Intelligent pushing strategy method and system for real-time user behavior monitoring

Abstract

The invention relates to the technical field of advertisement pushing, in particular to an intelligent pushing strategy method and system for real-time user behavior monitoring. The method comprises the following steps: acquiring user behavior data; performing feature extraction according to the user behavior data to obtain user behavior feature data; performing user behavior change capture according to the user behavior feature data to obtain user behavior change data, and performing model updating according to the user behavior change data and a preset user interest model to obtain a user real-time interest model; and generating advertisement content according to the user behavior data and the user real-time interest model to obtain user real-time advertisement content data so as to carry out real-time user behavior advertisement pushing operation. According to the method, the pushed content is optimized, so that the pushed content is highly matched with user requirements, interests and environments, and the click rate and the conversion rate of pushing are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of advertisement push technology, and in particular to an intelligent push strategy method and system for real-time user behavior monitoring. Background Art

[0002] In an era when the Internet and mobile devices are highly popular, smart push has become an important technology for personalized information recommendation and precise advertising. By analyzing user behavior data, inferring user interests and needs, and then pushing relevant content or advertisements, the user experience and advertising effect can be significantly improved. However, in the existing technology, user interest modeling is mostly based on static or historical behavior data, while ignoring the dynamic characteristics of user behavior, such as sudden changes in interest points or short-term hot spot demands. Model updates are usually delayed and cannot quickly respond to users' real-time needs, resulting in a large deviation between the pushed content and the user's actual interests. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an intelligent push strategy method and system for real-time user behavior monitoring to solve at least one of the above technical problems.

[0004] The present application provides an intelligent push strategy method for real-time user behavior monitoring, the method comprising: S1. Obtain user behavior data, wherein the user behavior data includes real-time user behavior data and external user behavior data. The real-time user behavior data includes user click record data, user browsing time data, user search query data, and user interaction history data. The external user behavior data includes user device data, user geographic location data, and user communication environment data. S2. Extract features based on the user behavior data to obtain user behavior feature data; S3, capturing user behavior changes according to the user behavior feature data to obtain user behavior change data, and updating the model according to the user behavior change data and a preset user interest model to obtain a user real-time interest model; S4. Generate advertising content based on user behavior data and the user's real-time interest model to obtain user real-time advertising content data to perform real-time user behavior advertising push operations.

[0005] The user behavior data in the present invention covers real-time behavior data (click records, browsing time, search queries, interaction history) and external data (device information, geographic location, communication environment), providing multi-dimensional user information and realizing comprehensive user portraits. Feature extraction is performed based on user behavior data to mine implicit relationships and patterns (such as click frequency, browsing depth, and geographic location preferences). By real-time monitoring of user behavior changes (such as interest point migration, behavioral mutations), and dynamically updating user interest representations in combination with preset interest models, user interest changes are dynamically captured, so that the pushed content can adapt to the user's latest needs in real time, avoiding the push of outdated or irrelevant content. The interest model is continuously iterated and optimized through the feedback mechanism of behavior change data, enhancing the adaptability to complex behavior changes. Advertising content is generated based on the user's real-time interest model and behavior data, dynamically matching user needs, and users are more likely to be attracted to advertisements related to their needs, thereby improving advertising click-through rate (CTR) and user experience satisfaction. The real-time generation and push strategy of advertising content is based on user behavior characteristics to avoid overly frequent or meaningless pushes. Optimize ad push to the peak period of user interest, and perform refined content matching based on user environment (such as device, geographic location) to improve ad conversion rate and optimize advertiser's ROI (return on investment).

[0006] Optionally, S1 includes: S11. Collect the user's click records, browsing time, search queries and interaction history in real time to obtain the user's real-time behavior data; S12. Utilize device API, GPS positioning service and network status detection tool to obtain user device information, geographic location and communication environment data, and obtain external data of user behavior; S13: Integrate the real-time user behavior data and external user behavior data to obtain user behavior data.

[0007] The present invention captures the user's operating behavior on the platform by collecting click records, browsing time, search queries and interaction history in real time. The multi-dimensional data coverage enables the system to accurately portray the user's current behavior status, thereby enhancing the accuracy of push notifications. The introduction of external data such as device information, geographic location, and communication environment provides a richer context for user portraits. External data supplements the deficiencies of behavioral data, allowing the system to further analyze the user's behavioral background (such as the influence of the user's location and the limitations of device capabilities) to optimize content and push notification strategies. Combining behavioral data with external data (such as user device type and network status) provides more comprehensive user background information.

[0008] Optionally, the user behavior characteristic data includes first user behavior characteristic data and second user behavior characteristic data, and S2 includes: S21, extracting click count features and geographic location features based on the user behavior data to obtain click count feature data and geographic location feature data; S22, combining features according to the click count feature data and the geographic location feature data to obtain initial cross feature data; S23, performing multi-feature expansion according to the initial cross-feature data to obtain first user behavior feature data; S24, extracting search keyword features and browsing time features from the user behavior data to obtain search keyword feature data and browsing time feature data respectively; S25, performing Gaussian mixture clustering based on the click count feature data, the search keyword feature data, and the browsing time feature data to obtain user behavior clustering data; S26, extracting key topics from the user behavior grouping data to obtain user interest topic data; S27, calculating the purchase intention according to the user's interest topic data to obtain the user's purchase intention data; S28. Structuring the user portrait according to the user behavior grouping data, the user interest topic data, and the user purchase intention data to obtain second user behavior feature data.

[0009] In the present invention, feature extraction is performed on the number of clicks, geographic location, search keywords and browsing time respectively, and multiple independent feature data are generated, ensuring that different dimensions of user behavior are fully covered, providing rich data input for user behavior modeling. By combining and expanding the number of click features and geographic location features, the potential associations between multiple features are mined, and the initial cross-features and the first user behavior feature data of the deep data that can reflect the correlation between the number of clicks and the geographic location are generated. Cluster analysis is performed based on the number of clicks, search keywords and browsing time features to capture the potential structure in the user behavior pattern. Interest topics are extracted from user behavior grouping data to discover the core driving force of user behavior. Purchase intention data enables the system to accurately lock in potential user groups with high conversion rates, thereby improving the efficiency of advertising delivery. By extracting and updating real-time features such as the number of clicks, geographic location and browsing time, the latest user behavior dynamics are captured, and the push content is matched with the user's current needs, improving the user experience and push effect. A structured user portrait is generated based on user grouping data, interest topic data and purchase intention data, providing comprehensive support for accurate recommendation and real-time push strategy optimization, and adapting to complex business scenario requirements.

