Information pushing method, apparatus, device, storage medium and program product

Abstract

The application provides an information pushing method and device, equipment, storage medium and program product, and relates to the field of financial technology or other related fields. The method comprises the following steps: after target recommendation information is pushed to a target user, a response delay time length of the target user for the target recommendation information is obtained, and a sensitivity parameter of a target recommendation information type corresponding to the target user is obtained. The target recommendation information belongs to the target recommendation information type. The response delay time length is adjusted by using the sensitivity parameter to obtain an adjusted response delay time length. If the adjusted response delay time length is greater than or equal to a preset time threshold, the pushing of recommendation information of the target recommendation information type to the user is stopped within a target time. The method of the application can accurately identify the current interest and acceptance state of the user, reduce repeated touch in a low-acceptance scenario, and improve the accuracy of pushing control, conversion effect and user experience.

Description

Technical Field

[0001] This application relates to the field of financial technology or other related fields, and in particular to an information push method, device, equipment, storage medium and program product. Background Technology

[0002] In digital marketing within banks, ad delivery is typically based on user profiles, historical behavioral tags, and preset rules, targeting users across mobile and web channels. However, existing solutions rely heavily on static profiles and fixed rules, making it difficult to promptly reflect user feedback such as continuous ad ignoring or disabling notifications in the delivery strategy. This can easily lead to a misalignment between the delivered content and current user needs. Furthermore, the lack of effective quantification regarding the varying levels of user acceptance across different ad types further exacerbates the problem of insufficient matching between push frequency and content, impacting conversion efficiency and user experience.

[0003] Therefore, how to respond to user feedback in a timely manner during the ad delivery process and improve the match between the push strategy and user acceptance is an urgent problem to be solved. Summary of the Invention

[0004] This application provides an information push method, apparatus, device, storage medium, and program product, which are used to respond to user feedback in a timely manner during the advertising push process and improve the matching between push strategy and user acceptance.

[0005] Firstly, this application provides an information push method, including:

[0006] After pushing target recommendation information to a target user, the response delay time of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user are obtained, and the target recommendation information belongs to the target recommendation information type.

[0007] The response delay duration is adjusted using the sensitivity parameter to obtain the adjusted response delay duration;

[0008] If the adjusted response delay duration is greater than or equal to a preset duration threshold, then the push of the target recommendation information type to the user will be stopped within the target duration.

[0009] Optionally, obtaining the response delay duration of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user includes:

[0010] Based on the target user's historical interaction behavior with the historical target recommendation information pushed by the bank system, a historical interaction behavior feature vector is obtained, which includes click behavior, ignore behavior, response delay, and complaint behavior.

[0011] Based on the historical interaction behavior feature vector, the initial sensitivity parameters of the target recommendation information type corresponding to the target user are updated and learned to obtain the sensitivity parameters of the target recommendation information type corresponding to the target user.

[0012] Optionally, the step of obtaining a historical interaction behavior feature vector based on the target user's historical interaction behavior with historical target recommendation information pushed by the bank system includes:

[0013] Based on the target user’s historical interaction behavior with the historical target recommendation information pushed by the bank system, the initial behavioral characteristics of the target user are obtained. The initial behavioral characteristics include the number of clicks, the number of ignored videos, the average response delay, and the historical complaint rate.

[0014] Cluster analysis is performed on the initial behavioral characteristics of the target user and the initial behavioral characteristics of other users to obtain at least two user groups and the cluster centers of each user group;

[0015] Based on the cluster center of the target user group to which the target user belongs, the behavioral characteristics of the target user are determined, and based on the behavioral characteristics of the target user, the historical interaction behavior feature vector is obtained.

[0016] Optionally, the step of updating and learning the initial sensitivity parameters of the target recommendation information type corresponding to the target user based on the historical interaction behavior feature vector to obtain the sensitivity parameters of the target recommendation information type corresponding to the target user includes:

[0017] The historical interaction behavior feature vector is weighted using the initial sensitivity parameter to obtain a weighted feature vector;

[0018] Calculate the predicted interaction feedback of the target user to the target recommendation information based on the weighted feature vector;

[0019] Based on the deviation between the target user's actual interaction feedback with the target recommendation information and the predicted interaction feedback, the initial sensitivity parameter is iteratively updated to obtain the target user's sensitivity parameter for the target recommendation information type.

[0020] Optionally, the method further includes:

[0021] Within the target duration, obtain sensitivity parameters for other recommendation information types corresponding to the target user, wherein the other recommendation information types are recommendation information types other than the target recommendation information type;

[0022] Recommendations of other recommendation information types are pushed to the target user according to the sorting order of the sensitivity parameters of the other recommendation information types.

[0023] Optionally, the method further includes:

[0024] When the target user is a new user, during the first preset number of pushes to the new user, test recommendation information of different recommendation information types or under different push strategies is pushed to the new user.

[0025] Collect feedback data on the interaction behavior of the newly added users during each test push process;

[0026] Based on the interaction behavior feedback data, the initial sensitivity parameter configuration for each type of recommended information for the new user is determined.

[0027] Optionally, the method further includes:

[0028] After the target duration ends, a preset number of recommendation messages of the target recommendation type will be pushed to the target user.

[0029] Obtain the click feedback and response delay duration of the target user for the recommended information of the target recommendation type;

[0030] If the click-through rate of the recommended information of the target recommendation type is greater than or equal to a preset click-through rate threshold, and the response delay is less than or equal to a preset cancellation threshold, then the recommended information of the target recommendation type will be pushed to the target user again.

