User portrait dynamic generation and advertisement targeted delivery system based on multi-source data fusion

Abstract

The application discloses a user portrait dynamic generation and advertisement targeted delivery system based on multi-source data fusion, relates to the technical field of advertisement targeted delivery, and comprises a multi-source data acquisition module, a general advertisement targeted delivery analysis module, a user image feature extraction module, a user portrait dynamic generation module, a small category advertisement targeted delivery analysis module and an advertisement targeted delivery module. The user portrait dynamic generation module utilizes user behavior data to construct a user dynamic portrait model. The advertisement targeted delivery module is used for calibrating advertisement targeted delivery results, constructing an advertisement targeted delivery model, and further delivering corresponding advertisements to users. The multi-source data acquisition technology, feature extraction technology, multi-source data fusion technology, user image dynamic generation technology and model construction technology in the system are closely combined with modern information technology, so that the precision and intelligent degree of advertisement targeted delivery are improved.

Description

Technical Field

[0001] This invention relates to the field of targeted advertising technology, specifically to a system for dynamically generating user profiles and targeted advertising based on multi-source data fusion. Background Technology

[0002] In the era of digital marketing, the accuracy and effectiveness of advertising have become key to enhancing corporate competitiveness. Traditional advertising models rely on single data sources or simple user attribute segmentation, making it difficult to comprehensively and deeply understand user needs and behavioral characteristics. This results in low advertising accuracy, insufficient conversion rates, and wasted resources. With the rapid development of internet technology, users generate massive amounts of behavioral data across multiple platforms such as e-commerce, social media, news, and games. This data contains rich user information, but due to the dispersed data sources and diverse formats, it forms data silos that cannot be used to build accurate user profiles. At the same time, user interests and behaviors are constantly changing, and traditional static user profiles cannot reflect the latest user characteristics in a timely manner, making it difficult to meet the real-time and personalized advertising needs. Therefore, there is an urgent need for a system that can integrate multi-source data, dynamically generate accurate user profiles, and achieve efficient targeted advertising based on these profiles. This system aims to solve the problems of poor advertising accuracy and insufficient timeliness of user profiles in existing technologies, thereby improving advertising effectiveness and user experience.

[0003] Although existing technologies have made significant progress in advertising delivery, some issues still need to be optimized. Current advertising delivery technologies rely heavily on historical user preference data, which cannot accurately grasp users' real-time preferences and thus dynamically adjust the type of ads delivered. This leads to a mismatch between ads and user preferences, affecting the effectiveness of advertising. Therefore, how to integrate multi-source data and user image feature data to build a user profile model and achieve targeted advertising is the problem we need to solve. Thus, we propose a user profile dynamic generation and targeted advertising delivery system based on multi-source data fusion. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solution: a user profile dynamic generation and advertising targeted delivery system based on multi-source data fusion, comprising a multi-source data acquisition module, a major category advertising targeted delivery analysis module, a user image feature extraction module, a user profile dynamic generation module, a minor category advertising targeted delivery analysis module, and an advertising targeted delivery module, wherein the various modules are interconnected.

[0005] The multi-source data acquisition module collects basic user data, user browsing data, user behavior data, and user image data, providing data preparation for the implementation of subsequent module functions;

[0006] The category-based advertising targeting analysis module calculates the category-based advertising targeting index based on user basic data and user browsing data, and obtains the category-based advertising targeting results. This provides a general direction for subsequent analysis of the sub-category advertising targeting results and reduces the difficulty of subsequent analysis.

[0007] The user image feature extraction module extracts features from user image data to obtain user image feature data, providing a data foundation for calibrating the user dynamic profile model and improving the accuracy of the user dynamic profile model.

[0008] The user profile dynamic generation module uses user behavior data to construct a user dynamic profile model, and combines user image feature data to calibrate the user dynamic profile model, thereby improving the accuracy of the analysis results of targeted advertising in the sub-category.

[0009] The category ad targeting analysis module analyzes the category ad targeting results through a calibrated user dynamic profile model.

[0010] The advertising targeting module is used to calibrate advertising targeting results, construct advertising targeting models, and then target corresponding advertisements to users. Through the advertising targeting model, the system can input basic user data and output corresponding sub-category advertising targeting results, thereby improving the intelligence level in the dynamic generation of user profiles and advertising targeting processes based on multi-source data fusion.

[0011] A further improvement to the technical solution of the present invention is that the advertising targeted delivery module includes a calibration unit, a model building unit, and an execution unit, wherein the functions of each unit are as follows:

[0012] The calibration unit uses a convolutional neural network algorithm to evaluate the calibration coefficient of the targeted placement of subcategories of advertisements by using user browsing data, user behavior data, and user image feature data, thereby calibrating the targeted placement results of subcategories of advertisements.