[0010] Optionally, the click count feature extraction is specifically as follows: Perform click distribution based on user behavior data to obtain click distribution feature data; Perform click frequency analysis based on click distribution characteristic data to obtain click frequency characteristic data; Click behavior clustering is performed according to the click frequency characteristic data to obtain click behavior clustering data; Get user ambient sound data; Extracting the user ambient sound change feature and the user ambient sound decibel feature according to the user ambient sound data to obtain the user ambient sound change feature data and the user ambient sound decibel feature data; Mapping the click behavior clustering data according to the user ambient sound change characteristic data and the user ambient sound decibel characteristic data to obtain click behavior clustering ambient sound association data; Perform internal clustering on the click behavior clustering ambient sound associated data to obtain ambient sound click behavior clustering data; Feature cross-calculation is performed based on the ambient sound click behavior clustering data to obtain click count feature data.

[0011] In the present invention, the distribution characteristics of user click behavior are statistically analyzed to capture the click distribution patterns of users in different time periods, geographical locations and other dimensions. Click behavior is further quantified, and the dynamic characteristics of user behavior and short-term interest fluctuations are revealed through frequency changes. Clustering technology is used to group user click behaviors and identify potential categories of click patterns (such as high-frequency click users and random click users). By extracting user ambient sound changes and decibel characteristics, the potential impact of the user's environment (such as quiet, noisy) on click behavior is understood, and user behavior modeling is extended from single behavior data to the correlation analysis of behavior and environment, thereby improving the context perception ability of the model. By mapping ambient sound and click behavior, the relationship between user behavior and the external environment is revealed (such as high click frequency in noisy environments and high click depth in quiet environments). Based on the clustering of click behavior, the ambient sound dimension is introduced to perform secondary subdivision on the clustering clusters. By performing high-order cross calculations on click behavior features and ambient sound features, multi-dimensional and deep click frequency feature data is generated.

[0012] Optionally, the geographic location feature extraction is specifically: According to the user's geographic location data and user click record data, click location aggregation is performed to obtain location click distribution data; Performing location-behavior association on the location-click distribution data to obtain location-click-behavior association data; Calculate the location purchasing power based on the location click behavior association data to obtain the location purchasing power data; Performing location classification on the location purchasing power data, the user location data, and the user click record data to obtain location classification data; The location classification data is cross-analyzed based on the user search query data to obtain the location feature data.

[0013] In the present invention, based on user geographic location data and click record data, user behavior is mapped to a specific geographic area to form location click distribution data. The location click distribution is associated with behavioral characteristics to identify user behavior patterns in a specific area (such as high-frequency clicks, high-conversion areas). By analyzing the location click behavior association data, the purchase potential of each geographic location is quantified (combined with indicators such as click frequency and conversion rate). The location purchasing power data, user geographic location data and click record data are combined, and the geographic locations are grouped through a classification model (such as high-consumption areas and low-consumption areas). The user search query data is combined with the location classification data to explore the association between user search behavior and geographic location characteristics (such as the search distribution of specific keywords in different areas). Through the interactive analysis of click records and location data, a dynamic connection between user behavior and geographic location is constructed. By combining location purchasing power and location classification data, customized advertising content (such as regional specific discount information) is pushed to users in different regions. Through location purchasing power calculation and classification analysis, advertisers can accurately place resources in high purchasing power areas and improve the return on investment (ROI). Location classification and feature cross-analysis support advertisers to explore regional market demand and discover potential opportunities.

[0014] Optionally, S3 includes: S31, performing feature diversion according to the user behavior feature data to obtain user behavior feature diversion data; S32, performing feature aggregation according to the user behavior feature diversion data to obtain user behavior feature aggregation data; S33, performing feature change calculation on the user behavior feature aggregation data to obtain user behavior feature change data; S34, performing change point detection according to the user behavior feature change data to obtain user behavior change data; S35. Update the model according to the user behavior change data and the preset user interest model to obtain a real-time user interest model.

[0015] In the present invention, different features are assigned to corresponding calculation channels, and independent optimization processing is performed for feature types. A global user behavior view is formed by aggregating the correlations of different features. By calculating the change amplitude and rate of user behavior characteristics (such as a sudden increase in click frequency or a shift in interest points), the user behavior dynamics are reflected in real time. Using change point detection algorithms (such as CUSUM or sliding window methods), the mutation points of user behavior are accurately identified. The interest model is dynamically updated through user behavior change data to adapt it to the continuous evolution of user behavior. Taking into account both short-term interests and long-term preferences, the pushed content has both immediate appeal and meets the overall needs of users.

[0016] Optionally, the feature split is specifically: Perform feature stratification according to user behavior feature data to obtain user behavior feature stratification data; Performing feature classification on the user behavior feature hierarchical data to obtain user behavior feature classification data; Perform advertising scene feature weighting on user behavior feature classification data to obtain feature classification weighted data; Perform multi-channel feature segmentation based on feature classification weighted data to obtain feature diversion channel data; The intra-channel features are reconstructed based on the feature diversion channel data to obtain the user behavior feature diversion data.

[0017] In the present invention, the user behavior feature data is layered (such as time dimension, behavior type, geographic location, etc.) to perform structured processing on the features. The layered feature data is further classified and the features are divided into multiple types (such as click features, search features, and browsing features). The weighting process highlights the importance of key features and improves the matching degree between advertising push and user behavior, thereby improving the click-through rate (CTR) and conversion rate (CVR) of advertisements. Sorting combines user behavior data (such as click frequency, browsing depth) and real-time interest models to make advertising recommendations more personalized. The diversion channel and reconstruction mechanism support a variety of business scenarios (such as e-commerce advertising, video advertising), enhancing system flexibility.

