Method, device and storage medium for displaying message identifiers

Abstract

This application proposes a method, apparatus, and storage medium for controlling the display of message identifiers. It predicts user identifiers whose predicted click probability for clicking to eliminate a first message identifier displayed in a first scenario for a communication client is greater than a first probability threshold within future time windows. When a communication service device receives an access request from any communication client, it can directly determine the preset time window to which the access request's access time belongs, and detect whether the first user identifier contained in the access request belongs to the first predicted user identifier associated with that preset time window. Based on the detection result, when responding to the access request, it determines whether to control the display of the first message identifier in the first scenario for that communication client. This achieves selective and personalized display control of the first message identifier for the user, avoiding frequent disturbance to the first user by the first message identifier and improving the efficiency of user traffic generation for push information in the first scenario.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method, apparatus, and storage medium for displaying message identifiers. Background Technology

[0002] With the rapid development of internet technology, various applications (apps) on the market offer increasingly diverse functions. For example, they use notification icons like red dots to alert users to unread new messages in specific scenarios. Taking WeChat's Moments, Discover, and Video Channels as examples, when a new message is received in a particular scenario, such as… Figure 1 As shown, a red dot will appear at the scene entrance to remind users to check for new messages in that scene.

[0003] However, with the rapid increase in the amount of information on the Internet, the information in this scenario is constantly being updated. Users need to click repeatedly to clear the red dot at the corresponding scenario entrance, which can easily cause user resistance and annoyance. As a result, users click on the red dot displayed in the scenario only to clear it, without actually reading the information content in that scenario. They may even choose to close the application for that scenario and stop receiving information from that scenario, which greatly affects the promotion of the information in that scenario. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, apparatus, and storage medium for displaying message identifiers. To achieve selective display control of message identifiers (such as red dots) in any scenario of a communication client, and to meet the personalized viewing needs of users for different scenario information within different time windows, embodiments of this application provide the following technical solutions:

[0005] On the one hand, this application proposes a method for controlling the display of message identifiers, the method comprising:

[0006] Obtain an access request from a communication client, wherein the access request includes a first user identifier that logged into the communication client and the access request time;

[0007] Query the first predicted user identifier stored in association with the requested access time, wherein the first predicted user identifier refers to the predicted user identifier whose click prediction probability of the first message identifier displayed in the first scenario of the communication client is greater than a first probability threshold within the preset time window to which the requested access time belongs, and the first message identifier is used to indicate that there are unviewed new messages in the first scenario;

[0008] Detect whether the first user identifier belongs to the first predicted user identifier, and obtain the corresponding detection result;

[0009] In response to the access request, and based on the detection result, determine whether to control the first message identifier to be displayed in the first scenario of the communication client.

[0010] In some embodiments, the process of obtaining the predicted user identifier includes:

[0011] The system acquires historical user behavior data, user interest data, and user attribute data related to the communication client; wherein the historical user behavior data is generated based on the user's operational behavior in different message identifier display scenarios of the communication client within different preset time windows;

[0012] The historical user behavior data, user interest data, and user attribute data are input into the message identifier elimination prediction model for processing, and the predicted click probability of the user clicking on the first message identifier displayed in the first scenario within different preset time windows is output.

[0013] If the predicted click probability is greater than the first probability threshold, the user identifier of the user who logged into the communication client will be determined as the predicted user identifier for the corresponding preset time window.

[0014] The predicted user identifier is associated with and stored in relation to the corresponding preset time window.

[0015] In some embodiments, the training process of the message identifier elimination prediction model includes:

[0016] Obtain sample user behavior data, sample user interest data, and sample user attribute data for each sample user;

[0017] Feature extraction is performed on the sample user behavior data, the sample user interest data, and the sample user attribute data respectively to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features.

[0018] Based on the preset time window with a time period, spatial transformation is performed on the sample user behavior features, sample user interest features, and sample user attribute features respectively to obtain corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features with the time period.

[0019] The undetermined behavioral features, undetermined interest features, and undetermined attribute features are merged to obtain the corresponding sample user encoding vector;

[0020] Obtain the target behavioral characteristics of each sample user in the first scenario;

[0021] The network model is trained using the sample user encoding vector and the target behavior features to obtain the message identifier elimination prediction model for the first scenario.

[0022] In some embodiments, the request access time consists of a first time in a first time dimension and a second time in a second time dimension, wherein the first time dimension has a first time period, the second time dimension has a second time period, and the second time period is a time unit of the first time period.

[0023] Therefore, different preset time windows include a first time window consisting of a time unit of the first time period and a second time window consisting of a time unit of the second time period;

[0024] The query is associated with the first predicted user identifier stored in relation to the requested access time, including:

[0025] Determine the target first time window to which the first time belongs, and the target second time window to which the second time belongs;

[0026] The predicted user identifier associated with the second time window of the target under the first time window of the target is determined as the first predicted user identifier.

[0027] In some embodiments, the step of spatially transforming the sample user behavior features, sample user interest features, and sample user attribute features based on the preset time window with a time period to obtain corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features with the time period includes:

[0028] Based on the first time window with a first time period, spatial transformation is performed on the sample user behavior features, the sample user interest features, and the sample user attribute features respectively to obtain the corresponding first undetermined behavior features, first undetermined interest features, and first undetermined attribute features with the first time period.

[0029] Based on the second time window with a second time period, spatial transformation is performed on the sample user behavior features, the sample user interest features, and the sample user attribute features respectively to obtain the corresponding second undetermined behavior features, second undetermined interest features, and second undetermined attribute features with the second time period.

[0030] The process of merging the undetermined behavioral features, undetermined interest features, and undetermined attribute features to obtain the corresponding sample user encoding vector includes:

[0031] The first undetermined behavioral feature, the first undetermined interest feature, the first undetermined attribute feature, the second undetermined behavioral feature, the second undetermined interest feature, and the second undetermined attribute feature are merged to obtain the corresponding sample user encoding vector.

[0032] In some embodiments, the spatial transformation process includes:

[0033] The preset time window within a time period is normalized to obtain a normalized time vector;

[0034] Using the normalized time vector and the parameters of the first training matrix, spatial transformation parameters are constructed;

[0035] Using the spatial transformation parameters, the corresponding feature values ​​of the input features are spatially transformed to obtain the corresponding undetermined input features;

[0036] The input features are the sample user behavior features, the sample user interest features, or the sample user attribute features, and the undetermined input features are the corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features.

[0037] In some embodiments, the step of extracting features from the sample user behavior data, the sample user interest data, and the sample user attribute data to obtain corresponding sample user behavior features, sample user interest features, and sample user attribute features includes:

[0038] The sample user behavior data is input into the first feature extraction network, which outputs the sample user behavior features.

[0039] The sample user interest data is input into the second feature extraction network, and the sample user interest features are output; wherein, the network structures of the first feature extraction network and the second feature extraction network are different, and the first feature extraction network has a first network parameter to be trained, and the second feature extraction network has a second network parameter to be trained;

[0040] The sample user attribute data is vectorized to obtain sample user attribute features;

[0041] The step of training a network model using the sample user encoding vector and the target behavior features to obtain a message identifier removal prediction model includes:

[0042] The sample user encoding vector is input into the pooling layer of the network model for processing to obtain the click prediction probability of the corresponding sample user clicking the first message identifier displayed in the first scene;

[0043] Obtain the loss value between the predicted click probability and the corresponding feature value in the target behavior feature;

[0044] If the loss value is detected to not meet the training constraints, the parameters of the first training matrix, the parameters of the first training network, and / or the parameters of the second training network are adjusted, and the step of extracting features from the sample user behavior data, the sample user interest data, and the sample user attribute data to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features is returned to continue execution.

[0045] If the loss value is found to satisfy the training constraints, the finally trained network model is determined as the message identifier elimination prediction model for the first scenario.

[0046] In some embodiments, the method further includes:

[0047] Detect whether the first message identifier has been displayed in the first scenario of the communication client within the preset time window to which the requested access time belongs;

[0048] If so, the first scenario controlling the communication client will prevent the display of the first message identifier within the preset time window to which the requested access time belongs;

[0049] If not, proceed with the step of determining whether to control the first message identifier to be displayed in the first scenario of the communication client based on the detection result.

[0050] Furthermore, this application also proposes a display control device for message identifiers, the device comprising:

[0051] An access request acquisition module is used to acquire access requests from communication clients, wherein the access request includes a first user identifier that logged into the communication client and the access request time;

[0052] The first predicted user identifier query module is used to query the first predicted user identifier stored in association with the requested access time. The first predicted user identifier refers to the predicted user identifier whose click prediction probability of the first message identifier displayed in the first scenario of the communication client is greater than the first probability threshold within the preset time window to which the requested access time belongs. The first message identifier is used to indicate that there are unviewed new messages in the first scenario.

[0053] The detection module is used to detect whether the first user identifier belongs to the first predicted user identifier and obtain the corresponding detection result;

[0054] The first message identifier display control module is used to respond to the access request and, based on the detection result, determine whether to control the first message identifier to be displayed in the first scenario of the communication client.

[0055] In another aspect, this application also proposes a storage medium on which a computer program is stored, the computer program being executed by a processor to implement the steps of the message identifier display control method described above.

