (such as click-throughs or form collection), the engine checks whether the template provides corresponding interactive nodes (such as clickable areas or input box containers) to support these functions. In the specific calculation, a comprehensive scoring function is used, combining the template's fit in satisfying compatibility style information (with higher weight) and its fit with the ad rendering data to calculate a total score for each candidate template. The target style template with the highest score is selected as the target style template for this campaign. This target style template is usually not a specific ad that can be displayed directly, but a blueprint that defines the final ad component structure, style rules and interaction logic. It includes the HTML structure skeleton, CSS style definitions and JavaScript behavior placeholders.

[0089] Step S3430: Based on the target style template and the advertising rendering data, generate the corresponding advertising components and add them to the corresponding position of the candidate advertising position with the highest comprehensive score.

[0090] The selected target style template is parsed to create the corresponding Document Object Model (DOM) node tree. Then, the ad rendering data is traversed, and the specific data is populated into the corresponding slots in the template. For example, the image URL from the ad rendering data is assigned to the template. The `src` attribute of the tag; inserts the text content of the ad title and description into the corresponding template. <h3>and Within the element; assign the clickable link to the href or onclick event of the entire ad container or a specific button; if the ad contains video, configure... <video>The element's source address, poster image, and control parameters. Simultaneously, all monitoring code carried in the ad rendering data (used to track impressions, clicks, conversions, etc.) will be parsed and embedded into appropriate locations within the component, such as as the image's onload event or as a standalone event. <script>标签。

[0091] 其次，适配样式与尺寸。具体而言，根据候选广告位属性信息中精确的坐标和尺寸，对嵌入相应广告渲染数据的模板实例进行最终的样式调整，包括设置组件的width和height样式属性，使其严格匹配广告位的大小；还包括根据兼容样式信息微调内部元素的样式，如字体大小、边距、圆角半径等，以确保与周边页面元素的视觉协调，以确保了生成的广告组件在几何尺寸上能够严丝合缝地嵌入预留的位置。

[0092] 生成完整的广告组件DOM节点后，将该节点添加到页面的DOM树中，并定位到综合评分最高的候选广告位的相应位置。一种实施例中，预先在目标位置预先插入了具有特定标识（如id或data-*属性）的空占位元素，此时通过document.getElementById（）或类似方法确定该占位元素，然后用新生成的广告组件节点替换。另一种实施例中，精确计算目标位置在现有DOM结构中的插入点，然后使用DOM API直接将广告组件插入到该位置。

[0093] 通过上述实施例，首先确定最优广告位的兼容样式信息，确保广告在视觉与交互层面能无缝融入页面；进而从预设模板库中筛选出既能满足样式约束又能承载广告素材的目标样式模板；最终，通过动态合成与精准插入，将广告渲染数据注入模板，生成样式协调、尺寸匹配的广告组件，并准确放置在页面的最优位置。该实施例确保了广告的最终展示形式不仅是商业上最优的选择，也是视觉上最和谐、对用户体验干扰最小的呈现，从而将前序复杂的评分与决策结果，高效、高质地转化为用户端可见的、友好的广告实体，是提升广告投放整体效果与用户感知质量的关键最终环节。

[0094] 请参阅图2，适应本申请的目的之一而提供的一种广告展示装置，是对本申请的广告展示方法的功能化体现，该装置包括广告位池构建模块3100、数据包获取模块3200、综合评分计算模块3300，以及广告渲染模块3400，其中，所述广告位池构建模块3100，设置为响应广告展示事件，基于该事件对应目标页面的布局结构信息，构建包含多个候选广告位的广告位池；所述数据包获取模块3200，设置为获取待投放广告的广告数据包，所述广告数据包包括广告成效值以及广告渲染数据；所述综合评分计算模块3300，设置为根据所述广告成效值、所述广告位池中各候选广告位的属性信息以及所述目标页面的上下文特征信息，计算出每个候选广告位的综合评分；所述广告渲染模块3400，设置为将所述待投放广告基于所述广告渲染数据展示于所述综合评分最高的候选广告位中。

[0095] 在本申请的装置任意实施例的基础上，所述广告位池构建模块3100之前，包括：特征获取模块，设置为实时获取当前用户页面对应的实时交互特征，所述实时交互特征包括用户行为特征、当前页面内容标签以及页面停留时长；条件判断模块，设置为基于所述实时交互特征，判断当前是否满足预设的广告触发条件；事件触发模块，设置为当满足时，将所述当前用户页面作为目标页面触发所述广告展示事件。