[0018] Optionally, S4 includes: S41, screening advertisement content through a preset advertisement content pool according to user behavior data and a user real-time interest model to obtain a screened advertisement pool; S42, performing advertisement similarity calculation based on the screened advertisement pool, user behavior data, and user real-time interest model to obtain advertisement similarity data; S43, sorting the screened advertisement pool according to the advertisement similarity data to obtain sorted advertisement data; S44, calculating the push timing according to the sorted advertisement data to obtain push timing data; S45. Pack the advertisement content according to the push timing data and the sorted advertisement data to obtain the user's real-time advertisement content data, so as to perform a real-time user behavior advertisement push operation.

[0019] The present invention combines user behavior data with real-time interest models, and filters advertisements that are most relevant to user needs through a preset advertising content pool. The dynamic update of the real-time interest model enables the screening process to quickly adapt to changes in user interests. Based on the screened advertising pool, user behavior data and real-time interest model, an embedding model or a similarity algorithm (such as cosine similarity) is used to calculate the matching degree between the advertising content and the user's interests. Similarity is calculated in multiple dimensions based on user interests, historical click behaviors and advertising features. The screened advertising pool is sorted according to the advertising similarity data, and advertising content that the user is most likely to be interested in is recommended first. The sorting combines user behavior data (such as click frequency, browsing depth) and real-time interest models to make advertising recommendations more customized and reduce users' aversion to pushed content. Combining the sorted advertising data with user behavior patterns, a time series model or reinforcement learning method is used to calculate the best push timing.

[0020] Optionally, the push timing calculation includes: Get user ambient sound data; Extracting features from the user ambient sound data to obtain user ambient sound feature data; Evaluate the attractiveness of advertisements based on the ranked advertisement data and the user's ambient sound feature data to obtain ambient sound weighted advertisement list data; The push timing is predicted based on the ambient sound weighted advertisement list data to obtain the push timing data.

[0021] In the present invention, user environmental sound data is collected in real time to capture the user's situation (such as quiet, noisy, natural sound, etc.). The system's perception of the user's current environment is enhanced to provide contextual support for advertising content and push strategies. User environmental sound feature data (such as volume intensity, noise type) is generated through feature extraction methods such as spectrum analysis and volume fluctuation. Fine-grained environmental sound feature analysis enables the system to identify the potential impact of different environments on user behavior and improve the situational adaptability of the push strategy. Combined with the sorted advertising data and user environmental sound feature data, the attractiveness score of the advertising content in a specific environment is calculated, for example, visual ads are pushed first in a noisy environment, and detailed text or audio ads are pushed in a quiet environment. After weighting the sorted advertising data, a new advertising list is generated to highlight the most attractive ads in the current environment. Combined with the environmental sound weighted advertising list data, a time series model (such as LSTM or Transformer) is used to predict the best time for users to respond to advertisements (such as quiet periods or click peak periods). The push timing prediction model is dynamically adjusted according to the real-time updated environmental sound feature data to ensure that the push strategy adapts to the rapid changes in user behavior and situations.

[0022] Optionally, the present application further provides an intelligent push strategy system for real-time user behavior monitoring, which is used to execute the intelligent push strategy method for real-time user behavior monitoring as described above, and the intelligent push strategy system for real-time user behavior monitoring includes: A user behavior data collection module is used to obtain user behavior data, wherein the user behavior data includes real-time user behavior data and external user behavior data. The real-time user behavior data includes user click record data, user browsing time data, user search query data, and user interaction history data. The external user behavior data includes user device data, user geographic location data, and user communication environment data; A user behavior feature extraction module is used to extract features based on user behavior data to obtain user behavior feature data; A user real-time interest update module is used to capture user behavior changes based on user behavior feature data to obtain user behavior change data, and to update the model based on the user behavior change data and a preset user interest model to obtain a user real-time interest model; The user real-time advertising content generation module is used to generate advertising content according to user behavior data and user real-time interest model, obtain user real-time advertising content data, and perform real-time user behavior advertising push operations.

[0023] The purpose of the present invention is to achieve all-round data collection of user behavior by acquiring real-time user behavior data (click records, browsing time, search queries, interaction history) and external data (device information, geographic location, communication environment). While collecting real-time data, the system integrates static behavior data with dynamic change data to form an efficient dynamic data management framework. The present invention generates user behavior feature data by extracting key features (number of clicks, geographic location, search keywords, etc.) from user behavior data and combining feature cross calculation and high-order feature expansion. Change point detection is performed on user behavior feature change data to effectively identify significant mutations in user behavior (such as interest point transfer, behavior pattern change). Based on the captured user behavior change data, the user interest model is updated in real time, and a comprehensive interest representation is generated by combining short-term behavior and long-term preferences. According to the user's real-time interest model and user behavior data, the most relevant advertising content is screened from the advertising content pool, and combined with advertising similarity evaluation and sorting optimization. By acquiring user ambient sound data and extracting features such as volume and change mode, the push timing is predicted in combination with sorted advertising data, and advertising content is pushed at the time when the user is most likely to accept it. By combining the push timing with the ad content data, the ad content is packaged into a display format (such as video ads, text ads) that adapts to the user's current environment and device characteristics. Timing optimization increases the effective display time of ads, enhances the user's acceptance of push content, and reduces the user's negative emotions caused by inappropriate push. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings: Figure 1 A flowchart showing a method for intelligent push strategy for real-time user behavior monitoring according to an embodiment of the present invention is provided; Figure 2 A flowchart showing the steps of a method for collecting user behavior data according to an embodiment is shown; Figure 3 A flowchart showing a method for extracting user behavior features according to an embodiment of the present invention is shown; Figure 4 A flowchart showing a method for updating user interests in real time according to an embodiment of the present invention is provided; Figure 5 A flowchart showing a method for generating real-time advertisement content for a user according to an embodiment of the present invention is provided; The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0027] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0028] E-commerce platforms want to push advertising content to users and optimize the effect of advertising. User A has recently frequently clicked and browsed smartwatch-related products, searching for keywords such as "smartwatch brand rankings" and "cost-effective smartwatches". The user's current device is a high-performance smartphone, and the location is in a shopping mall. The ambient sound data shows that the user is in a noisy environment (such as the background sound of a shopping mall). The goal is to push ads related to smartwatches to user A, ensuring that the push timing and content are accurately matched.