[0031] Optionally, the method further includes:

[0032] Upon receiving a deactivation instruction from the target user regarding the target recommendation information, and / or, if the number of consecutive ignore actions by the target user regarding the target recommendation information reaches a preset ignore count threshold, the push of recommendation information of the target recommendation information type to the target user will be stopped within the target duration.

[0033] Secondly, embodiments of this application provide an information push device, including:

[0034] The acquisition module is used to acquire, when pushing target recommendation information to a target user, the response delay duration of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user, wherein the target recommendation information belongs to the target recommendation information type;

[0035] The processing module is used to adjust the response delay duration using the sensitivity parameter to obtain the adjusted response delay duration;

[0036] The control module is configured to stop pushing the target recommendation information type to the user within the target duration if the adjusted response delay duration is greater than or equal to a preset duration threshold.

[0037] Thirdly, embodiments of this application provide an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0038] The memory stores the instructions that the computer executes;

[0039] The processor executes computer execution instructions stored in memory to implement the methods provided above.

[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided above.

[0041] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0042] The information push method, apparatus, device, storage medium, and program product provided in this application, after pushing target recommendation information to a target user, obtains the response delay time of the target user to the target recommendation information and the sensitivity parameter corresponding to the type of target recommendation information, and adjusts the response delay time using the sensitivity parameter. When the adjusted response delay time is greater than or equal to a preset time threshold, the push of recommendation information of the target recommendation information type to the user is stopped within the target time. This can target and suppress the push of specific types of recommendation information based on the actual feedback of users to different types of recommendation information, thereby improving the matching between the push strategy and user acceptance, and reducing the adverse effects of invalid or interfering pushes on user experience and recommendation conversion effect. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0044] Figure 1 A flowchart illustrating an information push method provided in an embodiment of this application;

[0045] Figure 2 A flowchart illustrating another information push method provided in an embodiment of this application;

[0046] Figure 3 A flowchart illustrating yet another information push method provided in an embodiment of this application;

[0047] Figure 4 A flowchart illustrating another information push method provided in an embodiment of this application;

[0048] Figure 5 A flowchart illustrating another information push method provided in an embodiment of this application;

[0049] Figure 6 This is a schematic diagram of the structure of an information push device provided in an embodiment of this application;

[0050] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0051] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0052] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0053] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0054] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0055] It should be noted that the information push method, apparatus, equipment, storage medium and program products provided in this application can be used in the field of fintech or other related fields, or in any field other than fintech or other related fields. The application fields of the information push method, apparatus, equipment, storage medium and program products in this application are not limited.

[0056] Push advertising is primarily used for delivering recommended information through mobile banking applications, web pages, and self-service terminals. In actual business operations, the back-end marketing system typically combines user profiles, historical behavioral tags, and delivery rules to distribute different types of advertisements, such as loans, credit cards, wealth management, or insurance, to target users, and adjusts subsequent outreach arrangements based on user interactions.

[0057] Current bank advertising push solutions typically determine whether to continue pushing similar content to the user after an ad display, based on the user's basic attributes, historical click records, and preset frequency control rules. Their operation is largely centered on existing tags and static rules, determining ad content and push frequency based on the user's customer group, past preferences, and preset priorities, thereby achieving multi-channel marketing reach.

[0058] However, existing solutions are relatively weak in responding to user feedback, especially when users show prolonged response delays, continuous ignoring, or significant low interest in certain types of recommendations. It's often difficult to promptly identify the discrepancy between this type of information and the user's current level of acceptance, and the same type of advertising content may continue to be pushed at the original pace. Furthermore, different ad types have varying levels of psychological acceptance from users. Without quantitative processing to quantify the sensitivity of ad types, relying solely on uniform durations or rules to judge user intent can easily lead to inaccurate judgments. This results in highly sensitive information being repeatedly delivered to users in a low-acceptance state, reducing conversion rates and negatively impacting user experience.

[0059] In view of this, this application provides an information push method. After pushing target recommendation information to a target user, the method obtains the response delay duration of the target user to the target recommendation information and a sensitivity parameter corresponding to the type of target recommendation information. The response delay duration is then adjusted using this sensitivity parameter. When the adjusted response delay duration is greater than or equal to a preset duration threshold, the method stops pushing recommendation information of that type to the user for the target duration. This method can be deployed in a bank marketing architecture consisting of a user behavior collection module, a silent period control engine, an advertising sensitivity grading module, and an advertising distribution system to achieve timely control over recommendation information of the same type.

[0060] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0061] Figure 1 This is a flowchart illustrating an information push method provided in an embodiment of this application. Figure 1 As shown, the method may include the following steps:

[0062] S101. After pushing target recommendation information to the target user, obtain the response delay time of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user.

[0063] Among them, target recommendation information belongs to the target recommendation information type.

[0064] In this application, the target user is the user who receives targeted recommendation information and interacts with it. The processing can be based on the user's current display behavior, historical interaction records, and subsequent push status. Target recommendation information can be, for example, advertisements, prompts, or other recommended information. Target recommendation information is the specific content actually delivered to the target user. This content has been categorized by the marketing system before entering the delivery process. The type of target recommendation information can be loan, credit card, wealth management, insurance, or other target recommendation information categories pre-configured in the bank's digital marketing system.

[0065] The response delay duration is used to characterize the time it takes for a target user to respond to the target recommendation information, while the sensitivity parameter is used to quantify the sensitivity of the target recommendation information type to the target user. The two types of data together constitute the input for subsequent control judgments.

