Information pushing method and device, equipment, computer medium and product

Abstract

The invention discloses an information pushing method, device and equipment, a computer medium and a product, which are applied to the field of information processing, and the method comprises the following steps: obtaining real-time operation behavior data of a target user; performing event processing on the real-time operation behavior data to obtain an operation event; extracting a multi-dimensional feature vector of the operation event, and inputting the multi-dimensional feature vector into a pre-trained event value classification model to obtain a detection result of whether the operation event is a noise event; under the condition that the operation event is not the noise event, updating a user portrait of the target user based on the operation event to obtain an updated user portrait; inputting the updated user portrait into a preset information push model to obtain push information of the target user; and pushing the push information to the target user. According to the method, the accuracy of information pushing is improved, and rapid and millisecond-level pushing can be realized.

Description

Technical Field

[0001] This application belongs to the field of information processing technology, and in particular relates to an information push method, apparatus, device, computer storage medium and computer program product. Background Technology

[0002] With the rapid development of the internet, the ability to accurately recommend information to users has become particularly important.

[0003] Currently, most existing information recommendation methods use batch information recommendation. However, batch push cannot accurately reach users' needs, resulting in low accuracy of information push and affecting user experience.

[0004] Therefore, how to provide an information push method to improve the accuracy of information push is an urgent problem to be solved. Summary of the Invention

[0005] This application provides an information push method, apparatus, device, computer storage medium, and computer program product, which can not only improve the accuracy of information push, but also achieve fast, millisecond-level push.

[0006] In a first aspect, embodiments of this application provide an information push method, the method comprising: Acquire real-time operational behavior data of the target user; The real-time operation data is processed into event-based events to obtain operation events; Extract the multidimensional feature vector of the operation event, input the multidimensional feature vector into a pre-trained event value classification model, and obtain the detection result of whether the operation event is a noise event; If the operation event is not a noise event, the user profile of the target user is updated based on the operation event to obtain the updated user profile; The updated user profile is input into a preset information push model to obtain the push information for the target user; The push notification is sent to the target user.

[0007] In some possible implementations, the step of updating the target user's user profile based on the operation event, when the operation event is not a noise event, to obtain the updated user profile, includes: Analyze the event type and event attributes of the operation event; Based on a preset event weight-time decay model, the real-time impact value of the operation event on the user tag in the user profile is calculated; the event weight-time decay model is used to configure initial weights for different types of events, and the weights decay as a function over time. Based on the real-time impact value, user tags are updated in real-time in an incremental manner; Based on the updated user tags, the user profile of the target user is updated to obtain the updated user profile.

[0008] In some possible implementations, the calculation of the real-time impact value of the operation event on the user tag in the user profile based on a preset event weight-time decay model includes: Based on the event type of the operation event, query the initial weight W0 and decay coefficient λ corresponding to the operation event; Based on the difference Δt between the occurrence time of the operation event and the current time, the current effective weight Wt is calculated according to the decay function f(Δt,λ), where f is a decreasing function; The real-time impact value Δ = V * Wt is obtained by multiplying the event attribute value V corresponding to the event attribute of the operation event with the current effective weight Wt.

[0009] In some possible implementations, the multidimensional feature vector includes at least one of the following: temporal features, sequence features, event intensity features, and environmental context features; The step of inputting the multidimensional feature vector into a pre-trained event value classification model to obtain a detection result of whether the operation event is a noise event includes: The multidimensional feature vector is input into a pre-trained event value classification model to obtain the predicted probability value that the operation event is a valid signal event; If the predicted probability value is lower than a preset threshold, the operation event is determined to be a noise event; If the predicted probability value is equal to or higher than a preset threshold, the operation event is determined to be a noise event.

[0010] In some possible implementations, the method further includes: Input the updated user profile into the time recommendation model to obtain the target recommended time; Input the updated user profile into the channel recommendation model to obtain the target recommendation channel; The step of pushing the push information to the target user includes: Upon reaching the target recommendation time, the push information is sent to the target user through the target recommendation channel.

[0011] In some possible implementations, the push information is product push information, and the step of inputting the updated user profile into a preset information push model to obtain the push information for the target user includes: Based on the updated user profile, a user-product interaction matrix is ​​constructed, where the rows of the matrix represent users, the columns of the matrix represent products, and the values ​​of the matrix represent the historical interaction intensity of users with products. The interaction matrix is ​​decomposed using a matrix factorization algorithm to obtain the latent feature vector of the target user and the latent feature vector of each product. The user latent feature vector and the product latent feature vector are input into the information push model to obtain the predicted score for each product; Products that meet the preset criteria and are corresponding to the target recommended products are identified. Based on the target recommended product, the product push information is determined.

[0012] In some possible implementations, the step of inputting the updated user profile into the time recommendation model to obtain the target recommendation time includes: Based on the updated user profile, extract the target user's historical behavioral events and the corresponding times of the historical behavioral events; Based on historical behavioral events and the corresponding times of those events, a behavioral time series of the target user is constructed. Feature extraction is performed on the behavioral time series to obtain a time series feature vector; The time-series feature vector is input into the time recommendation model to obtain the acceptance probability distribution representing the target user's acceptance of the recommended event within a future preset time period; The time corresponding to the probability of satisfying the preset conditions is determined as the target recommended time.

[0013] In some possible implementations, the step of inputting the updated user profile into the channel recommendation model to obtain the target recommendation channel includes: Based on the updated user profile, the channel selection features of the target user are extracted. The channel selection features include at least one of the following: the target user's attribute features, historical channel preference features, and features of the product to be recommended. The channel selection features are input into the channel recommendation model to obtain the target user's predicted feedback rating for each alternative channel; The channels corresponding to the predicted feedback scores that meet the preset scoring conditions are identified as the target recommendation channels.

