A method and system for intelligent online advertising push based on big data

Abstract

The present invention discloses an online advertising intelligent push method and system based on big data, which relates to the technical field of advertising push, including collecting user behavior data to construct a user interest model to generate a user interest score, dividing the advertising push area into a high-sales area and a low-sales area; pushing advertisements to users in different areas according to the user interest score, and dynamically adjusting the advertisement content according to the changes in the advertisement performance in different areas. The present invention can capture the dynamic interest changes of users by constructing a user interest model and generating a user interest score, can push advertisements in real time and accurately in response to user needs, divide the advertising push area into a high-sales area and a low-sales area to improve the problem of unreasonable resource allocation, and make advertising push more effective. Pushing advertisements to users in different areas according to the user interest score makes advertising push more efficient, and improves the user's advertising acceptance and conversion rate.

Description

Technical Field

[0001] The present invention relates to the field of advertisement push technology, and in particular to a method and system for intelligent online advertisement push based on big data. Background Art

[0002] With the rapid development of Internet technology and big data analysis, online advertising has become one of the important means for enterprises to promote products and services. In the early days of Internet advertising, the way of advertising was relatively simple, usually based on page content or simple user behavior. Although this method has improved the relevance of advertising to a certain extent, its accuracy and user experience still have a lot of room for improvement. With the extensive collection of user behavior data and the advancement of analysis technology, advertising push has gradually shifted from the traditional wide-ranging push to a more personalized and precise direction. However, most existing technologies are still relatively extensive in the analysis of user behavior and the formulation of advertising push strategies, and cannot fully consider the complexity and diversity of user behavior. Existing user interest models often rely on single behavioral characteristics or simple statistical methods, which makes it difficult to accurately capture the multi-dimensional interest changes of users. The division of advertising push areas lacks flexibility, and it is impossible to dynamically adjust the push strategy to adapt to the changes in user needs in different areas. Therefore, the accuracy and user experience of existing advertising push still need to be improved. Summary of the invention

[0003] In view of the problems existing in the above-mentioned existing online advertising intelligent push method and system based on big data, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is that the accuracy of existing advertisement push and user experience still need to be improved.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: an online advertising intelligent push method based on big data, which includes collecting user behavior data to construct a user interest model to generate a user interest score, and dividing the advertising push area into a high-sales area and a low-sales area; pushing advertisements to users in different areas according to the user interest score, and dynamically adjusting the advertisement content according to changes in advertisement performance in different areas; storing the data generated during the advertisement push process and visually displaying it.

[0006] As a preferred solution of the online advertising intelligent push method based on big data described in the present invention, wherein: the collection of user behavior data to build a user interest model and generate a user interest score refers to collecting historical user behavior data related to advertising, real-time user behavior data and advertising content data, cleaning the collected data and filling in missing values ​​to standardize the data, and mapping each user behavior with the corresponding advertising ID;

[0007] Construct a search-based user interest model, including general search units and precise search units;

[0008] Using a general search unit to extract sub-behavior sequences related to target ads from long-term user behavior data includes building an ad behavior tree, defining the root node as the initial state of the user before ad interaction, the middle node as the user's specific behavior event on the ad, each node storing the ad ID, behavior type, behavior time, and the leaf node as the termination state of the user's ad behavior;

[0009] Create a root node for each user as the starting point of the advertising behavior path. Insert the user's advertising behavior as a new node into the advertising behavior tree according to the time sequence of the advertising behavior.

[0010] Add indexes to nodes of high-frequency behavior paths and mark paths with excellent performance in the advertising behavior tree;

[0011] According to the target advertisement content, the behavior nodes related to the target advertisement are screened out from the advertisement behavior tree, and the advertisement behavior tree is traversed using a depth-first search algorithm to extract the sub-behavior sequences related to the target advertisement;

[0012] Use an embedding layer to map the behavior data in the sub-behavior sequence into a high-dimensional vector to form a behavior vector sequence;

[0013] Perform linear transformation on the behavior vector to generate query matrix, key matrix and value matrix, set multiple attention heads, input query, key and value matrices into multiple attention heads, use query matrix and key matrix to perform dot product calculation in each attention head to obtain the correlation score between behaviors, normalize the correlation score through softmax function, apply the normalized weight to the value matrix to obtain the weighted behavior vector output, merge the weighted behavior vectors to generate the user's interest embedding vector;