[0013] The model building unit uses the random forest algorithm to build an advertising targeted delivery model;

[0014] The execution unit delivers relevant advertisements to users based on the output of the advertising targeting model.

[0015] A further improvement to the technical solution of this invention lies in that: the multi-source data acquisition module's acquisition process for user basic data, user browsing data, user behavior data, and user image data includes:

[0016] Data entry technology is used to collect basic user data, user browsing data, user behavior data, and user image data;

[0017] The user basic data includes the user's gender, age, and occupation, where occupation includes students, employed, and retired; the user browsing data includes the number of searches, page views, and dwell time on different types of pages within 24 hours, where different types of pages include e-commerce platform product pages, social media pages, news report pages, game experience pages, and lifestyle hobby pages;

[0018] The user behavior data includes e-commerce behavior data, social behavior data, information behavior data, gaming behavior data, and lifestyle and hobby data; real-time images of users browsing different types of advertisements;

[0019] The e-commerce behavior data includes the number of times users search for different types of products, the number of times they purchase, the number of positive reviews and negative reviews, as well as users' brand preferences and promotional preferences. The different types of products include fashion apparel, digital home appliances, beauty and skincare, outdoor sports, home furnishings, maternal and infant care, audio and video products, and food tasting. The brand preferences include high-end brands and mass-market brands. The promotional preferences include not paying attention to any promotional methods, discount promotions, and full-reduction promotions.

[0020] The social behavior data includes the user's social platform usage time, the number of groups the user has joined and the number of interactions, the number of friends the user has and the number of interactions, and the duration of the user's live stream participation.

[0021] The information behavior data includes news viewing types, knowledge and science browsing types, lifestyle information browsing types, and industry news click counts. Among them, news viewing types include domestic and foreign news; knowledge and science browsing types include natural science, health and wellness, culture and art, legal knowledge, and financial knowledge; lifestyle information browsing types include real estate, recruitment, automobiles, food, travel, fashion, home furnishings, and pets; and industry news includes technology industry news, financial industry news, automotive industry news, fashion industry news, catering industry news, and medical industry news.

[0022] The game behavior data includes the types of games the user participates in and their duration. The game types include role-playing, action, strategy, simulation, casual, and competitive games. The user's lifestyle data includes the user's hobbies, such as sports and fitness, culture and art, food, handicrafts, pets, and travel.

[0023] Specifically, data entry technology, combined with questionnaires, is used to obtain users' gender, age and occupation, user browsing data, user behavior data and user image data;

[0024] The collected data is cleaned and normalized, and user basic data, user browsing data and user behavior data are integrated to generate a dynamic analysis dataset of user profiles. The dynamic analysis dataset of user profiles is divided into a training set and a test set with a ratio of 8:2.

[0025] A further improvement to the technical solution of this invention lies in the following: the process by which the category-based advertising targeting analysis module calculates the category-based advertising targeting index and obtains the category-based advertising targeting results includes:

[0026] The major advertising categories include e-commerce ads, social media ads, news ads, gaming ads, and lifestyle ads; the major advertising category targeting indices include e-commerce ad targeting indices, social media ad targeting indices, news ad targeting indices, gaming ad targeting indices, and lifestyle ad targeting indices.

[0027] Based on user base data, users are initially assigned major ad categories. Then, combined with user browsing data, a targeting index for each major ad category is calculated. The calculation process includes:

[0028]

[0029]

[0030]

[0031]

[0032]

[0033] in, , , , and These are the targeting indices for e-commerce ads, social media ads, news ads, gaming ads, and lifestyle / hobbies ads. , , , and The number of searches a user makes within 24 hours on product pages, social media pages, news report pages, game experience pages, and lifestyle and hobby pages on e-commerce platforms. , , , and This refers to the number of times users browse product pages, social media pages, news report pages, game experience pages, and lifestyle and hobby pages on e-commerce platforms within 24 hours. , , , and The time users spend on product pages, social media pages, news report pages, game experience pages, and lifestyle and hobby pages on e-commerce platforms within 24 hours; , and These represent the number of searches, page views, and dwell time of users on different types of pages within 24 hours.

[0034] Compare the size of the targeting indices for each major category of ads to obtain the targeting results for each major category. Specifically, analyze the targeting indices for e-commerce ads, social media ads, news ads, games ads, and lifestyle / hobbies ads, and select the corresponding major category of ads as the targeting results for each major category.

[0035] A further improvement to the technical solution of the present invention is that the user image feature extraction module, the process of acquiring user image feature data includes:

[0036] The user image feature data includes the user's person feature data, object feature data, and scene feature data;

[0037] Feature extraction is performed on real-time images of users browsing different types of advertisements. Using functions and algorithms in the Matlab computer vision toolbox, combined with a Haar cascade detector, facial features of users are extracted. An edge detection algorithm is used to extract the posture features of users. The human feature data includes facial features and posture features of users.

