Intelligent pushing method and device based on multi-dimensional user portrait and scene perception

Abstract

The invention discloses an intelligent pushing method and device based on multi-dimensional user portraits and scene perception, and relates to the technical field of Internet, and the method comprises the steps: collecting and obtaining user multi-dimensional data, and carrying out privacy protection processing; constructing a multi-dimensional user portrait based on the collected user multi-dimensional data, and dynamically updating a portrait label weight through a machine learning model; acquiring current scene data of a user terminal in real time, and converting the scene data into a scene label; based on the user portrait and the current scene label of the user terminal, a dynamic weight algorithm is adopted to calculate a push priority, and a push priority list corresponding to the current scene label is generated; and screening the to-be-pushed queue based on the pushing priority list, and performing real-time pushing. The invention provides a message pushing technology which can combine multi-dimensional features of the user with a real-time scene, dynamically optimize a pushing strategy and consider both accuracy and privacy protection, so as to solve the defects in the prior art.

Description

Technical Field

[0001] This invention relates to the field of Internet technology, and in particular to an intelligent push method, device, terminal equipment, and storage medium based on multi-dimensional user profiles and scene awareness. Background Technology

[0002] With the rapid development of mobile internet technology, various applications (APPs) have become the main carriers for users to obtain information, engage in social interactions, and make purchases. Push notifications, as a core means of connecting applications and users, directly impact user activity and commercial value.

[0003] However, existing push notification technology has the following significant drawbacks: 1) Insufficient accuracy: Most push systems generate push content based only on basic user tags (such as age, region) or single behavioral data (such as historical clicks), without considering the user's real-time scenario (such as geographical location, device status, and usage time) and dynamic changes in needs. This results in push content not matching the user's current interests, and a large number of invalid pushes causing user resentment.

[0004] 2) Inappropriate push timing: Existing technologies mostly use fixed time intervals (such as 8 am and 8 pm every day) or trigger-based push (such as when users log in), without analyzing the "golden interaction time" in user behavior habits, such as pushing long content during fragmented time such as users' commuting and lunch break, or pushing high-frequency content during users' work hours, which reduces users' willingness to open the push.

[0005] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0006] To address the technical problems of low accuracy and unreasonable timing in existing message push technologies, this invention provides an intelligent push method, device, terminal equipment, and storage medium based on multi-dimensional user profiles and scene awareness. This invention provides a message push technology that can combine multi-dimensional user characteristics with real-time scenes, dynamically optimize push strategies, and balance accuracy and privacy protection, thereby overcoming the shortcomings of existing technologies.

[0007] The technical solution of this application is as follows: An intelligent push method based on multi-dimensional user profiles and scene awareness, comprising: Collect and acquire multidimensional user data, including basic information data, historical interaction data, and device status data, and perform privacy protection processing; Multi-dimensional user profiles are constructed based on collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and the profile tag weights are dynamically updated through machine learning models. Real-time acquisition of current scene data from user terminals, and conversion of the scene data into scene tags; Based on the user profile and the current scene tag of the user terminal, a dynamic weighting algorithm is used to calculate the push priority and generate a push priority list corresponding to the current scene tag; The system filters the queue of users to be pushed based on the priority list, pushes data in real time, monitors the push status in real time, and feeds back the data to the constructed multi-dimensional user profile to update the user profile.

[0008] The intelligent push method based on multi-dimensional user profiles and scene awareness, wherein the step of calculating push priority based on the user profile and the current scene tag of the user terminal using a dynamic weight algorithm, and generating a push priority list corresponding to the current scene tag includes: A content-profile matching model is pre-built. Based on the matching degree between the user profile's interest tags and the keywords of the pushed content, a basic weight is assigned. A scenario-timing adaptation model is pre-built. Based on the scenario tags, the push timing weight that matches the scenario tags is dynamically calculated. By combining the base weight and the timing weight, a push priority list is generated, and only content with a predetermined priority value is selected to enter the push queue.

[0009] The intelligent push method based on multi-dimensional user profiles and scene awareness includes the following steps: collecting and acquiring multi-dimensional user data, including basic information data, historical interaction data, and device status data, and performing privacy protection processing. The system collects basic user information, historical interaction data, and device status data to obtain multidimensional user data. The collected multidimensional user data is then synchronized in real time to the privacy protection module for encryption. The user basic information includes: such as age, gender, registration time and / or data obtained with user authorization; the historical interaction data includes message opening records, clicked content, dwell time and / or unfollowing behavior data; the device status data includes device model, network type, battery data and / or non-privacy data.

[0010] The intelligent push method based on multi-dimensional user profiles and scene awareness, wherein the step of acquiring the current scene data of the user terminal in real time and converting the scene data into scene tags includes: Real-time acquisition of current scene data of user terminals, including geographical location, time scene, and device scene, and conversion of scene data into scene tags; The geographic location is a fuzzy location based on user authorization, not precise coordinates, and includes whether the user is in a shopping mall or office; the time scenario includes weekdays, weekends, commuting hours, and / or rest periods; and the device scenario includes network type and device battery level.

[0011] The intelligent push method based on multi-dimensional user profiles and scene awareness includes the following steps: filtering the queue to be pushed based on a push priority list, performing real-time pushes, monitoring the push status in real-time, feeding back data to the constructed multi-dimensional user profile, and updating the user profile. The system filters the queue of items to be pushed based on a priority list and then pushes them in real time. The message format is adjusted according to the user terminal network type. When the user terminal network type is detected to be in a Wi-Fi environment, rich media messages with pictures and / or videos are pushed in real time. When the user terminal network type is detected to be in a 5G environment, simplified text messages are pushed. Monitor push status in real time, provide feedback data to the constructed multi-dimensional user profile, and update the user profile.