[0014] Normalize the generated interest embedding vector;

[0015] Construct a user-ad interaction matrix, map the standardized user interest embedding vector to the user-ad interaction matrix, where rows represent users, columns represent ad content, and each value in the matrix represents the interaction frequency between users and ads;

[0016] The SVD algorithm is used to perform singular value decomposition on the user-ad interaction matrix to obtain three sub-matrices, including the user feature matrix, the singular value matrix, and the ad feature matrix;

[0017] To calculate the interest score of user u for advertisement p, the corresponding user features of user u and advertisement features of advertisement p are extracted from the user feature matrix and the advertisement feature matrix;

[0018] The user features, singular value matrix and advertisement features are multiplied by matrix multiplication to obtain the interest score of user u for advertisement p.

[0019] As a preferred solution of the online advertising intelligent push method based on big data described in the present invention, wherein: the division of the advertising push area into high-sales area and low-sales area refers to collecting sales data of the corresponding products after historical advertising push and historical advertising push data, and cleaning and standardizing the collected data, performing median calculation on the standardized advertising sales data, and dividing the advertising push area into high-sales area and low-sales area according to the median calculation result.

[0020] As a preferred solution of the online advertising intelligent push method based on big data described in the present invention, wherein: the pushing of advertisements to users in different regions according to user interest scores refers to setting an interest threshold Q for pushing advertisements, and comparing it with the user interest score. If the user interest score is greater than or equal to the interest threshold Q, the user is classified as a high-interest user; if the user interest score is less than the interest threshold Q, the user is classified as a low-interest user;

[0021] If the user is a high-interest user and is located in a high-sales area, relevant advertising content will be pushed directly to the user. If the user is in a low-sales area, the advertising content will be optimized and promotional activities and coupons will be added;

[0022] If the user is a low-interest user and is located in a high-sales area, priority will be given to pushing advertising content with broad appeal. If the user is located in a low-sales area, the frequency of ad push will be controlled and a new advertising combination will be constructed.

[0023] As a preferred solution of the method for intelligent online advertising push based on big data described in the present invention, wherein: the dynamic adjustment of advertising content according to changes in advertising performance in different regions refers to real-time collection of user behavior data on clicks, browsing and conversions of advertisements, and analysis of the performance of push advertisements in high-sales areas and low-sales areas. If the user responds poorly to the push advertisements, the advertising materials are redesigned and the display format of the advertisements is adjusted. After optimizing the push advertisements, the advertisements are pushed again to the target users, and the user's interest score is updated based on the real-time collected user behavior data on the push advertisements.

[0024] As a preferred solution of the online advertising intelligent push method based on big data described in the present invention, the data generated during the storage of advertising push refers to storing user behavior data, advertising content data, advertising push results and advertising optimization data in a database, partitioning the data according to the advertising push area, and regularly backing up and access controlling the stored data.

[0025] As a preferred solution of the online advertising intelligent push method based on big data described in the present invention, the visual display refers to using the Tableau visualization tool to design an interactive visualization interface, using a line chart to display the changing trend of the advertisement click-through rate and conversion rate over time, and using a bar chart to display the click status of different advertisements in various regions, and only authorized users are allowed to view the data.

[0026] Another object of the present invention is to provide an online advertising intelligent push system based on big data, which comprises:

[0027] A data collection module, used to collect real-time and historical user behavior data and advertising content data and pre-process the data;

[0028] The interest score calculation module is used to build a search-based user interest model and advertising behavior tree, capture the interaction between users and advertisements from the tree, generate user interest embedding vectors and build a user-ad interaction matrix, and use the SVD algorithm to calculate the user's interest score for advertisements;

[0029] Advertisement push module, used to divide the advertisement push area, push advertisements to users according to user interest scores and regional divisions, and optimize and adjust advertisement content according to real-time advertisement feedback data;

[0030] The data storage module is used to store the data generated during the advertisement push process and to visualize the data.

[0031] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of an online advertising intelligent push method based on big data when executing the computer program.

[0032] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of an online advertising intelligent push method based on big data.