[0038] Using the Faster R-CNN algorithm based on deep learning, object feature data is extracted; using Matlab software, semantic segmentation is performed on user image data to distinguish the background region from the user, thereby extracting scene feature data, and integrating the obtained user image feature data into the user profile dynamic analysis dataset.

[0039] A further improvement to the technical solution of this invention lies in the following: the user profile dynamic generation module constructs a user dynamic profile model, and the process of calibrating the user dynamic profile model by combining user image feature data includes:

[0040] Analyze user behavior data to extract e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, and lifestyle and hobby characteristics;

[0041] The extracted features are categorized based on dimensions, including consumption, social, information, entertainment, and lifestyle dimensions.

[0042] User behavior data is extracted from the dynamic analysis dataset of user profiles. A clustering analysis algorithm is used to assign users to different dimensions. The consumption, social, information, entertainment, and lifestyle dimensions are respectively mapped to e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, and lifestyle hobby characteristics. The training set data is used as input to obtain the statistics of users in each dimension. The statistics of each dimension are the specific values ​​corresponding to various types of user behavior data. Then, users are divided into different clusters, and the corresponding dynamic profiles of users are output, thus realizing the construction of the dynamic user profile model.

[0043] Input the test set data into the user dynamic profile model, adjust the parameters of the user dynamic profile model, optimize the performance of the user dynamic profile model, and obtain the final user dynamic profile model.

[0044] By using clustering analysis algorithms, user profile data is integrated with lifestyle and entertainment dimensions, user object data is integrated with consumption dimensions, and user scenario data is integrated with social and information dimensions, thereby calibrating the dynamic user profile model.

[0045] A further improvement to the technical solution of this invention lies in the following: the analysis process of the targeted delivery results for the sub-category ads in the sub-category ad delivery module includes:

[0046] The subcategories of ads are obtained by further subdividing the major categories of ads. Specifically, e-commerce ads are subdivided into product category ads, brand ads, and promotional ads; social media ads are subdivided into social platform promotion ads, interest group ads, friend recommendation ads, and live streaming ads; news ads are subdivided into news ads, knowledge popularization ads, lifestyle information ads, and industry news ads; gaming ads are subdivided into role-playing ads, action ads, strategy ads, simulation management ads, leisure ads, and competitive ads; and lifestyle and hobby ads are subdivided into sports and fitness ads, culture and art ads, food ads, handicraft ads, pet ads, and travel ads.

[0047] Combining current user behavior data, the user dynamic profile model outputs corresponding user dynamic profiles. Based on the results of major category advertising targeting, it analyzes the e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, lifestyle hobby characteristics, and user image characteristics corresponding to different dimensions in the user's dynamic profile, and then obtains the results of minor category advertising targeting.

[0048] A further improvement to the technical solution of the present invention is that: the calibration unit evaluates the calibration coefficient for targeted placement of sub-categories of advertisements, and then calibrates the targeted placement results of sub-categories of advertisements, including the following process:

[0049] Extract user browsing data, user behavior data, and user image feature data from the user profile dynamic analysis dataset. Combine this with a convolutional neural network, take the training set data as input, and the sub-category ad targeting calibration coefficient as output. Learn the non-linear relationship between user browsing data, user behavior data, and user image feature data and the sub-category ad targeting calibration coefficient, and train the sub-category ad targeting calibration model.

[0050] The test set data is input into the category ad targeting calibration model. The parameters of the category ad targeting calibration model are adjusted to optimize its performance. The category ad targeting calibration model is then deployed into the system. Combining current user browsing data, user behavior data, and user image feature data, the corresponding category ad targeting calibration coefficients are output. Based on these calibration coefficients, the category ad targeting results are adjusted. The calibrated category ad targeting results are then encoded and integrated into the user profile dynamic analysis dataset.

[0051] Specifically, the calibration coefficient for this category of advertising targeting is an integer between 1 and 19. When the calibration coefficient for the category of advertising targeting is between 1 and 19, it corresponds to product category advertising, brand advertising, promotional advertising, social platform promotion advertising, interest group advertising, friend recommendation advertising, live streaming advertising, news and information advertising, knowledge popularization advertising, lifestyle information advertising, industry news advertising, role-playing advertising, action advertising, strategy advertising, simulation management advertising, leisure advertising, competitive advertising, sports and fitness advertising, culture and art advertising, food advertising, handicraft advertising, pet advertising, and travel advertising. According to the corresponding results, the original category advertising targeting results are adjusted to achieve calibration of the category advertising targeting results.