[0012] The intelligent push method based on multi-dimensional user profiles and scene awareness includes the following steps: collecting and acquiring multi-dimensional user data, including basic information data, historical interaction data, and device status data, and performing privacy protection processing. The collected user multidimensional data was anonymized. A three-tier access control system was established: ordinary employees can only access the anonymized profile tags, administrators need to approve before they can view the original data, and third-party partners have no data access permissions.

[0013] The intelligent push method based on multi-dimensional user profiles and scene awareness, wherein, after the steps of filtering the queue to be pushed based on the push priority list, performing real-time push, monitoring the push status in real time, feeding back data to the constructed multi-dimensional user profile, and updating the user profile, it further includes: Regularly clean up non-core interaction data that has exceeded a specified time.

[0014] An intelligent push notification device based on multi-dimensional user profiles and scene awareness, wherein the device includes: The user data collection module is used to collect and acquire multi-dimensional user data, including basic information data, historical interaction data, and device status data, and to perform privacy protection processing. The user profile building module is used to build multi-dimensional user profiles based on collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and dynamically update profile tag weights through machine learning models; The scene perception module is used to acquire the current scene data of the user terminal in real time and convert the scene data into scene tags; The push strategy generation module is used to calculate the push priority based on the user profile and the current scene tag of the user terminal using a dynamic weight algorithm, and generate a push priority list corresponding to the current scene tag. The message push execution module is used to filter the queue to be pushed based on the push priority list, push in real time, monitor the push status in real time, feed back data to the constructed multi-dimensional user profile, and update the user profile.

[0015] A terminal device includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs including steps for performing any of the methods described herein.

[0016] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it enables an electronic device to perform the steps of any of the methods described.

[0017] As can be seen from the above, the present application provides an intelligent push method, device, terminal equipment and storage medium based on multi-dimensional user profile and scene awareness. The present invention achieves accurate matching between push content and user needs, optimizes push timing, reduces resource consumption, and ensures user data security. It improves push accuracy, effectively ensures more reasonable push timing, and provides convenience for users. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the intelligent push method based on multi-dimensional user profiles and scene awareness according to Embodiment 1 of the present invention.

[0020] Figure 2 This is a schematic diagram of the overall framework of the intelligent push method based on multi-dimensional user profiles and scene awareness in Embodiment 2 of the present invention.

[0021] Figure 3 The present invention provides a schematic diagram of the principle of an intelligent push device embodiment based on multi-dimensional user profiles and scene awareness.

[0022] Figure 4 This is a block diagram illustrating the internal structure of the terminal device provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0024] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0025] With the development of technology and the continuous improvement of people's living standards, various applications (APPs) have become the main carriers for users to obtain information, conduct social interactions, and make purchases. Push notifications, as a core means of connecting applications and users, directly impact user activity and commercial value. However, existing push notification technologies have the following significant drawbacks: 1) Insufficient accuracy: Most push systems generate push content based only on basic user tags (such as age, region) or single behavioral data (such as historical clicks), without considering the user's real-time scenario (such as geographical location, device status, and usage time) and dynamic changes in needs. This results in push content not matching the user's current interests, and a large number of invalid pushes causing user resentment.

[0026] 2) Inappropriate push timing: Existing technologies mostly use fixed time intervals (such as 8 am and 8 pm every day) or trigger-based push (such as when users log in), without analyzing the "golden interaction time" in users' behavior habits. For example, pushing long content during fragmented time such as users' commuting and lunch break, or pushing high-frequency content during users' work hours, reduces users' willingness to open the push.

[0027] 3) Serious waste of resources: Due to the lack of accurate prediction of user needs, the server needs to send messages to a large number of potential users who have no needs. This not only consumes network bandwidth and server storage resources, but may also lead to a further decline in the effective message delivery rate of the application due to users frequently turning off push notifications.

[0028] 4) Lack of user privacy protection: In order to improve accuracy, some push systems excessively collect user privacy data (such as browsing history and location trajectory) and have not established a sound data encryption and access control mechanism, which poses a risk of data leakage and does not comply with the requirements of the Personal Information Protection Law and other regulations.

[0029] Therefore, there is an urgent need for a message push technology that can combine multi-dimensional user characteristics with real-time scenarios, dynamically optimize push strategies, and balance accuracy and privacy protection to address the shortcomings of existing technologies.

[0030] Therefore, in order to solve the above-mentioned technical problems, this invention provides an intelligent push method based on multi-dimensional user profiles and scene awareness, as described in the following embodiments.

[0031] Example 1 like Figure 1 As shown in the figure, an intelligent push method based on multi-dimensional user profiles and scene awareness according to an embodiment of the present invention includes the following steps: Step S100: Collect and obtain multi-dimensional user data, including basic information data, historical interaction data and device status data, and perform privacy protection processing. The multidimensional data in this embodiment refers to comprehensive data that characterizes user features from different dimensions. Unlike single-dimensional data, it can comprehensively reflect user attributes, behaviors, and environmental conditions.

[0032] The basic information data in this embodiment refers to the user's static basic attribute data, which has the characteristics of high stability and low change frequency, and is the basis for building user profiles.

[0033] The historical interaction data in this embodiment refers to the records of past interactions between the user and the application, which can reflect the user's long-term interests and behavioral patterns. The device status data refers to the real-time operating status and related information of the user's terminal device, which can indirectly reflect the user's usage scenarios and behavioral habits.

[0034] The privacy protection measures implemented in this embodiment refer to the security measures taken throughout the entire process of data collection, storage, and use, which aim to prevent the leakage and misuse of user privacy data and comply with relevant laws and regulations.

