A user recall method, device, equipment and computer readable storage medium

Abstract

The application provides a user recall method and device, equipment and a computer readable storage medium; the user recall method comprises the following steps: determining a plurality of user relationship pairs between a first user and a plurality of second users to be recalled; traversing the plurality of user relationship pairs, predicting the click probability value of the first user clicking the second user and the number of blog posts published by the second user; according to the predicted click probability value and the number of blog posts, determining a target user relationship pair in the plurality of user relationship pairs, wherein the recall benefit of the target user relationship pair meets a first preset condition; according to the second user information in the target user relationship pair, recalling the blog information of the corresponding second user for the first user. Through the application, the reasonable number of recall materials of the first user can be ensured, the invalid calculation problem caused by too much recall quantity can be reduced, and the efficiency of recall can be improved.

Description

Technical Field

[0001] This application relates to computer processing technology, and more particularly to a user recall method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] Recommendation systems help users discover content they are interested in from large amounts of data and recommend this content to the appropriate users. Recommendation systems typically include two phases: recall and ranking. Currently, the recall phase mostly employs full recall or multi-path recall with identical constraints for each user.

[0003] However, while full recall theoretically maximizes the pool of potential users, the computational cost in practical applications is too high. Conversely, multi-path recall using the same constraints cannot be applied to every unique user, resulting in an excessive number of invalid recalls. Summary of the Invention

[0004] This application provides a user recall method, apparatus, and computer-readable storage medium to reduce computational overhead while ensuring recall effectiveness.

[0005] The technical solution of this application is implemented as follows:

[0006] In a first aspect, this application provides a user recall method, comprising: determining multiple user relationship pairs between a first user and multiple second users to be recalled, wherein a user relationship pair includes a first user and an associated second user; traversing multiple user relationship pairs and predicting the click probability value of the first user clicking on the second user and the number of blog posts published by the second user; determining, based on the predicted click probability value and the number of blog posts, a target user relationship pair among the multiple user relationship pairs whose recall benefits satisfy a first preset condition; and recalling the corresponding blog post information of the second user for the first user based on the second user information in the target user relationship pair.

[0007] In some possible implementations, determining multiple user relationship pairs between a first user and multiple second users to be recalled includes: acquiring historical interaction data of the first user; determining relationship information between the first user and each of the multiple third users based on the historical interaction data, wherein the relationship information is used to indicate the strength of the interaction relationship between the first user and each third user; determining, based on the relationship information, the third users whose interaction relationship strength meets a second preset condition as multiple second users; and associating the first user with each of the multiple second users to obtain multiple user relationship pairs.

[0008] In some possible implementations, predicting the click probability value of the first user clicking the second user includes: acquiring the first user's first user data and the second user's second user data; extracting the interaction feature data between the first user and the second user based on the first user data and the second user data; and obtaining the click probability value of the first user clicking the second user based on the first user data, the second user data, and the interaction feature data using a pre-trained click probability prediction model.

[0009] In some possible implementations, predicting the number of blog posts published by a second user includes: acquiring record data of blog posts published by each second user within a preset time period; extracting the posting features of each second user based on the record data; and inputting the posting features of each second user into a blog posting number prediction model to obtain the predicted number of blog posts published by each second user.

[0010] In some possible implementations, before recalling the corresponding blog posts of the second user for the first user based on the second user information in the target user relationship pair, the method further includes: determining the predicted number of blog posts published by the recalled second user for the first user; and suspending the recall of the corresponding blog posts of the second user for the corresponding first user when the total number of predicted blog posts of all recalled second users is greater than a first threshold.

[0011] In some possible implementations, based on the predicted click probability value and the number of blog posts, target user relationship pairs among multiple user relationship pairs whose recall benefits meet the first preset condition are determined, including: determining the recall benefits of recalling the second user from the first user based on the predicted click probability value of the first user clicking the second user and the predicted number of blog posts published by the second user; and determining the user relationship pairs that meet the first preset condition from multiple user relationship pairs as target user relationship pairs.

[0012] In some possible implementations, the user recall method further includes: determining whether the total number of predicted posts by second users recalled by each first user is greater than a second threshold; if the total number of predicted posts is greater than the second threshold, ending the user recall; otherwise, continuing to execute the user recall method.

[0013] Secondly, this application provides a user recall device, comprising:

[0014] The first determining module is used to determine multiple user relationship pairs between a first user and multiple second users to be recalled, wherein a user relationship pair includes a first user and an associated second user; the prediction module is used to traverse multiple user relationship pairs and predict the click probability value of the first user clicking on the second user and the number of blog posts published by the second user; the second determining module is used to determine the target user relationship pairs among the multiple user relationship pairs whose recall benefits meet the first preset condition based on the predicted click probability value and the number of blog posts; the recall module is used to recall the corresponding blog post information of the second user for the first user based on the second user information in the target user relationship pair.