[0052] A further improvement to the technical solution of this invention lies in that: the model building unit, the process of building the advertising targeted delivery model includes:

[0053] Extract basic user data and calibrated sub-category ad targeting result codes from the user profile dynamic analysis dataset. Combine the training set data with the random forest algorithm, using the training set data as input and the calibrated sub-category ad targeting result codes as output, to learn the non-linear relationship between basic user data and calibrated sub-category ad targeting result codes, and train the ad targeting model.

[0054] The test set data is input into the ad targeting model. The output results of the ad targeting model are compared with the actual calibrated sub-category ad targeting results to evaluate the performance of the ad targeting model. The parameters of the ad targeting model are adjusted to optimize the ad targeting model and obtain the final ad targeting model.

[0055] A further improvement to the technical solution of the present invention is that: the execution unit, based on the output results of the advertising targeting model, performs the following process for targeting users with corresponding advertisements:

[0056] Input basic user data into the ad targeting model, obtain the calibrated sub-category ad targeting result code, and then match the corresponding calibrated sub-category ad targeting result;

[0057] Based on the calibrated sub-category advertising targeting results, relevant ads are targeted to users.

[0058] The beneficial effects of this invention are as follows: Compared to traditional multi-source data fusion-based user profile dynamic generation and advertising targeting systems, the multi-source data acquisition technology, feature extraction technology, multi-source data fusion technology, user image dynamic generation technology, and model building technology in this invention are closely integrated with modern information technology. This allows for the accurate capture of basic user data, user browsing data, user behavior data, and user image data, achieving real-time and comprehensive evaluation of user behavior and interests. This, in turn, yields broad-category and sub-category advertising targeting results. Furthermore, by constructing and calibrating a dynamic user profile model, it provides a more suitable... This invention provides technical support to improve the accuracy of targeted advertising due to its dynamic nature. It addresses the problem that existing advertising technologies rely heavily on historical user preference data, failing to accurately grasp real-time user preferences and thus hinder dynamic adjustments to ad types. This leads to mismatches between ads and user preferences, negatively impacting advertising effectiveness. The method in this invention ensures that the dynamic monitoring standards for user profile generation and targeted advertising systems based on multi-source data fusion can be refined to a more precise range, making the monitored data more accurate indicators under the same conditions. The development and application of this method significantly enhances the intelligence level of the user profile generation and targeted advertising system based on multi-source data fusion. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0060] Figure 1 This is a block diagram of the user profile dynamic generation and targeted advertising system based on multi-source data fusion of the present invention;

[0061] Figure 2 A flowchart illustrating the components of different ad targeting types;

[0062] Figure 3 A block diagram of the components of a dynamic user profile model. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] like Figures 1 to 3 As shown, the present invention provides a user profile dynamic generation and advertising targeted delivery system based on multi-source data fusion, including a multi-source data acquisition module, a major category advertising targeted delivery analysis module, a user image feature extraction module, a user profile dynamic generation module, a minor category advertising targeted delivery analysis module, and an advertising targeted delivery module, wherein the modules are interconnected.

[0065] The multi-source data acquisition module collects basic user data, user browsing data, user behavior data, and user image data, providing data preparation for the implementation of subsequent module functions;

[0066] The category-based advertising targeting analysis module calculates the category-based advertising targeting index based on user basic data and user browsing data, and obtains the category-based advertising targeting results. This provides a general direction for subsequent analysis of sub-category advertising targeting results and reduces the difficulty of subsequent analysis.

[0067] The user image feature extraction module extracts features from user image data to obtain user image feature data, providing a data foundation for calibrating the user dynamic profile model and improving the accuracy of the user dynamic profile model.

[0068] The user profile dynamic generation module uses user behavior data to build a dynamic user profile model. It combines user image feature data to calibrate the dynamic user profile model, thereby improving the accuracy of the analysis results of targeted advertising in small categories.

[0069] The category ad targeting analysis module analyzes the results of category ad targeting through a calibrated user dynamic profile model.

[0070] The ad targeting module is used to calibrate ad targeting results, build ad targeting models, and then target corresponding ads to users. Through the ad targeting model, the module can input basic user data and output corresponding ad targeting results for specific categories, thereby improving the intelligence level of the dynamic generation of user profiles and ad targeting process based on multi-source data fusion.

[0071] The ad targeting module includes a calibration unit, a model building unit, and an execution unit. The functions of each unit are as follows:

[0072] The calibration unit uses convolutional neural network algorithms to evaluate the calibration coefficient of sub-category advertising targeting by using user browsing data, user behavior data, and user image feature data, thereby calibrating the sub-category advertising targeting results;

[0073] The model building unit uses the random forest algorithm to build an ad targeting model;

[0074] The execution unit delivers relevant ads to users based on the output of the ad targeting model.