[0033] The beneficial effects of the present invention are as follows: the present invention can capture the dynamic interest changes of users by constructing a user interest model and generating a user interest score, can respond to user needs in real time and accurately to push advertisements, and divides the advertisement push areas into high-sales areas and low-sales areas to improve the problem of unreasonable resource allocation, making advertisement push more effective. Pushing advertisements to users in different areas according to user interest scores makes advertisement push more efficient, thereby improving users' advertisement acceptance and conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0035] Figure 1 The figure is a flowchart of an online advertising intelligent push method based on big data.

[0036] Figure 2 This is a structural diagram of the online advertising intelligent push system based on big data. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0038] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0039] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0040] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and which provides an online advertising intelligent push method based on big data. The online advertising intelligent push method based on big data includes:

[0041] S1. Collect user behavior data to build a user interest model to generate user interest scores, and divide the advertising push areas into high-sales areas and low-sales areas;

[0042] Specifically, collecting user behavior data to build a user interest model and generate a user interest score refers to collecting historical user behavior data and real-time user behavior data related to advertising, including ad clicks, ad views, ad conversions (purchases and registrations), and ad content data, cleaning the collected data, filling in missing values, and then standardizing the data, mapping each user behavior with the corresponding ad ID;

[0043] Construct a search-based user interest model, including general search units and precise search units;

[0044] Using a general search unit to extract sub-behavior sequences related to target ads from long-term user behavior data includes: constructing an ad behavior tree, defining the root node as the initial state of the user before ad interaction, the intermediate nodes as the user's specific behavior events on the ad (ad browsing, clicking, interaction, conversion), each node storing the ad ID, behavior type, behavior time, and the leaf node as the termination state of the user's ad behavior;

[0045] Create a root node for each user and use it as the starting point of the advertising behavior path. According to the time sequence of advertising behavior, the user's advertising behavior is inserted into the advertising behavior tree as a new node. If the user interacts with the same advertisement multiple times, the nodes will be arranged in the order of behavior to form a behavior path. If the user interacts with different advertisements, different behavior paths will be formed. Each path represents a different interaction sequence between the user and the advertisement. If the advertising interaction behavior paths of different users are the same, the behavior paths are merged, and the weight of the node is adjusted according to the frequency of the user's behavior in different paths.

[0046] Add indexes to nodes of high-frequency behavior paths and mark paths with excellent performance (high click-through rate, high conversion rate paths) in the advertising behavior tree;

[0047] According to the target advertisement content, the behavior nodes related to the target advertisement are screened out from the advertisement behavior tree. The advertisement behavior tree is traversed using a depth-first search algorithm to extract the sub-behavior sequences related to the target advertisement. Each data point in the behavior sequence contains the timestamp of the user behavior, the behavior type (such as click, browse, purchase, etc.) and the corresponding advertisement ID.

[0048] Through the depth-first search (DFS) algorithm, the sub-behavior sequence related to the target ad can be extracted from the ad behavior tree. This process recursively traverses the nodes of the tree, checks whether each node is related to the target ad, and stores the information of the related nodes, which can effectively capture the user's interaction behavior with a specific ad.

[0049] Use an embedding layer to map the behavior data in the sub-behavior sequence into a high-dimensional vector:

[0050] Define an embedding matrix E with a size of (Y, y), where Y is the vocabulary size of the behavior type, i.e. the number of possible different behavior types, and y is the dimension of the embedding vector;

[0051] Map each discrete behavior feature (click, browse, purchase) to an index. The index points to the row in the embedding matrix to generate a behavior sequence. For each behavior feature index, find the i'th row in the embedding matrix E. This row is the embedding vector of the behavior: h i =E[i'],h iis the embedding vector of behavior i, E[i'] is the i'th row in the embedding matrix, and the entire behavior sequence is mapped to the corresponding embedding vector sequence through the embedding layer;

[0052] forming a sequence of behavior vectors;

[0053] Perform a linear transformation on the behavior vector to generate the query matrix, key matrix and value matrix. Set multiple attention heads, each of which is responsible for capturing the correlation of different dimensions in the behavior sequence. Input the query, key and value matrices into multiple attention heads. In each attention head, use the query matrix and key matrix to perform dot product calculation to obtain the correlation score between the behaviors. Normalize the correlation score through the softmax function to ensure that the weight is between [0, 1]. Apply the normalized weight to the value matrix to obtain the weighted behavior vector output:

[0054]

[0055] Where Q is the query matrix with dimension n×d k , where n is the sequence length, d k is the feature dimension, K is the key matrix, with dimension n×d k , V is the value matrix, dimension is n×d v , the dimension of the weighted summation is still n×d v , d v is the dimension of each vector of the value matrix V;

[0056] Merge the weighted behavior vectors to generate the user's interest embedding vector;

[0057] Normalize the generated interest embedding vector;

[0058] Construct the user-ad interaction matrix R:

[0059]

[0060] In the formula, r mn represents the interaction value of the mth user to the nth advertisement;

[0061] Map the standardized user interest embedding vector to the user-ad interaction matrix, where rows represent users, columns represent ad content, and each value in the matrix represents the interaction frequency between users and ads;

[0062] Use the SVD algorithm to perform singular value decomposition on the user-ad interaction matrix (use NumPy's numpy.linalg.svd method to directly calculate the SVD decomposition) to obtain three sub-matrices, including the user feature matrix, the singular value matrix, and the ad feature matrix:

[0063] R=UIPT ,

[0064] Where U is the user feature matrix with dimension m×k, where m is the number of users and k is the number of potential features. I is the singular value matrix with dimension k×k, which is a diagonal matrix containing k singular values. T is the transpose of the ad feature matrix, with dimension k×n, where n is the number of ads;

[0065] To calculate the interest score of user u for advertisement p, the corresponding user features of user u and advertisement features of advertisement p are extracted from the user feature matrix and the advertisement feature matrix;

[0066] Multiply the user features, singular value matrix and ad features by matrix multiplication to get the interest score of user u for ad p

[0067]

[0068] Where U u is the feature vector of user u extracted from the user feature matrix, I is the singular value matrix used to scale the feature vector, P p is the feature vector of ad j extracted from the ad feature matrix.

[0069] By collecting historical and real-time behavior data of users, the system can dynamically capture changes in user interests. The model can identify users' interests in different products or services at different time points, thereby ensuring that advertising push is more accurate and timely. For example, when users frequently browse a certain type of product, the system can immediately identify this behavioral trend and push relevant ads. The relevance and timeliness of advertising push are improved, and the click-through rate and conversion rate of ads are greatly increased. The advertising behavior tree records all the interaction paths of users in advertising push by presenting the interaction behavior between users and ads in a structured manner. Combined with the user interest model, the advertising behavior tree can not only better understand the user's behavior pattern, but also conduct a detailed analysis of the user's behavior path. In this way, the system can more accurately identify the key nodes of users in the advertising interaction process and optimize the display form of advertising content. For example, if the user's interest model shows that they have a high interest in a certain type of advertisement, the advertising behavior tree can help the system determine the best time and form for advertising display, thereby further improving the click-through rate and conversion effect of the advertisement. The addition of the advertising behavior tree makes advertising push not only accurate, but also personalized to adapt to the user's behavior path, bringing deeper user interaction.

[0070] The depth-first search algorithm can quickly extract key behavior sequences related to target ads from the ad behavior tree, helping the system identify user groups that show a high interest in specific ads. The introduction of the depth-first search algorithm makes the system more efficient in processing large-scale data and can quickly respond to changes in user behavior. For example, during the shopping festival, the system quickly screened out users who were interested in promotional ads through the depth-first search algorithm and gave priority to pushing these ads, further improving the conversion effect of ads. The depth-first search algorithm makes ad push more real-time and targeted. The embedding matrix captures the complex relationship between behavioral features by converting user behavioral features into high-dimensional vectors. This processing method enables the system to analyze user behavior patterns more comprehensively and generate more accurate user interest scores. Combined with the depth-first search algorithm, the embedding matrix can not only extract the key behaviors of users, but also convert these behavioral features into a form that is easier for the model to process through high-dimensional vectors, thereby optimizing the display of advertising content. For example, the system can calculate the user's potential interest in ads through the embedding matrix and further personalize the form of ad push, thereby improving the user's advertising experience and interactive effect. This combination further enhances the accuracy and effect optimization capabilities of ad push.

