User scene perception method for advertisement material AI generation

Abstract

The invention relates to the technical field of information retrieval, in particular to a user scene perception method for advertisement material AI generation. The method comprises the following steps: constructing a predetermined classification tree, and collecting user browsing information; calculating query deviation according to the browsing information features of the maximum classification, and determining the demand degree by combining the rising degree of the query deviation of the maximum classification in the adjacent time period; based on the number of the type progressive records in the maximum classification, the minimum classification distance between the minimum classification and the predetermined classification tree and the duration to the current moment, obtaining the type progressive performance; obtaining the neglect degree of the maximum classification in combination with the number of times of simultaneous purchase of the maximum classification and the browsing duration of the rest maximum classifications; calculating the favorite degree according to the demand degree and the search frequency in the analysis time period, determining attention classes, determining the purchase bias of the adjacent time period by analyzing the number and the neglect degree of the attention classes in the time period, and then obtaining the advertisement recommendation right; and performing advertisement putting through the advertisement recommendation right. According to the invention, advertisement recommendation is more accurate.

Description

Technical Field

[0001] This application relates to the field of information retrieval technology, specifically to a user scenario perception method for AI-generated advertising creatives. Background Technology

[0002] Precise ad recommendations are crucial for user experience and business efficiency. By analyzing user interests, behaviors, and needs, platforms can reduce irrelevant advertising interference, improve click-through rates and conversion rates, and enhance user satisfaction. For advertisers, precise targeting reduces customer acquisition costs and increases ROI; for platforms, it optimizes traffic monetization efficiency, achieving a win-win-win situation. AI-generated ads can be designed according to user preferences, further encouraging user spending.

[0003] Traditional methods determine user needs based on the frequency of browsing a certain type of product within a given time period and then recommend corresponding advertisements, without considering the impact of users frequently changing the type of product they search for on the weight of advertisement recommendations. Summary of the Invention

[0004] To address the technical issue of inaccurate ad recommendations caused by frequent user browsing changes, this application provides a user scenario-aware method for AI-generated ad creatives. The specific technical solution adopted is as follows: This application proposes a user-scene-aware method for AI-generated advertising creatives, which includes the following steps: A predetermined classification tree is constructed using a classification database, and user browsing information is collected; the browsing information includes product type, number of webpage clicks, and browsing duration. The largest category of product types is designated as the largest category, and all other categories are designated as subcategories. During each product search, the query bias of the largest category is calculated based on the ratio of page clicks for each largest category to all largest categories, as well as the continuous browsing time for each largest category. A preset adjacent time period is used to calculate the first-order difference of all query biases. The degree of increase in query bias is determined based on the number of positive differences and the sum of the positive differences. The demand level of the largest category is determined based on the query bias of the largest category and the degree of increase in query bias across all product searches. Within a nearby time period, the type progression records and their lengths of the largest category are statistically analyzed, and the smallest subcategory with the longest length is recorded as the smallest category. The type progression performance of the largest category is calculated based on the proportion of type progression records for each largest category, the length difference between the smallest category and the smallest subcategory in the predetermined category tree, and the duration of the smallest category up to the current time. The neglect level of the largest category is obtained based on the type progression performance of the largest category in a nearby time period, the number of times the largest category and other largest categories are purchased simultaneously, and the browsing time of other largest categories. Select an analysis period and calculate the preference for the largest category based on the percentage of searches for the largest category among all searches and the demand for the largest category across all analysis periods. Record the largest category with an ignore rate less than a preset threshold in adjacent time periods as the category of interest. Calculate the purchase bias in adjacent time periods based on the percentage of interest categories and the sum of ignore rates within the analysis period. Calculate the advertising recommendation weight for the largest category based on its preference, purchase bias, and demand. The recommendation weight is calculated by using advertising recommendation rights, and advertising is then delivered based on the recommendation weight.

[0005] In the above solution, this invention analyzes and obtains the user's current level of product neglect by examining the progression of different product levels within the same type and the degree of relevance between the user's browsing of other product types and the current product as the levels evolve. Furthermore, it determines the overall purchase preference of the current customer based on the user's attention to different products, and uses this as a weighting reference to analyze the impact of user preference preferences on the recommendation of different product types. Ultimately, it obtains recommendation weights for different product types, making the recommended products more aligned with user needs, increasing the time users spend browsing the purchase page, and thus improving the likelihood of user purchases, resulting in more accurate advertising recommendations.