[0014] Secondly, embodiments of this application provide an information push device, the device comprising: The acquisition module is used to acquire real-time operational behavior data of the target user; The event processing module is used to process the real-time operation behavior data into event-based data to obtain operation events. The update module is used to update the user profile of the target user based on the operation event, when the operation event is not a noise event, to obtain the updated user profile; The input module is used to input the updated user profile into a preset information push model to obtain the push information for the target user; The push module is used to push the push information to the target user.

[0015] Thirdly, embodiments of this application provide an information push device, the device comprising: A processor and a memory storing computer program instructions; an information push method that implements any of the above when the processor executes the computer program instructions.

[0016] Fourthly, embodiments of this application provide a computer storage medium on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the information push method described above is implemented.

[0017] Fifthly, embodiments of this application provide a computer program product in which instructions are executed by the processor of an electronic device, enabling the electronic device to execute any of the above-mentioned information push methods.

[0018] The information push method, apparatus, device, computer storage medium, and computer program product of this application, after acquiring real-time operation behavior data of a target user, perform event-based processing on the real-time operation behavior data to obtain operation events; then, further extract multi-dimensional feature vectors of the operation events, input the multi-dimensional feature vectors into a pre-trained event value classification model, and obtain a detection result indicating whether the operation event is a noise event; if the operation event is not a noise event, the user profile of the target user can be updated based on the operation event to obtain an updated user profile; then, the updated user profile is input into a preset information push model to obtain push information for the target user, and the push information is pushed to the target user. On the one hand, this method can obtain operation events based on the user's real-time operation behavior data, and determine whether the operation event is a noise event through the multi-dimensional feature vector of the operation event. By removing noise events, not only is the data quality improved, but also, through noise removal, invalid computation can be significantly reduced, and response efficiency can be improved, thereby achieving fast, millisecond-level push; on the other hand, using the noise-removed operation events to update the user profile can not only accurately capture user needs, but also make the push information determined based on the updated user profile more accurate. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an information push method provided in one embodiment of this application; Figure 2 This is a schematic flowchart of S130 provided in one embodiment of this application; Figure 3 This is a schematic flowchart of S140 provided in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of an information push device provided in another embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation

[0021] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are intended only to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0022] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0023] It should be noted that the acquisition, storage, use, and processing of data in this application embodiment all comply with the relevant provisions of national laws and regulations.

[0024] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0025] Currently, most existing information recommendation methods use batch information recommendation. However, batch push cannot accurately reach users' needs, resulting in low accuracy of information push and affecting user experience.

[0026] To address the problems of the prior art, embodiments of the present invention provide an information push method, apparatus, device, and computer storage medium.

[0027] The information push method provided in the embodiments of the present invention will be introduced first below.

[0028] Figure 1 A flowchart illustrating an embodiment of the information push method provided by the present invention is shown. Figure 1 As shown, the method may include the following steps: S110. Obtain real-time operational behavior data of the target user.

[0029] For example, the real-time operational behavior data of the target user may include internal data and external data.

[0030] Internal data can include: transaction data (transfers, payments, withdrawals, investment purchases, etc.); account data (account opening, closing, account balance changes, etc.); behavioral data (app logins, page views, function usage, etc.); and basic user information (age, gender, occupation, income, etc.).

[0031] External data can include social media data: user behavior on social media (such as likes, comments, shares, etc.); e-commerce platform data: user shopping behavior, consumption preferences, etc.; and credit data: user credit scores, loan records, etc.

[0032] For example, real-time user behavior data can be collected through real-time streaming data. For instance, transaction and behavioral data can be collected in real-time using tools like Kafka and Flume. Once a user action occurs, it is immediately published as a message to a Kafka topic with a latency in the millisecond range. Batch data collection can be performed by periodically importing basic user information and external data using ETL tools.

[0033] For example, raw logs can also be continuously collected from various data sources. One approach is through a tracking SDK: embedding code in mobile banking apps and web pages to capture every click, swipe, and browse by the user. Another approach is log collection: collecting transaction logs, account changes, and API call logs from backend servers. Then, Apache Kafka, Flume, or similar software are used as message queues / streaming data buses to receive these data streams in real time using a publish-subscribe model.

[0034] For example, user Zhang San browses the details page of a wealth management product on a mobile banking app and stays there for 2 minutes. This behavior will be captured by the app's event tracking, generating a JSON log containing {User ID: Zhang San, Event Type: page_view, Page: / product / xxx, Stay Duration: 120000ms, Timestamp: ...}, which is then sent to Kafka.

[0035] S120. Perform event-based processing on real-time operation behavior data to obtain operation events.

[0036] For example, after obtaining real-time user behavior data, it is also necessary to standardize the raw, messy, and unstructured logs into structured event objects rich in business semantics. On the one hand, this can be achieved through data cleaning to remove garbled characters, complete missing fields, and standardize the time format. On the other hand, data transformation can be performed to map data from different sources to a unified format (such as a unified JSON Schema). Afterward, events are categorized and tagged with business labels based on their content.

[0037] For example, events can be categorized based on business scenarios, such as transaction events (large transfers, frequent payments, etc.), behavioral events (login, browsing financial products, clicking ads, etc.), and external events (negative comments on social media, shopping behavior on e-commerce platforms, etc.). A stream processing engine (Apache Flink) is used to process real-time events. For example, the logs from the previous step can be categorized as "high-intent financial product browsing events."