[0075] The multi-source data acquisition module collects user basic data, user browsing data, user behavior data, and user image data. The process includes:

[0076] Data entry technology is used to collect basic user data, user browsing data, user behavior data, and user image data;

[0077] User base data includes users' gender, age, and occupation, with occupation including students, employed, and retired; user browsing data includes the number of searches, page views, and dwell time on different types of pages within 24 hours, with different types of pages including e-commerce platform product pages, social media pages, news report pages, game experience pages, and lifestyle hobby pages;

[0078] User behavior data includes e-commerce behavior data, social behavior data, information behavior data, gaming behavior data, and lifestyle and hobby data; real-time images of users browsing different types of advertisements;

[0079] Among them, e-commerce behavior data includes the number of times users search for different types of products, the number of times they purchase, the number of positive reviews and the number of negative reviews, as well as users' brand preferences and promotional method preferences. The different types of products include fashion apparel, digital home appliances, beauty and skin care, outdoor sports, home furnishings, maternal and infant care, image and speaker products, and food tasting products. The brand preferences include high-end brands and mass-market brands. The promotional method preferences include not paying attention to any promotional methods, discount promotions, and full reduction promotions.

[0080] Social behavior data includes users' social platform usage time, number of groups joined and interaction times, number of friends and interaction times, and live stream participation time;

[0081] Information behavior data includes news viewing types, knowledge and science browsing types, lifestyle information browsing types, and industry news click counts. Among them, news viewing types include domestic and foreign news; knowledge and science browsing types include natural science, health and wellness, culture and art, legal knowledge, and financial knowledge; lifestyle information browsing types include real estate, recruitment, automobiles, food, travel, fashion, home furnishings, and pets; and industry news includes technology industry news, financial industry news, automotive industry news, fashion industry news, catering industry news, and medical industry news.

[0082] Game behavior data includes the types of games users participate in and the duration of those games. These game types include role-playing, action, strategy, simulation, casual, and competitive games. User lifestyle data includes the user's hobbies, such as sports and fitness, culture and art, food, handicrafts, pets, and travel.

[0083] Specifically, data entry technology, combined with questionnaires, is used to obtain users' gender, age and occupation, user browsing data, user behavior data and user image data;

[0084] The collected data is cleaned and normalized, and user basic data, user browsing data and user behavior data are integrated to generate a dynamic analysis dataset of user profiles. The dynamic analysis dataset of user profiles is divided into a training set and a test set with a ratio of 8:2.

[0085] The category-based advertising targeting analysis module calculates the category-based advertising targeting index and obtains the category-based advertising targeting results. The process includes:

[0086] The major advertising categories include e-commerce ads, social media ads, news ads, gaming ads, and lifestyle ads; the major advertising category targeting indices include e-commerce ad targeting indices, social media ad targeting indices, news ad targeting indices, gaming ad targeting indices, and lifestyle ad targeting indices.

[0087] Based on user base data, users are initially assigned major ad categories. Then, combined with user browsing data, a targeting index for each major ad category is calculated. The calculation process includes:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] in, , , , and These are the targeting indices for e-commerce ads, social media ads, news ads, gaming ads, and lifestyle / hobbies ads. , , , and The number of searches a user makes within 24 hours on product pages, social media pages, news report pages, game experience pages, and lifestyle and hobby pages on e-commerce platforms. , , , and This refers to the number of times users browse product pages, social media pages, news report pages, game experience pages, and lifestyle and hobby pages on e-commerce platforms within 24 hours. , , , and The time users spend on product pages, social media pages, news report pages, game experience pages, and lifestyle and hobby pages on e-commerce platforms within 24 hours; , and These represent the number of searches, page views, and dwell time of users on different types of pages within 24 hours.

[0094] Compare the size of the targeting indices for each major category of ads to obtain the targeting results for each major category. Specifically, analyze the targeting indices for e-commerce ads, social media ads, news ads, games ads, and lifestyle / hobbies ads, and select the corresponding major category of ads as the targeting results for each major category.

[0095] The user image feature extraction module, the process of acquiring user image feature data includes:

[0096] User image feature data includes user character feature data, object feature data, and scene feature data;

[0097] Feature extraction is performed on real-time images of users browsing different types of advertisements. Using functions and algorithms in the Matlab computer vision toolbox, combined with a Haar cascade detector, facial features of users are extracted. An edge detection algorithm is used to extract the posture features of users. The human feature data includes facial features and posture features of users.

[0098] Using the Faster R-CNN algorithm based on deep learning, object feature data is extracted; using Matlab software, semantic segmentation is performed on user image data to distinguish the background region from the user, thereby extracting scene feature data, and integrating the obtained user image feature data into the user profile dynamic analysis dataset.