[0006] In one embodiment, the query bias of the largest category is positively correlated with a first ratio and continuous browsing time, respectively; the first ratio is the ratio of the number of clicks on the webpage corresponding to each largest category to the number of clicks on the webpages of all largest categories.

[0007] In one embodiment, the degree of increase in query bias is positively correlated with the proportion of values ​​greater than 0 in the bias difference sequence and the sum of values ​​greater than 0 in the bias difference sequence. The bias difference sequence is a first-order difference sequence of the sequence composed of all query biases.

[0008] In one embodiment, the degree of demand is positively correlated with the degree of increase in query bias and the sum of query biases.

[0009] In one embodiment, the method for calculating the type progression performance of the maximum category based on the proportion of type progression records in each maximum category, the length difference between the minimum category and the minimum subclass of the predetermined classification tree, and the time elapsed from the minimum category to the current moment is as follows: , This represents the number of records in the largest category k. This indicates the number of records of all types in a progressive manner. This represents the distance interval of the j-th minimum category. Let represent the time elapsed from the j-th smallest category to the current time. This represents the number of the smallest categories. This represents the type progression of the maximum category k, where the distance interval of the minimum category is the difference between the length of the minimum category and the length of the minimum subclass in the predetermined classification tree.

[0010] In one embodiment, the ignore degree is positively correlated with the number of other major categories besides the major category and the browsing time of the other major categories, respectively. The ignore degree is negatively correlated with the type progression performance of the major category and the sales ratio of the major category and the other major categories. The sales ratio of the other categories is the number of times the two major categories are purchased at the same time divided by the total number of shopping.

[0011] In one embodiment, the method for calculating the preference level of the largest category based on the proportion of searches for the largest category among all searches and the demand level of the largest category across all analysis time periods is as follows: , This indicates the proportion of the largest category k. This represents the maximum percentage of all the largest categories. This represents the demand level of the largest category k in the r-th analysis period. Indicates the number of time periods analyzed. This represents the degree of preference for the largest category k.

[0012] In one embodiment, the purchase bias is positively correlated with the degree of ignoring the focus category and negatively correlated with the proportion of the focus category.

[0013] In one embodiment, the method for calculating the advertising recommendation weight of the largest category based on the degree of liking, purchase bias, and demand of the largest category is as follows: , This represents the degree of preference for the largest category k. This indicates the degree of demand for the largest category k. This indicates purchasing preferences within a recent time period. Represents the normalization function. This represents the advertising recommendation weight of the largest category k.

[0014] In one embodiment, the method for calculating recommendation weights through advertising recommendation rights and delivering advertisements based on those recommendation weights is as follows: After normalizing all ad recommendation weights, when the normalized ad recommendation weight is greater than the recommendation threshold, the corresponding maximum category is recorded as the ad category. The ad recommendation weight of each ad category is then divided by the ad recommendation weights of all ad categories to obtain the recommendation weight of each maximum category. Based on the recommendation weight, the recommendation time for different ad categories is given.

[0015] The beneficial effects of this application are as follows: This invention analyzes the degree to which a user is currently ignoring a product by examining the progression of different product tiers within the same product category and the correlation between this product and other product types browsed during this tier evolution. Furthermore, it determines the overall purchase preference of the current customer based on the user's attention to different products, and uses this as a weighting reference to analyze the impact of user preference preferences on the recommendation of different product types. Ultimately, it obtains recommendation weights for different product types, making the recommended products more aligned with user needs, increasing the time users spend browsing the purchase page, and thus improving the likelihood of a purchase, resulting in more accurate advertising recommendations. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a user scene perception method for AI-generated advertising creatives, provided as an embodiment of this application. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the user scene perception method for AI generation of advertising materials proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] Example of a user-scene-aware method for AI-generated advertising creatives: The following description, in conjunction with the accompanying drawings, details the specific scheme of the user scene perception method for AI generation of advertising creatives provided in this application.

[0021] Please see Figure 1 The diagram illustrates a flowchart of a user scene perception method for AI-generated advertising creatives according to an embodiment of this application. The method includes the following steps: Step S001: Construct a predetermined classification tree and collect user browsing information.

[0022] Obtain the user's access path under the current domain using the browser's history API (such as window.history). Read the page access records cached in the browser's local storage (such as localStorage, IndexedDB). Extract the user's page request records from the server access logs (such as Nginx, Apache logs) (which need to be associated with the user ID or session).