[0035] In the specific implementation of this step, firstly, three types of core data can be collected synchronously through legitimate channels such as application authorization interfaces, device sensors, and server logs: basic information data is used to locate the user's basic attributes, historical interaction data is used to mine the user's long-term preferences, and device status data is used to capture the user's real-time status; among them, the basic information data includes user age data, gender data, registration time data, etc., which are obtained with the user's authorization; the historical interaction data includes message opening records, clicked content, dwell time, unfollowing behavior data, etc.; and the device status data includes device model, network type, battery level, and non-privacy data, etc.

[0036] Subsequently, the collected data is cleaned, deduplicated, and standardized to remove invalid data and outliers. Finally, privacy protection measures are implemented to ensure that data use complies with the Personal Information Protection Law and to avoid the risk of privacy leaks.

[0037] For example, when a shopping app executes step S100, it collects basic user information, such as age (35 years old), city (specific city), and anonymized registered phone number; it also collects historical interaction data, such as browsing home appliance products 12 times in the past 3 months, ordering one refrigerator, adding 3 smart home appliances to favorites, and deleting promotional push notifications twice; and it collects device status data, such as currently using an Android phone, device battery level (68%), network mode (Wi-Fi), and location permissions enabled. Then, a privacy protection process is implemented, including hashing and encrypting the phone number, obfuscating the location information (retaining only the city level, omitting specific street addresses), clearly informing the user of the data collection purpose and obtaining authorization, and storing the data using encrypted storage.

[0038] This step, through multi-dimensional data collection, overcomes the limitations of single-data sources, providing a comprehensive data foundation for subsequent precise profiling. Simultaneously, data cleaning improves data quality, preventing invalid data from interfering with subsequent analysis. Furthermore, this invention employs privacy protection processing to address the lack of privacy protection in existing technologies, reducing the risk of data leakage, complying with regulatory requirements, and increasing user trust in the application. Moreover, this invention preprocesses data, reducing computational costs in subsequent profiling and push decision-making stages, and improving the overall efficiency of the technical process.

[0039] Step S200: Construct a multi-dimensional user profile based on the collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and dynamically update the profile tag weights through a machine learning model; The multi-dimensional user profile in this embodiment refers to a virtual user model built based on multi-dimensional data. It comprehensively depicts user interests, behaviors, needs, and other characteristics through a hierarchical tagging system, differing from a single-tag profile in that it is comprehensive and dynamic. The interest tagging layer is the core level of the user profile, used to label content areas, product types, and service directions that the user is interested in, such as home appliance enthusiasts or science fiction readers.

[0040] In this embodiment, the behavior habit layer refers to marking the patterns of user behavior when using the application, including usage time, interaction methods, and dwell time, reflecting user behavioral preferences. The demand priority layer refers to prioritizing potential user needs based on historical user behavior and attributes, clarifying the type of demand that the user is most likely to respond to at present.

[0041] The machine learning model in this embodiment refers to an algorithmic model used to analyze user data and update tag weights, such as logistic regression and collaborative filtering models, which can automatically learn the patterns of user behavior changes and achieve dynamic optimization of user profiles.

[0042] In this embodiment, the tag weight refers to the numerical value that measures the importance of each tag in characterizing the user's features. The higher the weight, the better the corresponding tag reflects the user's current core features.

[0043] In the specific implementation of this step S200, firstly, based on the multidimensional data processed in step S100, user profiles are constructed in layers, including: the interest tag layer, which extracts core interests by analyzing historical interaction data, such as tagging home appliances if a user frequently browses home appliances; the behavior habit layer, which summarizes patterns by mining user usage records, such as tagging evening activity if the user uses the APP from 8-9 pm every day; and the demand priority layer, which sorts demands by combining basic information and historical behavior, such as prioritizing home appliance purchases over clothing purchases for users who have just renovated their houses.

[0044] Subsequently, the hierarchical tags are integrated into a complete user profile. Machine learning models are used to analyze newly generated user interaction data in real time and dynamically adjust the weight of each tag. For example, if a user no longer browses home appliances, the weight of the home appliance tag is reduced; if a user frequently browses maternity and baby products, the weight of the maternity and baby tag is increased, ensuring that the profile matches the user's latest status.

[0045] For example, based on the shopping app data from step S100, a user profile is constructed. The interest tag layer assigns a weight of 0.7 to smart home appliances and 0.3 to kitchen appliances; the behavior habit layer assigns a weight of 0.8 to activity between 8-9 PM and 0.6 to usage in a Wi-Fi environment; the demand priority layer prioritizes smart home appliance purchases (priority 1), home appliance after-sales consultations (priority 2), and kitchen appliance recommendations (priority 3). One week later, if frequent browsing of baby and maternity products is detected, the machine learning model automatically updates the tags: a new tag for baby and maternity products with a weight of 0.6 is added, the weight of the smart home appliance tag is reduced to 0.4, and the demand priority layer is adjusted to prioritize baby and maternity product purchases (priority 1), thus achieving dynamic profile updates.

[0046] As can be seen, the hierarchical tagging system in this embodiment overcomes the limitations of existing technologies that rely solely on basic tags or single behavioral data, characterizing user features from multiple dimensions and improving the matching degree between pushed content and user needs. Furthermore, this invention uses a machine learning model to dynamically update tag weights, enabling real-time capture of changes in user needs and avoiding ineffective push notifications caused by fixed user profiles, thus solving the problem of insufficient accuracy in existing technologies. Moreover, the demand priority layer of this invention provides a clear basis for subsequent push ordering, reducing blind push notifications and improving push efficiency.

[0047] Step S300: Acquire the current scene data of the user terminal in real time, and convert the scene data into scene tags; The terminal current scene data in this embodiment refers to the real-time environment and status data obtained when the user uses the terminal device. It can directly reflect the user's current scene and behavioral intentions, and is different from historical static data.