[0066] In this step, the executing entity can be, for example, an information push control server deployed in the bank's marketing architecture, or a backend processing cluster that communicates with mobile banking applications (Apps), web pages, or self-service terminals. After the targeted recommendation information is delivered, the start time of the push, user ID, target recommendation information ID, and target recommendation information type ID are recorded, and the push is written to the behavior collection queue.

[0067] The system continuously monitors user interactions after a given display. These interactions can include clicks, closing, swiping, remaining on the page without clicking, and leaving the page. The first valid interaction corresponding to the target recommendation information is taken as the response time; the time difference between the display start time and the response time is determined as the response delay duration. If no interaction is detected within the preset monitoring window, the end time of the monitoring window or the page exit time is taken as the response time, thus forming the corresponding response delay duration. The resulting response delay duration is stored in seconds, milliseconds, or a system-wide unified time unit and associated with the user identifier and the target recommendation information type.

[0068] The sensitivity parameter for the type of target recommendation information corresponding to the target user can be obtained by the system. This sensitivity parameter can be obtained, for example, by pre-configuration of the system, based on historical data analysis, or through other feasible methods. A higher value indicates that the target user is more sensitive to this type of information; a lower value indicates that the user is relatively more accepting of this type of information.

[0069] Based on the above analysis, this step provides a quantifiable data basis for subsequent push control by simultaneously acquiring real-time feedback data and type-sensitive data after a single push is completed.

[0070] S102. Adjust the response delay duration using the sensitivity parameter to obtain the adjusted response delay duration.

[0071] Among them, adjusting the response delay time using sensitivity parameters refers to combining time data that characterizes the user's current response speed with parameters that characterize the sensitivity of the target recommendation information type to obtain a more suitable time result for determining whether to stop pushing.

[0072] The response delay time reflects the user's immediate reaction speed after the target recommendation information is displayed, while the sensitivity parameter reflects the user's acceptance characteristics of the target recommendation information type. The combination of the two can form a judgment on whether to continue pushing the same type of information.

[0073] In this step, the response delay duration and sensitivity parameter obtained in step S101 can be read, and the response delay duration can be corrected according to a preset adjustment rule to obtain the adjusted response delay duration. The adjusted response delay duration is the basis for triggering the stop push notification. For example, the response delay duration can be multiplied by the sensitivity parameter to correct the response delay duration and obtain the adjusted response delay duration.

[0074] This step introduces a sensitivity parameter to categorize and correct the time results, resulting in a higher degree of alignment between subsequent threshold judgments and actual user acceptance. Therefore, the adjusted response delay in this application is no longer a simple time difference, but a comprehensive measure that includes the relationship between the target user's immediate feedback and their acceptance of that information type.

[0075] S103. If the adjusted response delay duration is greater than or equal to the preset duration threshold, then stop pushing the target recommendation information type to the user within the target duration.

[0076] The preset duration threshold is used to limit the critical conditions for triggering the cessation of push notifications, while the target duration is used to limit the duration of the cessation of push notifications.

[0077] After obtaining the adjusted response delay duration in step S102, the result can be compared with a preset duration threshold. When the adjusted response delay duration reaches or exceeds the threshold, it can be determined that the target user is currently in a low-acceptance state for the target recommendation information type, and the push of target recommendation information of the target recommendation information type to the target user will be stopped within the target duration.

[0078] In this step, the preset duration threshold and target duration can be pre-set by the system. Within the target duration, subsequent recommendation information belonging to the target recommendation information type will stop being pushed to the target user.

[0079] If the adjusted response delay is less than a preset threshold, the system maintains the pushable status of this type of target recommendation information. For example, the preset threshold can be set to 15 seconds. If the adjusted response delay is greater than or equal to 15 seconds, it indicates that the target user has a very low willingness to interact after receiving this type of recommendation information and a strong tendency to resist. Continuous pushes are likely to cause user resentment and churn. Therefore, it is necessary to suspend the push of this type of recommendation information to the target user within the target duration. If the adjusted response delay is less than 15 seconds, it indicates that the target user has a good acceptance of this type of recommendation information and no obvious rejection behavior. Therefore, it is necessary to maintain the normal push permission for this type of recommendation information.

[0080] By comparing the adjusted response delay with a preset threshold, and stopping the push of target recommendation information of the target type to the target user within the target duration when the conditions are met, the system can respond promptly to user feedback on low acceptance after a single push, and form targeted quiet period control for the corresponding type of information, instead of continuing to deliver repeatedly at the original pace. This ensures that the bank's outreach strategy is consistent with the user's current acceptance status.

[0081] The method provided in this application involves obtaining the response delay duration of the target user to the target recommendation information and a sensitivity parameter of the target recommendation information type corresponding to the target user after pushing target recommendation information to the target user. The response delay duration is adjusted using the sensitivity parameter to obtain an adjusted response delay duration. When the adjusted response delay duration is greater than or equal to a preset duration threshold, the push of target recommendation information of the target recommendation information type to the user is stopped within the target duration. In this application, real-time user feedback data and type-sensitive data are combined for the quiet period control of the same type of target recommendation information. Push control no longer relies solely on static frequency control rules but is based on a linkage judgment based on the response delay and sensitivity parameter after the current display, so that the stop-push condition can match the user's acceptance state corresponding to different target recommendation information types.

[0082] The following section provides a detailed explanation of how to obtain the response delay time of the target user to the target recommendation information and the sensitivity parameters of the target recommendation information type corresponding to the target user in step S101. Figure 2 This is a flowchart illustrating another information push method provided in an embodiment of this application. Figure 2 As shown, the aforementioned step S101 may specifically include the following steps:

[0083] S201. Based on the target user's historical interaction behavior with the historical target recommendation information pushed by the bank system, obtain the historical interaction behavior feature vector.