[0023] The system uses JavaScript on the front end to listen for events such as `window.onload` and `window.beforeunload`, recording page URLs, dwell time, and scroll depth. Session storage or cookies are used to maintain user sessions and associate them with multiple page browsing sequences.

[0024] Based on the existing classification database Wikidata, a predefined classification tree is constructed to establish multi-level categories, such as "electronic products → mobile phones → mobile phone brands → mobile phone brand models". The product models, browsing duration, and page click counts of users are transmitted to the database for storage and then sent to the central processing unit for ad recommendation analysis.

[0025] At this point, multiple data points from the user's browsing experience have been obtained.

[0026] Step S002: Calculate the query bias based on the browsing information characteristics of the largest category, and determine the demand level of the largest category by combining the degree of increase of the query bias of the largest category in the adjacent time period.

[0027] Because users have varying levels of demand for different products at different times, advertisements should prioritize products with higher user demand to provide more suitable products. When a user searches for a particular product type more frequently than other product types within a certain timeframe, and the duration of their searches for that product type is longer, the user's demand for that product type is greater during that period. Furthermore, if a user fails to find a satisfactory product type after several consecutive searches, and their demand for that product type increases, their demand for that product will gradually increase with each subsequent search.

[0028] A single product search is defined as the time from when a user opens their browser to when they close it. Based on a predefined category tree, the category with the most viewed products is designated as the maximum category, such as "electronic products," and other categories are designated as subcategories. The ratio of the number of clicks on the webpage corresponding to each maximum category to the total number of clicks on the webpages of all maximum categories is recorded as the first ratio. The continuous browsing time for each maximum category is calculated. If a maximum category is interrupted by other maximum categories, the average continuous browsing time for each instance of browsing that maximum category is taken as the continuous browsing time. For example, in a product search, if a user browses one maximum category (k) for 5 minutes, then accidentally clicks on another maximum category or for other reasons, and then switches back to maximum category k for 3 minutes, then the continuous browsing time for maximum category A is 4 minutes.

[0029] During each product search, the longer the continuous browsing time, the larger the first ratio, indicating a greater user preference for that major category. This helps to determine the user's search preference for that major category.

[0030] The query bias of the largest category is positively correlated with the first ratio and the continuous browsing time, respectively.

[0031] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large. The specific relationship is determined by the actual application, and this application does not impose any special restrictions.

[0032] The preferred expression for the maximum category query bias is: , This represents the number of clicks on the webpage of the largest category k under the p-th product search. This represents the total number of page clicks in the p-th product search. This represents the continuous browsing duration of the largest category k under the p-th product search. This indicates the query bias of the largest category k in the p-th product search.

[0033] Analyze the user's historical search records and make advertising recommendations based on the user's historical search behavior. Therefore, the data of the previous preset days are counted and the time period of the preset days is recorded as the adjacent time period. In this embodiment, the data of the previous 5 days are taken.

[0034] Within a short timeframe, for each product search, the query bias for each major category is calculated. All query biases for each major category within the short timeframe are then sorted chronologically to obtain a bias sequence for each major category. A first-order difference is performed on this bias sequence to obtain a bias difference sequence. The higher the proportion of values ​​greater than 0 in the first-order difference and the larger the sum of these values, the more it indicates that the query bias for that major category is gradually increasing. Based on this, the degree of increase in the query bias for the major category is calculated.

[0035] The degree of increase in query bias is positively correlated with the proportion of values ​​greater than 0 in the biased difference sequence and the sum of values ​​greater than 0 in the biased difference sequence.

[0036] Preferably, in this embodiment, the expression for querying the degree of upward bias is: , This represents the percentage of values ​​greater than 0 in the biased difference sequence of the largest class k. This represents the first-order difference value of the query bias for the largest category k in the (p+1)th product search. This represents the number of values ​​greater than 0 in the biased difference sequence of the largest class k. This indicates the degree of query bias towards the largest category k.

[0037] When a user performs multiple product searches consecutively, the degree to which the user's search preference for a particular product increases. The larger the value and the greater the frequency of queries, the greater the user demand for this type of product, making it more appropriate to recommend this product type through advertising.

[0038] Therefore, the demand for each maximum category is obtained based on the degree of increase in query bias in the vicinity of the maximum category and the sum of query biases.

[0039] The degree of demand is positively correlated with the degree of increase in query bias and the sum of query biases.

[0040] Preferably, in this embodiment, the expression for the degree of demand is: , This indicates the query bias of the largest category k in the p-th product search. This represents the number of values ​​greater than 0 in the biased difference sequence of the largest class k. This indicates the degree of query bias towards the largest category k. This indicates the degree of demand for the largest category k.