[0071] The user-ad interaction matrix records the frequency of interaction between users and ads, providing a quantitative data basis for ad push. Through SVD decomposition, the system can extract the potential feature relationship between users and ads and generate user interest scores for ads. This process enables the system to allocate advertising resources more accurately based on the previous steps to ensure that users who are most likely to generate conversions receive relevant ads. For example, the system prioritizes ads to highly relevant users based on interest scores, reducing invalid pushes and further improving the click-through rate and conversion effect of ads. SVD decomposition optimizes the system's resource allocation based on the interaction matrix to ensure the effectiveness and efficiency of ad push. Standardization ensures the balance of different behavioral characteristics in the model and avoids the excessive influence of certain characteristics on the results. The weight adjustment mechanism dynamically optimizes the ad push strategy based on the user's behavior frequency, enabling the system to adjust the ad display order in real time, optimize the ad content, and ensure the efficient use of ad resources. For example, when the system detects that an ad performs well in a specific area, it automatically adjusts the ad push priority to display it to more users who may be interested. Such a mechanism ensures flexibility and optimization effects in the ad push process and provides greater commercial returns.

[0072] Furthermore, dividing the advertising push area into high-sales area and low-sales area means collecting sales data of products corresponding to historical advertising pushes and historical advertising push data, performing data cleaning and standardization on the collected data, performing median calculation on the standardized advertising sales data, and dividing the advertising push area into high-sales area and low-sales area according to the median calculation result.

[0073] Dividing the advertising push area into high-sales area and low-sales area can solve the problem of unreasonable advertising resource allocation. By analyzing the corresponding product sales data after the historical advertising push, we can identify which areas performed well after the advertising push and define these areas as high-sales areas. This division method can ensure that advertising resources are concentrated in the efficient market and maximize the return on advertising investment. At the same time, for low-sales areas, the advertising effect can be optimized by adjusting the advertising content or strategy to avoid resource waste. The median calculation plays a key role in regional division. By calculating the median of the standardized advertising sales data, the influence of extreme values ​​on regional division can be avoided, making the division result more robust and reliable. This division method can not only identify the best performing areas in the market, but also help companies identify areas that need to be optimized and improved. By dividing high-sales areas and low-sales areas, more targeted advertising strategies can be formulated. In high-sales areas, companies can further strengthen advertising and push more advertising content for related products to increase users' purchase conversion rate. In low-sales areas, companies can adopt more flexible strategies, such as launching discounts or promotions, to attract users' attention and improve advertising effects.

[0074] S2: Push advertisements to users in different regions based on user interest scores, and dynamically adjust advertisement content based on changes in advertisement performance in different regions;

[0075] Specifically, pushing advertisements to users in different regions according to user interest scores means setting an interest threshold Q for pushing advertisements and comparing it with the user interest score:

[0076] Q=q min +α×(q max -q min ),

[0077] In the formula, q min is the minimum value of interest score, q max is the maximum value of the interest score, α is the percentage coefficient, ranging from [0, 1], which is used to dynamically set the threshold according to business needs;

[0078]

[0079] In the formula, z is the target coverage, which refers to the proportion of users that the advertiser hopes to cover. For example, if the goal is to cover the top 20% of high-interest users, the target coverage is 0.2, and a is the actual coverage, which refers to the proportion of users actually covered under the current push strategy. It can be measured by comparing actual advertising performance data (such as click-through rate and conversion rate);

[0080] If the user's interest score is greater than or equal to the interest threshold Q, the user is classified as a high-interest user; if the user's interest score is less than the interest threshold Q, the user is classified as a low-interest user;

[0081] If the user is a high-interest user and is located in a high-sales area, relevant advertising content will be pushed directly to the user. If the user is in a low-sales area, the advertising content will be optimized and promotional activities and coupons will be added;

[0082] If the user is a low-interest user and is located in a high-sales area, priority will be given to pushing advertising content with broad appeal. If the user is located in a low-sales area, the frequency of ad push will be controlled and a new advertising combination will be constructed.

[0083] By setting the interest threshold Q, users are divided into high-interest users and low-interest users. This process enables the advertising push strategy to more accurately match user needs. For high-interest users, the system can quickly identify and push relevant advertising content to avoid wasting resources. For low-interest users, the system can optimize the advertising effect by adjusting the advertising content or reducing the push frequency. This classification strategy improves the accuracy and effectiveness of advertising and ensures that advertising resources are maximized.