[0029] User A has recently clicked on smartwatch-related products 5 times. Each browsing time is about 2 minutes, indicating that the user has a high interest in related products. The keyword "smartwatch ranking" was searched 3 times. Smartwatch products are frequently added to the shopping cart, but the purchase has not been completed. The user's device is a high-performance smartphone. The geographical location is in a shopping mall. The communication environment is a stable Wi-Fi connection. The environmental sound data is noisy (background sound peak > 70 decibels).

[0030] Extract the click frequency (5 times a day) from the click records. Aggregate the user's mall location and click behavior to extract the "electronics product area" with high purchasing power. Extract the search keyword "smart watch" and browsing time (2 minutes) features to indicate high interest correlation. Classify the number of clicks, search keywords, and browsing time as the "smart watch feature layer". Assign a high weight (0.8) to the "smart watch" feature. The click frequency of smart watches has increased by 20% recently, and the mutation point indicates that the user's interest is concentrated. Update "smart watch" as the core interest to the real-time interest model of user A. Update the user's real-time interest model and mark "smart watch" as a high priority interest.

[0031] We screened ads related to "smart watches" from the ad content pool and selected 10 ads in total. We calculated the similarity between ads and users based on the user interest model: Ad 1 (smart watches of brand X): 0.95, Ad 2 (smart watches of brand Y): 0.85, Ad 3 (smart watch accessories): 0.6.

[0032] Sorting ad pool: Ad 1 > Ad 2 > Ad 3. The current noisy environment is suitable for visual content push. Combined with user active time and ambient sound data, select 3 pm (high response probability period). Select Ad 1 and Ad 2, and attach short video display mode. The push time is 3 pm, and the push content is Ad 1: Brand X smartwatch, with short video display function and "Today's Special Offer" promotion information, Ad 2: Brand Y smartwatch, with free earphone discount activity.

[0033] See also Figures 1 to 5 The present application provides an intelligent push strategy method for real-time user behavior monitoring, the method comprising: S1. Obtain user behavior data, wherein the user behavior data includes real-time user behavior data and external user behavior data. The real-time user behavior data includes user click record data, user browsing time data, user search query data, and user interaction history data. The external user behavior data includes user device data, user geographic location data, and user communication environment data. In one embodiment, on an e-commerce platform, user behavior data is acquired in real time through an embedded data acquisition module. Real-time user behavior data includes the following: click record data, which records the product links, page elements and corresponding timestamps clicked by users. Browsing time data, which monitors the time users stay on each product page and records it with millisecond time resolution. Search query data, which captures the keywords entered by users in the search box and the time when the search request is submitted. Interaction history data, which tracks users' likes, comments, favorites and sharing behaviors to form an interaction log. The collection of external user behavior data includes the following: user device data, which identifies the hardware type, operating system and browser version through the user terminal device. User geographic location data, which uses the GPS module or IP address to resolve the user's geographic location. User communication environment data, which detects the network type (such as WiFi or 4G) and the current signal strength, and evaluates the communication stability.

[0034] S2. Extract features based on the user behavior data to obtain user behavior feature data; In one embodiment, multi-level feature extraction is performed on the user's behavior pattern. Click feature extraction counts the user's click frequency in different time periods to identify the user's active time periods and key focus products. Search keyword feature extraction analyzes the frequency and semantics of user search terms to determine the user's interest category. Browsing behavior feature extraction screens pages with a high user browsing depth based on the user's stay time to determine the user's interest concentration. Device and environment feature extraction combines the user's device type and network status to evaluate whether the user is in an environment suitable for receiving advertisements. Generate user behavior feature data, including indicators such as click heat, interest keyword distribution, and browsing depth.

[0035] S3, capturing user behavior changes according to the user behavior feature data to obtain user behavior change data, and updating the model according to the user behavior change data and a preset user interest model to obtain a user real-time interest model; In one embodiment, the preset user interest model is generated by extracting user interest features and building a model through user historical behavior data. Compare the user's behavior characteristics in the last hour with the average behavior characteristics in the past week to capture significant changes in behavior. For example, if the number of clicks on a certain type of product by a user surges in a short period of time, it is marked as a change in interest. Based on the captured behavioral changes, the weights in the user interest model are dynamically adjusted. For example, if a user has frequently browsed smartphone-related content recently, the weight of the smartphone category is increased, while the weights of other categories are reduced to form a real-time interest model.

[0036] S4. Generate advertising content based on user behavior data and the user's real-time interest model to obtain user real-time advertising content data to perform real-time user behavior advertising push operations.

[0037] In one embodiment, advertising content that is highly matched with the user's real-time interest model is screened from the advertising library. For example, if the user's interest model shows that the user has a high degree of interest in "smartphones", advertisements of smartphone brands are given priority. Customized advertising content is generated based on the user's geographic location, language preference, and device type. For example, if the user is located in City S, the advertising content generates "Limited time promotion in City S, new smartphones are discounted by 500 yuan!". If the user uses a mobile device, advertising material adapted to the mobile terminal interface is generated. Combined with the user's active time period and communication environment, the time period when the user is most active and the network is stable is selected for advertising push. The generated advertising content is sent to the user terminal through a push service to ensure that the advertising content is accurately displayed at the most suitable time for the user.

[0038] Optionally, S1 includes: S11. Collect the user's click records, browsing time, search queries and interaction history in real time to obtain the user's real-time behavior data; In one embodiment, on an e-commerce platform, user behavior data is recorded in real time through an embedded monitoring module, and the user's click operation is captured through the page tracking code when the user browses the web page. For example, when a user clicks a link to a product detail page, the time of the click and the target page are recorded. When a user enters a page, the entry time is recorded; when a user leaves the page, the exit time is recorded. The user's stay time on the page is calculated by the difference. When a user enters and submits a keyword in the search box, the search content and the time point of the search are recorded. When a user collects, likes, or comments on a product, these interactive operations and related time and object information are recorded in real time.