[0041] At this point, the demand level for each major category has been obtained.

[0042] Step S003: Obtain the type progression performance based on the number of type progression records in the maximum category, the distance between the minimum category and the minimum category of the predetermined category tree, and the duration up to the current time; then, combine the number of times the maximum category and other maximum categories were purchased at the same time, and the browsing duration of the other maximum categories to obtain the neglect degree of the maximum category.

[0043] When purchasing a product, users often delve deeper into the product's main category and then into its subcategories to determine the product they want to buy. Therefore, when the types of products a user clicks or searches for become increasingly detailed over time, it indicates that the user is placing greater emphasis on purchasing that product, and the advertising recommendation should give greater weight to that type of product.

[0044] When a user searches for multiple products within a short timeframe, the system stores the maximum category and the minimum subcategory of each product. The minimum subcategory refers to the smallest subcategory of the searched product, not necessarily the smallest subcategory in a predefined classification tree. For example, if the searched product is a mobile phone of a certain brand, its maximum category is "electronic products," while its minimum subcategory could be a specific model of that brand of mobile phone or simply the brand itself. The system then stores the searched products according to their corresponding subcategories based on the search sequence.

[0045] Retrieve products belonging to the same maximum category from the stored product time sequence. Determine the progressive relationship of products within the same maximum category as they delve deeper into smaller categories. Compare the type names of the multi-level categories in the constructed predefined classification tree with the corresponding smaller categories of products belonging to the same maximum category, ensuring that each smallest subcategory corresponds to a type progression record with a length equal to the number of types. For example, Electronic Products → Mobile Phones → Mobile Phone Brands → Mobile Phone Brand Models has a length of 4.

[0046] The algorithm analyzes all type progression records and their corresponding lengths within the maximum category. The smallest subclass with the longest type progression record is designated as the minimum category. The difference between the length of the minimum category and the length of that smallest subclass in the predefined category tree is used as the distance interval for each minimum category. The algorithm also calculates the time elapsed between the minimum categories and the current time. A smaller distance interval indicates a more specific category, while a closer interval to the current time suggests a higher likelihood of user purchase of the product. Furthermore, the larger the proportion of type progression records in the maximum category, the more ad recommendations should be given to that category.

[0047] Based on this, the type progression of the maximum classification is calculated, and the expression is: , This represents the number of records in the largest category k. This indicates the number of records of all types in a progressive manner. This represents the distance interval of the j-th minimum category. Let represent the time elapsed from the j-th smallest category to the current time. This represents the number of the smallest categories. This represents the progressive representation of the largest category k.

[0048] When users search for products, they don't necessarily target only one type. They may also search for other types of products along the way. The less focused a user is on a particular product type, the more frequently they will search for other products. Furthermore, the correlation between the other products searched and the current product type will be weaker.

[0049] Therefore, for two different maximum categories, the number of times a user purchased from both categories simultaneously in their historical shopping history is counted, and then compared with the total number of purchases to obtain the percentage of sales related to the two maximum categories.

[0050] Within a short time period, count the number of searches for other major categories besides the major category k; and count the browsing time for the other major categories besides the major category k. The smaller the correlation between the other major categories interspersed with the major category k, the more major categories interspersed, the longer the browsing time for other major categories when searching for the major category k, and the smaller the type progression of the major category k, the more the user considers other products when searching for the major category k, the less attention they pay to the major category k, and the less purchase preference they have for the major category k. Therefore, the recommendation weight of the major category k should be reduced when recommending products.

[0051] This allows us to obtain the user's degree of neglect for the largest category. The degree of neglect is positively correlated with the number of other largest categories besides the largest category and the browsing time of the other largest categories. The degree of neglect is negatively correlated with the type progression of the largest category and the sales ratio of the largest category and the other largest categories.

[0052] It should be noted that negative correlation means that when one variable increases, the other variable decreases accordingly, and the two variables change in opposite directions. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small. The specific relationship is determined by practical application, and this application does not impose any special restrictions.

[0053] The preferred expression for the degree of ignoring is: , This indicates the number of the largest category in the nearest time period. This represents the sales ratio associated with the largest category k and the largest category i. This represents the browsing time for all categories other than the largest category k. This represents the progressive representation of the largest category k. This indicates the degree of ignoring the largest category.