[0056] Based on the above technical solution, this application pre-predicts the corresponding predicted user identifiers whose click prediction probability for clicking to eliminate the first message identifier displayed in the first scenario of the communication client is greater than a first probability threshold within future time windows. When the communication service device receives an access request from any communication client, it can directly determine the time window to which the access time of the access request belongs, and detect whether the first user identifier contained in the access request belongs to the first predicted user identifier associated with the time window. In order to control whether the first message identifier is displayed in the first scenario of the communication client when responding to the access request, based on the obtained detection result, this achieves selective and personalized display control of the first message identifier for users, avoids frequent disturbance of the first user by the first message identifier, and improves the user traffic efficiency of push information in the first scenario, thereby achieving the purpose of promoting push information in the first scenario. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0058] Figure 1 This is a schematic diagram illustrating message identifiers for different scenarios on a communication page of an optional communication client.

[0059] Figure 2 A schematic diagram of the system structure for the application environment of the message identifier display control method and apparatus proposed in this application;

[0060] Figure 3 This is a flowchart illustrating an optional example of the message identifier display control method proposed in this application;

[0061] Figure 4 A flowchart illustrating an optional training method for the message identifier elimination prediction model in the message identifier display control method proposed in this application;

[0062] Figure 5This is a flowchart illustrating an optional example of the feature space transformation method during the training process of the message identifier elimination prediction model in the message identifier display control method proposed in this application.

[0063] Figure 6 This is a flowchart illustrating another optional training method for the message identifier elimination prediction model in the message identifier display control method proposed in this application.

[0064] Figure 7 This is a flowchart illustrating an optional method for obtaining predicted user identifiers associated with different time windows in the message identifier display control method proposed in this application.

[0065] Figure 8 This is a schematic diagram of an optional application scenario for the message identifier display control method proposed in this application;

[0066] Figure 9 A schematic diagram of an optional example of the message identifier display control device proposed in this application;

[0067] Figure 10 A schematic diagram of another optional example of the display control device for the message identifier proposed in this application;

[0068] Figure 11 This is a schematic diagram of the hardware structure of an optional example of a communication service device for the message identifier display control method and apparatus proposed in the embodiments of this application. Detailed Implementation

[0069] Regarding the description of the background technology section, we will still use the WeChat application as an example. This application aims to constrain the number of red dots displayed in each scenario (i.e., a message identifier, but not limited to red dots, and may also include numerical markers, etc., used to indicate the existence of unread new messages in the corresponding scenario. This application does not restrict the content of the message identifier or its display method, and can be determined as appropriate). Specifically, this can be achieved by estimating the user's current desire to eliminate the displayed red dots (which can be expressed by the probability of eliminating red dots; the higher the probability, the higher the user's desire to eliminate red dots). Based on this, the communication client (such as the WeChat application) can selectively display red dots to guide users to normally view the updated information content in the target scenario (such as the "Discover" scenario), thereby bringing truly effective user clicks to view the target scenario and achieving a virtuous cycle.

[0070] Based on the above analysis, this application considers that different users have different tolerance levels for displayed message identifiers, and that message identifiers are bound to scenarios (such as "Discover," "Moments," and "Video Channels" in WeChat). Users have different needs for different scenarios, and their attitudes towards message identifiers displayed at the corresponding scenario entry points will also differ. Furthermore, for the same user, their attitude towards message identifiers in the same scenario will change over time. This application needs to model, in conjunction with but not limited to, these aspects, to predict each user's tendency to eliminate message identifiers in different scenarios, in order to achieve selective control over the display of message identifiers.

[0071] For the above modeling and analysis process, this application can utilize artificial intelligence (AI) technology to achieve the display control of message markers such as red dots using the constructed prediction model. The AI ​​technology used in modeling can include foundational AI technologies such as cloud computing, distributed storage, big data processing, and operating / interactive systems, and / or AI software technologies such as natural language processing (NLP), machine learning (ML) / deep learning, and computer vision (CV). Specifically, appropriate AI technologies can be selected for modeling based on actual needs; the specific implementation process is not detailed in this application.

[0072] Based on the above research and analysis, this paper proposes using recommendation algorithms to predict message identifiers corresponding to scene information content that users are likely to click on and view, and controlling the display of these message identifiers at the corresponding scene entry points in the communication client. Specifically, click-through rate (CTR) prediction methods, transfer learning methods, and multi-objective learning methods can be used for modeling. CTR prediction is mainly applied to various deep neural network models. By considering each user-content pair, the model can predict the probability of a user clicking on the content based on user basic characteristics, behavioral characteristics, content, and contextual characteristics, such as the predicted CTR score belonging to [0,1]. Multi-objective learning defines the product's prediction objective as multiple independent objective results and trains the model through objective fusion or objective round-robin, such as the MMOE (Multi-gate Mixture-of-Experts) algorithm, to predict user behavior and interests.

[0073] As analyzed above, the modeling process based on click-through rate prediction and multi-objective learning methods considers the relationship between user behavior, interests, and target values. It directly treats user behavior as a feature to achieve model learning of the fit between the behavior sequence and the target value, but it does not consider the changing attitudes of users towards eliminating message icons displayed in different scenarios over time. Furthermore, the model construction process requires calculation with a candidate set, which can lead to the model's prediction results being influenced by the candidate set's own biases such as popularity and interest. However, the application scenario proposed in this application, which involves eliminating message icons such as red dots, does not have a candidate set. It requires predicting the user's tendency to eliminate message icons displayed in the scenario. Therefore, the recommendation principles of the aforementioned click-through rate prediction and multi-objective learning methods are not applicable to the application scenario of this application.

[0074] Transfer learning learns the parameter relationship between two models by limiting the similarity between the behavior distributions of users in the source domain and users in the target domain, thereby transferring user behavior from the source domain to the target domain to help achieve reliable prediction of user behavior in the target domain. However, in the message identifier removal scenarios proposed in this application, such as WeChat red dot removal, there is no auxiliary task with a similar distribution, so this transfer learning algorithm cannot be used to predict users' tendency to remove each message identifier.

[0075] Based on the analysis of several modeling implementation methods above, this application proposes incorporating a time factor into the modeling process to learn the changes in user interest expression under different time factors. This allows for modeling the impact of time periodicity on the user's message tag elimination ability, thereby improving the user elimination rate of message tags. Specifically, this can be achieved by combining user behavior, user interests, scene interests, and user attribute characteristics in the target scenario. By modeling weakened temporal causal relationships through time factor modeling, the expression of time features is enhanced, thus improving the accuracy of predicting the likelihood of message tag elimination in the target scenario at a given time. Taking the red dot display control in the WeChat app's "Discover" scenario as an example, this application can perform online red dot distribution based on the "Discover" scenario, improving the efficiency of attracting users to the "Discover" recommendation targets and reducing the frequent disturbance to users caused by the red dot display at the "Discover" scenario entry point.

[0076] The modeling process after introducing the aforementioned time factor, and the process of using the constructed prediction model to control message identification display, will not be detailed here. The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0077] It should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0078] It should be understood that the terms "system," "apparatus," "unit," and / or "module" used in this application are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0079] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "a," and / or "the" are not specifically singular and may include the plural. Generally, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements. An element defined by the phrase "comprising an..." does not exclude the presence of other identical elements in the process, method, product, or apparatus that includes the element.

[0080] In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" can explicitly or implicitly include one or more of that feature.

[0081] Furthermore, flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Additionally, other operations can be added to these processes, or one or more steps can be removed from them.

[0082] Reference Figure 2 This is a schematic diagram of the system structure for an application environment suitable for the message identifier display control method and apparatus provided in this application. In this application environment, it typically includes, but is not limited to, a communication client 11, a communication service device 12, and a data storage device 13, wherein:

[0083] The communication client 11 can be a communication application installed on an electronic device. Specifically, it can be a standalone instant messaging application, such as social software like WeChat or QQ downloaded and installed from an app store, or it can be a web application that does not require downloading and installation and can be launched and run directly by the user through a browser or other applications. This application does not describe in detail the application category and working principle of the communication client 11.

[0084] The aforementioned electronic devices may include, but are not limited to, smartphones, tablets, wearable devices, personal computers (PCs), netbooks, personal digital assistants (PDAs), smartwatches, in-vehicle devices, robots, desktop computers, etc. Users can choose to use electronic devices according to their actual needs, and launch the operation interface of the communication client 11 to view message content pushed to the communication client in different scenarios. (As above) Figure 1 As shown, after a message status update in a certain scenario, a message icon can be displayed at the corresponding scenario entry point to remind the user to view the updated message content.

[0085] The communication service device 12 can be a service device that provides communication services to communication clients. Specifically, it can be an independent physical server or a service cluster composed of multiple physical servers; it can also be a cloud server with certain cloud computing capabilities, etc. The communication service device 12 can communicate with electronic devices through wired or wireless networks to realize data interaction with the communication client 11 of the electronic device and meet the user's communication needs. The specific implementation process can be determined as appropriate, and will not be described in detail here.

[0086] In this embodiment, the aforementioned communication service device can execute the message identifier display control method and apparatus proposed in the following embodiments of this application, controlling the communication client 11 launched by the user to selectively display message identifiers in corresponding scenarios. For example, if it is predicted that the user will be interested in viewing new messages pushed in the scenario within the current time window, then the message identifier will be displayed in that scenario; otherwise, even if the user launches the communication client and enters a communication page containing that scenario, the scenario will not display message identifiers, thereby avoiding frequent display of message identifiers in that scenario, which could cause users to feel annoyed and resistant to the scenario and the displayed message identifiers, and prevent them from actually viewing the message content, thus failing to achieve the purpose of information promotion in that scenario. The specific implementation process will not be detailed here. The data storage device 13 can be a database server used to store data, specifically it can be used to store various behavioral data generated by the user during the use of the communication client, communication logs between the communication client and the communication server, various cached data from other application platforms, attribute information of each user, and pre-trained and constructed message identifier elimination prediction models, etc. This application does not limit the data content to be stored by the data storage device 13 and its storage method, and it can be determined as appropriate.