[0084] Historical interaction behaviors include click behavior, ignore behavior, response delay, and complaint behavior.

[0085] Historical interaction behavior refers to the actions of target users in response to historical target recommendation information pushed by the bank system over a past period. Click behavior refers to the target user clicking on historical target recommendation information; ignore behavior refers to the target user not clicking or interacting after seeing historical target recommendation information; response latency refers to the time interval between the target user seeing historical target recommendation information and taking an interaction; complaint behavior refers to the target user complaining about historical target recommendation information. The historical interaction behavior feature vector is a vector obtained by vectorizing the various features of historical interaction behavior, and its number of dimensions corresponds to the number of features of the historical interaction behavior.

[0086] One possible approach is to statistically analyze the target user's historical interaction behavior data, extract various features, and vectorize them to obtain a historical interaction behavior feature vector. For example, one could collect all interaction logs corresponding to four categories of historical target recommendation information pushed by the bank system to the target user within the past 30 days: loans, wealth management, credit cards, and insurance. Raw data on click behavior, ignore behavior, response delay, and complaint behavior corresponding to each push could be extracted. After data cleaning, missing value imputation, and removal of abnormal delay data, the corresponding behavioral indicators could be summarized and statistically analyzed. Then, a unified normalization and standardization process could be performed to eliminate differences in time and push frequency, ultimately completing the feature vectorization to generate the historical interaction behavior feature vector.

[0087] Another possible implementation can be achieved through the following steps:

[0088] S2011. Based on the target user's historical interaction behavior with the historical target recommendation information pushed by the bank system, the initial behavioral characteristics of the target user are obtained.

[0089] Initial behavioral characteristics include the number of clicks, the number of ignored videos, the average response delay, and the historical complaint rate.

[0090] Click count is the total number of times a target user clicks on historical target recommendations over a past period; Ignore count is the total number of times a target user ignores historical target recommendations over a past period; Average response latency is the average of all response latencyes for historical target recommendations over a past period; Historical complaint rate is the ratio of the number of complaints a target user has made about historical target recommendations over a past period to the total number of historical target recommendations pushed to them.

[0091] In this step, the original interaction logs of the target user can be streamed in real time based on a preset sliding window (the window duration can be set to 5 seconds, for example), and the number of clicks and the number of ignored videos within the window period can be summarized and calculated respectively; the response delay of each valid push is normalized, and the response delay of each valid push can be calculated by the following formula (1):

[0092] (1)

[0093] in, For example, it can be set to 30 seconds. For example, it can be set to 0 seconds.

[0094] The response latency for each valid push was calculated. Then, the average response delay can be obtained by calculating the mean value.

[0095] The historical complaint rate can be calculated by dividing the total number of complaints by the total number of times the target recommendation information was pushed during the same period.

[0096] The above four values ​​together constitute the initial behavioral characteristics of the target user.

[0097] S2012. Perform cluster analysis on the initial behavioral characteristics of the target user and the initial behavioral characteristics of other users to obtain at least two user groups and the cluster centers of each user group.

[0098] Cluster analysis is a data analysis method that groups users with similar characteristics. User groups are user groups obtained after cluster analysis, and cluster centers are feature centers of each user group.

[0099] In this step, cluster analysis is performed on the initial behavioral characteristics of the target user and other users. For example, K-means clustering or hierarchical clustering algorithms can be used. For instance, using K-means clustering, users can be divided into different user groups based on the similarity of their initial behavioral characteristics, and the cluster center for each user group can be calculated.

[0100] For example, the number of clusters K=3 can be preset, corresponding to the highly sensitive and resistant user group, the moderately tolerant ordinary user group, and the low-sensitive and high-conversion user group, respectively. The initial behavioral characteristics of all users are input into the K-means model, and the Euclidean distance between each user sample and the cluster center is calculated iteratively. The cluster centers are continuously updated until the center values ​​converge and no longer change. Finally, three user groups are output, along with the four-dimensional cluster center corresponding to each group (i.e., the mean of four dimensions: number of clicks, number of ignored videos, average response latency, and historical complaint rate, respectively).

[0101] S2013. Based on the cluster center of the target user group to which the target user belongs, determine the behavioral characteristics of the target user, and based on the behavioral characteristics of the target user, obtain the historical interaction behavior feature vector.

[0102] The target user group to which the target user belongs is the user group to which the target user belongs after cluster analysis. The behavioral characteristics of the target user can be determined based on the cluster centers of this user group. For example, the feature values ​​of the cluster centers can be used as the behavioral feature values ​​of the target user. The historical interaction behavior feature vector obtained based on the behavioral characteristics of the target user can be arranged into a vector form according to a certain order.

[0103] For example, suppose the target user is classified into a highly sensitive and resistant user group. The four-dimensional values ​​of the cluster center of this group are as follows: number of clicks 12, number of ignored videos 46, average response delay 22 seconds, and historical complaint rate 0.18. Then, the four-dimensional values ​​of the cluster center can be directly used as the behavioral features of the target user, and arranged in order of [number of clicks, number of ignored videos, average response delay, historical complaint rate] to generate a corresponding four-dimensional historical interaction behavior feature vector.

[0104] S202. Based on the historical interaction behavior feature vector, update and learn the initial sensitivity parameters of the target recommendation information type corresponding to the target user to obtain the sensitivity parameters of the target recommendation information type corresponding to the target user.