[0099] The user profile dynamic generation module constructs a user dynamic profile model and calibrates the model by combining user image feature data. The process includes:

[0100] Analyze user behavior data to extract e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, and lifestyle and hobby characteristics;

[0101] The extracted features are categorized based on dimensions, including consumption, social, information, entertainment, and lifestyle dimensions.

[0102] User behavior data is extracted from the dynamic analysis dataset of user profiles. A clustering analysis algorithm is used to assign users to different dimensions. The consumption, social, information, entertainment, and lifestyle dimensions are respectively mapped to e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, and lifestyle hobby characteristics. The training set data is used as input to obtain the statistics of users in each dimension. The statistics of each dimension are the specific values ​​corresponding to various types of user behavior data. Then, users are divided into different clusters, and the corresponding dynamic profiles of users are output, thus realizing the construction of the dynamic user profile model.

[0103] Input the test set data into the user dynamic profile model, adjust the parameters of the user dynamic profile model, optimize the performance of the user dynamic profile model, and obtain the final user dynamic profile model.

[0104] By using clustering analysis algorithms, user profile data is integrated with lifestyle and entertainment dimensions, user object data is integrated with consumption dimensions, and user scenario data is integrated with social and information dimensions, thereby calibrating the dynamic user profile model.

[0105] The sub-category advertising targeting module analyzes the results of sub-category advertising targeting, including:

[0106] Subcategories of ads are acquired by further subdividing the major ad categories. Specifically, e-commerce ads are subdivided into product category ads, brand ads, and promotional ads; social media ads are subdivided into social platform promotion ads, interest group ads, friend recommendation ads, and live streaming ads; news ads are subdivided into news ads, knowledge popularization ads, lifestyle information ads, and industry news ads; gaming ads are subdivided into role-playing ads, action ads, strategy ads, simulation management ads, casual ads, and competitive ads; and lifestyle and hobby ads are subdivided into sports and fitness ads, culture and art ads, food ads, handicraft ads, pet ads, and travel ads.

[0107] Combining current user behavior data, the user dynamic profile model outputs corresponding user dynamic profiles. Based on the results of major category advertising targeting, it analyzes the e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, lifestyle hobby characteristics, and user image characteristics corresponding to different dimensions in the user's dynamic profile, and then obtains the results of minor category advertising targeting.

[0108] The calibration unit evaluates the calibration coefficient for sub-category ad targeting, and then calibrates the sub-category ad targeting results. This process includes:

[0109] Extract user browsing data, user behavior data, and user image feature data from the user profile dynamic analysis dataset. Combine this with a convolutional neural network, take the training set data as input, and the sub-category ad targeting calibration coefficient as output. Learn the non-linear relationship between user browsing data, user behavior data, and user image feature data and the sub-category ad targeting calibration coefficient, and train the sub-category ad targeting calibration model.

[0110] The test set data is input into the category ad targeting calibration model. The parameters of the category ad targeting calibration model are adjusted to optimize its performance. The category ad targeting calibration model is then deployed into the system. Combining current user browsing data, user behavior data, and user image feature data, the corresponding category ad targeting calibration coefficients are output. Based on these calibration coefficients, the category ad targeting results are adjusted. The calibrated category ad targeting results are then encoded and integrated into the user profile dynamic analysis dataset.

[0111] Specifically, the calibration coefficient for this category of advertising targeting is an integer between 1 and 19. When the calibration coefficient for the category of advertising targeting is between 1 and 19, it corresponds to product category advertising, brand advertising, promotional advertising, social platform promotion advertising, interest group advertising, friend recommendation advertising, live streaming advertising, news and information advertising, knowledge popularization advertising, lifestyle information advertising, industry news advertising, role-playing advertising, action advertising, strategy advertising, simulation management advertising, leisure advertising, competitive advertising, sports and fitness advertising, culture and art advertising, food advertising, handicraft advertising, pet advertising, and travel advertising. According to the corresponding results, the original category advertising targeting results are adjusted to achieve calibration of the category advertising targeting results.

[0112] The model building unit, the process of building an ad targeting model, includes:

[0113] Extract basic user data and calibrated sub-category ad targeting result codes from the user profile dynamic analysis dataset. Combine the training set data with the random forest algorithm, using the training set data as input and the calibrated sub-category ad targeting result codes as output, to learn the non-linear relationship between basic user data and calibrated sub-category ad targeting result codes, and train the ad targeting model.

[0114] The test set data is input into the ad targeting model. The output results of the ad targeting model are compared with the actual calibrated sub-category ad targeting results to evaluate the performance of the ad targeting model. The parameters of the ad targeting model are adjusted to optimize the ad targeting model and obtain the final ad targeting model.