[0048] The scene tags refer to the results of standardizing and tagging real-time scene data, which are used to quickly associate user scenarios with push content, such as commuting scenarios and lunch break scenarios.

[0049] The core of this step is to achieve real-time scene perception and scene data transformation, providing support for push strategies to adapt to the current scene and solving the deficiency of existing technologies that ignore the user's real-time scene. In the specific implementation of step S300, firstly, scene data is collected in real time through the user terminal's positioning module, sensors, network status monitoring, and other functions. The core data includes geographical location data (e.g., subway station, office), time data (e.g., 10 AM on weekdays), network status (e.g., 4G / 5G, Wi-Fi), and device usage status (e.g., screen locked, background running). Subsequently, the collected real-time scene data is quickly parsed and classified, transforming abstract scenes into standardized scene tags to ensure that the tags accurately represent the user's current state. Simultaneously, the real-time nature of data acquisition is guaranteed; the delay control of this invention can be achieved at the second level, adapting to dynamic changes in the user's scene.

[0050] For example, when a news app executes step S300, it acquires real-time user terminal data, including: geolocation detection showing a uniform change in movement trajectory to Metro Line 2, time 7:30 AM on a weekday, network status 4G, device screen on, and the app running in the foreground. Based on this data, the scenario is tagged as a morning rush hour commuting scenario (metro). If, one hour later, the user's geolocation changes to an office building, the network switches to office Wi-Fi, and the time is 8:30 AM, the scenario tag is quickly updated to "office scenario" during the workday.

[0051] As can be seen, through real-time scene awareness and tagging in this step, the pushed content can be matched with the user's current scene, solving the problem of inaccurate push notifications caused by existing technologies not considering real-time scenes. Furthermore, the second-level scene parsing and tag conversion of this invention can quickly adapt to changes in user scenarios, ensuring real-time adjustments to the push strategy and improving user experience. Moreover, this invention uses standardized scene tags to reduce the complexity of subsequent push priority calculations, improving the efficiency of push decision-making.

[0052] Step S400: Based on the user profile and the current scene tag of the user terminal, a dynamic weight algorithm is used to calculate the push priority and generate a push priority list corresponding to the current scene tag; The dynamic weighting algorithm in this embodiment dynamically adjusts the weight ratio of each tag based on real-time changes in user profile tags and scenario tags. Unlike fixed weighting algorithms, it can adapt to the differences in user needs in different scenarios. The push priority in this embodiment is an indicator that measures the attractiveness and suitability of the pushed content to the user at present; the higher the priority, the more frequently it is pushed to the user.

[0053] The push priority list is a list formed by sorting the content to be pushed from high to low priority, providing a clear basis for subsequent push filtering.

[0054] This step is the core of the push notification decision-making process, enabling push notification ranking driven by both user profiles and scenarios, thus addressing the issue of inappropriate timing and content in existing technologies. Specifically, in step S400, the multi-dimensional user profile constructed in step S200 and the current scenario tags generated in step S300 are retrieved to clarify the user's core characteristics and current scenario needs. Then, a dynamic weighting algorithm is used to adjust the weight ratio of each profile tag according to the current scenario; for example, in a commuting scenario, the weight of the short-content preference behavior tag is increased, while the weight of the in-depth reading tag is decreased. Next, the matching degree between the content to be pushed and the user profile tags and scenario tags is calculated, and combined with the dynamically adjusted weights, the push priority of each piece of content is determined. Finally, the content is sorted from highest to lowest priority to generate a push priority list corresponding to the current scenario tags, ensuring that the content in the list is adapted to the user's current scenario and core needs.

[0055] For example, based on the user profile and scene tags of the shopping app mentioned earlier, when the user's scene tag is detected as an evening home scenario, the dynamic weighting algorithm increases the proportion of smart home appliance interest tags (weight 0.7) and evening active behavior tags (weight 0.8). The content to be pushed includes smart home appliance promotional information, kitchenware recommendations, maternity and baby product coupons, and workplace course advertisements. After calculating the matching degree and priority, the generated list is: 1. Smart home appliance promotional information (priority 0.92); 2. Kitchenware recommendations (priority 0.65); 3. Maternity and baby product coupons (priority 0.58); 4. Workplace course advertisements (priority 0.21). If the user switches to an office scenario, the algorithm adjusts the weights, reducing the weight of home appliance-related tags and increasing the weight of tags such as efficient office tools, regenerating the priority list, and removing home-related push content.

[0056] As can be seen from the above, this invention employs a dual-dimensional approach of user profile and scenario-based optimization, combined with a dynamic weighting algorithm, to ensure that the pushed content matches both long-term user preferences and current scenario needs, significantly improving push accuracy and reducing invalid pushes. Furthermore, this invention uses scenario-based tagging to adjust push priority, avoiding pushing corresponding content in inappropriate scenarios, such as not pushing long content in an office setting, thus solving the problem of unreasonable push timing in existing technologies. Moreover, this invention clearly defines push priorities, allowing resources to be concentrated on pushing high-priority content, reducing invalid pushes of low-priority content, and minimizing resource waste.

[0057] Step S500: Filter the queue to be pushed based on the push priority list, push in real time, monitor the push status in real time, feed back the data to the constructed multi-dimensional user profile, and update the user profile.

[0058] The push queue in this embodiment refers to the set of content selected from the push priority list that meets the current push conditions (such as push frequency limits and content compliance), and is the content pool that will eventually be pushed to the user.