[0084] When the user's interest score reaches or exceeds the interest threshold Q and is located in a high-sales area, relevant advertising content will be pushed directly to the user. This strategy can make full use of the user's purchase intention and the market demand of the region to ensure that advertising push can produce the maximum conversion effect. For high-interest users in low-sales areas, by optimizing advertising content and adding promotional activities and coupons to further stimulate the user's desire to buy, not only improve the attractiveness of the advertisement, but also effectively promote sales growth in low-sales areas. For low-interest users, the advertising push strategy needs to be more cautious. When low-interest users are located in high-sales areas, priority is given to pushing advertising content with broad appeal, such as popular products or popular goods, to increase user interest and promote sales. For low-interest users in low-sales areas, the system will control the frequency of advertising pushes to avoid frequent interruptions to users, and at the same time build new advertising combinations to test and optimize advertising effects. This strategy can effectively reduce the waste of advertising resources and improve the overall effect of advertising by continuously optimizing advertising content. In low-sales areas, by optimizing advertising content and adding promotional activities and coupons, users' desire to buy can be stimulated and the conversion rate of advertising can be improved. The introduction of promotional activities and coupons can, to a certain extent, make up for the lack of user interest in low-sales areas, thereby driving sales growth. In addition, advertising content optimization can be dynamically adjusted according to user interests and behavioral characteristics, making the advertisement more in line with user needs, thereby increasing user acceptance of advertisements.

[0085] Furthermore, dynamically adjusting the advertising content according to changes in advertising performance in different regions refers to collecting real-time behavioral data on users' clicks, browsing and conversions on advertisements, analyzing the performance of push ads in high-sales areas and low-sales areas, and if users respond poorly to push ads, redesigning the advertising creatives and adjusting the display format of the ads. After optimizing the push ads, the ads are pushed again to the target users, and the user's interest score is updated based on the real-time collected user behavioral data on push ads.

[0086] By collecting user behavior data such as clicks, browsing, and conversions in real time, the system can continuously monitor the performance of advertisements in different regions. This ensures that ad push can reflect the latest needs and behavior dynamics of users, avoiding the problem that traditional ad push lags behind market changes. The collection of real-time data provides a basis for subsequent ad optimization, making ad push more intelligent and personalized. Analyzing the advertising effects in high-sales areas and low-sales areas can help identify the performance differences of ads in different market environments and provide data support for the adjustment of advertising strategies. For example, high-sales areas may indicate that the advertising content is highly matched with user needs, while low-sales areas may indicate that the advertising content needs to be optimized. When users respond poorly to an ad, the attractiveness and conversion rate of the ad can be improved by redesigning the ad material (such as copy, pictures, videos) and adjusting the display format (such as ad location, display time, ad format). This dynamic adjustment process can help the advertising system better adapt to changes in user needs and improve the overall performance of the ad through optimized ad content. After the optimization of ad materials and display formats is completed, the ad is re-pushed to the target user. This process ensures that users see the latest and most relevant ad content, further improving the click-through rate and conversion rate of the ad. Based on the user behavior data collected in real time, the system continuously updates the user's interest score to ensure that the advertising push strategy can continue to reflect the user's interest changes. The dynamic adjustment of the interest score enables the advertising system to more accurately target the target user and push advertising content that meets their needs, thereby improving user engagement and advertising conversion effects.

[0087] S3, stores the data generated during the advertising push process and displays it visually;

[0088] Specifically, storing data generated during the advertising push process means storing user behavior data, advertising content data, advertising push results, and advertising optimization data in a database, partitioning the data according to the advertising push area, and regularly backing up and access controlling the stored data.

[0089] Storing user behavior data in the database can provide the advertising push system with continuous and analyzable historical behavior data, laying a data foundation for personalized advertising push. Through the accumulation and analysis of long-term behavior data, the user's interest changes can be predicted more accurately, thereby improving the effect of advertising. The storage of advertising content data can not only help advertisers record and manage the specific forms of advertising, but also provide a reference for subsequent advertising optimization. By analyzing the performance data of different advertising content, it is possible to identify which content is more attractive and apply it in future advertising, thereby improving the click-through rate and conversion rate of advertising. By storing advertising push results, the system can conduct a comprehensive analysis of the effect of advertising. It can timely understand the effect of advertising, find out the reasons for poor advertising performance, and make targeted optimization adjustments. Through continuous result feedback and adjustment, the accuracy and efficiency of advertising push will continue to improve. Partition storage can classify and manage data according to the advertising push area, which can not only improve the efficiency of data query, but also reduce conflicts during data access, and ensure the stable operation of the system under high concurrency.