[0115] The execution unit, based on the output of the advertising targeting model, performs the following process to deliver relevant advertisements to users:

[0116] Input basic user data into the ad targeting model, obtain the calibrated sub-category ad targeting result code, and then match the corresponding calibrated sub-category ad targeting result;

[0117] Based on the calibrated sub-category advertising targeting results, relevant ads are targeted to users.

[0118] First, data entry technology is used to collect basic user data, user browsing data, user behavior data, and user image data. Second, using the basic user data and browsing data, a category-based advertising targeting index is calculated to obtain the category-based advertising targeting results. Then, feature extraction is performed on the user image data to obtain user image feature data. Using user behavior data, a dynamic user profile model is constructed, and this model is calibrated based on the user image feature data. Next, the calibrated dynamic user profile model is used to analyze the sub-category advertising targeting results. Then, a convolutional neural network algorithm is employed, combining user browsing data, user behavior data, and user image feature data, to evaluate the sub-category advertising targeting calibration coefficient, thereby calibrating the sub-category advertising targeting results. A random forest algorithm is used to construct an advertising targeting model, improving the intelligence of dynamic user profile generation and advertising targeting based on multi-source data fusion. Finally, based on the output of the advertising targeting model, corresponding advertisements are targeted to users.

[0119] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A user profile dynamic generation and ad targeting system based on multi-source data fusion, comprising a multi-source data acquisition module, a category-based ad targeting analysis module, a user image feature extraction module, a user profile dynamic generation module, a sub-category ad targeting analysis module, and an ad targeting module, wherein... The various modules are connected for communication, characterized by: The multi-source data acquisition module collects user basic data, user browsing data, user behavior data, and user image data; The category-based advertising targeting analysis module calculates the category-based advertising targeting index and obtains the category-based advertising targeting results through user basic data and user browsing data; The user image feature extraction module extracts features from user image data to obtain user image feature data; The user profile dynamic generation module uses user behavior data to construct a user dynamic profile model and combines it with user image feature data to calibrate the user dynamic profile model. The category ad targeting analysis module analyzes the category ad targeting results through a calibrated user dynamic profile model. The ad targeting module is used to calibrate ad targeting results, build an ad targeting model, and then target the corresponding ads to users. The category-based advertising targeting analysis module calculates the category-based advertising targeting index and obtains the category-based advertising targeting results, including the following process: The major advertising categories include e-commerce ads, social media ads, news ads, gaming ads, and lifestyle ads; the major advertising category targeting indices include e-commerce ad targeting indices, social media ad targeting indices, news ad targeting indices, gaming ad targeting indices, and lifestyle ad targeting indices. Based on user base data, we initially assign major ad categories to users, and then calculate the targeting index for each major ad category by combining user browsing data. Compare the size of the advertising targeting index for each major category to obtain the advertising targeting results for each major category. The analysis process of the targeted advertising results in the sub-category ad delivery analysis module includes: The sub-categories of ads are obtained by subdividing various major categories of ads. Combined with current user behavior data, the user dynamic profile model outputs the corresponding user dynamic profile. Based on the targeting results of major category ads, the model analyzes the e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, lifestyle and hobby characteristics, and user image characteristics corresponding to different dimensions in the user's dynamic profile, thereby obtaining the targeting results of sub-category ads. The calibration unit evaluates the calibration coefficient for sub-category ad targeting, and then calibrates the sub-category ad targeting results. This process includes: Extract user browsing data, user behavior data, and user image feature data from the user profile dynamic analysis dataset. Combine this with a convolutional neural network, take the training set data as input, and the sub-category ad targeting calibration coefficient as output. Learn the non-linear relationship between user browsing data, user behavior data, and user image feature data and the sub-category ad targeting calibration coefficient, and train the sub-category ad targeting calibration model. The test set data is input into the category ad targeting calibration model. The parameters of the category ad targeting calibration model are adjusted to optimize its performance. The category ad targeting calibration model is then deployed into the system. Combining current user browsing data, user behavior data, and user image feature data, the corresponding category ad targeting calibration coefficients are output. Based on these calibration coefficients, the category ad targeting results are adjusted. The calibrated category ad targeting results are then encoded and integrated into the user profile dynamic analysis dataset.

2. The user profile dynamic generation and targeted advertising system based on multi-source data fusion according to claim 1, characterized in that: The ad targeting module includes a calibration unit, a model building unit, and an execution unit, the functions of which are as follows: The calibration unit uses a convolutional neural network algorithm to evaluate the calibration coefficient of the targeted placement of subcategories of advertisements by using user browsing data, user behavior data, and user image feature data, thereby calibrating the targeted placement results of subcategories of advertisements. The model building unit uses the random forest algorithm to build an advertising targeted delivery model; The execution unit delivers relevant advertisements to users based on the output of the advertising targeting model.