[0059] The push status refers to the entire process of pushing content, including whether the push was successful, whether the user clicked on it, whether the push was deleted, and whether the user stayed to read it. The feedback data refers to the user's interactive behavior data regarding the push content, which is the core basis for measuring push effectiveness and updating user profiles.

[0060] In the specific implementation of this step, firstly, based on the push priority list and combined with push frequency limits such as no more than 5 pushes per user per day and content compliance review, a push queue is selected, and high-priority content in the queue is pushed first. Then, content is pushed to the user's terminal in real time through the push interface, while simultaneously activating the status monitoring module to collect push status and user interaction feedback data in real time. Finally, the feedback data is synchronized to the multi-dimensional user profile constructed in step S200, triggering the machine learning model to update the profile tag weights again. For example, if a user clicks on a smart home appliance promotion push, the smart home appliance tag weight increases; if a baby product push is deleted, the baby product tag weight decreases, forming a closed-loop optimization mechanism.

[0061] For example, a shopping app, based on the priority list in step S400, filters the queue of items to be pushed, removes workplace course ads, retains the top 3 high-priority items that meet the daily push frequency limit of 3, and pushes smart home appliance promotions, kitchen utensil recommendations, and baby product coupons to users. The push status is monitored in real time: if a user clicks on the smart home appliance promotion and stays for 5 minutes, the kitchen utensil recommendation is ignored, and the baby product coupon is deleted. This feedback data is fed back into the user profile, and the machine learning model updates the tag weights: the smart home appliance tag weight is increased from 0.4 to 0.6, the kitchen utensil tag weight is decreased from 0.3 to 0.2, and the baby product tag weight is decreased from 0.6 to 0.3, completing the user profile update and providing optimized data support for the next push.

[0062] As can be seen, through the embodiments of this invention, a closed-loop mechanism of push-feedback-update is formed, ensuring that user profiles continuously align with users' latest behavioral habits and needs, gradually improving push accuracy and solving the problem of fixed user profiles in existing technologies. Furthermore, this invention uses priority filtering of the push queue to avoid invalid content pushes, reducing network bandwidth and server resource consumption, thus solving the problem of serious resource waste in existing technologies. Moreover, this invention employs real-time monitoring of push status to avoid high-frequency and invalid pushes, reducing user aversion, and simultaneously optimizes subsequent pushes based on feedback, improving user activity and application stickiness.

[0063] In a further embodiment of the present invention, the intelligent push method based on multi-dimensional user profiles and scene awareness, wherein step S400 includes: S401. Pre-build a content-profile matching model. Based on the matching degree between the user profile's interest tags and the keywords of the pushed content, assign basic weights using the pre-built content-profile matching model. S402. Pre-build a scenario-timing adaptation model. Based on the scenario tags, dynamically calculate the push timing weight that matches the scenario tags. S403. Based on the combined basic weight and timing weight, generate a push priority list, and select only the content with the highest priority value to enter the push queue.

[0064] In a detailed embodiment of the present invention, the push strategy generation uses user profiles and scene tags as inputs and employs a dynamic weighting algorithm to calculate the push priority. Specifically, it includes: A content-profile matching degree model is pre-constructed. This model is a quantitative analysis model based on machine learning, which takes multi-dimensional user profile tags and push content feature tags as inputs and trains the model to obtain a quantitative matching degree value. Specifically, the full amount of content data pushed by the application in history is collected first, along with the multi-dimensional profile data and interaction feedback data of the corresponding users. The push content is decomposed into content feature tags, such as the following tags for shopping apps: smart home appliances, promotions, short video explanations, and less than 50 characters; and the following tags for news apps: technology, short news, images and text, and less than 100 characters. User interaction results (clicks / stays / converted into positive samples, ignored / deleted as negative samples) are used as tags to construct a training sample set.

[0065] Then, model training and parameter optimization are performed. For example, gradient boosting tree (GBDT) or deep neural network (DNN) can be selected as the basic algorithm framework. The user profile labels (including weights) at each level and the content feature labels are used as the model input features, and whether the user generates effective interaction is used as the output target. The model is trained on the sample set. Through cross-validation and hyperparameter tuning, the model accuracy is optimized so that the model can output accurate matching degree values ​​based on the input features.

[0066] Then, the model is deployed and invoked in real time. The trained content-profile matching model is deployed to the push system algorithm engine and invoked in real time in the subsequent S400 steps. When calculating the push priority, the model automatically retrieves the user profile tags and the feature tags of the content to be pushed, and quickly outputs the matching degree value of the two as the core basic indicator for push priority calculation.

[0067] This invention establishes a content-profile matching model in advance, assigning a basic weight of 1-5 points based on the matching degree between user interest tags and push content keywords. For example, if a user prefers maternal and infant products, pushing baby formula promotional content will earn 5 points, while pushing digital products will earn 1 point.

[0068] In this embodiment of the invention, a scenario-timing adaptation model is also pre-built. The scenario-timing adaptation model is a machine learning model that quantifies and calculates the adaptation of different push timings based on user scenario characteristics and behavioral habits, such as specific time points, push duration, content presentation format, etc. It is designed to solve the problem of what scenario, at what time, and in what form to push.

[0069] The construction steps of the scenario-timing adaptation model include: 1) Feature system construction: Construct a dual-input feature system, one is scenario feature labels (including scenario type, time, network status, device status, etc., such as morning rush hour commuting, 4G network, screen on), and the other is user behavior habit features (including historical golden interaction time periods, content format preferences, single interaction duration, etc., such as 8-9 pm activity, short content preference, single interaction not exceeding 1 minute); At the same time, construct a push timing feature library, such as push time point: 7:30 am, 8:00 pm; content format: short image and text, plain text; push quantity: 1 push per time, 3 push per time).