[0038] S130. Extract the multidimensional feature vector of the operation event, and input the multidimensional feature vector into the pre-trained event value classification model to obtain the detection result of whether the operation event is a noise event. If the operation event is not a noise event, proceed to S140; if the operation event is a noise event, proceed to S170.

[0039] S140. Based on the operation event, update the user profile of the target user to obtain the updated user profile.

[0040] For example, a user profile includes static and dynamic tags. Static tags could be age, gender, occupation, income, etc. Dynamic tags could be consumption habits, risk preferences, financial needs, etc. This user profile is dynamic rather than static; the user's dynamic tags are refreshed in real time through user actions. For example, if Zhang San browses financial products with a high intention, the weight of dynamic tags such as "recently interested in financial management" and "preferred product type: conservative" in the corresponding user profile will be increased in real time.

[0041] S150. Input the updated user profile into the preset information push model to obtain the push information for the target user.

[0042] For example, when the push notification is for a product, the target recommended product information for the target user can be obtained based on the updated user profile.

[0043] S160, Push the information to the target user.

[0044] For example, targeted product recommendations can be pushed to target users.

[0045] If the operation event is a noise event, execute S170.

[0046] S170, Noise Removal Event.

[0047] By removing noisy events, we can prevent them from contaminating the data, saving a lot of resources that would otherwise be spent on computing and storing invalid events. At the same time, we can ensure the quality of user profiles and model training data, which is beneficial to improving the accuracy of information push.

[0048] The information push method, apparatus, device, computer storage medium, and computer program product of this application, after acquiring real-time operation behavior data of a target user, perform event-based processing on the real-time operation behavior data to obtain operation events; then, further extract multi-dimensional feature vectors of the operation events, input the multi-dimensional feature vectors into a pre-trained event value classification model, and obtain a detection result indicating whether the operation event is a noise event; if the operation event is not a noise event, the user profile of the target user can be updated based on the operation event to obtain an updated user profile; then, the updated user profile is input into a preset information push model to obtain push information for the target user, and the push information is pushed to the target user. On the one hand, this method can obtain operation events based on the user's real-time operation behavior data, and determine whether the operation event is a noise event through the multi-dimensional feature vector of the operation event. By removing noise events, not only is the data quality improved, but also, through noise removal, invalid computation can be significantly reduced, and response efficiency can be improved, thereby achieving fast, millisecond-level push; on the other hand, using the noise-removed operation events to update the user profile can not only accurately capture user needs, but also make the push information determined based on the updated user profile more accurate.

[0049] In some embodiments, such as Figure 2 As shown, S130 may include: S1301. Analyze the event type and event attributes of the operation event.

[0050] For example, event types are used to categorize events for business purposes. For instance, "large-amount transfer received," "browsing high-risk financial products for more than 3 minutes," and "credit card overdue bill reminder read" are dimensions for judging the quality of an event. Event attributes are used to describe the specific parameters of an event. For example, the transfer amount, browsing duration, and number of clicks are dimensions for judging the quantity of an event.

[0051] For example, the event can be parsed as follows: Type: Purchase of financial products, Attributes: {Product risk level: R4, Purchase amount: 100,000 yuan, Purchase time: 2023-10-27 14:30}.

[0052] S1302. Based on the preset event weight-time decay model, calculate the real-time impact value of the operation event on the user tag in the user profile.

[0053] For example, the event weight-time decay model is used to configure initial weights for different types of events, and the weights decay as a function over time.

[0054] In some embodiments, when calculating the real-time impact value of an operation event on user tags in a user profile based on a preset event weight-time decay model, the initial weight W0 and decay coefficient λ corresponding to the operation event can be queried first according to the event type of the operation event; based on the difference Δt between the event occurrence time and the current time, the current effective weight Wt can be calculated according to the decay function f(Δt, λ), where f is a decreasing function; based on the product of the event attribute value V corresponding to the event attribute and the current effective weight Wt, the real-time impact value Δ=V * Wt can be obtained.

[0055] For example, the initial weights (W) Assign a base importance score to different event types. For example, the W score for "completing a financial purchase" is... Possibly 1.0 (strong signal). The W in "Browse the details page of the financial product" It might be 0.3 (medium signal). The W value for "quickly scrolling through the product list" is... It could be 0.1 (weak signal). The weighting configuration reflects user preferences; for example, "buying" reflects user preferences more than "browsing".

[0056] The time decay function (f(t)) describes how the influence of an event diminishes over time. For example, a user's purchase yesterday is less representative of their "current" preferences than their purchase today. The function typically takes the form of exponential decay, W_t = W * e^(-λt) or linear decay. λ is the decay coefficient, and t is the time difference from the event to the present. For example, an event of "financial purchase" that occurred 3 days ago (W =1.0, λ=0.1), its current effective weight W_t may decay to 1.0 * e^(-0.1*3)≈ 0.74.

[0057] Real-time impact value (Δ) calculation formula: Δ = V * W_t Where V is the result of quantifying the event attribute value (e.g., purchase amount / 10000, browsing time / 60 seconds), and W_t is the current effective weight.

[0058] For example, a user purchased 100,000 yuan worth of high-risk financial products today.

[0059] Event type: W = 1.0 (Purchase Behavior) Time decay: just occurred, t≈0, W_t≈1.0 Attribute value V: Amount 100,000 / Base value 10,000 = 10 The real-time impact value Δ = 10 * 1.0 = 10 (a strong positive impact of +10 on the "high-risk preference" label). By using the decay function f(Δt, λ), a "timestamp" and "half-life" are built into the impact of each event, enabling user profiles to keenly reflect the latest and most relevant behavioral patterns of users and always maintain high-fidelity synchronization with the user's real state, thus solving the problem of outdated profiles.

[0060] S1303. Based on real-time impact values, update user tags in an incremental manner in real time.