[0096] 在本申请的装置任意实施例的基础上，所述数据包获取模块3200之前，包括：历史数据获取模块，设置为获取待投放广告的投放历史数据，所述投放历史数据包括历史展示次数、历史点击次数、历史转化数据及历史收益数据；基础值计算模块，设置为基于所述历史展示次数与历史收益数据，计算出所述待投放广告的基础成效值；调优系数计算模块，设置为基于预设的成效值调优模型，根据所述历史点击次数以及所述历史转化数据，得到成效值调优系数；成效值生成模块，设置为基于所述成效值调优系数，对所述基础成效值进行调整，生成所述广告成效值。

[0097] 在本申请的装置任意实施例的基础上，所述广告位池构建模块3100，包括：信息解析模块，设置为解析所述目标页面的布局结构信息，确定其中的各个用户界面组件及其位置与尺寸信息；候选广告位确定模块，设置为基于预设的广告位确定规则，根据所述位置及尺寸信息，确定出所述目标页面的候选广告位；池确定模块，设置为基于确定出的候选广告位的属性信息，构建所述广告位池。

[0098] 在本申请的装置任意实施例的基础上，所述综合评分计算模块3300，包括：收益评分计算模块，设置为根据所述广告成效值以及曝光潜力系数，确定出相应的收益评分，所述曝光潜力系数基于所述候选广告位的属性信息中包含的尺寸范围以及相对位置信息确定；体验评分计算模块，设置为根据所述候选广告位的属性信息中包含的基础干扰度值，以及场景干扰修正因子，计算出所述候选广告位的体验评分，所述场景干扰修正因子基于所述上下文特征信息中包含的页面视觉特征、实时交互特征以及场景敏感度确定；综合评分确定模块，设置为根据所述收益评分和所述体验评分，基于预设的收益权重系数以及体验权重系数，计算得到相应候选广告位的综合评分。

[0099] 在本申请的装置任意实施例的基础上，所述综合评分计算模块3300之前，包括：视觉特征构建模块，设置为基于所述目标页面的页面色彩分布特征、字体风格特征和图形元素特征，构建出相应的页面视觉特征；敏感度生成模块，设置为基于所述目标页面的页面类型和用户当前操作行为，生成场景敏感度；特征信息构建模块，设置为将所述页面视觉特征、当前用户页面对应的实时交互特征以及所述场景敏感度，构建为所述目标页面的上下文特征信息。

[0100] 在本申请的装置任意实施例的基础上，所述广告渲染模块3400，包括：样式信息确定模块，设置为基于所述综合评分最高的候选广告位的属性信息，确定出该候选广告位的兼容样式信息；模板匹配模块，设置为基于所述兼容样式信息，结合所述广告渲染数据，从预设的样式模板库中匹配出目标样式模板；组件生成模块，设置为基于该目标样式模板以及所述广告渲染数据，生成相应的广告组件添加至所述综合评分最高的候选广告位的相应位置。

[0101] 为解决上述技术问题，本申请实施例还提供计算机设备。如图3所示，计算机设备的内部结构示意图。该计算机设备包括通过系统总线连接的处理器、计算机可读存储介质、存储器和网络接口。其中，该计算机设备的计算机可读存储介质存储有操作系统、数据库和计算机可读指令，数据库中可存储有控件信息序列，该计算机可读指令被处理器执行时，可使得处理器实现一种广告展示方法。该计算机设备的处理器用于提供计算和控制能力，支撑整个计算机设备的运行。该计算机设备的存储器中可存储有计算机可读指令，该计算机可读指令被处理器执行时，可使得处理器执行本申请的广告展示方法。该计算机设备的网络接口用于与终端连接通信。本领域技术人员可以理解，图3中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。

[0102] 本实施方式中处理器用于执行图2中的各个模块及其子模块的具体功能，存储器存储有执行上述模块或子模块所需的程序代码和各类数据。网络接口用于向用户终端或服务器之间的数据传输。本实施方式中的存储器存储有本申请的广告展示装置中执行所有模块 / 子模块所需的程序代码及数据，服务器能够调用服务器的程序代码及数据执行所有子模块的功能。

[0103] 本申请还提供一种存储有计算机可读指令的存储介质，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行本申请任一实施例的广告展示方法的步骤。

[0104] 本领域普通技术人员可以理解实现本申请上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，该计算机程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体（Read-Only Memory，ROM）等计算机可读存储介质，或随机存储记忆体（Random Access Memory，RAM）等。

[0105] 本技术领域技术人员可以理解，本申请中已经讨论过的各种操作、方法、流程中的步骤、措施、方案可以被交替、更改、组合或删除。进一步地，具有本申请中已经讨论过的各种操作、方法、流程中的其他步骤、措施、方案也可以被交替、更改、重排、分解、组合或删除。进一步地，现有技术中的具有与本申请中开源的各种操作、方法、流程中的步骤、措施、方案也可以被交替、更改、重排、分解、组合或删除。

[0106] 以上所述仅是本申请的部分实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本申请的保护范围。< / script> < / video> < / h3> < / button> < / article> < / section> < / article>

Claims

1. An advertising display method, characterized in that, include: In response to an ad display event, an ad pool containing multiple candidate ad slots is constructed based on the layout structure information of the target page corresponding to the event. Obtain the advertising data package of the advertisement to be delivered, wherein the advertising data package includes the advertising effectiveness value and the advertising rendering data; Based on the advertising effectiveness score, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page, a comprehensive score for each candidate ad slot is calculated. The advertisement to be delivered will be displayed in the candidate advertisement position with the highest comprehensive score based on the advertisement rendering data.