[0087] In some embodiments, data of different categories or for different purposes can be written to different databases for storage. Therefore, the data storage device 13 of this application may include one or more databases, such as key-value databases, MySQL databases, etc. This application does not describe in detail the categories of each database and their data storage methods.

[0088] Among them, one or more of the databases listed above can be integrated with communication server 12; or they can exist as independent servers. The specific choice depends on the system architecture and application requirements of the applicable application platform and is not limited to this. Figure 2 The structural relationships are shown.

[0089] It should be understood that, Figure 2 The system architecture of the application environment shown does not constitute a limitation on the application environment in the embodiments of this application. In practical applications, the application environment applicable to the message identifier display control method and apparatus provided in this application may include, but is not limited to, the following: Figure 2 The more or fewer devices shown, such as other multimedia servers, are not listed here.

[0090] Combined with Figure 2 The system structure diagram shown is for reference. Figure 3 This is a flowchart illustrating an optional embodiment of the message identifier display control method proposed in this application. This embodiment can be applied to, but is not limited to, the aforementioned communication service equipment, such as... Figure 3 As shown, the method may include:

[0091] Step S11: Obtain the access request from the communication client;

[0092] In this embodiment, based on the above analysis, in addition to meeting users' instant messaging needs and enabling conversations with other users' communication clients, the communication client, with its diversified functions, can typically also display articles, updates, and other information posted by friends, as well as the latest news, articles, short videos, and other multimedia content, to enrich the information content published by the communication application platform. This multimedia content is usually pushed to corresponding scenarios, allowing users to click on a scenario to view the various multimedia contents obtained within that scenario. The specific implementation process is not detailed in this application.

[0093] In this process, when a user launches a communication client and accesses the information that the communication client can output, they usually first log in to the registered account of the communication client. The specific login process will not be detailed in this application. Based on this, the access request of the communication client obtained in step S11 can be a login request generated and sent by the first user (any user) when launching the communication client.

[0094] For example, a first user can click the application icon of the communication client displayed on the electronic device's screen to enter the login interface of the communication client, enter the account and login password registered by the first user, and click the "Login" button. In response to the input operation of the communication client, the electronic device generates a corresponding login request (i.e., an access request) and sends the login request to the corresponding communication service device.

[0095] It is understandable that since the communication client can usually run in the background of the electronic device after the user registers an account and logs in, when the user launches the communication client again, he / she does not need to enter the account and login password again. He / she can directly click the corresponding application icon to generate an access request to the corresponding communication service device. The communication client will send the generated access request to the corresponding communication service device. The specific implementation process will not be described in detail in this application.

[0096] In this case, the login request sent by the aforementioned communication client can typically include: the first user identifier (such as account ID) of the account registered by the first user on the communication application platform where the communication client is located, and information such as the login time. This application does not limit the specific content contained in the login request and can be determined as appropriate.

[0097] In addition, after the first user starts the communication client, they can access various scenarios displayed on the output communication pages. They can generate access requests for the corresponding scenarios, and the communication client will send the access requests to the communication service device to obtain scenario information for the requested access scenario. For example, clicking on the "Moments" scenario in the WeChat application can obtain the dynamic information posted by the first user's friends' accounts and display it on the Moments display interface; or, in response to the first user's refresh operation of the current interface, a new access request can be generated for the current interface to request the latest interface information and display it, etc.

[0098] As can be seen, in different application scenarios, the first user performs different operations on the communication client, and the content of the resulting access request is often different. In order to ensure that the communication service device can respond to the access content requested by the user, the access request sent usually includes a request content identifier in addition to the first user identifier and the request access time, in order to identify the content requested for access. This application does not limit the specific content of the request content identifier, and it can be determined as appropriate.

[0099] Step S12: Query the first predicted user identifier stored in association with the access time of the request;

[0100] In this embodiment of the application, the aforementioned first predicted user identifier may refer to a predicted user identifier whose predicted click probability of the first message identifier displayed in the first scenario of the communication client is greater than a first probability threshold (this application does not limit its value) within a preset time window (such as which hour of the day the access request is initiated) within the time window to which the request access time belongs.

[0101] As can be seen, this application can predict in advance that if the first scenario displays a message identifier within different preset time windows in the future, there is a high probability that the user identifier of the communication client logged in by the predicted user in the first scenario will be clicked, and the user identifier will be identified as the predicted user identifier of the corresponding preset time window. The user identifier will be associated with the preset time window and stored so as to query one or more first predicted user identifiers associated with the current preset time window in the future. This application does not restrict the specific acquisition process of each predicted user identifier associated with different preset time windows.

[0102] The aforementioned preset time window can be determined based on the characteristics of the user's information operation behavior under the display scenarios of various message identifiers output by the communication client. Specifically, it can be a day, an hour, etc. This application does not limit the specific duration of the preset time window, but since message identifiers are usually not sent on a second-by-second basis, the preset time window is usually not one second. It is understood that for different preset time windows under the same scenario, the content of the predicted user identifier may be different or the same, depending on the situation.

[0103] The first scenario mentioned above can refer to any scenario in the communication page of the communication client that can display message identifiers, such as... Figure 1 The application does not limit the category of the first scenario or the implementation method of displaying message identifiers, as shown in the examples of "Discover" and "Moments" scenarios; these can be determined as appropriate. The first message identifier can represent the message identifier displayed in the first scenario, indicating the existence of unread new messages within that scenario, such as... Figure 1 The red dot markers at the entrances of the various scenes shown in the illustrations are not limited in this application to the content or display location of the first message marker, nor are they restricted to it. Figure 1 The example shown.

[0104] Step S13: Detect whether the first user identifier belongs to the first predicted user identifier, and obtain the corresponding detection result;

[0105] Following the above description, after retrieving the first predicted user identifier associated with the preset time window to which the requested access time belongs, this application can further detect whether the first user identifier that initiated the access request matches any of the queried first predicted user identifiers. Specifically, the first user identifier can be compared with each of the queried first predicted user identifiers. If there is a first predicted user identifier that is the same as the first user identifier among the queried first predicted user identifiers, it means that if the first message identifier is displayed in the first scenario, the first user is very likely to click on the displayed first message identifier, clear the displayed first message identifier, and enter the information display page of the first scenario to view the latest information content displayed on the information display page.

[0106] Conversely, if the first predicted user identifiers found are all different from the first user identifier, it means that the first user identifier does not belong to any of the first predicted user identifiers. Even if the first message identifier is displayed in the first scenario, the first user is unlikely to click on the first message identifier to eliminate it. In fact, the first message identifier displayed in the first scenario may even annoy the first user. In this case, even if the first user clicks on the first message identifier, it will only be to eliminate the first message identifier and will not be used to view the information content in the first scenario.

[0107] Step S14: In response to the above access request, based on the detection result, determine whether to control the first scenario of the communication client to display the first message identifier.

[0108] Normally, when a communication service device receives an access request from any user-logged-in communication client, it can provide corresponding access information based on the specific requested access object, so that the communication client outputs a communication interface containing the received access information. However, in this embodiment, during the response to an access request, the communication service device, in addition to obtaining the requested access information, also needs to determine whether to display a configuration instruction for a message identifier in the corresponding scenario. The display control process for the message identifier in any scenario of the communication client is similar to that in this application; this application will only describe the display control process for the first message identifier in the first scenario as an example.

[0109] As analyzed above, in the process of predicting whether the first message identifier will be displayed in the first scenario of the communication client used by the user, this application introduces a time window as a time factor to perform feature space transformation on the behavioral characteristics, interest characteristics, and basic attribute characteristics of each user, so as to obtain a feature expression with time periodicity. Based on this, it predicts the predicted user who will click on the first scenario to display the first message identifier and actually view the information content in the first scenario (such as clicking to read) within a certain time window in the future. The predicted user identifier of the predicted user who logs into the communication client is recorded and the predicted user identifier is associated with the time window and stored.

[0110] Thus, when any user (referred to as the first user) uses the communication client, it can be detected in the above manner whether the first user is the predicted user associated with the time window where the current request access time is located. In the process of responding to the access request initiated by the first user, based on the detection result, it can be determined whether to control the first user's communication client to display the first message identifier in the first scenario. This application does not describe in detail the specific implementation method of the communication service device controlling the communication client to display the message identifier in each scenario.

[0111] In some embodiments proposed in this application, if the above detection result is yes, it can also be detected first whether the first message identifier of the communication client has been controlled to be displayed in the first scene within the preset time window to which the requested access time belongs. For example, whether a red dot has been sent to the user's WeChat "Discover" scene within the preset time window (such as the current hour). If yes, the first message identifier of the communication client can no longer be controlled to be displayed in the first scene within the preset time window. For example, the red dot will no longer be displayed in the "Discover" scene within the preset time window. If no, the first message identifier of the communication client will be controlled to be displayed in the first scene. For example, the first message identifier information will be pushed to the communication client so that the communication client executes the first message identifier information and displays the first message identifier in the first scene. This application does not limit the specific implementation method of the communication client displaying the corresponding message identifier in the scene, and it can be determined as appropriate.