[0070] 2) Then, sample training and peak time mining are carried out. User interaction data in different scenarios and at different push times are collected. The scenario feature labels and user behavior habit features are used as input, and the push time features and user effective interaction rate are used as output. K-means clustering + logistic regression algorithm can be used for model training. During the training process, the model automatically mines the peak interaction time and optimal push time features in different scenarios. For example, in the commuting scenario, the peak time is 7:00-8:30 am and 5:30-7:00 pm, and the optimal form is short text and image, one message per push.

[0071] The trained scenario-timing adaptation model is then deployed to the push system. After the scenario tags are generated in the S300, it can be called in real time. The model output includes two core results: first, the optimal push timing features under the current scenario and user behavior habits, including the recommended time point, content format, and number of pushes; second, the timing adaptation score between the selected push timing and the current scenario, which provides a quantitative basis for timing-level calculation of push priority in the S400 and push execution in the S500.

[0072] This invention uses a scenario-timing adaptation model to adjust the push timing weight based on scenario tags. For example, in a weekday-office-Wi-Fi scenario, pushes are prioritized during the user's historically active lunch break (e.g., 12:00-13:00), increasing the weight by 20%. In a scenario where the device battery is less than 20%, non-urgent messages are delayed, decreasing the weight by 50%.

[0073] Then, this invention combines the basic weight and the timing weight to generate a push priority list, and selects only the top 3 priority items to enter the push queue.

[0074] In a further embodiment of the present invention, the intelligent push method based on multi-dimensional user profiles and scene awareness, wherein step S100 specifically includes: The system collects basic user information, historical interaction data, and device status data to obtain multidimensional user data. The collected multidimensional user data is then synchronized in real time to the privacy protection module for encryption. The user basic information includes: such as age, gender, registration time and / or data obtained with user authorization; the historical interaction data includes message opening records, clicked content, dwell time and / or unfollowing behavior data; the device status data includes device model, network type, battery data and / or non-privacy data.

[0075] In this embodiment of the invention, a user data collection module can collect basic user information (such as age, gender, and registration time, obtained with user authorization), historical interaction data (such as message opening records, clicked content, dwell time, and unfollowing behavior), and device status data (such as device model, network type, and battery level, which are not private data). The data collection process is synchronized in real time to the privacy protection module for encryption processing.

[0076] In a further embodiment of the present invention, the intelligent push method based on multi-dimensional user profiles and scene awareness, wherein step S300 includes: S301. Acquire the current scene data of the user terminal in real time, including geographical location, time scene, and device scene, and convert the scene data into scene tags; The geographic location is a fuzzy location based on user authorization, not precise coordinates, and includes whether the user is in a shopping mall or office; the time scenario includes weekdays, weekends, commuting hours, and / or rest periods; and the device scenario includes network type and device battery level.

[0077] In this embodiment of the invention, a scene perception module can be set up to obtain the user's current scene data in real time, including geographical location (such as whether the user is in a shopping mall or office, based on user-authorized fuzzy positioning, not precise coordinates), time scene (such as weekday / weekend, commuting time / rest time), and device scene (such as network type Wi-Fi / 5G, device battery ≥50% / <20%). The scene data is then converted into scene tags, such as weekday-office-Wi-Fi, weekend-shopping mall-5G scene.

[0078] In a further embodiment of the present invention, the intelligent push method based on multi-dimensional user profiles and scene awareness, wherein step S500 includes: S501: Filter the queue to be pushed based on the push priority list and push in real time; S502. Adjust the message format according to the user terminal network type. When the user terminal network type is detected to be in a Wi-Fi environment, control the real-time push of rich media messages with pictures and / or videos. When the user terminal network type is detected to be in a 5G environment, control the push of simplified text messages. S503: Monitor push status in real time, provide feedback data to the constructed multi-dimensional user profile, and update the user profile.

[0079] In this embodiment of the invention, the push strategy generation module receives the push queue, adjusts the message format according to the user device network type, such as detecting when the user terminal device pushes rich media messages with pictures and / or videos in a Wi-Fi environment, and detecting when the user terminal device pushes simplified text messages in a 5G environment to save data, and monitors the push status in real time, such as whether it has been delivered, whether it has been opened, and whether it has been clicked, and sends the feedback data back to the user profile building module for updating the user profile.

[0080] In a further embodiment of the present invention, the intelligent push method based on multi-dimensional user profiles and scene awareness further includes the following steps: S601. De-identify the collected user multidimensional data; In this embodiment of the invention, the collected user data will be anonymized, such as converting the precise location into a region label and hiding the middle 4 digits of the phone number.

[0081] S602. Establish a three-level access control system. Ordinary employees can only access the de-identified profile tags, while administrators need to approve before they can view the original data. Third-party partners have no data access permissions.

[0082] In this embodiment of the invention, a three-level access control system will be established. For example, ordinary employees can only access the de-identified profile tags, administrators need to approve before they can view the original data, and third-party partners have no data access permissions.

[0083] S603 periodically cleans up non-core interactive data that has exceeded a specified time.

[0084] For example, regularly clean up non-core interaction data that is older than 6 months to avoid excessive data storage.