[0105] The initial sensitivity parameter is a pre-set parameter used to measure the sensitivity of target users to the type of target recommendation information. Update learning refers to the process of adjusting and optimizing the initial sensitivity parameter based on historical interaction behavior feature vectors. The initial sensitivity parameter can be set based on historical experience and / or expert opinions; this application does not limit its specific value. For example, suppose the initial sensitivity parameters can be: loan recommendation information = 1.5 (representing highly sensitive recommendation information), financial recommendation information = 1.0 (representing moderately sensitive recommendation information), and credit card recommendation information = 0.8 (representing low-sensitivity recommendation information).

[0106] One possible approach is to use machine learning algorithms to train the historical interaction behavior feature vectors and initial sensitivity parameters to obtain updated sensitivity parameters.

[0107] Another possible implementation can be achieved through the following steps:

[0108] S2021. The historical interaction behavior feature vector is weighted using the initial sensitivity parameter to obtain a weighted feature vector.

[0109] In this step, weighting refers to multiplying the initial sensitivity parameter by each dimension of the historical interaction behavior feature vector. The weighted feature vector is the vector obtained after weighting. For example, suppose the historical interaction behavior feature vector is [N1, N2, T, P], where N1 is the number of clicks, N2 is the number of ignored videos, T is the average response delay, and P is the historical complaint rate. Then the weighted feature vector is Feature = [N1, N2, T, P] * γ, where γ is the initial sensitivity parameter.

[0110] S2022. Calculate the target user's predicted interaction feedback on the target recommendation information based on the weighted feature vector.

[0111] Among them, predicting interactive feedback is to predict the possible interactive behavior of the target user in response to the target recommendation information based on the weighted feature vector.

[0112] In this step, the predicted interaction feedback values ​​can be calculated using methods such as regression analysis, logistic regression, and linear prediction models. For example, a weighted feature vector can be input into a linear prediction model, which outputs a predicted interaction feedback score in the range of 0-1. The closer the score is to 1, the more likely the predicted user is to generate a positive interaction by clicking; the closer the score is to 0, the more likely the predicted user is to generate negative interactions such as ignoring or experiencing high latency.

[0113] S2023. Based on the deviation between the target user's actual interaction feedback and the predicted interaction feedback to the target recommendation information, iteratively update the initial sensitivity parameters to obtain the target user's sensitivity parameters for the target recommendation information type.

[0114] Actual interaction feedback is the numerical value corresponding to the actual interaction behavior of the target user with the target recommendation information. Deviation is the difference between the actual interaction feedback and the predicted interaction feedback. Iterative update refers to adjusting the initial sensitivity parameter multiple times based on the deviation until the deviation is less than a preset threshold. The parameter obtained at this point is the sensitivity parameter of the target user for the target recommendation information type.

[0115] For example, an online regularized learning algorithm (Follow the RegularizedLeader, FTRL) can be used, which uses the difference between the actual interaction feedback and the predicted interaction feedback as the loss bias, corrects the initial sensitivity parameter γ with the back gradient, and continuously updates it in a streaming manner until the loss bias is less than a preset bias threshold or the preset number of iterations is reached, and then stops iterating. Finally, the corrected sensitivity parameter that is adapted to the target user's target recommendation information type is output.

[0116] The method provided in this application analyzes and processes the historical interaction behavior of the target user to obtain a historical interaction behavior feature vector, and updates the initial sensitivity parameters based on the vector to obtain more accurate sensitivity parameters, thereby improving the accuracy of pushing recommendation information to the target user.

[0117] Figure 3 This is a flowchart illustrating another information push method provided in an embodiment of this application. Figure 3 As shown, the method may also include the following steps:

[0118] S301. Within the target duration, obtain the sensitivity parameters of other recommendation information types corresponding to the target user.

[0119] Among these, "other recommendation information types" refers to recommendation information types other than the target recommendation information type. The target duration refers to the length of time during which recommendation information of the target type will no longer be pushed to the target user. "Other recommendation information types" are those other than the target recommendation information type. For example, if the target recommendation information type is financial product recommendations, then "other recommendation information types" could be lifestyle service recommendations, news, etc.

[0120] In this step, the sensitivity parameters for other recommendation information types corresponding to the target user can be obtained by calling the sensitivity parameter data for other recommendation information types of the target user, which is pre-stored in the data storage unit. The calculation method for the sensitivity parameters of other recommendation information types is the same as that for the sensitivity parameters of the target recommendation information type, and will not be repeated here.

[0121] S302. Push other recommendation information types to the target user according to the sorting order of the sensitivity parameters of other recommendation information types.

[0122] The sorting order is based on the magnitude of the sensitivity parameters of other recommendation information types. For example, it can be sorted in descending order of sensitivity parameters.

[0123] Pushing other types of recommendations to the target user in sorted order can ensure that the target user receives recommendations that better match their interests and needs.

[0124] The method provided in this application embodiment enriches the recommendation information content for the target user and improves the user experience by acquiring and pushing recommendation information of other recommendation information types according to the sorting order of sensitivity parameters of other recommendation information types within a target time period during which the push of recommendation information of the target recommendation information type is stopped.

[0125] Figure 4 This is a flowchart illustrating another information push method provided in an embodiment of this application. Figure 4 As shown, the method may also include the following steps:

[0126] S401. When the target user is a new user, during the first preset number of push notifications to the new user, push test recommendation information of different recommendation information types or under different push strategies to the new user.

[0127] "Target users are new users" refers to users receiving recommendation information for the first time. "Preset number of times" is a pre-set number of test recommendation messages. "Different recommendation information types" refers to various categories of recommendation information. "Different push strategies" refers to different push timing, frequency, and methods. Test recommendation information is used to test the feedback of new users to different recommendation information types and push strategies.