[0112] In summary, this application pre-predicts the corresponding predicted user identifiers whose predicted click probability for clicking to eliminate the first message identifier displayed in the first scenario of the communication client is greater than a first probability threshold within future time windows. When the communication service device receives an access request from any communication client, it can directly determine the preset time window to which the access time of the access request belongs, and detect whether the first user identifier contained in the access request belongs to the first predicted user identifier associated with the preset time window. In order to determine whether to control the display of the first message identifier in the first scenario of the communication client based on the obtained detection result when responding to the access request, this achieves selective and personalized display control of the first message identifier for users, avoids frequent disturbance of the first user by the first message identifier, and improves the user traffic efficiency of push information in the first scenario, thereby achieving the purpose of promoting push information in the first scenario.

[0113] In some embodiments proposed in this application, the predicted user identifiers stored in association with different preset time windows can be determined based on a pre-trained message identifier elimination prediction model, which processes information such as behavioral data, interest data, and attribute data of different users. In conjunction with the above description of the technical concept of this application, the message identifier elimination prediction model trained in this application can be called an UNPTF (User Notification Probability Based on Time Factor) prediction model. That is, in the process of training this message identifier elimination prediction model, this application introduces a time factor to achieve a non-linear spatial transformation of input features such as behavioral features, interest features, and attribute features, resulting in a feature expression with time periodicity. Based on this, model training is achieved to improve the accuracy of the model's prediction results.

[0114] Based on this, refer to Figure 4 This is a flowchart illustrating another optional embodiment of the message identifier display control method proposed in this application. This application mainly describes the training process of the above-mentioned message identifier elimination prediction model, but is not limited to the training method described in this embodiment. In practical applications, in order to improve prediction efficiency, this application can train and store the message identifier elimination prediction model offline by the communication service device, so that when it is necessary to obtain the predicted user identifier corresponding to the current preset time window or its next adjacent preset time window (i.e., time windows with the same duration in the same time dimension), it can directly call the message identifier elimination prediction model.

[0115] Since users have different needs for different scenarios and different attitudes towards the removal of message icons displayed in each scenario, the predicted probability of removal of message icons displayed at different times in each scenario (i.e., the above-mentioned click prediction probability) is different. In order to ensure the accuracy of the prediction results, this application can train a corresponding message icon removal prediction model for each scenario. The training process of message icon removal prediction models for different scenarios is similar. This application only takes the training process of the message icon removal prediction model in the first scenario as an example for explanation.

[0116] like Figure 4 As shown, the training process of the above message identifier elimination prediction model may include, but is not limited to, the following steps:

[0117] Step S21: Obtain sample user behavior data, sample user interest data, and sample user attribute data for each sample user.

[0118] The training process of the message identifier removal prediction model proposed in this application can be divided into a user behavior statistics stage and a model training stage in the message identifier display scenario. The user behavior statistics stage mainly involves collecting training sample data. Specifically, this application can summarize the behavior data of each sample user in the message identifier display scenario to construct user behavior features that characterize the corresponding sample user's ability and motivation to remove message identifiers.

[0119] Specifically, this application can statistically analyze the behavioral data generated by sample users' operational behaviors in different message identifier display scenarios of the communication client within different preset time windows to obtain sample user behavioral data within different preset time windows. The duration represented by the preset time window can be determined as appropriate, and this application does not impose any restrictions on it.

[0120] In some embodiments proposed in this application, taking WeChat as an example, since it is an instant messaging tool, it mainly carries the user's social needs. However, the behavioral data for scenarios other than social interaction, such as reading, video, and Moments, is often sparse. Therefore, for this type of communication client, the duration represented by the predetermined preset time window is not very long. Statistical analysis shows that the distribution of user behavior in this communication client is mostly within a week, and the changes in user behavior within a day are usually periodic. This application can use a day as a statistical time window to fully characterize the user's message tag removal habits in a scenario. Therefore, this application can statistically analyze the behavioral data of each sample user in the message tag display scenario for each day in the past week or several weeks to obtain the corresponding sample user behavior data. The specific statistical process is not detailed in this application.

[0121] Meanwhile, in order to more accurately predict the probability of users clicking on message icons displayed in the corresponding scenarios, the communication service device can also obtain the interest data of each sample user, which is referred to as sample user interest data. Specifically, it can be obtained by processing user behavior data based on a pre-trained interest model. Specifically, it can be represented by interest tags (such as the ID of the object of interest). This application does not restrict the acquisition method and the content of the interest tags it contains, and can be determined as appropriate.

[0122] In addition, basic attribute information such as age, gender, and activity level of each sample user can be obtained and recorded as sample user attribute information. This information can be obtained through the registration information obtained by the corresponding sample user during the registration process of the communication client, as well as through the statistical analysis of the historical usage logs of the corresponding sample user on the communication client, to obtain the activity level and other attribute information of the sample user. This application does not restrict the content of each sample user attribute information or the method of obtaining it, and can be determined as appropriate.

[0123] Step S22: Extract features from the sample user behavior data, sample user interest data, and sample user attribute data respectively to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features.

[0124] For the sample user behavior data, sample user interest data, and sample user attribute data obtained above, different processing logics can be used to extract the features of the corresponding data and obtain the corresponding feature vectors. However, this application does not restrict the specific processing logic used for feature extraction of each data, and it can be determined according to the situation.

[0125] In some embodiments proposed in this application, the sample user behavior data for each message identifier display scenario of the communication client is typically a sequence of behaviors generated by the corresponding sample user's operation behavior in each message identifier display scenario, combined according to the order of generation time. This application embodiment can utilize deep learning networks such as Recurrent Neural Networks (RNNs) to extract features from this data to obtain the corresponding sample user behavior features, i.e., behavior feature vectors. Therefore, this application can denote the deep learning network used for feature extraction of sample user behavior data as the first feature extraction network. After obtaining the sample user behavior data, it can first be normalized and encoded, and then the resulting encoded vector can be input into the first feature extraction network for processing to obtain the corresponding sample user behavior features. The network structure of this first feature extraction network can include, but is not limited to, network structures such as LSTM (Long Short-Term Memory) and GRU (Gate Recurrent Unit), depending on the situation.

[0126] Similarly, for the sample user interest data obtained above, its features can be extracted using a pre-defined second feature extraction network to obtain the corresponding sample user interest features. Specifically, the interest tags contained in the sample user interest data can be vectorized, and the obtained interest vectors can be input into the pooling layer of a deep learning network for processing to obtain sample user interest features expressing user interests. In this case, the second feature extraction network can also be based on a deep learning network, but it is not limited to the feature extraction implementation method proposed in this application.

[0127] Based on the above analysis, in this embodiment, the communication service device can input sample user behavior data into a first feature extraction network and output sample user behavior features; input sample user interest data into a second feature extraction network and output sample user interest features; and perform vectorization processing on sample user attribute data to obtain sample user attribute features. The network structures of the first and second feature extraction networks are typically different, but the specific network structures are not limited. The first feature extraction network has first network parameters to be trained, and the second feature extraction network has second network parameters to be trained. During the training process of the message identifier elimination prediction model, the first and / or second network parameters to be trained can be continuously optimized and adjusted to improve the accuracy of the output prediction results of the training model until the model constraints are met. This application does not detail the optimization and adjustment methods for the first and second network parameters to be trained; these methods can be determined as needed.

[0128] Step S23: Based on a preset time window with a time period, perform spatial transformation on the sample user behavior features, sample user interest features, and sample user attribute features respectively to obtain the corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features with the time period.

[0129] Based on the above description of the technical concept of this application, the time factor with a time period introduced in the model training process can be a preset time window t with a time period. For example, if the time window duration is one day, the time period can be 7 days, i.e., one week; if the time window duration is one hour, the time period can be 24 hours, i.e., one day. This application can perform spatial transformation on the input features according to a preset calculation logic, using the preset time window with a time period, to obtain the corresponding feature representation with time periodicity, so that the finally trained model can be sensitive to time representation and enhance the modeling representation of time periodicity.

[0130] In some embodiments proposed in this application, the computational logic for implementing the input feature space transformation can be represented by the following technical formula:

[0131] Time(f,t)=f*e -W*S(t) (1)

[0132] In the above formula (1), f can represent the feature input to the calculation logic, such as the sample user behavior feature, sample user interest feature, or sample user attribute feature obtained above; t represents the input time factor, that is, the time represented by the preset time window with time periodicity, such as Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, and Sunday; or the hours between 1 o'clock and 24 o'clock, that is, the 24 hours contained in a day, etc.; Time(f,t) can represent the feature expression of the corresponding input feature f after spatial transformation within the corresponding preset time window t; W can represent a matrix parameter, that is, the model parameter to be trained and learned. In this application, it is denoted as the first matrix parameter to be trained. In this application, the method of obtaining its initial value is not restricted, and it can be a default value, etc.; S can represent a time factor normalization method, as shown in the following implementation:

[0133]

[0134] In formula (2) above, S(t) represents the normalization method based on the sin function. Generally, S(t)∈(0,1). max() represents the maximum value function. It should be noted that the normalization method for the time factor is not limited to the method described in formula (2). Compared with other normalization processing functions, the normalization processing method based on the sine function in this application has better smoothness and is conducive to improving the accuracy of the prediction results.

[0135] For the sample user behavior features, sample user interest features and sample user attribute features obtained above, this application can use the calculation logic recorded in the above formulas (1) and (2) to perform spatial transformation to obtain the corresponding undetermined behavior features, undetermined interest features and undetermined attribute features with the time period. The specific implementation process will not be described in detail here.