[0085] The present invention will be further described in detail below through specific application examples: This specific application embodiment presents an intelligent push method based on multi-dimensional user profiles and scene awareness, which is based on... Figure 2 The system framework shown, and the intelligent push method based on multi-dimensional user profiles and scene awareness described in this specific application embodiment, include the following steps: S11, the user equipment (user terminal equipment) agrees to the authorization and generates interaction / device data, and then proceeds to S11; S12. The user data collection module collects basic user information (such as age, gender, and registration time, which are obtained with user authorization), historical interaction data (such as message opening records, clicked content, dwell time, and unfollowing behavior), and device status data (such as device model, network type, and battery level, which are not privacy data). The authorized data is transmitted in encrypted form. That is, the data collection process is synchronized to the privacy protection module in real time for encryption processing, and then proceeds to S13. S13. After temporarily decrypting the data, transmit the data to the image module, and then proceed to S14; S14. Provide real-time scene data (location / time / device); that is, obtain the user's current scene data in real time, including geographical location (such as whether in a shopping mall or office, fuzzy positioning based on user authorization, not precise coordinates), time scene (such as weekday / weekend, commuting time / rest time), device scene (such as network type Wi-Fi / 5G, device battery ≥50% / <20%), and convert the scene data into scene tags; S15. Generate scene labels and transmit them, then proceed to S16; S16. Generate and transmit multi-dimensional user profiles, then proceed to S17; S17. Calculate the push priority, output the push strategy (timing / format), and then proceed to S18. S18. Push messages according to the strategy; then proceed to S19. S19. Generate interactive feedback (open / click); and proceed to S20; S20: Send back feedback data to update the user profile; and proceed to S21; S21. Adjust privacy settings as needed, and proceed to S22; S22. Provide de-identified data / permission audit results regularly.

[0086] Exemplary device like Figure 3 As shown, this embodiment of the invention provides an intelligent push device based on multi-dimensional user profiles and scene awareness. The device includes: The user data collection module 310 is used to collect and acquire multi-dimensional user data, including basic information data, historical interaction data and device status data, and to perform privacy protection processing. The user profile building module 320 is used to build multi-dimensional user profiles based on collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and dynamically update the profile tag weights through machine learning models. Scene perception module 330 is used to acquire the current scene data of the user terminal in real time and convert the scene data into scene tags; The push strategy generation module 340 is used to calculate the push priority based on the user profile and the current scene tag of the user terminal using a dynamic weight algorithm, and generate a push priority list corresponding to the current scene tag. The message push execution module 350 is used to filter the queue to be pushed based on the push priority list, push in real time, monitor the push status in real time, feed back data to the constructed multi-dimensional user profile, and update the user profile, as described above.

[0087] Based on the above embodiments, the present invention also provides a terminal device, which can be a smart TV, and its principle block diagram can be as follows. Figure 4 As shown. The terminal device includes a processor, memory, network interface, display screen, and database connected via a system bus.

[0088] The memory stores one or more programs configured to be executed by a processor to implement the intelligent push method based on multi-dimensional user profiles and scene awareness described in the above embodiments.

[0089] In this context, terminal devices refer to intelligent computers and similar devices with data processing capabilities. The memory can be internal memory, flash memory, hard disk, or cloud storage, used to store program code and various data, including collected multi-dimensional user data. The processor can be a central processing unit (CPU), used to execute the algorithmic logic within the program. The program includes intelligent push methods based on multi-dimensional user profiles and scene awareness.

[0090] In a further embodiment, a terminal device of this embodiment includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations: Collect and acquire multidimensional user data, including basic information data, historical interaction data, and device status data, and perform privacy protection processing; Multi-dimensional user profiles are constructed based on collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and the profile tag weights are dynamically updated through machine learning models. Real-time acquisition of current scene data from user terminals, and conversion of the scene data into scene tags; Based on the user profile and the current scene tag of the user terminal, a dynamic weighting algorithm is used to calculate the push priority and generate a push priority list corresponding to the current scene tag; The system filters the queue of devices to be pushed based on the priority list, pushes them in real time, monitors the push status in real time, feeds back the data to the constructed multi-dimensional user profile, and updates the user profile, as described above.

[0091] The step of calculating the push priority based on the user profile and the current scene tag of the user terminal using a dynamic weighting algorithm, and generating a push priority list corresponding to the current scene tag, includes: A content-profile matching model is pre-built. Based on the matching degree between the user profile's interest tags and the keywords of the pushed content, a basic weight is assigned. A scenario-timing adaptation model is pre-built. Based on the scenario tags, the push timing weight that matches the scenario tags is dynamically calculated. By combining the base weight and the timing weight, a push priority list is generated, and only content with a predetermined priority value is selected to enter the push queue.

[0092] The steps of collecting and acquiring multidimensional user data, including basic information data, historical interaction data, and device status data, and performing privacy protection processing include: The system collects basic user information, historical interaction data, and device status data to obtain multidimensional user data. The collected multidimensional user data is then synchronized in real time to the privacy protection module for encryption. The user basic information includes: such as age, gender, registration time and / or data obtained with user authorization; the historical interaction data includes message opening records, clicked content, dwell time and / or unfollowing behavior data; the device status data includes device model, network type, battery data and / or non-privacy data.

[0093] The step of acquiring the current scene data of the user terminal in real time and converting the scene data into scene tags includes: Real-time acquisition of current scene data of user terminals, including geographical location, time scene, and device scene, and conversion of scene data into scene tags; The geographic location is a fuzzy location based on user authorization, not precise coordinates, and includes whether the user is in a shopping mall or office; the time scenario includes weekdays, weekends, commuting hours, and / or rest periods; and the device scenario includes network type and device battery level.

[0094] The steps of filtering the queue to be pushed based on the push priority list, pushing in real time, monitoring the push status in real time, feeding back data to the constructed multi-dimensional user profile, and updating the user profile include: The system filters the queue of items to be pushed based on a priority list and then pushes them in real time. The message format is adjusted according to the user terminal network type. When the user terminal network type is detected to be in a Wi-Fi environment, rich media messages with pictures and / or videos are pushed in real time. When the user terminal network type is detected to be in a 5G environment, simplified text messages are pushed. Monitor push status in real time, provide feedback data to the constructed multi-dimensional user profile, and update the user profile.