[0061] For example, the impact of new real-time events can be smoothly integrated into a user's historical state.

[0062] New label value = α * old label value + (1-α) * Δ α is the forgetting factor (0 < α < 1), which determines the degree to which historical memory is retained. The larger α is, the smoother the label changes and the stronger the resistance to noise; the smaller α is, the more sensitive the label is to the latest events. This avoids abrupt changes in label values, allowing them to stably and continuously reflect the changing trends of user preferences.

[0063] For example, a user's "high-risk preference" tag originally had a value of 5.0 and α=0.8, and the Δ=10 after receiving the above purchase event.

[0064] The new label value = 0.8 * 5.0 + 0.2 * 10 = 4.0 + 2.0 = 6.0. The label value smoothly increases from 5.0 to 6.0.

[0065] S1304. Based on the updated user tags, update the user profile of the target user to obtain the updated user profile.

[0066] For example, the calculated new label value can be written back to the user profile database, and the user's digital profile now accurately reflects the latest state that "his risk appetite has increased after he has just made a large, high-risk investment".

[0067] On the one hand, by instantly converting user behavior events into incremental updates of profile tags, the data lag problem caused by the traditional batch (T+1) update mode is solved, reducing the latency of user intent perception and response to "seconds," truly achieving synchronization with user behavior. On the other hand, the "initial weight × time decay" model is used to quantify the impact of events, making profile tags no longer simple Boolean values ​​or static scores, but dynamic indicators with clear business meanings that change continuously over time. This effectively distinguishes the influence of recent strong signal events from long-term weak signal events, ensuring that the profile can more realistically and nuancedly reflect the user's current state and short-term preferences at any time, providing a high-fidelity data foundation for accurate decision-making.

[0068] To improve the accuracy of information push, before S140, it is necessary to determine whether the operation event is a noise event. Specifically, this can be achieved by extracting a multi-dimensional feature vector of the operation event, which includes at least one of the following: time-series features, sequence features, event intensity features, and environmental context features. When inputting the multi-dimensional feature vector into a pre-trained event value classification model to obtain the detection result of whether the operation event is a noise event, specifically, the multi-dimensional feature vector is input into the pre-trained event value classification model to obtain the predicted probability value of the operation event as a valid signal event. If the predicted probability value is lower than a preset threshold, the operation event is determined to be a noise event; if the predicted probability value is equal to or higher than the preset threshold, the operation event is determined not to be a noise event.

[0069] It's important to note that among the massive amounts of user behavior events, there is a large amount of data that has no business value, interferes with decision-making, and can even mislead the system—this is "noise." Examples include: accidental touches (users clicking unintentionally), invalid browsing (extremely short page dwell time, <1 second), script / bot behavior (high-frequency, regular non-human operations), tentative operations (users clicking randomly without genuine intent), and failed / interrupted business processes (e.g., payment failure due to insufficient balance). If these noisy events are not filtered out and directly used to update user profiles, it not only pollutes the profiles, making user tags chaotic and unreliable, but also wastes resources, as valuable computing resources are used to process invalid data. Furthermore, it reduces the accuracy of push notifications and the user experience; push notifications triggered by noise lead to a decline in user experience and even complaints.

[0070] When determining whether an operation event is a noise event, a multi-dimensional feature vector is first extracted, including temporal features, sequence features, event intensity features, and environmental context features. Temporal features measure the pattern of the event over time. For example, the time interval since the last similar event: if the interval is extremely short (e.g., clicking again within 1 second), it may be a misclick or machine behavior. The cumulative number of similar events today / this week: used to identify abnormally high-frequency behavior. The absolute time of the event (e.g., 3 AM): abnormal operations during inactive periods. For example, if user A checks their balance 5 times consecutively within 2 seconds, the temporal features show an extremely high frequency and very short intervals, which is a typical noise pattern.

[0071] Sequence features are used to measure the position and order of events within a user session or business process. Examples include the current event's sequence number within the session, the pattern of preceding event sequences (e.g., login → homepage → logout immediately, this sequence has a weaker intent), and whether key business processes have been completed (e.g., browsing product details → adding to cart → not paying). For example, if user B only stays on a financial product page for 0.5 seconds before immediately returning to the homepage, the sequence feature indicates a very short stay on a deep page, which may be an invalid browsing experience.

[0072] Event intensity features are used to measure the business intensity or level of investment in an event. Examples include browsing duration, page scroll depth; transaction amount, transfer amount; and the ratio of time spent on a product details page to the average time spent on similar products. For example, if user C reads a complex fund contract PDF for 10 minutes, this indicates extremely high event intensity, signifying strong intent and is a highly effective signal.

[0073] Contextual features are used to measure the external environment in which an event occurs. Examples include device type (phone / PC), operating system; network environment (Wi-Fi / 4G / 5G); geographical location (whether in a familiar location); and application version. For example, if user D logs in on an unfamiliar device or from a different location and immediately performs a large transaction → the contextual features are abnormal, potentially indicating risk or fraud, and the transaction needs to be filtered or transferred to the risk control process.

[0074] After extracting the multidimensional feature vector of the operation event, the multidimensional feature vector is input into a pre-trained event value classification model to obtain the predicted probability value of the operation event as a valid signal event.

[0075] For example, the event value classification model is a binary classifier (typically a gradient boosting tree such as XGBoost / LightGBM, or a neural network). Training data can be historical data, labeled by operations experts or based on subsequent conversion results. For example, positive samples (valid signals): events that ultimately led to valuable behaviors such as product purchases, customer service inquiries, or in-depth browsing. Negative samples (noise): events with no subsequent action or that were clearly erroneous. The model output is a predicted probability value between 0 and 1, representing the confidence level that the event is a "valid signal".