[0039] S12. Utilize device API, GPS positioning service and network status detection tool to obtain user device information, geographic location and communication environment data, and obtain external data of user behavior; In one embodiment, the user's hardware and software information, such as device type (mobile phone, tablet or PC), operating system version and browser type, is obtained through the device API in the browser or application. The user's latitude and longitude information is obtained using the GPS positioning service, or the user's geographic location is reversely resolved through the IP address. If the user's device allows precise positioning, the system can identify the business district or city where the user is currently located. The user's network connection type (such as WiFi, 4G or 5G) and the stability of the current network signal are identified through the network status detection tool.

[0040] S13: Integrate the real-time user behavior data and external user behavior data to obtain user behavior data.

[0041] In one embodiment, the user's real-time behavior data (clicks, browsing, searches, interaction history) and the user's behavior external data (device information, geographic location, communication environment) are synchronized and integrated according to the time dimension. For example, the user's click record at 10 am is bound to the geographic location data of that period.

[0042] Optionally, the user behavior characteristic data includes first user behavior characteristic data and second user behavior characteristic data, and S2 includes: S21, extracting click count features and geographic location features based on the user behavior data to obtain click count feature data and geographic location feature data; In one embodiment, the number of clicks by users on page elements (such as product detail page links, advertising banners, etc.) within a specified time period is recorded to calculate the user's click frequency, for example, the user's click behavior is counted within an hour to extract the number of click features. The user's longitude and latitude information is obtained by using the GPS data or IP address resolution of the user's device, and the user's main activity area is extracted based on the user's historical activity range, for example, the business district locations visited by the user in the past day are counted.

[0043] S22, combining features according to the click count feature data and the geographic location feature data to obtain initial cross feature data; In one embodiment, the click count feature and the geographic location feature are associated to analyze the relationship between the user's behavior and their location. For example, the click distribution of users in different business districts is counted to generate initial cross-feature data.

[0044] S23, performing multi-feature expansion according to the initial cross-feature data to obtain first user behavior feature data; In one embodiment, by analyzing the interaction data between user clicks and location features, multi-dimensional features are expanded, for example: calculating the click ratio of users in different locations and extracting their attention to certain business districts; inferring their main consumption scenarios (such as office areas and shopping areas) based on the frequency of user activities.

[0045] S24, extracting search keyword features and browsing time features from the user behavior data to obtain search keyword feature data and browsing time feature data respectively; In one embodiment, the search keyword feature extraction extracts high-frequency keywords from the user's search query records and generates keyword distribution features. The user's attention to the product is calculated based on the user's stay time on the product page.

[0046] S25, performing Gaussian mixture clustering based on the click count feature data, the search keyword feature data, and the browsing time feature data to obtain user behavior clustering data; In one embodiment, the extracted feature data is clustered and divided into different groups according to the user's behavior pattern. For example, user A is clustered as a "high consumption interest user" because of the high number of clicks and the focus of search on high-priced goods; user B is clustered as an "information search user" because of the long browsing time but no purchase behavior.

[0047] S26, extracting key topics from the user behavior grouping data to obtain user interest topic data; In one embodiment, the main areas of interest of users are extracted based on the keywords and behavior characteristics in the user grouping data. For example, by analyzing high-frequency keywords and the content of the browsed pages, the user's interests are classified into "smartphones" and "electronic accessories".

[0048] In one embodiment, a topic model analysis is performed on the search keywords of each group of users to extract the main topics of interest, for example, by term frequency-inverse document frequency (TF-IDF) calculation.

[0049] S27, calculating the purchase intention according to the user's interest topic data to obtain the user's purchase intention data; In one embodiment, the user's purchase intention is evaluated based on the user's interest topics and behavior characteristics and weight calculation. For example, after browsing multiple smartphones, if the user spends more time on a certain model, it indicates that the user has a strong purchase intention.

[0050] S28. Structuring the user portrait according to the user behavior grouping data, the user interest topic data, and the user purchase intention data to obtain second user behavior feature data.

[0051] In one embodiment, the user's grouping results, interest topics, and purchase intentions are integrated into a structured portrait for precision marketing and advertising push.

[0052] Optionally, the click count feature extraction is specifically as follows: Perform click distribution based on user behavior data to obtain click distribution feature data; In one embodiment, on an online shopping platform, the click behaviors of users on different pages or time periods are recorded. For example, the number of clicks of users on product category pages is counted, and user A clicks 20 times on the "electronic products" category and 10 times on the "clothing" category.

[0053] In one embodiment, the click data is grouped and counted by time window or page area. For example, one hour is used as a time window, and the number of clicks in each time period is counted.

[0054] Perform click frequency analysis based on click distribution characteristic data to obtain click frequency characteristic data; In one embodiment, the click distribution data is subjected to frequency analysis in the time dimension. For example, the click frequency of the user in the past hour is counted and compared with the change of the click frequency throughout the day. The click frequency of user A in the past hour is 2 times per minute, while the click frequency throughout the day is 1 time per minute.

[0055] Click behavior clustering is performed according to the click frequency characteristic data to obtain click behavior clustering data; In one embodiment, the user's click frequency characteristics are combined with other behavior characteristics (such as click categories) to perform behavior pattern classification. For example, the click behavior of user A is classified as "highly active shopping type" because the click frequency is high and concentrated in a specific category.

[0056] In one embodiment, the click behavior feature of each user may be represented as a matrix, and clustering is performed using the feature matrix to divide the user click behavior groups. For example, the Euclidean distance between two samples is calculated to output the click behavior category of each user.

[0057] Get user ambient sound data; In one embodiment, real-time ambient sound data is collected through the microphone of the user device. For example, the current ambient sound of user A is displayed as "background music" with a low volume, while the current ambient sound of user B is "noisy human voice" with a high volume.

[0058] Extracting the user ambient sound change feature and the user ambient sound decibel feature according to the user ambient sound data to obtain the user ambient sound change feature data and the user ambient sound decibel feature data; In one embodiment, by analyzing the changes in the types of ambient sound of the user at different times, it is identified whether there is a significant change in the environment. For example, user A switches from a "quiet" environment to a "noisy" environment. Ambient sound decibel feature extraction, based on the collected ambient sound amplitude, extracts decibel features to determine the volume level of the environment. For example, the ambient sound decibel of user A is 40 dB.