[0090] Furthermore, visual display refers to using the Tableau visualization tool to design an interactive visualization interface, using a line chart to display the changing trend of the advertisement click-through rate and conversion rate over time, and using a bar chart to display the click status of different advertisements in various regions, and only authorized users are allowed to view the data.

[0091] Tableau's interactive interface design enables flexible exploration and analysis of data. Users can conduct in-depth mining and interpretation of advertising data through operations such as clicking and filtering. This interactive experience not only improves the efficiency of data analysis, but also enhances the user's sense of control over the data, making the effect of advertising more transparent and controllable. The line chart can dynamically reflect the time series data of advertising click-through rate and conversion rate, helping users to identify fluctuations in advertising performance within a specific time period. This is of great significance for the adjustment of advertising strategies. Users can adjust the timing of advertising delivery according to time changes to maximize advertising effects. The bar chart can clearly show the click status of different advertisements in various regions, helping users to intuitively compare the performance of advertisements in different markets. Through this comparative analysis, efficient and inefficient advertising areas can be identified and optimized in a targeted manner. By setting an authorized user mechanism, the system can effectively control data access rights, protect data security, and prevent data leakage or misuse.

[0092] Example 2, reference Figure 2 , which is the second embodiment of the present invention, and which is different from the previous embodiment, provides an online advertising intelligent push system based on big data, which includes:

[0093] A data collection module, used to collect real-time and historical user behavior data and advertising content data and pre-process the data;

[0094] The interest score calculation module is used to build a search-based user interest model and advertising behavior tree, capture the interaction between users and advertisements from the tree, generate user interest embedding vectors and build a user-ad interaction matrix, and use the SVD algorithm to calculate the user's interest score for advertisements;

[0095] Advertisement push module, used to divide the advertisement push area, push advertisements to users according to user interest scores and regional divisions, and optimize and adjust advertisement content according to real-time advertisement feedback data;

[0096] The data storage module is used to store the data generated during the advertisement push process and to visualize the data.

[0097] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0098] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0099] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0100] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

Claims

1. A method for intelligent online advertising push based on big data, characterized by: include, Collect user behavior data to build a user interest model to generate user interest scores, and divide advertising push areas into high-sales areas and low-sales areas; Push ads to users in different regions based on user interest scores, and dynamically adjust ad content based on changes in ad performance in different regions; Store the data generated during the advertising push process and display it visually; The collecting of user behavior data to construct a user interest model and generate a user interest score refers to collecting historical user behavior data, real-time user behavior data and advertising content data related to advertising, cleaning the collected data, filling in missing values, and then standardizing the data, and mapping each user behavior with a corresponding advertising ID; Construct a search-based user interest model, including general search units and precise search units; Using a general search unit to extract sub-behavior sequences related to target ads from long-term user behavior data includes building an ad behavior tree, defining the root node as the initial state of the user before ad interaction, the middle node as the user's specific behavior event on the ad, each node storing the ad ID, behavior type, behavior time, and the leaf node as the termination state of the user's ad behavior; Create a root node for each user as the starting point of the advertising behavior path. Insert the user's advertising behavior as a new node into the advertising behavior tree according to the time sequence of the advertising behavior. Add indexes to nodes of high-frequency behavior paths and mark paths with excellent performance in the advertising behavior tree; According to the target advertisement content, the behavior nodes related to the target advertisement are screened out from the advertisement behavior tree, and the advertisement behavior tree is traversed using a depth-first search algorithm to extract the sub-behavior sequences related to the target advertisement; Use an embedding layer to map the behavior data in the sub-behavior sequence into a high-dimensional vector to form a behavior vector sequence; Perform linear transformation on the behavior vector to generate query matrix, key matrix and value matrix, set multiple attention heads, input query, key and value matrices into multiple attention heads, use query matrix and key matrix to perform dot product calculation in each attention head to obtain the correlation score between behaviors, normalize the correlation score through softmax function, apply the normalized weight to the value matrix to obtain the weighted behavior vector output, merge the weighted behavior vectors to generate the user's interest embedding vector; Normalize the generated interest embedding vector; Construct a user-ad interaction matrix, map the standardized user interest embedding vector to the user-ad interaction matrix, where rows represent users, columns represent ad content, and each value in the matrix represents the interaction frequency between users and ads; The SVD algorithm is used to perform singular value decomposition on the user-ad interaction matrix to obtain three sub-matrices, including the user feature matrix, the singular value matrix, and the ad feature matrix; To calculate the interest score of user u for advertisement p, the corresponding user features of user u and advertisement features of advertisement p are extracted from the user feature matrix and the advertisement feature matrix; The user features, singular value matrix and advertisement features are multiplied by matrix multiplication to obtain the interest score of user u for advertisement p.