[0076] Next, the predicted probability value is compared with a preset threshold. If the predicted probability value is lower than the preset threshold, the operation event is determined to be a noise event, and in this case, the noise event can be removed. That is, it will not be used to update the user profile, nor will it trigger subsequent push notifications; it may be discarded or stored in a dedicated log for analysis. If the predicted probability value is equal to or higher than the preset threshold, the operation event is determined not to be a noise event. In this case, the user profile can be updated based on this event.

[0077] By filtering out noise upstream in the data pipeline, the waste of resources in all subsequent complex calculations (profile updates, model inference) is avoided, significantly reducing computing and storage costs. At the same time, it ensures that the data flowing into user profiles and models are valid signals, fundamentally improving the accuracy of user profiles, thereby further improving the precision of information push.

[0078] To further improve the accuracy of information push, in practice, the updated user profile can be input into the time recommendation model to obtain the target recommendation time; the updated user profile can be input into the channel recommendation model to obtain the target recommendation channel; and when pushing information to the target user, specifically, when the target recommendation time is reached, the information is pushed to the target user through the target recommendation channel.

[0079] By using a time-based recommendation model, push notifications can be synchronized with users' personal biological clocks and behavioral rhythms, ensuring that information appears "when users want to see it," thus improving the accuracy of push notifications. A channel-based recommendation model ensures that information reaches users through their most preferred and comfortable mediums. The combination of these two approaches enhances the effectiveness of information delivery and the user experience without excessively intruding on users.

[0080] For example, the time-based recommendation model is used to determine the target recommendation time. In some embodiments, when the updated user profile is input into the time-based recommendation model to obtain the target recommendation time, the historical behavioral events of the target user and the corresponding times of the historical behavioral events can be extracted based on the updated user profile; a behavioral time series of the target user can be constructed based on the historical behavioral events and the corresponding times of the historical behavioral events; features are extracted from the behavioral time series to obtain a time-series feature vector; the time-series feature vector is input into the time-based recommendation model to obtain the acceptance probability distribution representing the target user's acceptance of the recommended event within a preset time period in the future; and the time corresponding to the probability that meets the preset conditions is determined as the target recommendation time.

[0081] In practice, the updated user profile (including user ID) can be used as an index, and the user ID as the key. A series of events with precise timestamps can be retrieved from the user's historical behavioral event database, outputting a time-sorted list. This list is then transformed into a standard time series data structure used in machine learning or statistics. This can be a discrete event sequence, directly using event type and timestamps, or an aggregated time series, dividing time into fixed windows (e.g., hourly) and counting the frequency of specific event types (e.g., "login," "transaction," "browse") within each window to form a continuous numerical sequence. Ultimately, this generates a "daily login activity" curve or a "weekly financial browsing hotspot" distribution chart.

[0082] For example, the extracted time-series feature vectors typically include: periodic features, trend features, statistical features, and event-related features.

[0083] The aforementioned time-series feature vectors are input into a time-based recommendation model. Commonly used time-series prediction models include: LSTM (Long Short-Term Memory Network): It excels at capturing long-term dependencies and complex patterns, making it ideal for such tasks.

[0084] Transformer (temporal attention model): can better handle long-range dependencies in a sequence.

[0085] Prophet: An open-source model from Facebook that performs very well in forecasting business time series with strong periodicity.

[0086] The time-based recommendation model predicts a user's "acceptance level" of a push notification at each point in time within a future period (e.g., the next 24 hours) based on time-series feature vectors, outputting an "acceptance probability distribution curve." For example, the horizontal axis represents the time points in the next 24 hours (8:00, 8:30, ..., 7:30 the next day), and the vertical axis represents the predicted probability value (between 0 and 1) for each time point. This curve visually displays the user's "golden reach window" throughout the day.

[0087] Next, select time periods where the probability value exceeds a certain minimum confidence threshold (e.g., > 0.5). Within these threshold-compliant time periods, select the highest global or local probability points to ultimately determine one or more "target recommendation times." For example, the curve shows that the peak probability for users at 8:15 PM tonight is 0.72, with a secondary peak of 0.65 at 9:30 AM tomorrow morning. Therefore, 8:15 PM tonight can be chosen as the primary push time.

[0088] By continuously learning each user's unique time behavior, the system automatically and continuously optimizes and updates its "best time to reach" to ensure that push notifications are always synchronized with the user's latest life rhythm, which helps improve the accuracy of information push and user experience. At the same time, by outputting probability distributions rather than single points in time, the system improves the flexibility of push notifications.

[0089] For example, the channel recommendation model is used to determine the target recommendation channel. In some embodiments, when the updated user profile is input into the channel recommendation model to obtain the target recommendation channel, the channel selection features of the target user can be extracted based on the updated user profile. The channel selection features include at least one of the following: the target user's attribute features, historical channel preference features, and features of the product to be recommended; the channel selection features are input into the channel recommendation model to obtain the target user's predicted feedback rating for each candidate channel; the channel corresponding to the predicted feedback rating that meets the preset rating conditions is determined as the target recommendation channel.

[0090] When determining target recommendation channels based on updated user profiles, the channel selection characteristics of target users can be extracted first, including at least one of the following: target user attribute characteristics, historical channel preference characteristics, and characteristics of the product to be recommended. User attribute characteristics include, for example, age, occupation, and education level. For instance, young white-collar workers prefer app push notifications, while older customers may be more sensitive to SMS or phone calls. Historical channel preference characteristics (behavioral feedback data) can be determined through channel response rates, including the open rate, click-through rate, conversion rate, ignore rate, and complaint / blocking rate of marketing information from various channels. For example, the average open rate for app push notifications is 40%, for SMS is 5%, and for email is 1%. Characteristics of the product to be recommended (information content attributes) include, for example, product complexity. Simple information (such as interest rate adjustment notifications) is suitable for SMS; complex products (such as fund prospectuses) are suitable for emails or rich text pages within apps. These characteristics are vectorized to form a comprehensive channel selection feature vector, which serves as the input to the channel recommendation model.