[0136] Based on the above analysis, the communication service equipment can specifically implement the above spatial conversion processing in the following ways, but is not limited to the spatial conversion processing method described in the embodiments of this application, such as... Figure 5 As shown, the spatial transformation process may include:

[0137] Step A1: Normalize the preset time window within a time period to obtain a normalized time vector;

[0138] Based on the description of the above formula (2), for multiple preset time windows representing different time points under the same time dimension, such as 7 days, each day is a preset time window, these multiple preset time windows can be substituted into the above formula (2) to obtain a normalized time vector composed of values ​​between 0 and 1, but it is not limited to the normalization processing method described above.

[0139] Step A2: Construct spatial transformation parameters using the normalized time vector and the parameters of the first matrix to be trained;

[0140] Referring to the right side of the equation (1) above, e -W*S(t) This computational logic yields the normalized time vector S(t) and the first training matrix parameter W. Following this logic, spatial transformation parameters can be constructed, which involves substituting the obtained S(t) and W into the computational logic to obtain the output computational result.

[0141] Step A3: Using the spatial transformation parameter, perform spatial transformation on the corresponding feature values ​​of the input features to obtain the corresponding undetermined input features;

[0142] The spatial transformation calculation logic represented by the above formula (1) can be obtained by multiplying the spatial transformation parameters obtained above with the input features to obtain the time-periodic feature expression of the input features within different preset time windows, but it is not limited to the feature spatial transformation implementation method described in this application.

[0143] It is understood that the input features in step A3 can specifically be the aforementioned sample user behavior features, sample user interest features, or sample user attribute features, and the pending input features can specifically be the corresponding pending behavior features, pending interest features, and pending attribute features. Since the spatial transformation process for each input feature is similar, this embodiment will not describe them in detail here.

[0144] Step S24: Merge the undetermined behavioral features, undetermined interest features, and undetermined attribute features to obtain the corresponding sample user coding vector;

[0145] In this embodiment, the Concat function can be used to merge the undetermined behavioral features, undetermined interest features, and undetermined attribute features obtained after the above feature space transformation to obtain a unified feature output. In this application, the merged feature vector is denoted as the sample user coding vector.

[0146] Step S25: Obtain the target behavior features of each sample user in the first scenario;

[0147] Step S26: Train the network model using the sample user encoding vector and target behavior features to obtain the message identifier elimination prediction model.

[0148] The process of obtaining the target behavior features can be referred to the process of obtaining the sample user behavior features described above, and will not be detailed here. It should be noted that the elements in the target behavior features can be the corresponding sample user clicking on the first message icon displayed in the first scene within the corresponding preset time window, which can be represented by 1; or not clicking, which can be represented by 0. Therefore, the target behavior features can be a vector composed of 0 and 1. It can be seen that each feature value contained in the target behavior features is the actual operation behavior of the corresponding sample user on the first message icon displayed in the first scene within the corresponding preset time window, such as clicking to clear the first message icon, or not clicking the first message icon to maintain the display state of the first message icon, etc.

[0149] After obtaining the sample user's encoded vector, further processing in the network model, such as pooling layers and output layers, can yield a predicted score (i.e., click prediction probability) indicating whether the sample user will click on the first message icon displayed in the first scene within the corresponding time window. A higher click prediction probability value indicates that the sample user is more likely to click on the first message icon displayed in the first scene within the corresponding time window, thus eliminating the first message icon. Subsequently, a preset loss function can be used to calculate the loss between the click prediction probability output by the network model and the feature values ​​contained in the actual statistically obtained target behavior features, thereby enabling the learning and training of the network model.

[0150] Based on this, step S26 can specifically include: inputting the sample user encoding vector into the pooling layer of the network model for processing to obtain the click prediction probability of the corresponding sample user clicking the first message identifier displayed in the first scene; obtaining the loss value between the click prediction probability and the corresponding feature value in the target behavior feature; detecting that the loss value does not meet the training constraints, adjusting the network model parameters such as the first training matrix parameters, the first training network parameters and / or the second training network parameters, and returning to step S22 to continue execution, that is, using the network model with the adjusted network model parameters, continuing to learn the network model from the above-mentioned data until the loss value meets the training constraints, and finally determining the finally trained network model as the message identifier elimination prediction model for the first scene.

[0151] It should be noted that this application does not limit the specific content of the above training constraints, such as whether the obtained loss value is less than the preset loss threshold, or whether it tends to be stable; the above loss function may include but is not limited to the cross-entropy loss function, depending on the situation.

[0152] In summary, during the training process of the message identifier elimination prediction model described in this application embodiment, a preset time window with a time period is introduced to realize the spatial transformation of sample user behavior features, interest features, and attribute features. Then, the model is trained using the feature expression with time periodicity, so that the trained message identifier elimination prediction model is more sensitive to time expression, enhances the modeling expression of time periodicity, and thus improves the accuracy of the prediction results of the user's click operation on the first message identifier displayed in the first scenario within different preset time windows.

[0153] Based on the model training process described above, and combined with the description of the technical concept of this application above, in some embodiments proposed in this application, the above modeling process can introduce two time factors, that is, analyze and model the above user features through two different time dimensions. Therefore, the preset time window involved in the above embodiments can include two time windows of different durations, referred to as the first time window and the second time window. It should be noted that in the feature space transformation process based on multiple preset time windows, it is implemented based on multiple preset time windows of the same duration (or under the same time dimension).

[0154] In this case, the request access time of the communication client's access request obtained by the communication service device can be specifically composed of a first time of a first time dimension and a second time of a second time dimension. The first time dimension has a first time period, the second time dimension has a second time period, and the second time period is a time unit of the first time period. Therefore, the above-mentioned different preset time windows include a first time window composed of a time unit of the first time period and a second time window composed of a time unit of the second time period.

[0155] Specifically, for instant messaging clients like WeChat, the distribution of user behavior is mostly within a week, and changes in user behavior within a day are usually periodic. Since message markers like red dots are not sent in real-time at the second level, this application can define the first time dimension as a day-level dimension and the second time dimension as an hour-level dimension. That is, one unit of time in the first time dimension is defined as a day, with a seven-day (weekly) cycle under this time dimension, i.e., the first time cycle is 7 days. Similarly, one unit of time in the second time dimension can be defined as an hour, with a 24-hour (dayly) cycle under this time dimension, i.e., the second time cycle is 24 hours. Therefore, the value of the first time window can be an integer in the range [1, 7], and the value of the second time window can be an integer in the range [0, 23].

[0156] Therefore, the first time t1 in the above-mentioned request access time can represent the i-th day within the first time period, i.e., weekday i, where i is a positive integer less than or equal to 7, and its corresponding preset time window is determined as the target first time window; the second time t2 can represent the j-th hour within the second time period, where j is an integer less than 24, and its corresponding preset time window is determined as the target second time window. In other words, the request access time that generates the above-mentioned access request can be the j-th hour of weekday i, such as 3 PM on Tuesday. This application does not restrict the specific times of the first time t1 and the second time t2. After obtaining the actual request access time, the request access time can be composed of day-level and hour-level times based on its specific time. For example, an access request initiated at 2:26 PM on Wednesday will generate a request access time of 2 PM on Wednesday. In this case, the first time t1 is Wednesday, and the second time t2 is 2 PM. It can be understood that when a user clicks to access a communication page of a communication client, the first time in the obtained request access time usually refers to the current day, and the second time can be the current hour, but is not limited to this.

[0157] Based on the above analysis, referring to Figure 6 The flowchart shown illustrates the training method for the message identifier elimination prediction model. By introducing the two time windows mentioned above during the modeling process, this application can model these two time windows separately during the time-periodic feature space transformation, obtaining 7 (days) + 24 (hours) = 31 time features. This is in contrast to the splicing modeling method based on the first time window * the second time window. As can be seen, compared to the 7 * 24 = 168 time features obtained by this splicing modeling method, the feature space is significantly compressed. This solves the technical problem of some time features being very sparse and reducing the accuracy of model prediction results, thus enhancing the generalization ability of the model's training features.

[0158] Therefore, step S23 in the above embodiment may specifically include: based on a first time window with a first time period, performing spatial transformation on the sample user behavior features, sample user interest features, and sample user attribute features respectively to obtain corresponding first undetermined behavior features, first undetermined interest features, and first undetermined attribute features with the first time period; based on a second time window with a second time period, performing spatial transformation on the sample user behavior features, sample user interest features, and sample user attribute features respectively to obtain corresponding second undetermined behavior features, second undetermined interest features, and second undetermined attribute features with the second time period.

[0159] As above Figure 6As shown, for each of the above input features, in the feature space transformation process based on the first time window with a first time period, the preset time window in the calculation logic of the above formulas (1) and (2) is specifically the first time window, that is, belonging to [1,7]; while in the feature space transformation process based on the second time window with a second time period, the preset time window in the calculation logic of the above formulas (1) and (2) is specifically the second time window, that is, belonging to [0,23]. Figure 6 The spatial transformation process for different input features based on different time windows can be determined in conjunction with the description of the corresponding parts of the above embodiments, and will not be repeated here.

[0160] As can be seen, based on the calculation logic corresponding to different time factors, this application can perform time-periodic feature space transformation on the input features of sample users to enhance the temporal characteristics of the features and have time-periodic generalization, thereby making the trained message identifier elimination prediction model more sensitive to time expression, enhancing the time-periodic modeling expression, and improving the accuracy and efficiency of prediction results.