[0095] The steps of collecting and acquiring multidimensional user data, including basic information data, historical interaction data, and device status data, and performing privacy protection processing include: The collected user multidimensional data was anonymized. A three-tier access control system was established: ordinary employees can only access the anonymized profile tags, administrators need to approve before they can view the original data, and third-party partners have no data access permissions.

[0096] The steps of filtering the queue to be pushed based on the push priority list, pushing in real time, monitoring the push status in real time, feeding back data to the constructed multi-dimensional user profile, and updating the user profile further include: Regularly clean up non-core interaction data that has exceeded a specified time, as described above.

[0097] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables an electronic device to perform the steps of any of the methods described above, specifically as described above.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An intelligent push method based on multi-dimensional user profiles and scene awareness, characterized in that, include: Collect and acquire multidimensional user data, including basic information data, historical interaction data, and device status data, and perform privacy protection processing; Multi-dimensional user profiles are constructed based on collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and the profile tag weights are dynamically updated through machine learning models. Real-time acquisition of current scene data from user terminals, and conversion of the scene data into scene tags; Based on the user profile and the current scene tag of the user terminal, a dynamic weighting algorithm is used to calculate the push priority and generate a push priority list corresponding to the current scene tag; The system filters the queue of users to be pushed based on the priority list, pushes data in real time, monitors the push status in real time, and feeds back the data to the constructed multi-dimensional user profile to update the user profile.

2. The intelligent push method based on multi-dimensional user profiles and scene awareness according to claim 1, characterized in that, The step of calculating the push priority based on the user profile and the current scene tag of the user terminal, and generating a push priority list corresponding to the current scene tag, includes: A content-profile matching model is pre-built. Based on the matching degree between the user profile's interest tags and the keywords of the pushed content, a basic weight is assigned. A scenario-timing adaptation model is pre-built. Based on the scenario tags, the push timing weight that matches the scenario tags is dynamically calculated. By combining the base weight and the timing weight, a push priority list is generated, and only content with a predetermined priority value is selected to enter the push queue.

3. The intelligent push method based on multi-dimensional user profiles and scene awareness according to claim 1, characterized in that, The steps of collecting and acquiring multidimensional user data, including basic information data, historical interaction data, and device status data, and performing privacy protection processing include: The system collects basic user information, historical interaction data, and device status data to obtain multidimensional user data. The collected multidimensional user data is then synchronized in real time to the privacy protection module for encryption. The user basic information includes: such as age, gender, registration time and / or data obtained with user authorization; the historical interaction data includes message opening records, clicked content, dwell time and / or unfollowing behavior data; the device status data includes device model, network type, battery data and / or non-privacy data.

4. The intelligent push method based on multi-dimensional user profiles and scene awareness according to claim 1, characterized in that, The step of acquiring the current scene data of the user terminal in real time and converting the scene data into scene tags includes: Real-time acquisition of current scene data of user terminals, including geographical location, time scene, and device scene, and conversion of scene data into scene tags; The geographic location is a fuzzy location based on user authorization, not precise coordinates, and includes whether the user is in a shopping mall or office; the time scenario includes weekdays, weekends, commuting hours, and / or rest periods; and the device scenario includes network type and device battery level.

5. The intelligent push method based on multi-dimensional user profiles and scene awareness according to claim 1, characterized in that, The steps of filtering the queue to be pushed based on the push priority list, pushing in real time, monitoring the push status in real time, feeding back data to the constructed multi-dimensional user profile, and updating the user profile include: The system filters the queue of items to be pushed based on a priority list and then pushes them in real time. The message format is adjusted according to the user terminal network type. When the user terminal network type is detected to be in a Wi-Fi environment, rich media messages with pictures and / or videos are pushed in real time. When the user terminal network type is detected to be in a 5G environment, simplified text messages are pushed. Monitor push status in real time, provide feedback data to the constructed multi-dimensional user profile, and update the user profile.

6. The intelligent push method based on multi-dimensional user profiles and scene awareness according to claim 1, characterized in that, The steps of collecting and acquiring multidimensional user data, including basic information data, historical interaction data, and device status data, and performing privacy protection processing include: The collected user multidimensional data was anonymized. A three-tier access control system was established: ordinary employees can only access the anonymized profile tags, administrators need to approve before they can view the original data, and third-party partners have no data access permissions.

7. The intelligent push method based on multi-dimensional user profiles and scene awareness as described in claim 1, characterized in that, The steps of filtering the queue to be pushed based on the push priority list, pushing in real time, monitoring the push status in real time, feeding back data to the constructed multi-dimensional user profile, and updating the user profile also include: Regularly clean up non-core interaction data that has exceeded a specified time.

8. An intelligent push device based on multi-dimensional user profiles and scene awareness, characterized in that, The device includes: The user data collection module is used to collect and acquire multi-dimensional user data, including basic information data, historical interaction data, and device status data, and to perform privacy protection processing. The user profile building module is used to build multi-dimensional user profiles based on collected multi-dimensional user data, including: interest tag layer, behavior habit layer, and demand priority layer, and dynamically update profile tag weights through machine learning models; The scene perception module is used to acquire the current scene data of the user terminal in real time and convert the scene data into scene tags; The push strategy generation module is used to calculate the push priority based on the user profile and the current scene tag of the user terminal using a dynamic weight algorithm, and generate a push priority list corresponding to the current scene tag. The message push execution module is used to filter the queue to be pushed based on the push priority list, push in real time, monitor the push status in real time, feed back data to the constructed multi-dimensional user profile, and update the user profile.

9. A terminal device, characterized in that, It includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors, wherein the one or more programs include steps for performing the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it enables the electronic device to perform the steps of the method as described in any one of claims 1-7.

Images