[0091] For example, channel recommendation models can be multi-class or ranking models. Commonly used ones include: Gradient boosting trees (such as XGBoost and LightGBM) are good at handling tabular features, can capture non-linear relationships between features well, and the results are interpretable to a certain extent.

[0092] Multilayer perceptron: May perform better in scenarios with very complex feature interactions.

[0093] For a given user and product, predict their "expected feedback score" on each alternative channel (e.g., SMS, push notifications, email, phone). This score is a composite metric, which can be an estimated click-through rate, conversion rate, or a combined utility score that considers both positive feedback and negative risk. Output a list of scores, for example: {"Push notifications": 0.85, "SMS": 0.30, "Email": 0.10, "Phone": 0.65}.

[0094] Finally, based on the predicted feedback score list, you can directly select the channel with the highest score, or you can combine the threshold with the highest score, requiring that the score of the preferred channel must exceed the minimum validity threshold (e.g., >0.5), otherwise the alternative plan will be activated or the push will be abandoned.

[0095] Based on precise matching of user profiles, behavioral history, and specific content, it achieves a deep fit with user reach habits, which helps improve the accuracy of information push; by outputting clear predictive scores, it quantifies the channel selection process and improves interpretability; by predicting the conversion probability of different channels, it can automatically select the channel combination with the highest "cost-effectiveness" and improve configuration efficiency.

[0096] In some embodiments, such as Figure 3 As shown, when the push information is product push information, S140 may include: S1401. Based on the updated user profile, construct a user-product interaction matrix.

[0097] For example, the rows of the matrix represent users, the columns represent products, and the values ​​represent the intensity of a user's historical interaction with a product. For instance: Purchase = 5 points; In-depth browsing (duration > 3 minutes) = 4 points; Add to favorites / compare = 3 points; Simple browsing = 2 points; Ignore / quickly swipe = 1 point or 0 points. This sparse matrix mathematically and globally characterizes the preference distribution and behavioral patterns of the entire user group.

[0098] S1402. Use matrix factorization algorithm to decompose the interaction matrix to obtain the potential feature vector of the target user and the potential feature vector of each product.

[0099] By decomposing the interaction matrix, we can identify the underlying factors driving user preferences and product appeal. For example, we can use Singular Value Decomposition (SVD) or Alternating Least Squares (ALS) to approximate the large user-product interaction matrix R (m×n) as the product of two low-dimensional matrices: R ≈ U × V^T U (m×k): User latent feature matrix, where each row represents a user's k-dimensional latent feature vector. These k features can be understood as abstract traits that cannot be directly observed but determine user preferences, such as "risk preference level", "liquidity demand intensity", "acceptance of digital products", etc.

[0100] V (n×k): Product latent feature matrix, where each row represents a k-dimensional latent feature vector of a product. Similarly, these k features represent the abstract attributes that attract users to the product, such as "risk-return characteristics", "product complexity", "innovation", etc.

[0101] k is the dimension of the potential space, which is much smaller than the number of users m and the number of products n.

[0102] The final output can be, for the target user, the user's potential feature vector u; for all candidate products, the product's potential feature vector v_i.

[0103] S1403. Input the user's potential feature vector and the product's potential feature vector into the information push model to obtain the predicted score for each product.

[0104] The predicted interaction intensity (rating) for a target user u and a product i is calculated as follows: Predicted score = u · v_i^T (i.e., the dot product of user feature vector and product feature vector).

[0105] The larger the dot product value, the more consistent the directions of the user feature vector and the product feature vector are in the latent space, and the smaller the angle between them, which means the higher the "matching degree".

[0106] S1404. Products that meet the preset conditions and are corresponding to the target recommended products are identified.

[0107] For example, the predicted ratings of target users for all products they have not interacted with can be calculated, and then filtered and sorted according to preset conditions. This can be based on threshold filtering, retaining only products with ratings higher than a certain threshold; or it can be selecting the N products with the highest ratings (such as Top-3). The final list is the "target recommended products".

[0108] S1405. Based on the target product recommendation, determine the product push information.

[0109] The selected list of recommended products is combined with the user's personalized information (such as name and title) and business rules to generate the final product push notification. For example, "Dear Mr. Zhang, based on your investment preferences, we have selected the following three financial products for you..."

[0110] Matrix factorization models, by learning the latent feature vectors of users and products, can deeply understand the abstract reasons why users have preferences and why products are chosen. By calculating the matching degree (dot product) between user vectors and product vectors, they can accurately uncover unique combinations of preferences that match the target users, further improving the accuracy of product information push.

[0111] The information push method, apparatus, device, computer storage medium, and computer program product of this application, after acquiring real-time operation behavior data of a target user, perform event-based processing on the real-time operation behavior data to obtain operation events; then, further extract multi-dimensional feature vectors of the operation events, input the multi-dimensional feature vectors into a pre-trained event value classification model, and obtain a detection result indicating whether the operation event is a noise event; if the operation event is not a noise event, the user profile of the target user can be updated based on the operation event to obtain an updated user profile; then, the updated user profile is input into a preset information push model to obtain push information for the target user, and the push information is pushed to the target user. On the one hand, this method can obtain operation events based on the user's real-time operation behavior data, and determine whether the operation event is a noise event through the multi-dimensional feature vector of the operation event. By removing noise events, not only is the data quality improved, but also, through noise removal, invalid computation can be significantly reduced, and response efficiency can be improved, thereby achieving fast, millisecond-level push; on the other hand, using the noise-removed operation events to update the user profile can not only accurately capture user needs, but also make the push information determined based on the updated user profile more accurate.