[0161] After obtaining the message identifier elimination prediction model corresponding to the first scenario based on the above embodiment, it can be stored in a data storage device. This way, during daily use of the communication client, the user can invoke the message identifier elimination prediction model to extract the first message identifier that is highly likely to be clicked in the first scenario within different preset time windows, query the predicted user identifier of the information content in the first scenario, and write it to the corresponding storage space for storage. This allows the system to determine whether to control the display of the first message identifier in the first scenario when the user accesses various communication pages of the communication client. To obtain the predicted user identifier associated with each preset time window, such as... Figure 7 As shown, this application proposes a flowchart of another optional embodiment of the message identifier display control method. This application mainly describes the process of obtaining the predicted user identifier. For the model training process and the query and application process of the predicted user identifier, please refer to the description of the corresponding part of the above embodiment. The embodiments of this application will not be repeated here.

[0162] like Figure 7 As shown, the process of obtaining the predicted user identifier described above may include, but is not limited to:

[0163] Step S31: Obtain historical user behavior data, user interest data, and user attribute data of different users on the communication client;

[0164] Based on the description of the corresponding parts of the above embodiments, the historical user behavior data can be generated based on the user's operation behavior on different message identifier display scenarios of the communication client within different preset time windows. This application does not describe in detail the specific acquisition process of historical user behavior data, user interest data and user attribute data.

[0165] Step S32: Input historical user behavior data, user interest data and user attribute data into the message identifier elimination prediction model for processing, and output the click prediction probability of the corresponding user clicking on the first message identifier displayed in the first scenario within different time windows;

[0166] The process of obtaining the click prediction probability for different users can be referred to the description of the corresponding part in the above-mentioned message identifier elimination prediction model training embodiment, which will not be detailed here. In practical applications, since the message identifier elimination prediction model is pre-trained, when it is necessary to obtain the click prediction probability of different users clicking on the first message identifier displayed in the first scene within different time windows, the message identifier elimination prediction model corresponding to the first scene can be directly called. The data obtained for each user as described above can be input into the message identifier elimination prediction model, and the click prediction probability of the corresponding user clicking on the first message identifier displayed in the first scene within different time windows can be output quickly and accurately.

[0167] It is understood that this application may also obtain only the click prediction probability of different users clicking on the first message identifier displayed in the first scenario within the current time window and the next adjacent time window, in order to determine the predicted user identifier corresponding to the corresponding time window, that is, to determine the communication client currently used by the user who needs to control the display of the first message identifier within the corresponding time window.

[0168] Step S33: If the predicted probability of the click is greater than the first probability threshold, the user identifier of the user who logs into the communication client is determined as the predicted user identifier for the corresponding time window.

[0169] If the predicted click probability obtained above is less than or equal to the first probability threshold, it can be considered that the user may not want the first message identifier to be displayed in the first scenario of the communication client within the corresponding preset time window. The first message identifier displayed within this time period is regarded as interference. In this case, when the user logs into the communication client and outputs a communication page containing the first scenario, the first message identifier does not need to be displayed. For example, the first message identifier does not need to be sent to the communication client logged into by the user within the corresponding time period. The specific implementation method is not limited in this application.

[0170] Step S34: Associate and store the predicted user identifier with the corresponding time window.

[0171] As can be seen, this application utilizes the message identifier elimination prediction model corresponding to the first scenario to predict the user identifier that needs to be controlled to display the first message identifier in the first scenario within each preset time window, such as the account ID of the communication client, and writes it into the corresponding data storage device, such as a key-value database, for storage. In this way, when a user initiates an access request at any time point within a preset time window, the system can determine whether to control the communication client of the user to display the first message identifier by detecting whether the user's user identifier belongs to the predicted user identifier associated with the corresponding preset time window. This achieves selective display control of message identifiers and reduces frequent interference to users caused by displaying message identifiers.

[0172] In some embodiments of this application, if the message identifier elimination prediction model introduces the two time factors mentioned above to achieve learning and training, combined with the relevant description of the predicted user identifier above, this application can predict in advance the predicted user identifier that is likely to click on the first message identifier displayed in the first scene within each hour of the day (usually within each hour of the future), and associate it with the corresponding hour of that day for storage. In this way, after this application determines the request access time of this access request in the above manner, it can query the data storage device to see if there is a first predicted user identifier stored in association with the second time in the first time, such as whether there is a first predicted user identifier stored in association with 13:00 of the day (which can be the current hour or a future hour). If so, it then checks whether the first user identifier belongs to the first predicted user identifier, and determines whether the first scene controlling the communication client used by the first user displays the first message identifier.

[0173] Reference Figure 8 The flowchart shown illustrates the scenario. Following the above method, the predicted user identifier u, associated with different second time windows within the first time window, is predicted and stored in the corresponding key-value database according to a data structure such as (predicted user identifier, first time window, second time window). When a user logs into a communication client and initiates an access request, obtaining the requested access times t1 and t2, they can query the first predicted user identifier u associated with t1 and t2. This determines the target first time window to which the first time belongs and the target second time window to which the second time belongs. The predicted user identifier associated with the target second time window under the target first time window is then identified as the first predicted user identifier. Next, it is checked whether the user's first user identifier u1 belongs to the first predicted user identifier u. If so, the first message identifier is displayed in the first scenario of the communication client used by the user, such as a red dot in WeChat's "Discover" scenario. At this time, the user typically clicks the displayed first message identifier to view the information content in the first scenario.

[0174] Subsequently, the communication service device can write the user's click operation behavior on the first message identifier displayed this time into the corresponding message identifier removal log to update the user's behavior data, so as to optimize the message identifier removal prediction model of the first scenario and improve the accuracy of the prediction results. The model optimization implementation process is similar to the above model training process, and will not be described in detail here.

[0175] Reference Figure 9 This application provides a schematic diagram of an optional embodiment of a message identifier display control device, which can be applied to the aforementioned communication service equipment, such as... Figure 9 As shown, the device may include:

[0176] Access request acquisition module 21 is used to acquire access requests from communication clients, wherein the access request includes a first user identifier that logged into the communication client and the access request time;

[0177] The first predicted user identifier query module 22 is used to query the first predicted user identifier stored in association with the requested access time;

[0178] Wherein, the first predicted user identifier refers to the predicted user identifier whose predicted click probability of the first message identifier displayed in the first scenario of the communication client is greater than the first probability threshold within the time window to which the requested access time belongs; the first message identifier is used to indicate that there are unviewed new messages in the first scenario.

[0179] Detection module 23 is used to detect whether the first user identifier belongs to the first predicted user identifier and obtain the corresponding detection result;

[0180] The first message identifier display control module 24 is used to respond to the access request and, based on the detection result, determine whether to control the first message identifier to be displayed in the first scenario of the communication client.

[0181] Optionally, the device may also include:

[0182] The message identifier display detection module is used to detect whether the first message identifier has been displayed in the first scene of the communication client within the preset time window to which the requested access time belongs; if not, the first message identifier display control module 24 is triggered to determine whether to control the first scene of the communication client to display the first message identifier based on the detection result;

[0183] The control module is configured to, when the detection result of the message identifier display detection module is yes, control the first scenario of the communication client to prevent the display of the first message identifier within the preset time window to which the requested access time belongs.

[0184] In some embodiments, in order to obtain the predicted user identifiers associated with different time windows, such as Figure 10 As shown, the above-mentioned device may further include:

[0185] The first data acquisition module 25 is used to acquire historical user behavior data, user interest data and user attribute data of the user on the communication client;

[0186] The historical user behavior data is generated based on the user's operational behavior in different message identifier display scenarios of the communication client within different preset time windows;

[0187] Click prediction module 26 is used to input the historical user behavior data, the user interest data and the user attribute data into the message identifier elimination prediction model for processing, and output the click prediction probability of the user clicking the first message identifier displayed in the first scenario within different time windows.

[0188] The predictive user identifier determination module 27 is used to determine the user identifier of the user who logged into the communication client as the predictive user identifier for the corresponding time window when the click prediction probability is greater than a first probability threshold.

[0189] Storage module 28 is used to associate and store the predicted user identifier with the corresponding time window.

[0190] In some embodiments, to train the aforementioned message identifier removal prediction model, such as... Figure 10 As shown, the above-mentioned device may further include:

[0191] The second data acquisition module 29 is used to acquire sample user behavior data, sample user interest data and sample user attribute data of each sample user.

[0192] Feature extraction module 210 is used to extract features from the sample user behavior data, the sample user interest data and the sample user attribute data respectively, to obtain corresponding sample user behavior features, sample user interest features and sample user attribute features.

[0193] The feature space transformation module 211 is used to perform spatial transformation on the sample user behavior features, the sample user interest features and the sample user attribute features based on a preset time window with a time period, to obtain the corresponding undetermined behavior features, undetermined interest features and undetermined attribute features with the time period.

[0194] The feature merging processing module 212 is used to merge the undetermined behavior features, the undetermined interest features, and the undetermined attribute features to obtain the corresponding sample user encoding vector;

[0195] The target behavior feature acquisition module 213 is used to acquire the target behavior features of each sample user in the first scenario.

[0196] The model training module 214 is used to train a network model using the sample user encoding vector and the target behavior features to obtain a message identifier elimination prediction model for the first scenario.