[0128] In this step, information related to the target user's account, such as registration duration and historical information recommendations, can be used to determine whether the target user is a new user. If the target user is a new user, test recommendations of different types or under different push strategies can be pushed to the target user first. This is to quickly collect the new user's interaction preferences for various types of recommendation information, rapidly establish basic behavioral samples specific to the new user, and then generate initial sensitivity parameters for various types of recommendation information for the new user. This reduces the risk of user resistance caused by an imbalance in the push strategy due to a lack of historical user behavior data and excessive push of highly sensitive recommendation information.

[0129] S402. Collect feedback data on the interactive behavior of new users during each test push.

[0130] Interaction behavior feedback data refers to the data related to the interactive behaviors of new users after receiving test recommendation information, such as click behavior, ignore behavior, response delay, and complaint behavior. This data can be collected by recording and storing the interactive behaviors of new users.

[0131] S403. Based on the interaction behavior feedback data, determine the initial sensitivity parameter configuration for each type of recommended information for new users.

[0132] The initial sensitivity parameter configuration is the initial sensitivity parameter set for each type of recommended information for new users.

[0133] In this step, all collected interaction feedback data can be summarized and statistically analyzed to calculate the number of clicks, ignored videos, average response delay, and complaint rate for each type of recommendation information for new users, thereby constructing exclusive initial behavioral characteristics for new users.

[0134] Then, the initial behavioral features are multiplied by the baseline initial sensitivity parameters corresponding to various recommendation information types to obtain a weighted feature vector. Based on the user clustering group matching rules completed in the pre-training, the corresponding customer groups are matched, and then a round of parameter iteration is completed through FTRL to output the initial sensitivity parameter configuration of each recommendation information type adapted to the new user.

[0135] The method provided in this application pushes test recommendation information and collects interaction behavior feedback data during the initial preset number of pushes for new users. Based on the feedback data, it determines the initial sensitivity parameter configuration for each type of recommendation information for new users, so as to achieve accurate initialization of recommendation information push for new users.

[0136] Figure 5 This is a flowchart illustrating another information push method provided in an embodiment of this application. Figure 5As shown, the method may also include the following steps:

[0137] S501. After the target duration ends, push a preset number of target recommendation information types to the target user.

[0138] The target duration is the control period for stopping the push of the corresponding type of recommendation information. This can include durations such as 1 day, 7 days, or 30 days, which can be customized in the backend. The preset number of times is the number of test pushes used to verify user interaction preferences after the target duration expires; this can include, for example, 3 consecutive times. The target recommendation information type is the category of recommendation information that previously triggered the control and suspension of pushes; this can include loan-related, wealth management, and credit card-related information.

[0139] In this step, the control expiration time stored in the system can be read. After the target duration expires, the verification process will be automatically started, the pre-configured number of test pushes will be read, and recommendation information consistent with the original control type will be generated and pushed to the target user through the regular push channel.

[0140] For example, assuming the target duration is 7 days and the preset number of times is set to 3 consecutive times, then after the 7-day control period ends, 3 similar loan recommendation messages can be pushed to the target user as verification test material.

[0141] S502. Obtain the target user's click feedback and response delay time for the target recommendation information type.

[0142] Click feedback records positive user interactions with the test push notification, such as clicking or viewing details. Response latency is the time interval between displaying the recommendation information and the user's interaction, such as 2 seconds, 10 seconds, or 20 seconds.

[0143] In this step, the display logs and user operation logs of each test push can be collected in real time. The streaming behavior data is standardized using a 5-second sliding window, and the click mark and original response delay time corresponding to each test push are recorded respectively.

[0144] For example, the dwell time of each of the three test pushes can be recorded. If the dwell time of the user is less than 3 seconds, it is marked as no click feedback. At the same time, the original response delay corresponding to each push is calculated, and then the delay data is standardized by the normalization formula shown in the above formula (1).

[0145] S503. If the click-through rate of the target recommendation information type is greater than or equal to the preset click-through rate threshold, and the response delay is less than or equal to the preset cancellation threshold, then the push of the target recommendation information type to the target user will resume.

[0146] The click-through rate (CTR) is the ratio of the number of clicks generated in the test pushes to the total number of test pushes; for example, it can include percentages such as 20% or 30%. The preset CTR threshold is the minimum conversion standard required to remove push control, for example, it can be set to 20%. The preset removal threshold is the maximum average response delay for determining that users do not exhibit any resistance; for example, it can be set to 8 seconds.

[0147] In this step, the click records of all preset number of test pushes can be counted to calculate the overall click-through rate, and the average response delay time of multiple rounds of test pushes can be obtained. The two indicators are compared with the corresponding preset thresholds respectively. If both conditions are met at the same time, the push control restrictions for this category are lifted.

[0148] For example, suppose the preset click-through rate threshold is 20% and the preset cancellation threshold is 8 seconds. If one out of three test pushes to the target user generates a click, the click-through rate is approximately 33.3% (≥20%), and the average response delay of the three pushes is 6 seconds (≤8 seconds), then it is determined that the user's resistance has been eliminated, and the normal push of this type of recommendation information is restored.

[0149] The method provided in this application embodiment pushes a preset number of target recommendation information types after the target duration ends, obtains click feedback and response delay duration, and determines whether to resume pushing based on relevant thresholds, so as to realize the dynamic recovery of recommendation information push for target users and improve the flexibility of recommendation information push.

[0150] Optionally, in embodiments of this application, upon receiving a deactivation instruction from a target user regarding the target recommendation information, the push of recommendation information of the target recommendation information type may be stopped for a target duration.