[0059] Mapping the click behavior clustering data according to the user ambient sound change characteristic data and the user ambient sound decibel characteristic data to obtain click behavior clustering ambient sound association data; In one embodiment, the ambient sound features are time-aligned with the user's click behavior clustering data to analyze whether the user's click behavior is affected by the change in ambient sound. For example, the click frequency of user A in a "noisy" environment is reduced to once per minute.

[0060] Perform internal clustering on the click behavior clustering ambient sound associated data to obtain ambient sound click behavior clustering data; In one embodiment, each click behavior cluster is further subdivided into subgroups according to the environmental sound characteristics. For example, user A is classified as "stable behavior in noisy environment" because changes in environmental sound do not significantly affect his click behavior.

[0061] Feature cross-calculation is performed based on the ambient sound click behavior clustering data to obtain click count feature data.

[0062] In one embodiment, a more accurate click count feature is generated by cross-calculating the user's click behavior feature and the ambient sound feature. For example, user A clicks less frequently on the "electronic products" page in a noisy environment, but clicks more frequently on the "fast moving consumer goods" page.

[0063] Optionally, the geographic location feature extraction is specifically: According to the user's geographic location data and user click record data, click location aggregation is performed to obtain location click distribution data; In one embodiment, in an e-commerce platform, the user's geographical location is parsed through the GPS module or IP address of the user's device, and the number of clicks of the user in different geographical areas is counted in combination with the user's click records. For example, the record of user A shows that his click behavior in the following areas is 5 clicks in business district 1 and 3 clicks in business district 2.

[0064] Performing location-behavior association on the location-click distribution data to obtain location-click-behavior association data; In one embodiment, by analyzing the click location distribution data and combining the user's click behavior type (such as browsing, purchasing), the relationship between location and behavior is established. For example, user A's clicks in business district No. 1 are mainly concentrated on "electronic products"; user A's clicks in business district No. 2 are mainly concentrated on "clothing products".

[0065] Calculate the location purchasing power based on the location click behavior association data to obtain the location purchasing power data; In one embodiment, the purchasing power of each location is evaluated by counting the proportion of users' purchase behaviors in each geographical location. For example, among the click behaviors of user A in business district No. 1, 3 are converted into actual purchases; among the click behaviors of user A in business district No. 2, 1 is converted into actual purchase.

[0066] Performing location classification on the location purchasing power data, the user location data, and the user click record data to obtain location classification data; In one embodiment, the geographical location data, purchasing power data, and click records of the user are combined to classify the geographical location into high purchasing power area, low purchasing power area, or other categories. For example, business district No. 1 is classified as "high purchasing power area"; business district No. 2 is classified as "medium purchasing power area". The feature vectors are classified, and similarity metrics (such as Euclidean distance) or clustering algorithms are used to classify the locations into categories such as high purchasing power area and ordinary area.

[0067] The location classification data is cross-analyzed based on the user search query data to obtain the location feature data.

[0068] In one embodiment, by analyzing the user's search keywords, the user's search interests are combined with the location classification. For example, user A searches for "smartphones" and "high-end watches", and his search interests match the high purchasing power characteristics of business district 1; user A searches for "discounted clothes", and his search interests match the medium purchasing power characteristics of business district 2.

[0069] Optionally, S3 includes: S31, performing feature diversion according to the user behavior feature data to obtain user behavior feature diversion data; In one embodiment, on a streaming platform, the user's behavior feature data includes viewing time, viewing category, interaction frequency, etc. The system divides the features according to the behavior type, for example, viewing time is divided into "viewing behavior features"; viewing categories are divided into "interest classification features"; interaction frequency is divided into "interaction behavior features".

[0070] S32, performing feature aggregation according to the user behavior feature diversion data to obtain user behavior feature aggregation data; In one embodiment, the traffic data is aggregated by category to evaluate the user's behavior characteristics as a whole. For example, the user's "viewing behavior characteristics" are aggregated into the total viewing time and viewing category distribution in the past week; the user's "interaction behavior characteristics" are aggregated into the average number of likes and comments per day.

[0071] S33, performing feature change calculation on the user behavior feature aggregation data to obtain user behavior feature change data; In one embodiment, the behavior characteristics of the current time period are compared with historical aggregated data to capture changes in user behavior. For example, the current user's viewing time in the "Drama" category has increased from 40% in the past to 60%; the number of daily comments by the user has increased from an average of 3 times to 5 times.

[0072] S34, performing change point detection according to the user behavior feature change data to obtain user behavior change data; In one embodiment, significant behavior changes are identified by analyzing feature change data. For example, when the percentage of watching "drama" increases by more than a preset threshold (such as 20%), it is marked as interest transfer; when the user's interactive behavior increases significantly, it is marked as high engagement.

[0073] S35. Update the model according to the user behavior change data and the preset user interest model to obtain a real-time user interest model.

[0074] In one embodiment, the interest model is updated by linear weighting in combination with the user behavior change data. For example, the user's real-time interest model changes from a weight mainly based on "comedy" to a weight mainly based on "drama"; the system reduces the frequency of recommending content in the "comedy" category and increases the frequency of recommending content in the "drama" category.

[0075] Optionally, the feature split is specifically: Perform feature stratification according to user behavior feature data to obtain user behavior feature stratification data; In one embodiment, on an e-commerce platform, user behavior feature data includes click records, browsing time, search keywords, etc. The system stratifies the data according to the time dimension, such as short-term behavior, click and browse data in the past 24 hours; long-term behavior, click, search and browse statistics in the past 30 days.

[0076] Performing feature classification on the user behavior feature hierarchical data to obtain user behavior feature classification data; In one embodiment, the system classifies hierarchical data according to feature types, such as interactive features, number of clicks, number of likes, etc.; interest features, search keywords, browsing categories, etc.; environmental features, geographic location, communication status, etc.

[0077] Perform advertising scene feature weighting on user behavior feature classification data to obtain feature classification weighted data; In one embodiment, different types of features are weighted in combination with advertising scenarios (such as during e-commerce promotions). During shopping festivals, interest features have a higher weight; during ordinary periods, interactive features have a higher weight. For example, the user's interest weight in the "smartphone" category is increased to 0.8; the user's click weight is set to 0.6.