2. The method for intelligent online advertising push based on big data as claimed in claim 1, characterized in that: Dividing the advertising push area into high-sales area and low-sales area refers to collecting sales data of the products corresponding to the historical advertising push and the historical advertising push data, cleaning and standardizing the collected data, performing median calculation on the standardized advertising sales data, and dividing the advertising push area into high-sales area and low-sales area according to the median calculation result.

3. The method for intelligent online advertising push based on big data as claimed in claim 2, characterized in that: Pushing advertisements to users in different regions according to user interest scores refers to setting an interest threshold Q for pushing advertisements, comparing it with the user interest score, and if the user interest score is greater than or equal to the interest threshold Q, the user is classified as a high-interest user; if the user interest score is less than the interest threshold Q, the user is classified as a low-interest user; If the user is a high-interest user and is located in a high-sales area, relevant advertising content will be pushed directly to the user. If the user is in a low-sales area, the advertising content will be optimized and promotional activities and coupons will be added; If the user is a low-interest user and is located in a high-sales area, priority will be given to pushing advertising content with broad appeal. If the user is located in a low-sales area, the frequency of ad push will be controlled and a new advertising combination will be constructed.

4. The method for intelligent online advertising push based on big data as claimed in claim 3, characterized in that: Dynamically adjusting the advertising content according to changes in advertising performance in different regions refers to collecting real-time behavioral data on users' clicks, views, and conversions of advertisements, analyzing the performance of push ads in high-sales areas and low-sales areas, and if users respond poorly to push ads, redesigning the advertising creatives and adjusting the display format of the ads. After optimizing the push ads, the ads are pushed again to the target users, and the user's interest score is updated based on the real-time collected user behavioral data on the push ads.

5. The method for intelligent online advertising push based on big data as claimed in claim 4, characterized in that: The storing of data generated during the advertisement push process refers to storing user behavior data, advertisement content data, advertisement push results, and advertisement optimization data in a database, partitioning the data according to the advertisement push area, and regularly backing up and access controlling the stored data.

6. The method for intelligent online advertising push based on big data as claimed in claim 5, characterized in that: The visual display mentioned above refers to using the Tableau visualization tool to design an interactive visualization interface, using a line chart to display the changing trend of the advertisement click-through rate and conversion rate over time, and using a bar chart to display the click status of different advertisements in various regions. Only authorized users are allowed to view the data.

7. An online advertising intelligent push system based on big data based on the online advertising intelligent push method based on big data according to any one of claims 1 to 6, characterized in that: include, A data collection module, used to collect real-time and historical user behavior data and advertising content data and pre-process the data; The interest score calculation module is used to build a search-based user interest model and advertising behavior tree, capture the interaction between users and advertisements from the tree, generate user interest embedding vectors and build a user-ad interaction matrix, and use the SVD algorithm to calculate the user's interest score for advertisements; Advertisement push module, used to divide the advertisement push area, push advertisements to users according to user interest scores and regional divisions, and optimize and adjust advertisement content according to real-time advertisement feedback data; The data storage module is used to store the data generated during the advertisement push process and to visualize the data.

8. A computer device comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the online advertising intelligent push method based on big data described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the online advertising intelligent push method based on big data described in any one of claims 1 to 6 are implemented.

Images