[0151] Specifically, this can be achieved by receiving a proactive deactivation command from the target user. This command can be sent by the target user through relevant interfaces or channels within the bank's system, such as when the target user selects to deactivate recommendations of the target recommendation type in the recommendation settings interface of the bank's app. Receiving this command indicates that the target user clearly has a negative intention to reject this type of recommendation. If it continues to be pushed, it is likely to cause negative behaviors such as user complaints and refusing notification permissions. Therefore, the push of recommendations of this target recommendation type to the target user can be stopped for a target period of time.

[0152] Alternatively, if the number of times a target user ignores the target recommendation information reaches a preset threshold, the system can stop pushing the target recommendation information type to the target user for the target duration.

[0153] Specifically, the number of times a target user continuously ignores the target recommendation information can be recorded. When this number reaches a preset threshold, it indicates that the target user has no intention of interacting with this type of recommendation information for a long time and has a clear tendency to resist it. Continuous push will reduce the user experience. Therefore, a stop push mechanism can be triggered to stop pushing the target recommendation information of this type to the target user within the target time period.

[0154] Alternatively, if a user actively disables the target recommendation information upon receiving such a notification, and the number of times the user ignores the target recommendation information reaches a preset threshold, the system can stop pushing the target recommendation information type to the target user for the target duration.

[0155] Specifically, by simultaneously meeting two conditions—receiving a proactive disabling instruction and the number of consecutive ignore actions reaching a preset ignore threshold—it can be indicated that the target user has an extremely high level of resistance to this type of recommendation information, requiring a longer or standard duration of push restrictions. Therefore, a stop push mechanism can be triggered to stop pushing recommendation information of this target type to the target user for the target duration or longer.

[0156] Figure 6 This is a schematic diagram of the structure of an information push device provided in an embodiment of this application. Figure 6 As shown, the information push device may include: an acquisition module 11, a processing module 12, and a control module 13.

[0157] The acquisition module 11 is used to acquire, when pushing target recommendation information to target users, the response delay time of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user, wherein the target recommendation information belongs to the target recommendation information type.

[0158] Processing module 12 is used to adjust the response delay duration using a sensitivity parameter to obtain the adjusted response delay duration.

[0159] The control module 13 is used to stop pushing the target recommendation information type to the user within the target duration if the adjusted response delay duration is greater than or equal to the preset duration threshold.

[0160] Optionally, module 11 is specifically used to obtain a historical interaction behavior feature vector based on the target user's historical interaction behavior with the historical target recommendation information pushed by the bank system. The initial sensitivity parameters of the target recommendation information type corresponding to the target user are updated and learned based on the historical interaction behavior feature vector to obtain the sensitivity parameters of the target recommendation information type corresponding to the target user. The historical interaction behavior includes click behavior, ignore behavior, response delay, and complaint behavior.

[0161] Optionally, module 11 is specifically used to obtain the initial behavioral characteristics of the target user based on the target user's historical interaction behavior with historical target recommendation information pushed by the bank system. Cluster analysis is performed on the initial behavioral characteristics of the target user and the initial behavioral characteristics of other users to obtain at least two user groups and the cluster centers of each user group. Based on the cluster centers of the target user groups to which the target user belongs, the behavioral characteristics of the target user are determined, and based on the behavioral characteristics of the target user, a historical interaction behavior feature vector is obtained. The initial behavioral characteristics include the number of clicks, the number of ignored views, the average response latency, and the historical complaint rate.

[0162] Optionally, module 11 is specifically used to weight the historical interaction behavior feature vector with the initial sensitivity parameter to obtain a weighted feature vector. Based on the weighted feature vector, the predicted interaction feedback of the target user to the target recommendation information is calculated. Based on the deviation between the target user's actual interaction feedback and the predicted interaction feedback to the target recommendation information, the initial sensitivity parameter is iteratively updated to obtain the target user's sensitivity parameter for the type of target recommendation information.

[0163] Optionally, the acquisition module 11 is further configured to acquire sensitivity parameters of other recommendation information types corresponding to the target user within the target duration, wherein the other recommendation information types are recommendation information types other than the target recommendation information type. The control module 13 is further configured to push recommendation information of other recommendation information types to the target user according to the sorting order of the sensitivity parameters of the other recommendation information types.

[0164] Optionally, the control module 13 is further configured to, when the target user is a new user, push test recommendation information of different recommendation information types or under different push strategies to the new user during the first preset number of pushes. The acquisition module 11 is further configured to collect the interaction behavior feedback data of the new user during each test push. The processing module 12 is further configured to determine the initial sensitivity parameter configuration of the new user for each recommendation information type based on the interaction behavior feedback data.

[0165] Optionally, the control module 13 is further configured to push a preset number of recommended information of the target recommendation type to the target user after the target duration has ended. The acquisition module 11 is further configured to acquire the click feedback and response delay duration of the target user on the recommended information of the target recommendation type. The control module 13 is further configured to resume pushing recommended information of the target recommendation type to the target user if the click rate of the recommended information of the target recommendation type is greater than or equal to a preset click rate threshold and the response delay duration is less than or equal to a preset cancellation threshold.

[0166] Optionally, the control module 13 is also configured to stop pushing target recommendation information of the target recommendation information type to the target user within the target duration when it receives a deactivation instruction from the target user for the target recommendation information, and / or when the number of consecutive ignore behaviors of the target user for the target recommendation information reaches a preset ignore count threshold.

[0167] The information push device provided in this application embodiment can execute the information push method in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described again here.