[0112] Based on the information push method provided in the above embodiments, this application also provides specific implementations of the information push device. Please refer to the following embodiments.

[0113] First see Figure 4 The information push device 400 provided in this application embodiment includes: Module 410 is used to acquire real-time operation behavior data of the target user; The event processing module 420 is used to process real-time operation behavior data into event-based data to obtain operation events. The extraction module 430 is used to extract the multidimensional feature vector of the operation event, input the multidimensional feature vector into the pre-trained event value classification model, and obtain the detection result of whether the operation event is a noise event. The update module 440 is used to update the user profile of the target user based on the operation event, when the operation event is not a noise event, to obtain the updated user profile. The input module 450 is used to input the updated user profile into the preset information push model to obtain the push information for the target user. The push module 460 is used to push push information to target users.

[0114] In some possible implementations, update module 440 is also used for: Analyze the event type and event properties of the operation event; Based on a preset event weight-time decay model, the real-time impact of the operation event on the user tag in the user profile is calculated; the event weight-time decay model is used to configure the initial weight for different types of events, and the weight decays as a function over time. User tags are updated in real time in an incremental manner based on real-time impact values. Based on the updated user tags, the user profile of the target user is updated to obtain the updated user profile.

[0115] In some possible implementations, based on a preset event weight-time decay model, the real-time impact of the operation event on user tags in the user profile is calculated, including: Based on the event type of the operation event, query the initial weight W0 and decay coefficient λ corresponding to the operation event; Based on the difference Δt between the event occurrence time and the current time, the current effective weight Wt is calculated according to the decay function f(Δt, λ), where f is a decreasing function. The real-time impact value Δ = V * Wt is obtained by multiplying the event attribute value V corresponding to the event attribute of the operation event with the current effective weight Wt.

[0116] In some possible implementations, the multidimensional feature vector includes at least one of the following: temporal features, sequence features, event intensity features, and environmental context features; Extraction module 430 is also used for: The multidimensional feature vector is input into a pre-trained event value classification model to obtain the predicted probability value of the operation event as a valid signal event; If the predicted probability value is lower than a preset threshold, the operation event is determined to be a noise event; If the predicted probability value is equal to or higher than a preset threshold, the operation event is determined to be a noise event.

[0117] In some possible implementations, the input module 450 is also used for: Input the updated user profile into the time recommendation model to obtain the target recommended time; Input the updated user profile into the channel recommendation model to obtain the target recommendation channel; The push module 460 is also used for: Once the target recommendation time is reached, the push notification will be sent to the target users through the target recommendation channels.

[0118] In some possible implementations, the push information is product push information, and the push module 460 is also used for: Based on the updated user profile, a user-product interaction matrix is ​​constructed, where the rows of the matrix represent users, the columns represent products, and the values ​​of the matrix represent the historical interaction intensity of users with products. The interaction matrix is ​​decomposed using a matrix factorization algorithm to obtain the latent feature vectors of the target users and the latent feature vectors of each product. By inputting the latent feature vectors of users and products into the information push model, the predicted scores of each product are obtained. Products that meet the preset criteria and are rated accordingly are identified as target recommended products. Recommend products based on the target audience and determine the product push information.

[0119] In some possible implementations, the updated user profile is input into the time recommendation model to obtain the target recommended time, including: Based on the updated user profile, extract the target user's historical behavioral events and the corresponding times of these events. Construct a time series of target user behaviors based on historical behavioral events and the corresponding times. Feature extraction is performed on the behavioral time series to obtain the time series feature vector; By inputting the temporal feature vector into the time recommendation model, we can obtain the acceptance probability distribution representing the target user's acceptance of the recommended event within a future preset time period. The time corresponding to the probability of meeting the preset conditions is determined as the target recommended time.

[0120] In some possible implementations, the updated user profile is input into the channel recommendation model to obtain the target recommendation channel, including: Based on the updated user profile, extract the channel selection features of the target user. The channel selection features include at least one of the following: the target user's attribute features, historical channel preference features, and features of the product to be recommended. By inputting the channel selection features into the channel recommendation model, the predicted feedback ratings of target users for each alternative channel are obtained. The channels corresponding to the predicted feedback scores that meet the preset scoring conditions are identified as the target recommendation channels.

[0121] The various modules of the information push device provided in this application embodiment can achieve... Figure 1 It provides the functionality for each step of the information push method and achieves the corresponding technical effects. For the sake of brevity, it will not be elaborated here.

[0122] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0123] The electronic device 500 may include a processor 501 and a memory 502 storing computer program instructions.

[0124] Specifically, the processor 501 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0125] Memory 502 may include mass storage for data or instructions. For example, and not limitingly, memory 502 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 502 may include removable or non-removable (or fixed) media. Where appropriate, memory 502 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 502 is non-volatile solid-state memory.

[0126] In certain embodiments, the memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this application.

[0127] The processor 501 reads and executes computer program instructions stored in the memory 502 to implement any of the information push methods in the above embodiments.

[0128] In some examples, the electronic device 500 may also include a communication interface 503 and a bus 504. For example, Figure 5 As shown, the processor 501, memory 502, and communication interface 503 are connected through bus 504 and complete communication with each other.