3. The user profile dynamic generation and targeted advertising system based on multi-source data fusion according to claim 2, characterized in that: The multi-source data acquisition module's acquisition process for user basic data, user browsing data, user behavior data, and user image data includes: Data entry technology is used to collect basic user data, user browsing data, user behavior data, and user image data; The user basic data includes the user's gender, age, and occupation, where occupation includes students, employed, and retired; the user browsing data includes the number of searches, page views, and dwell time on different types of pages within 24 hours, where different types of pages include e-commerce platform product pages, social media pages, news report pages, game experience pages, and lifestyle hobby pages; The user behavior data includes e-commerce behavior data, social behavior data, information behavior data, gaming behavior data, and lifestyle and hobby data; real-time images of users browsing different types of advertisements; The e-commerce behavior data includes the number of times users search for different types of products, the number of times they purchase, the number of positive reviews and negative reviews, as well as users' brand preferences and promotional method preferences; The social behavior data includes the user's social platform usage time, the number of groups the user has joined and the number of interactions, the number of friends the user has and the number of interactions, and the duration of the user's live stream participation. The information behavior data includes news viewing type, knowledge popularization browsing type, lifestyle information browsing type, and industry news click count; The game behavior data includes the types of games the user participates in and their duration; the user's lifestyle data refers to the user's lifestyle preferences. The collected data is cleaned and normalized, and user basic data, user browsing data and user behavior data are integrated to generate a dynamic analysis dataset of user profiles. The dynamic analysis dataset of user profiles is divided into training set and test set.

4. The user profile dynamic generation and targeted advertising system based on multi-source data fusion according to claim 1, characterized in that: The user image feature extraction module, the process of acquiring user image feature data includes: The user image feature data includes the user's person feature data, object feature data, and scene feature data; Feature extraction is performed on real-time images of users browsing different types of advertisements. Using functions and algorithms in the Matlab computer vision toolbox, combined with a Haar cascade detector, facial features of users are extracted. An edge detection algorithm is used to extract the posture features of users. The human feature data includes facial features and posture features of users. Using the Faster R-CNN algorithm based on deep learning, object feature data is extracted; using Matlab software, semantic segmentation is performed on user image data to distinguish the background region from the user, thereby extracting scene feature data, and integrating the obtained user image feature data into the user profile dynamic analysis dataset.

5. The user profile dynamic generation and targeted advertising system based on multi-source data fusion according to claim 4, characterized in that: The user profile dynamic generation module constructs a user dynamic profile model and calibrates the user dynamic profile model by combining user image feature data. The process includes: Analyze user behavior data to extract e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics, and lifestyle and hobby characteristics; The extracted features are categorized based on dimensions, including consumption, social, information, entertainment, and lifestyle dimensions. User behavior data is extracted from the dynamic analysis dataset of user profiles. Clustering analysis algorithm is used to assign users to different dimensions. Consumption, social, information, entertainment and lifestyle dimensions are respectively mapped to e-commerce behavior characteristics, social behavior characteristics, information behavior characteristics, gaming behavior characteristics and lifestyle hobby characteristics. The training set data is used as input to obtain the statistical values ​​of users in each dimension. Then, users are divided into different clusters and the corresponding dynamic profiles of users are output, thus realizing the construction of the dynamic user profile model. Input the test set data into the user dynamic profile model, adjust the parameters of the user dynamic profile model, optimize the performance of the user dynamic profile model, and obtain the final user dynamic profile model. By using clustering analysis algorithms, user profile data is integrated with lifestyle and entertainment dimensions, user object data is integrated with consumption dimensions, and user scenario data is integrated with social and information dimensions, thereby calibrating the dynamic user profile model.

6. The user profile dynamic generation and targeted advertising system based on multi-source data fusion according to claim 5, characterized in that: The model building unit, the process of building an ad targeting model, includes: Extract basic user data and calibrated sub-category ad targeting result codes from the user profile dynamic analysis dataset. Combine the training set data with the random forest algorithm, using the training set data as input and the calibrated sub-category ad targeting result codes as output, to learn the non-linear relationship between basic user data and calibrated sub-category ad targeting result codes, and train the ad targeting model. The test set data is input into the ad targeting model. The output results of the ad targeting model are compared with the actual calibrated sub-category ad targeting results to evaluate the performance of the ad targeting model. The parameters of the ad targeting model are adjusted to optimize the ad targeting model and obtain the final ad targeting model.

7. The user profile dynamic generation and targeted advertising system based on multi-source data fusion according to claim 6, characterized in that: The execution unit, based on the output of the advertising targeting model, performs the following process to deliver relevant advertisements to users: Input basic user data into the ad targeting model, obtain the calibrated sub-category ad targeting result code, and then match the corresponding calibrated sub-category ad targeting result; Based on the calibrated sub-category advertising targeting results, relevant ads are targeted to users.

Images