2. The advertising display method according to claim 1, characterized in that, In response to an ad display event, before constructing an ad pool containing multiple candidate ad slots based on the layout structure information of the target page corresponding to the event, the following steps are included: Real-time interaction features corresponding to the current user page are obtained in real time, including user behavior features, current page content tags, and page dwell time. Based on the real-time interaction features, determine whether the preset advertising triggering conditions are met. When the conditions are met, the current user page is used as the target page to trigger the ad display event.

3. The advertising display method according to claim 1, characterized in that, Before obtaining the ad data package for the ad to be delivered, which includes ad performance values ​​and ad rendering data, the process includes: Obtain the historical data of the advertisement to be placed, including historical impressions, historical clicks, historical conversion data, and historical revenue data; Based on the historical display count and historical revenue data, the basic performance value of the advertisement to be placed is calculated; Based on the preset performance value optimization model, the performance value optimization coefficient is obtained according to the historical click count and the historical conversion data. Based on the performance value optimization coefficient, the basic performance value is adjusted to generate the advertising performance value.

4. The advertising display method according to claim 1, characterized in that, In response to an ad display event, based on the layout structure information of the target page corresponding to the event, an ad placement pool containing multiple candidate ad placements is constructed, including: The layout structure information of the target page is analyzed to determine the various user interface components and their positions and sizes. Based on the preset ad placement determination rules, candidate ad placements for the target page are determined according to the location and size information. The ad slot pool is constructed based on the attribute information of the identified candidate ad slots.

5. The advertising display method according to claim 1, characterized in that, Based on the advertising effectiveness score, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page, a comprehensive score is calculated for each candidate ad slot, including: Based on the advertising effectiveness value and exposure potential coefficient, a corresponding revenue score is determined, wherein the exposure potential coefficient is determined based on the size range and relative position information contained in the attribute information of the candidate advertising position; The experience score of the candidate ad slot is calculated based on the basic interference value contained in the attribute information of the candidate ad slot and the scene interference correction factor. The scene interference correction factor is determined based on the page visual features, real-time interaction features and scene sensitivity contained in the context feature information. Based on the revenue score and the experience score, and using preset revenue weighting coefficients and experience weighting coefficients, the comprehensive score of the corresponding candidate ad slot is calculated.

6. The advertising display method according to claim 5, characterized in that, Based on the basic interference value contained in the attribute information of the candidate ad slot, and the scene interference correction factor, the experience score of the candidate ad slot is calculated. The scene interference correction factor, before being determined based on the page visual features, real-time interaction features, and scene sensitivity contained in the context feature information, includes: Based on the page color distribution characteristics, font style characteristics, and graphic element characteristics of the target page, the corresponding page visual characteristics are constructed. Based on the page type of the target page and the user's current operation behavior, a scene sensitivity is generated; The visual features of the page, the real-time interaction features corresponding to the current user page, and the scene sensitivity are used to construct the contextual feature information of the target page.

7. The advertising display method according to any one of claims 1 to 6, characterized in that, Displaying the advertisement to be delivered in the candidate ad slot with the highest overall score based on the ad rendering data includes: Based on the attribute information of the candidate ad slot with the highest comprehensive score, the compatible style information of the candidate ad slot is determined; Based on the compatibility style information and combined with the advertising rendering data, the target style template is matched from the preset style template library; Based on the target style template and the ad rendering data, corresponding ad components are generated and added to the corresponding positions of the candidate ad slots with the highest overall scores.

8. An advertising display device, characterized in that, include: The ad placement pool construction module is configured to respond to ad display events and, based on the layout structure information of the target page corresponding to the event, construct an ad placement pool containing multiple candidate ad placements. The data packet acquisition module is configured to acquire the advertising data packet of the advertisement to be delivered, wherein the advertising data packet includes the advertising effectiveness value and the advertising rendering data; The comprehensive score calculation module is configured to calculate the comprehensive score of each candidate ad slot based on the ad performance value, the attribute information of each candidate ad slot in the ad slot pool, and the contextual feature information of the target page. The ad rendering module is configured to display the ad to be delivered in the candidate ad slot with the highest comprehensive score based on the ad rendering data.

9. A computer device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images