[0078] Perform multi-channel feature segmentation based on feature classification weighted data to obtain feature diversion channel data; In one embodiment, data is divided into different channels according to weighted features: Channel 1: advertisement recommendation based on interest features; Channel 2: advertisement optimization based on interactive features.

[0079] The intra-channel features are reconstructed based on the feature diversion channel data to obtain the user behavior feature diversion data.

[0080] In one embodiment, in each channel, features are reconstructed to generate targeted data. In the interest feature channel, the trend of user interest changes in the "smartphone" category is extracted; in the interactive feature channel, user activity time periods (such as click peaks) are generated.

[0081] Optionally, S4 includes: S41, screening advertisement content through a preset advertisement content pool according to user behavior data and a user real-time interest model to obtain a screened advertisement pool; In one embodiment, in an e-commerce platform, the advertising content pool contains advertisements of multiple categories, such as electronic products, household goods, clothing, etc. User behavior data shows that the user has recently clicked and browsed multiple smartphones, and the search keywords include "smartphone" and "accessories". The user's real-time interest model shows that the user has a high interest weight in the "electronic products" category. Based on these data, advertisements related to "electronic products" and "smartphones" are screened from the advertising content pool.

[0082] S42, performing advertisement similarity calculation based on the screened advertisement pool, user behavior data, and user real-time interest model to obtain advertisement similarity data; In one embodiment, the system performs similarity matching between the advertisement content of the filtered advertisement pool and the user behavior and interest model, and the user's interest weight for the "smartphone" category is higher than that for "accessories". The matching degree between each advertisement and the user's interest model is calculated, for example, the similarity score of "smartphone advertisement" is higher than that of "accessories advertisement".

[0083] S43, sorting the screened advertisement pool according to the advertisement similarity data to obtain sorted advertisement data; In one embodiment, the advertisements in the screened advertisement pool are sorted according to the advertisement similarity scores (sorted from high to low according to the similarity scores, and combined with advertisement priority factors (such as advertisement budget, promotion strategy, etc.)), and the advertisements are arranged from high to low according to the similarity scores to ensure that advertisements with higher matching degree are pushed to users first.

[0084] S44, calculating the push timing according to the sorted advertisement data to obtain push timing data; In one embodiment, the best time for advertising push is determined by combining user behavior data and real-time environmental characteristics (the user's high-activity time period is obtained based on the user's click and browsing behavior statistics). User behavior data shows that the user's click and browsing active time is 18:00-20:00 in the evening. User environmental characteristics indicate that the current network signal is stable and suitable for pushing multimedia advertising content.

[0085] S45. Pack the advertisement content according to the push timing data and the sorted advertisement data to obtain the user's real-time advertisement content data, so as to perform a real-time user behavior advertisement push operation.

[0086] In one embodiment, the sorted advertising data is combined with the push timing data to generate an advertising content package suitable for the user. The push time information is attached to each piece of advertising content; if the user uses a mobile device, the mobile-adapted advertising material is generated, such as optimizing the image size and loading speed. At the specified push time, the advertising content is sent to the user's device through the message push service. For example, at 18:00 in the evening, the user is shown an advertisement for "the latest 5G smartphone discounts"; at 18:30 in the evening, the user is shown an advertisement for "smartphone case specials".

[0087] Optionally, the push timing calculation includes: Get user ambient sound data; In one embodiment, in a mobile e-commerce application, the user device collects ambient sound data in real time through a microphone. The current ambient sound of user A is "background music" with a low volume; the current ambient sound of user B is "noisy human voice" with a high volume. These data are recorded in the form of audio samples and aligned with the user behavior timeline.

[0088] Extracting features from the user ambient sound data to obtain user ambient sound feature data; In one embodiment, features are extracted by analyzing the collected ambient sound data, including volume features, where the ambient sound volume of user A is at a low level, while the volume of user B is at a high level; audio type features, where the ambient sound of user A is continuous and smooth background music, while the ambient sound of user B is intermittent noise; and audio change features, where the ambient sound of user A changes smoothly, while the ambient sound of user B fluctuates greatly.

[0089] Evaluate the attractiveness of advertisements based on the ranked advertisement data and the user's ambient sound feature data to obtain ambient sound weighted advertisement list data; In one embodiment, the attractiveness of the advertisement in the current environment is evaluated through weight calculation in combination with the characteristics of the advertisement content and the characteristics of the user's ambient sound. For user A in a "low-volume background music" environment, it is more suitable to push detailed introduction advertisements, such as "smartphone review videos"; for user B in a "noisy human voice" environment, it is more suitable to push short, image-based advertisements, such as "smartphone discount banners."

[0090] The push timing is predicted based on the ambient sound weighted advertisement list data to obtain the push timing data.

[0091] In one embodiment, the best time to push an ad is predicted by combining the user's active time and the ambient sound characteristics, and calculating based on a regression function. User A's current environment is relatively quiet, and his behavior record shows that 18:00-19:00 is an active time period, so 18:10 is suitable for push; User B's current environment is noisy, and his behavior record shows that 20:00 is his peak click time, so 20:00 is a suitable push time.

[0092] Optionally, the present application further provides an intelligent push strategy system for real-time user behavior monitoring, which is used to execute the intelligent push strategy method for real-time user behavior monitoring as described above, and the intelligent push strategy system for real-time user behavior monitoring includes: A user behavior data collection module is used to obtain user behavior data, wherein the user behavior data includes real-time user behavior data and external user behavior data. The real-time user behavior data includes user click record data, user browsing time data, user search query data, and user interaction history data. The external user behavior data includes user device data, user geographic location data, and user communication environment data; A user behavior feature extraction module is used to extract features based on user behavior data to obtain user behavior feature data; A user real-time interest update module is used to capture user behavior changes based on user behavior feature data to obtain user behavior change data, and to update the model based on the user behavior change data and a preset user interest model to obtain a user real-time interest model; The user real-time advertising content generation module is used to generate advertising content according to user behavior data and user real-time interest model, obtain user real-time advertising content data, and perform real-time user behavior advertising push operations.