[0197] Wherein, if the requested access time is composed of a first time in a first time dimension and a second time in a second time dimension, the first time dimension has a first time period, the second time dimension has a second time period, and the second time period is a time unit of the first time period; therefore, different preset time windows include a first time window composed of a time unit of the first time period and a second time window composed of a time unit of the second time period. For example, the first time dimension is a day-level dimension, the second time dimension is an hour-level dimension, the first time period is seven days, and the second time period is twenty-four hours.

[0198] Based on this, the aforementioned first predicted user identifier query module 22 may specifically include:

[0199] The first determining unit is used to determine the target first time window to which the first time belongs, and the target second time window to which the second time belongs;

[0200] The first predicted user identifier determination unit is used to determine the predicted user identifier stored in association with the second time window of the target under the first time window of the target as the first predicted user identifier.

[0201] In addition, the aforementioned feature space conversion module 211 may include:

[0202] The first feature space transformation unit is used to perform spatial transformation on the sample user behavior features, the sample user interest features and the sample user attribute features based on a first time window with a first time period, to obtain corresponding first undetermined behavior features, first undetermined interest features and first undetermined attribute features with the first time period.

[0203] The second feature space transformation unit is used to perform spatial transformation on the sample user behavior features, the sample user interest features and the sample user attribute features based on a second time window with a second time period, to obtain corresponding second undetermined behavior features, second undetermined interest features and second undetermined attribute features with the second time period.

[0204] Accordingly, the feature merging processing module 212 can be used to merge the first undetermined behavior feature, the first undetermined interest feature, the first undetermined attribute feature, the second undetermined behavior feature, the second undetermined interest feature, and the second undetermined attribute feature to obtain the corresponding sample user encoding vector.

[0205] In one possible implementation, to achieve the above spatial transformation process, the spatial transformation module or unit may specifically include:

[0206] The normalization processing unit is used to normalize a preset time window within a time period to obtain a normalized time vector;

[0207] The spatial transformation parameter construction module is used to construct spatial transformation parameters using the normalized time vector and the parameters of the first training matrix;

[0208] A spatial transformation unit is used to perform spatial transformation on the corresponding feature values ​​of the input features using the spatial transformation parameters to obtain the corresponding undetermined input features;

[0209] The input features are the sample user behavior features, the sample user interest features, or the sample user attribute features, and the undetermined input features are the corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features.

[0210] In some embodiments, the feature extraction module 210 may include:

[0211] The first feature extraction unit is used to input the sample user behavior data into the first feature extraction network and output the sample user behavior features.

[0212] The second feature extraction unit is used to input the sample user interest data into the second feature extraction network and output the sample user interest features.

[0213] The third feature extraction unit is used to perform vectorization processing on the sample user attribute data to obtain sample user attribute features.

[0214] The first feature extraction network and the second feature extraction network have different network structures, and the first feature extraction network has a first network parameter to be trained, while the second feature extraction network has a second network parameter to be trained.

[0215] Accordingly, the model training module 214 described above may include:

[0216] The click prediction probability obtaining unit is used to input the sample user encoding vector into the pooling layer of the network model for processing, and obtain the click prediction probability of the corresponding sample user clicking on the first message identifier displayed in the first scene;

[0217] The loss value acquisition unit is used to acquire the loss value between the click prediction probability and the corresponding feature value in the target behavior feature;

[0218] The network model parameter adjustment unit is used to adjust the first training matrix parameter, the first training network parameter and / or the second training network parameter when the loss value is detected to not meet the training constraints, and to trigger the feature extraction module 210 to continue to extract features from the sample user behavior data, the sample user interest data and the sample user attribute data respectively, so as to obtain the corresponding sample user behavior features, sample user interest features and sample user attribute features.

[0219] The message identifier removal prediction model determination unit is used to determine the finally trained network model as the message identifier removal prediction model for the first scenario when the loss value is detected to meet the training constraints.

[0220] It should be noted that the various modules and units in the above-mentioned device embodiments can all be stored in the memory as program modules. The processor executes the above-mentioned program modules stored in the memory to realize the corresponding functions. The functions realized by each program module and its combination, as well as the technical effects achieved, can be referred to the description of the corresponding part of the above-mentioned method embodiments. This embodiment will not repeat them here.

[0221] This application embodiment also provides a storage medium storing a computer program, which is executed by a processor to implement the steps of the above-described message identifier display control method. The implementation process of the message identifier display control method can be referred to the description of the above method embodiment.

[0222] This application also proposes a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above-described message identifier display control method or message identifier display control device. Specific implementation processes can be referred to the descriptions of the corresponding embodiments above, and will not be repeated here.

[0223] Reference Figure 11 This is a schematic diagram of the hardware structure of an optional example of a communication service device for the message identification display control method and apparatus proposed in the embodiments of this application. The communication service device may include a communication module 31, a memory 32 and a processor 33.

[0224] In this embodiment, the communication module 31, memory 32, and processor 33 can communicate with each other through a communication bus, and the number of the communication module 31, memory 32, processor 33, and communication bus can be at least one.

[0225] Optionally, the communication module 31 may be a communication module that can realize data interaction using a wireless communication network, such as a WIFI module, a 5G / 6G (fifth generation mobile communication network / sixth generation mobile communication network) module, a GPRS module, etc. The communication module 31 may also include a communication interface that realizes data interaction between internal components of the computer device, such as a USB interface, a serial / parallel port, etc. This application does not limit the specific contents included in the communication module 31.

[0226] The memory 32 can be used to store a program that implements the message identifier display control method as described above. The processor 33 is used to load and execute the program stored in the memory 32 to implement each step of the message identifier display control method described in the corresponding method embodiments above. The specific implementation process will not be elaborated.

[0227] In this embodiment, memory 32 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device. Processor 33 may be a central processing unit (CPU), application-specific integrated circuit (ASIC), digital signal processor (DSP), application-specific integrated circuit (ASIC), off-the-shelf programmable gate array (FPGA), or other programmable logic device.

[0228] It should be understood that, Figure 11The structure of the communication service device shown does not constitute a limitation on the communication service device in the embodiments of this application. In practical applications, the communication service device may include more than Figure 11 The more or fewer components shown, or combinations of certain components, are not listed here.

[0229] Finally, it should be noted that the various embodiments in this specification are described in a progressive or parallel manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus and communication service equipment disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0230] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the core ideas or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for controlling the display of message identifiers, characterized in that, The method, applicable to controlling the display of message identifiers in any scenario of a communication client, includes: Obtain an access request from a communication client, wherein the access request includes a first user identifier who logged into the communication client and the access request time. After starting the communication client, perform access operations on each scenario displayed on each of the output communication pages to generate access requests for the corresponding scenarios. Query the first predicted user identifier stored in association with the requested access time, wherein the first predicted user identifier refers to the predicted user identifier whose click prediction probability is greater than a first probability threshold within the preset time window to which the requested access time belongs. The first message identifier is used to indicate that there are unviewed new messages in the first scenario. The preset time window is determined based on the information operation behavior characteristics of the user in the message identifier display scenarios output by the communication client. Detect whether the first user identifier belongs to the first predicted user identifier, and obtain the corresponding detection result; In response to the access request, and based on the detection result, determine whether to control the first message identifier to be displayed in the first scenario of the communication client.

2. The method according to claim 1, characterized in that, The process of obtaining the predicted user identifier includes: The system acquires historical user behavior data, user interest data, and user attribute data related to the communication client; wherein the historical user behavior data is generated based on the user's operational behavior in different message identifier display scenarios of the communication client within different preset time windows; The historical user behavior data, user interest data, and user attribute data are input into the message identifier elimination prediction model for processing, and the predicted click probability of the user clicking on the first message identifier displayed in the first scenario within different preset time windows is output. If the predicted click probability is greater than the first probability threshold, the user identifier of the user who logged into the communication client will be determined as the predicted user identifier for the corresponding preset time window. The predicted user identifier is associated with and stored in relation to the corresponding preset time window.

3. The method according to claim 2, characterized in that, The training process of the message identifier elimination prediction model includes: Obtain sample user behavior data, sample user interest data, and sample user attribute data for each sample user; Feature extraction is performed on the sample user behavior data, the sample user interest data, and the sample user attribute data respectively to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features. Based on the preset time window with a time period, spatial transformation is performed on the sample user behavior features, sample user interest features, and sample user attribute features respectively to obtain corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features with the time period. The undetermined behavioral features, undetermined interest features, and undetermined attribute features are merged to obtain the corresponding sample user encoding vector; Obtain the target behavioral characteristics of each sample user in the first scenario; The network model is trained using the sample user encoding vector and the target behavior features to obtain the message identifier elimination prediction model for the first scenario.

4. The method according to claim 3, characterized in that, The requested access time consists of a first time in a first time dimension and a second time in a second time dimension. The first time dimension has a first time period, the second time dimension has a second time period, and the second time period is a time unit of the first time period. Therefore, the preset time window includes a first time window consisting of one time unit of the first time period and a second time window consisting of one time unit of the second time period. The query is associated with the first predicted user identifier stored in relation to the requested access time, including: Determine the target first time window to which the first time belongs, and the target second time window to which the second time belongs; The predicted user identifier associated with the second time window of the target under the first time window of the target is determined as the first predicted user identifier.