[0054] The above operations are used to determine the level of attention ignored by a user for all product types searched within a certain period of time. When the obtained level of attention ignored... The larger the value, the less attention users pay to the largest category k, the more likely it is to be a mistaken click or a sudden impulse, and the lower the probability of purchase. Therefore, the recommendation weight for this product type should be smaller when recommending products in advertisements.

[0055] At this point, the maximum neglect level of the category has been obtained.

[0056] Step S004: Calculate the degree of preference based on the demand and search frequency of the largest category in the analysis period, determine the category of interest, and determine the purchase bias of adjacent time periods by analyzing the number and ignore rate of the category of interest in the analysis period, thereby obtaining the right to recommend advertisements.

[0057] Sometimes users may be browsing aimlessly without a clear purchase intention. In such cases, analyzing multiple search terms for different product types will result in a high rate of neglect. To address this, analyzing a user's browsing history over a longer period can reveal their preferences. Products that align with these preferences can then be recommended to the user, extending their browsing time and potentially increasing their desire to buy.

[0058] The analysis period was selected from historical time periods of equal length to those of neighboring time periods; the analysis was conducted based on the proportion of different maximum categories (k) in past users' search records. And the degree of user demand for the maximum classification k within several analysis periods. As a reference for users' liking of this type of product, the ratio of the largest category's percentage to the largest percentage within the largest category is used as the percentage of the largest category. The larger the percentage, the more popular the largest category is. The degree of liking of the largest category is calculated in combination with the degree of demand.

[0059] Preferably, in this embodiment, the expression for the degree of liking of the maximum category is: , This indicates the proportion of the largest category k. This represents the maximum percentage of all the largest categories. This represents the demand level of the largest category k in the r-th analysis period. Indicates the number of time periods analyzed. This represents the degree of preference for the largest category k. In this embodiment, the value of R is 5.

[0060] The number of the largest categories with an ignore degree less than a preset threshold within a nearby time period is counted. These largest categories are recorded as the attention categories. The lower the number of attention categories is relative to the total number of largest categories within a nearby time period, and the higher the ignore degree of the attention categories, the more it indicates the user's purchasing bias within the analysis time period.

[0061] The purchase bias is positively correlated with the degree of ignoring the categories of interest, and negatively correlated with the proportion of the number of categories of interest.

[0062] Preferably, in this embodiment, the expression for purchasing bias is: , This represents the number of all the largest categories within a nearby time period. This indicates the number of categories of interest within a nearby time period. This represents the degree of neglect of the e-th class of interest within the adjacent time period. This indicates purchasing preferences in the immediate time period.

[0063] When users' purchasing preferences for the largest categories searched within a nearby time period When the weighting is low, the recommendation weight of ads in the categories with higher user preference should be appropriately increased to increase the user's browsing time on the search page and improve the user's purchase preference. This will affect the user's current time frame. Demand Together with the adjusted level of liking, it serves as the current... Selection reference.

[0064] Based on this, the expression for calculating the ad recommendation weight of the largest category is: , This represents the degree of preference for the largest category k. This indicates the degree of demand for the largest category k. This indicates purchasing preferences within a recent time period. Represents the normalization function. This represents the advertising recommendation weight of the largest category k.

[0065] At this point, they acquired the advertising recommendation rights for the largest category.

[0066] Step S005: Place advertisements through advertising recommendation rights.

[0067] The user's browsing information is input into the central processing unit for the above analysis. The analysis is performed every 24 hours. After calculating the maximum category's advertising recommendation weight according to the above steps, the advertising recommendation weight is normalized. When the obtained advertising recommendation weight is greater than the recommendation threshold, this maximum category is used as the advertising category to recommend to the user, thereby promoting the user's consumption. The obtained advertising product types that should be recommended are stored in the database.

[0068] By using existing AI-powered algorithms for intelligent splicing and optimization of advertising creatives, the algorithm selects creatives from the creative library based on the ad category to be recommended, and uses legacy algorithms to combine and optimize different creative combinations to obtain suitable recommended ads.

[0069] The recommendation weight of each ad category is calculated by comparing the recommendation weight of all ad categories to obtain the recommendation weight of each category. Based on the recommendation weight, different time slots are allocated to each ad category to deliver ads to the user's browsing web pages.