[0093] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0094] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An intelligent push strategy method for real-time user behavior monitoring, characterized in that: The method comprises: S1. Obtain user behavior data, wherein the user behavior data includes real-time user behavior data and external user behavior data. The real-time user behavior data includes user click record data, user browsing time data, user search query data, and user interaction history data. The external user behavior data includes user device data, user geographic location data, and user communication environment data. S2. Extract features based on the user behavior data to obtain user behavior feature data; S3, capturing user behavior changes according to the user behavior feature data to obtain user behavior change data, and updating the model according to the user behavior change data and a preset user interest model to obtain a user real-time interest model; S4. Generate advertising content based on user behavior data and the user's real-time interest model to obtain user real-time advertising content data to perform real-time user behavior advertising push operations.

2. The method according to claim 1, characterized in that S1 includes: Collect users' click records, browsing time, search queries and interaction history in real time to obtain real-time user behavior data; Use device API, GPS positioning service and network status detection tools to obtain user device information, geographic location and communication environment data, and obtain external user behavior data; Integrate user real-time behavior data and external user behavior data to obtain user behavior data.

3. The method according to claim 1, characterized in that The user behavior feature data includes first user behavior feature data and second user behavior feature data, and S2 includes: Extract click count features and geographic location features based on user behavior data to obtain click count feature data and geographic location feature data; Perform feature combination according to the click count feature data and the geographic location feature data to obtain initial cross feature data; Perform multi-feature expansion based on the initial cross-feature data to obtain first user behavior feature data; Extract search keyword features and browsing time features from user behavior data to obtain search keyword feature data and browsing time feature data respectively; Gaussian mixture clustering is performed based on the click count feature data, search keyword feature data, and browsing time feature data to obtain user behavior clustering data; Extract key topics from user behavior clustering data to obtain user interest topic data; Calculate purchase intention based on user interest topic data to obtain user purchase intention data; The user portrait is structured according to the user behavior grouping data, the user interest topic data and the user purchase intention data to obtain the second user behavior feature data.

4. The method according to claim 3, characterized in that The specific feature extraction of click count is as follows: Perform click distribution based on user behavior data to obtain click distribution feature data; Perform click frequency analysis based on click distribution characteristic data to obtain click frequency characteristic data; Click behavior clustering is performed according to the click frequency characteristic data to obtain click behavior clustering data; Get user ambient sound data; Extracting the user ambient sound change feature and the user ambient sound decibel feature according to the user ambient sound data to obtain the user ambient sound change feature data and the user ambient sound decibel feature data; Mapping the click behavior clustering data according to the user ambient sound change characteristic data and the user ambient sound decibel characteristic data to obtain click behavior clustering ambient sound association data; Perform internal clustering on the click behavior clustering ambient sound associated data to obtain ambient sound click behavior clustering data; Feature cross-calculation is performed based on the ambient sound click behavior clustering data to obtain click count feature data.

5. The method according to claim 3, characterized in that: The specific extraction of geographic location features is as follows: According to the user's geographic location data and user click record data, click location aggregation is performed to obtain location click distribution data; Performing location-behavior association on the location-click distribution data to obtain location-click-behavior association data; Calculate the location purchasing power based on the location click behavior association data to obtain the location purchasing power data; Performing location classification on the location purchasing power data, the user location data, and the user click record data to obtain location classification data; The location classification data is cross-analyzed based on the user search query data to obtain the location feature data.

6. The method according to claim 1, characterized in that S3 includes: Perform feature diversion according to user behavior feature data to obtain user behavior feature diversion data; Perform feature aggregation based on the user behavior feature diversion data to obtain user behavior feature aggregation data; Perform feature change calculation on the aggregated data of user behavior features to obtain user behavior feature change data; Perform change point detection based on the user behavior feature change data to obtain the user behavior change data; The model is updated according to the user behavior change data and the preset user interest model to obtain the user's real-time interest model.

7. The method according to claim 6, characterized in that The specific feature splitting is: Perform feature stratification according to user behavior feature data to obtain user behavior feature stratification data; Performing feature classification on the user behavior feature hierarchical data to obtain user behavior feature classification data; Perform advertising scene feature weighting on user behavior feature classification data to obtain feature classification weighted data; Perform multi-channel feature segmentation based on feature classification weighted data to obtain feature diversion channel data; The intra-channel features are reconstructed based on the feature diversion channel data to obtain the user behavior feature diversion data.

8. The method according to claim 1, characterized in that: S4 includes: Screening advertisement content through a preset advertisement content pool according to user behavior data and a user real-time interest model to obtain a screened advertisement pool; Ad similarity calculation is performed based on the screened ad pool, user behavior data, and user real-time interest model to obtain ad similarity data; Sorting the filtered advertisement pool according to the advertisement similarity data to obtain sorted advertisement data; Calculate the push timing according to the sorted advertisement data to obtain push timing data; Advertisement content is packaged according to push timing data and sorted advertisement data to obtain user real-time advertisement content data for real-time user behavior advertisement push operations.

9. The method according to claim 8, characterized in that The push timing calculation includes: Get user ambient sound data; Extracting features from the user ambient sound data to obtain user ambient sound feature data; Evaluate the attractiveness of advertisements based on the ranked advertisement data and the user's ambient sound feature data to obtain ambient sound weighted advertisement list data; The push timing is predicted based on the ambient sound weighted advertisement list data to obtain the push timing data.

10. An intelligent push strategy system for real-time user behavior monitoring, characterized in that: The intelligent push strategy method for real-time user behavior monitoring according to claim 1, wherein the intelligent push strategy system for real-time user behavior monitoring comprises: A user behavior data collection module is used to obtain user behavior data, wherein the user behavior data includes real-time user behavior data and external user behavior data. The real-time user behavior data includes user click record data, user browsing time data, user search query data, and user interaction history data. The external user behavior data includes user device data, user geographic location data, and user communication environment data; A user behavior feature extraction module is used to extract features based on user behavior data to obtain user behavior feature data; A user real-time interest update module is used to capture user behavior changes based on user behavior feature data to obtain user behavior change data, and to update the model based on the user behavior change data and a preset user interest model to obtain a user real-time interest model; The user real-time advertising content generation module is used to generate advertising content according to user behavior data and user real-time interest model, obtain user real-time advertising content data, and perform real-time user behavior advertising push operations.

Images