5. The method according to claim 4, characterized in that, Based on the preset time window with a time period, spatial transformation is performed on the sample user behavior features, sample user interest features, and sample user attribute features to obtain corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features with the time period, including: Based on the first time window with a first time period, spatial transformation is performed on the sample user behavior features, the sample user interest features, and the sample user attribute features respectively to obtain the corresponding first undetermined behavior features, first undetermined interest features, and first undetermined attribute features with the first time period. Based on the second time window with a second time period, spatial transformation is performed on the sample user behavior features, the sample user interest features, and the sample user attribute features respectively to obtain the corresponding second undetermined behavior features, second undetermined interest features, and second undetermined attribute features with the second time period. The process of merging the undetermined behavioral features, undetermined interest features, and undetermined attribute features to obtain the corresponding sample user encoding vector includes: The first undetermined behavioral feature, the first undetermined interest feature, the first undetermined attribute feature, the second undetermined behavioral feature, the second undetermined interest feature, and the second undetermined attribute feature are merged to obtain the corresponding sample user encoding vector.

6. The method according to claim 3, characterized in that, The spatial transformation process includes: The preset time window within a time period is normalized to obtain a normalized time vector; Using the normalized time vector and the parameters of the first training matrix, spatial transformation parameters are constructed; Using the spatial transformation parameters, the corresponding feature values ​​of the input features are spatially transformed to obtain the corresponding undetermined input features; The input features are the sample user behavior features, the sample user interest features, or the sample user attribute features, and the undetermined input features are the corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features.

7. The method according to claim 6, characterized in that, The step of extracting features from the sample user behavior data, the sample user interest data, and the sample user attribute data to obtain corresponding sample user behavior features, sample user interest features, and sample user attribute features includes: The sample user behavior data is input into the first feature extraction network, which outputs the sample user behavior features. The sample user interest data is input into the second feature extraction network, and the sample user interest features are output; wherein, the network structures of the first feature extraction network and the second feature extraction network are different, and the first feature extraction network has a first network parameter to be trained, and the second feature extraction network has a second network parameter to be trained; The sample user attribute data is vectorized to obtain sample user attribute features; The step of training a network model using the sample user encoding vector and the target behavior features to obtain a message identifier removal prediction model includes: The sample user encoding vector is input into the pooling layer of the network model for processing to obtain the click prediction probability of the corresponding sample user clicking the first message identifier displayed in the first scene; Obtain the loss value between the predicted click probability and the corresponding feature value in the target behavior feature; If the loss value is detected to not meet the training constraints, the parameters of the first training matrix, the parameters of the first training network, and / or the parameters of the second training network are adjusted, and the step of extracting features from the sample user behavior data, the sample user interest data, and the sample user attribute data to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features is returned to continue execution. If the loss value is found to satisfy the training constraints, the finally trained network model is determined as the message identifier elimination prediction model for the first scenario.

8. The method according to any one of claims 1 to 7, characterized in that, The method further includes: Detect whether the first message identifier has been displayed in the first scenario of the communication client within the preset time window to which the requested access time belongs; If so, the first scenario controlling the communication client will prevent the display of the first message identifier within the preset time window to which the requested access time belongs; If not, proceed with the step of determining whether to control the first message identifier to be displayed in the first scenario of the communication client based on the detection result.

9. A message identifier display control device, characterized in that, The device, used for displaying and controlling message identifiers in any scenario of a communication client, includes: The access request acquisition module is used to acquire access requests from the communication client. The access request includes the first user identifier who logged into the communication client and the access request time. After the communication client is started, the module performs access operations on each scenario displayed on each of the output communication pages to generate access requests for the corresponding scenarios. The first predicted user identifier query module is used to query the first predicted user identifier stored in association with the requested access time. The first predicted user identifier refers to the predicted user identifier whose click prediction probability of the first message identifier displayed in the first scenario of the communication client is greater than a first probability threshold within the preset time window to which the requested access time belongs. The first message identifier is used to indicate that there are unviewed new messages in the first scenario. The preset time window is determined based on the information operation behavior characteristics of the user in the message identifier display scenarios output by the communication client. The detection module is used to detect whether the first user identifier belongs to the first predicted user identifier and obtain the corresponding detection result; The first message identifier display control module is used to respond to the access request and, based on the detection result, determine whether to control the first message identifier to be displayed in the first scenario of the communication client.

10. The apparatus according to claim 9, characterized in that, The device further includes: The first data acquisition module is used to acquire historical user behavior data, user interest data, and user attribute data of the user on the communication client; wherein, the historical user behavior data is generated based on the user's operation behavior on different message identifier display scenarios of the communication client within different preset time windows; The click prediction module is used to input the historical user behavior data, the user interest data, and the user attribute data into the message identifier elimination prediction model for processing, and output the click prediction probability of the user clicking the first message identifier displayed in the first scenario within different preset time windows. The predictive user identifier determination module is used to determine the user identifier of the user logging into the communication client as the predictive user identifier for the corresponding preset time window if the click prediction probability is greater than a first probability threshold. The storage module is used to associate and store the predicted user identifier with the corresponding preset time window.

11. The apparatus according to claim 10, characterized in that, The device further includes: The second data acquisition module is used to acquire sample user behavior data, sample user interest data, and sample user attribute data for each sample user. The feature extraction module is used to extract features from the sample user behavior data, the sample user interest data, and the sample user attribute data respectively, to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features. The feature space transformation module is used to perform spatial transformation on the sample user behavior features, sample user interest features, and sample user attribute features based on the preset time window with a time period, respectively, to obtain the corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features with the time period. The feature merging processing module is used to merge the undetermined behavior features, the undetermined interest features, and the undetermined attribute features to obtain the corresponding sample user encoding vector; The target behavior feature acquisition module is used to acquire the target behavior features of each sample user in the first scenario; The model training module is used to train a network model using the sample user encoding vector and the target behavior features to obtain a message identifier elimination prediction model for the first scenario.

12. The apparatus according to claim 11, characterized in that, The requested access time consists of a first time in a first time dimension and a second time in a second time dimension. The first time dimension has a first time period, the second time dimension has a second time period, and the second time period is a time unit of the first time period. Therefore, the preset time window includes a first time window consisting of one time unit of the first time period and a second time window consisting of one time unit of the second time period. The first predictive user identifier query module includes: The first determining unit is used to determine the target first time window to which the first time belongs, and the target second time window to which the second time belongs; The first predicted user identifier determination unit is used to determine the predicted user identifier stored in association with the second time window of the target under the first time window of the target as the first predicted user identifier.

13. The apparatus according to claim 12, characterized in that, The feature space transformation module includes: The first feature space transformation unit is used to perform spatial transformation on the sample user behavior features, the sample user interest features and the sample user attribute features based on the first time window with a first time period, to obtain the corresponding first undetermined behavior features, first undetermined interest features and first undetermined attribute features with the first time period. The second feature space transformation unit is used to perform spatial transformation on the sample user behavior features, the sample user interest features and the sample user attribute features based on the second time window with the second time period, to obtain the corresponding second undetermined behavior features, second undetermined interest features and second undetermined attribute features with the second time period. The feature merging processing module is specifically used for: The first undetermined behavioral feature, the first undetermined interest feature, the first undetermined attribute feature, the second undetermined behavioral feature, the second undetermined interest feature, and the second undetermined attribute feature are merged to obtain the corresponding sample user encoding vector.

14. The apparatus according to claim 11, characterized in that, The feature space transformation module includes: The normalization processing unit is used to normalize the preset time window within a time period to obtain a normalized time vector; The spatial transformation parameter construction module is used to construct spatial transformation parameters using the normalized time vector and the parameters of the first training matrix; A spatial transformation unit is used to perform spatial transformation on the corresponding feature values ​​of the input features using the spatial transformation parameters to obtain the corresponding undetermined input features; The input features are the sample user behavior features, the sample user interest features, or the sample user attribute features, and the undetermined input features are the corresponding undetermined behavior features, undetermined interest features, and undetermined attribute features.

15. The apparatus according to claim 14, characterized in that, The feature extraction module includes: The first feature extraction unit is used to input the sample user behavior data into the first feature extraction network and output the sample user behavior features. The second feature extraction unit is used to input the sample user interest data into the second feature extraction network and output the sample user interest features; wherein, the network structures of the first feature extraction network and the second feature extraction network are different, and the first feature extraction network has a first network parameter to be trained, and the second feature extraction network has a second network parameter to be trained. The third feature extraction unit is used to perform vectorization processing on the sample user attribute data to obtain sample user attribute features. The model training module includes: The click prediction probability obtaining unit is used to input the sample user encoding vector into the pooling layer of the network model for processing, and obtain the click prediction probability of the corresponding sample user clicking on the first message identifier displayed in the first scene; The loss value acquisition unit is used to acquire the loss value between the click prediction probability and the corresponding feature value in the target behavior feature; The network model parameter adjustment unit is used to adjust the first training matrix parameters, the first training network parameters, and / or the second training network parameters when the loss value is detected to not meet the training constraints. Then, the step of extracting features from the sample user behavior data, the sample user interest data, and the sample user attribute data to obtain the corresponding sample user behavior features, sample user interest features, and sample user attribute features is returned to continue execution. The message identifier removal prediction model determination unit is used to determine the finally trained network model as the message identifier removal prediction model for the first scenario when the loss value is detected to satisfy the training constraints.

16. A storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor to implement the steps of the message identifier display control method as described in any one of claims 1-8.

17. A computer program product, characterized in that, The computer program product includes computer instructions, and the processor of the computer device executes the computer instructions, causing the computer device to perform each step of the message identifier display control method according to any one of claims 1-8.

Images