[0070] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

[0071] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A user scene perception method for AI-generated advertising creatives, characterized in that, The method includes the following steps: A predetermined classification tree is constructed using a classification database, and user browsing information is collected; the browsing information includes product type, number of webpage clicks, and browsing duration. The category with the highest product type in the predefined search tree is designated as the maximum category, and all other categories are designated as subcategories. During each product search, the query bias of the maximum category is calculated based on the ratio of page clicks for each maximum category to all maximum categories and the continuous browsing duration of each maximum category. A pre-defined adjacent time period is used to calculate the first-order difference of all query biases. The degree of increase in query bias is determined based on the number of positive differences and the sum of the positive differences. The demand level of the maximum category is determined based on the query bias of the maximum category and the degree of increase in query bias under all product searches. Within a nearby time period, the type progression records and their lengths of the largest category are statistically analyzed, and the smallest subcategory with the longest length is recorded as the smallest category. The type progression performance of the largest category is calculated based on the proportion of type progression records for each largest category, the length difference between the smallest category and the smallest subcategory in the predetermined category tree, and the duration of the smallest category up to the current time. The neglect level of the largest category is obtained based on the type progression performance of the largest category in a nearby time period, the number of times the largest category and other largest categories are purchased simultaneously, and the browsing time of other largest categories. Select an analysis period and calculate the preference for the largest category based on the percentage of searches for the largest category among all searches and the demand for the largest category across all analysis periods. Record the largest category with an ignore rate less than a preset threshold in adjacent time periods as the category of interest. Calculate the purchase bias in adjacent time periods based on the percentage of interest categories and the sum of ignore rates within the analysis period. Calculate the advertising recommendation weight for the largest category based on its preference, purchase bias, and demand. The recommendation weight is calculated by using advertising recommendation rights, and advertising is then delivered based on the recommendation weight.

2. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The query bias of the largest category is positively correlated with the first ratio and the continuous browsing time, respectively; the first ratio is the ratio of the number of clicks on the webpage corresponding to each largest category to the number of clicks on the webpages of all largest categories.

3. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The degree of increase in query bias is positively correlated with the proportion of values ​​greater than 0 in the bias difference sequence and the sum of values ​​greater than 0 in the bias difference sequence; the bias difference sequence is the first-order difference sequence of the sequence composed of all query biases.

4. The user scene perception method for AI-generated advertising materials as described in claim 1, characterized in that, The degree of demand is positively correlated with the degree of increase in query bias and the sum of query biases.

5. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The method for calculating the type progression performance of the maximum category based on the proportion of type progression records in each maximum category, the length difference between the minimum category and the minimum subclass of the predetermined classification tree, and the time elapsed from the minimum category to the current time is as follows: , This represents the number of records in the largest category k. This indicates the number of records of all types in a progressive manner. This represents the distance interval of the j-th minimum category. Let represent the time elapsed from the j-th smallest category to the current time. This represents the number of the smallest categories. This represents the type progression of the maximum category k, where the distance interval of the minimum category is the difference between the length of the minimum category and the length of the minimum subclass in the predetermined classification tree.

6. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The ignore degree is positively correlated with the number of other major categories besides the major category and the browsing time of other major categories. The ignore degree is negatively correlated with the type progression of the major category and the sales ratio of the major category and other major categories. The sales ratio of the two major categories is the number of times the two major categories are purchased at the same time divided by the total number of shopping.

7. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The method for calculating the preference level of the largest category based on the proportion of searches for the largest category among all searches and the demand level of the largest category across all analysis time periods is as follows: , This indicates the proportion of the largest category k. This represents the maximum percentage of all the largest categories. This represents the demand level of the largest category k in the r-th analysis period. Indicates the number of time periods analyzed. This represents the degree of preference for the largest category k.

8. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The purchase bias is positively correlated with the degree of ignoring the categories of interest, and negatively correlated with the proportion of the number of categories of interest.

9. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The method for calculating the advertising recommendation weight of the largest category based on the degree of liking, purchase bias, and demand of the largest category is as follows: , This represents the degree of preference for the largest category k. This indicates the degree of demand for the largest category k. This indicates purchasing preferences within a recent time period. Represents the normalization function. This represents the advertising recommendation weight of the largest category k.

10. The user scene perception method for AI-generated advertising creatives as described in claim 1, characterized in that, The method for calculating recommendation weights through advertising recommendation rights and then placing ads based on those recommendation weights is as follows: After normalizing all ad recommendation weights, when the normalized ad recommendation weight is greater than the recommendation threshold, the corresponding maximum category is recorded as the ad category. The ad recommendation weight of each ad category is then divided by the ad recommendation weights of all ad categories to obtain the recommendation weight of each maximum category. Based on the recommendation weight, the recommendation time for different ad categories is given.

Images