[0168] Figure 7 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device is used to execute the aforementioned information push method. Figure 7 As shown, the electronic device 700 may include at least one processor 701, a memory 702, and a communication interface 703.

[0169] The memory 702 is used to store programs. Specifically, the program may include program code, which includes computer operation instructions.

[0170] The memory 702 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage.

[0171] The processor 701 is used to execute computer execution instructions stored in the memory 702 to implement the method described in the foregoing method embodiments. The processor 701 may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0172] The processor 701 can communicate and interact with external devices through the communication interface 703. In specific implementations, if the communication interface 703, memory 702, and processor 701 are implemented independently, they can be interconnected via a bus to complete communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0173] Optionally, in a specific implementation, if the communication interface 703, memory 702, and processor 701 are integrated on a single chip, then the communication interface 703, memory 702, and processor 701 can communicate through an internal interface.

[0174] This application also provides a computer-readable storage medium, which may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. Specifically, the computer-readable storage medium stores program instructions, which are used in the methods described in the above embodiments.

[0175] This application also provides a program product including executable instructions stored in a readable storage medium. At least one processor of a computing device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the computing device to implement the above-described information push method.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An information push method, characterized in that, The method includes: After pushing target recommendation information to a target user, the response delay time of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user are obtained, and the target recommendation information belongs to the target recommendation information type. The response delay duration is adjusted using the sensitivity parameter to obtain the adjusted response delay duration; If the adjusted response delay duration is greater than or equal to a preset duration threshold, then the push of the target recommendation information type to the user will be stopped within the target duration.

2. The method according to claim 1, characterized in that, The step of obtaining the response delay duration of the target user to the target recommendation information and the sensitivity parameters of the target recommendation information type corresponding to the target user includes: Based on the target user's historical interaction behavior with the historical target recommendation information pushed by the bank system, a historical interaction behavior feature vector is obtained, which includes click behavior, ignore behavior, response delay, and complaint behavior. Based on the historical interaction behavior feature vector, the initial sensitivity parameters of the target recommendation information type corresponding to the target user are updated and learned to obtain the sensitivity parameters of the target recommendation information type corresponding to the target user.

3. The method according to claim 2, characterized in that, The step of obtaining a historical interaction behavior feature vector based on the target user's historical interaction behavior with historical target recommendation information pushed by the bank system includes: Based on the target user’s historical interaction behavior with the historical target recommendation information pushed by the bank system, the initial behavioral characteristics of the target user are obtained. The initial behavioral characteristics include the number of clicks, the number of ignored videos, the average response delay, and the historical complaint rate. Cluster analysis is performed on the initial behavioral characteristics of the target user and the initial behavioral characteristics of other users to obtain at least two user groups and the cluster centers of each user group; Based on the cluster center of the target user group to which the target user belongs, the behavioral characteristics of the target user are determined, and based on the behavioral characteristics of the target user, the historical interaction behavior feature vector is obtained.

4. The method according to claim 2, characterized in that, The step of updating and learning the initial sensitivity parameters of the target recommendation information type corresponding to the target user based on the historical interaction behavior feature vector to obtain the sensitivity parameters of the target recommendation information type corresponding to the target user includes: The historical interaction behavior feature vector is weighted using the initial sensitivity parameter to obtain a weighted feature vector; Calculate the predicted interaction feedback of the target user to the target recommendation information based on the weighted feature vector; Based on the deviation between the target user's actual interaction feedback with the target recommendation information and the predicted interaction feedback, the initial sensitivity parameter is iteratively updated to obtain the target user's sensitivity parameter for the target recommendation information type.

5. The method according to claim 1, characterized in that, The method further includes: Within the target duration, obtain sensitivity parameters for other recommendation information types corresponding to the target user, wherein the other recommendation information types are recommendation information types other than the target recommendation information type; Recommendations of other recommendation information types are pushed to the target user according to the sorting order of the sensitivity parameters of the other recommendation information types.

6. The method according to claim 1, characterized in that, The method further includes: When the target user is a new user, during the first preset number of pushes to the new user, test recommendation information of different recommendation information types or under different push strategies is pushed to the new user. Collect feedback data on the interaction behavior of the newly added users during each test push process; Based on the interaction behavior feedback data, the initial sensitivity parameter configuration for each type of recommended information for the new user is determined.

7. The method according to claim 1, characterized in that, The method further includes: After the target duration ends, a preset number of recommendation messages of the target recommendation type will be pushed to the target user. Obtain the click feedback and response delay duration of the target user for the recommended information of the target recommendation type; If the click-through rate of the recommended information of the target recommendation type is greater than or equal to a preset click-through rate threshold, and the response delay is less than or equal to a preset cancellation threshold, then the recommended information of the target recommendation type will be pushed to the target user again.

8. The method according to claim 1, characterized in that, The method further includes: Upon receiving a deactivation instruction from the target user regarding the target recommendation information, and / or, if the number of consecutive ignore actions by the target user regarding the target recommendation information reaches a preset ignore count threshold, the push of recommendation information of the target recommendation information type to the target user will be stopped within the target duration.

9. An information push device, characterized in that, include: The acquisition module is used to acquire, when pushing target recommendation information to a target user, the response delay duration of the target user to the target recommendation information and the sensitivity parameter of the target recommendation information type corresponding to the target user, wherein the target recommendation information belongs to the target recommendation information type; The processing module is used to adjust the response delay duration using the sensitivity parameter to obtain the adjusted response delay duration; The control module is configured to stop pushing the target recommendation information type to the user within the target duration if the adjusted response delay duration is greater than or equal to a preset duration threshold.

10. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 8.

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

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images