[0129] The communication interface 503 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0130] Bus 504 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not as a limitation, bus 504 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 504 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0131] For example, the electronic device 500 can be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc.

[0132] The electronic device 500 can execute the information push method in the embodiments of this application, thereby achieving a combination Figure 1 The described information push method.

[0133] In addition, in conjunction with the information push method in the above embodiments, this application also provides a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the information push methods in the above embodiments. Examples of computer-readable storage media include non-transitory computer-readable storage media, such as portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, etc.

[0134] This application also provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, they implement any of the information push methods described in the above embodiments.

[0135] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0136] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0137] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. An information push method, characterized in that, include: Acquire real-time operational behavior data of the target user; The real-time operation data is processed into event-based events to obtain operation events; Extract the multidimensional feature vector of the operation event, input the multidimensional feature vector into a pre-trained event value classification model, and obtain the detection result of whether the operation event is a noise event; If the operation event is not a noise event, the user profile of the target user is updated based on the operation event to obtain the updated user profile; The updated user profile is input into a preset information push model to obtain the push information for the target user; The push notification is sent to the target user.

2. The information push method according to claim 1, characterized in that, When the operation event is not a noise event, updating the user profile of the target user based on the operation event to obtain the updated user profile includes: Analyze the event type and event attributes of the operation event; Based on a preset event weight-time decay model, the real-time impact value of the operation event on the user tag in the user profile is calculated; the event weight-time decay model is used to configure initial weights for different types of events, and the weights decay as a function over time. Based on the real-time impact value, user tags are updated in real-time in an incremental manner; Based on the updated user tags, the user profile of the target user is updated to obtain the updated user profile.

3. The information push method according to claim 2, characterized in that, The calculation of the real-time impact of the operation event on user tags in the user profile, based on a preset event weight-time decay model, includes: Based on the event type of the operation event, query the initial weight W0 and decay coefficient λ corresponding to the operation event; Based on the difference Δt between the occurrence time of the operation event and the current time, the current effective weight Wt is calculated according to the decay function f(Δt, λ), where f is a decreasing function; The real-time impact value Δ = V * Wt is obtained by multiplying the event attribute value V corresponding to the event attribute of the operation event with the current effective weight Wt.

4. The information push method according to claim 1, characterized in that, The multidimensional feature vector includes at least one of the following: temporal features, sequence features, event intensity features, and environmental context features; The step of inputting the multidimensional feature vector into a pre-trained event value classification model to obtain a detection result of whether the operation event is a noise event includes: The multidimensional feature vector is input into a pre-trained event value classification model to obtain the predicted probability value that the operation event is a valid signal event; If the predicted probability value is lower than a preset threshold, the operation event is determined to be a noise event; If the predicted probability value is equal to or higher than a preset threshold, the operation event is determined to be a noise event.

5. The information push method according to claim 1, characterized in that, The method further includes: Input the updated user profile into the time recommendation model to obtain the target recommended time; Input the updated user profile into the channel recommendation model to obtain the target recommendation channel; The step of pushing the push information to the target user includes: Upon reaching the target recommendation time, the push information is sent to the target user through the target recommendation channel.

6. The information push method according to claim 1, characterized in that, The push information is product push information. The step of inputting the updated user profile into a preset information push model to obtain the push information for the target user includes: Based on the updated user profile, a user-product interaction matrix is ​​constructed, where the rows of the matrix represent users, the columns of the matrix represent products, and the values ​​of the matrix represent the historical interaction intensity of users with products. The interaction matrix is ​​decomposed using a matrix factorization algorithm to obtain the latent feature vector of the target user and the latent feature vector of each product. The user latent feature vector and the product latent feature vector are input into the information push model to obtain the predicted score for each product; Products that meet the preset criteria and are corresponding to the target recommended products are identified. Based on the target recommended product, the product push information is determined.

7. The information push method according to claim 5, characterized in that, The step of inputting the updated user profile into the time recommendation model to obtain the target recommendation time includes: Based on the updated user profile, extract the target user's historical behavioral events and the corresponding times of the historical behavioral events; Based on historical behavioral events and the corresponding times of those events, a behavioral time series of the target user is constructed. Feature extraction is performed on the behavioral time series to obtain a time series feature vector; The time-series feature vector is input into the time recommendation model to obtain the acceptance probability distribution representing the target user's acceptance of the recommended event within a future preset time period; The time corresponding to the probability of satisfying the preset conditions is determined as the target recommended time.

8. The information push method according to claim 5, characterized in that, The step of inputting the updated user profile into the channel recommendation model to obtain the target recommendation channel includes: Based on the updated user profile, the channel selection features of the target user are extracted. The channel selection features include at least one of the following: the target user's attribute features, historical channel preference features, and features of the product to be recommended. The channel selection features are input into the channel recommendation model to obtain the target user's predicted feedback rating for each alternative channel; The channels corresponding to the predicted feedback scores that meet the preset scoring conditions are identified as the target recommendation channels.

9. An information push device, characterized in that, The device includes: The acquisition module is used to acquire real-time operational behavior data of the target user; The event processing module is used to process the real-time operation behavior data into event-based data to obtain operation events. An extraction module is used to extract a multidimensional feature vector of the operation event, input the multidimensional feature vector into a pre-trained event value classification model, and obtain a detection result of whether the operation event is a noise event. The update module is used to update the user profile of the target user based on the operation event, when the operation event is not a noise event, to obtain the updated user profile; The input module is used to input the updated user profile into a preset information push model to obtain the push information for the target user; The push module is used to push the push information to the target user.

10. An information push device, characterized in that, The device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the information push method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the information push method as described in any one of claims 1-8.

12. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device is able to perform the information push method as described in any one of claims 1-8.

Images