[0015] In some possible implementations, the first determining module includes: a first acquiring module, used to acquire historical interaction data of a first user; a third determining module, used to determine relationship information between the first user and each of a plurality of third users based on the historical interaction data, wherein the relationship information is used to indicate the strength of the interaction relationship between the first user and each third user; a fourth determining module, used to determine third users whose interaction relationship strength meets a second preset condition as a plurality of second users based on the relationship information; and an association module, used to associate the first user with each of the plurality of second users to obtain a plurality of user relationship pairs.

[0016] In some possible implementations, the prediction module is further configured to acquire first user data of a first user and second user data of a second user; extract interaction feature data between the first user and the second user based on the first user data and the second user data; and obtain the click probability value of the first user clicking the second user based on the first user data, the second user data, and the interaction feature data using a pre-trained click probability prediction model.

[0017] In some possible implementations, the prediction module is further configured to acquire record data of blog posts published by each second user within a preset time period; extract the posting features of each second user based on the record data; input the posting features of each second user into the posting number prediction model to obtain the predicted number of blog posts published by each second user.

[0018] In some possible implementations, the user recall device further includes: a fifth determining module, configured to determine the predicted number of blog posts published by the recalled second user who is the first user; and a termination module, configured to terminate the recall of blog posts corresponding to the second user for the corresponding first user when the total number of predicted blog posts of all recalled second users is greater than a first threshold.

[0019] In some possible implementations, the second determining module is further configured to determine the recall benefit of the first user recalling the second user based on the predicted click probability value of the first user clicking the second user and the predicted number of blog posts published by the second user; and to determine the user relationship pair that meets the first preset condition from multiple user relationship pairs as the target user relationship pair.

[0020] In some possible implementations, the user recall device further includes: a sixth determining module, configured to determine whether the total number of predicted posts by the second users recalled by each first user is greater than a second threshold; and a termination recall module, configured to terminate the user recall when the total number of predicted posts is greater than the second threshold; otherwise, to continue executing the user recall method.

[0021] Thirdly, this application provides a user recall device, comprising:

[0022] Memory, used to store executable instructions;

[0023] A processor, when executing executable instructions stored in memory, implements the method provided in the first aspect of this application.

[0024] Fourthly, this application provides a computer-readable storage medium storing executable instructions for implementing the method provided in the first aspect of this application when executed by a processor.

[0025] Fifthly, this application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the method provided in the first aspect of this application.

[0026] The advantages of the technical solution provided in this application compared with the prior art are as follows:

[0027] In this application, by identifying related user pairs and excluding second users who are not related to the first user, the computational load is greatly reduced while maintaining effectiveness. Furthermore, based on predictions of the probability of the first user clicking on the second user's post and the number of posts published by the second user, the recall benefits for the first user are comprehensively considered. The second user is recalled using a first preset condition that satisfies the recall benefits. This ensures that the first user has a reasonable amount of recall materials while reducing the problem of invalid calculations caused by excessive recall, thus improving recall efficiency. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a user recall method in the related technologies of this application;

[0029] Figure 2 This is a schematic diagram of the architecture of a user recall system provided in an embodiment of this application;

[0030] Figure 3 This is a schematic diagram of the server structure provided in an embodiment of this application;

[0031] Figure 4 This is a flowchart illustrating a user recall method provided in an embodiment of this application.

[0032] Figure 5 This is a flowchart illustrating another user recall method provided in an embodiment of this application;

[0033] Figure 6 This is a flowchart illustrating the process of predicting the probability of a first user clicking on a second user's click in an embodiment of this application.

[0034] Figure 7 This is a flowchart illustrating the process of predicting the number of posts made by the second user in an embodiment of this application.

[0035] Figure 8 This is a schematic diagram of another user recall method provided in the embodiments of this application.

[0036] Figure 9 This is a flowchart illustrating the user recall method provided in this application embodiment applied to social application software;

[0037] Figure 10 This is a schematic diagram of a computer device structure provided in an embodiment of this application. Detailed Implementation

[0038] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0039] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0040] In the following description, the terms "first / second / third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0042] To illustrate the technical solution described in this application, specific embodiments are provided below.

[0043] With the rapid development of technology, especially the widespread use of mobile communication networks and mobile terminals, various blog post data exists on the Internet. Users can also produce blog post data anytime and anywhere. Different manufacturers provide platforms, which on the one hand collect some of the blog post data produced by users, and on the other hand, filter high-quality blog post data through different methods, and push corresponding blog post data to users based on different user interest types, so as to increase the probability of users clicking to view the pushed data.

[0044] Various online platforms typically store large amounts of blog post data, forming blog post databases. This data can include multimedia data in various formats, such as video, audio, images, and text. The format varies depending on the business scenario, including live broadcasts, short videos, songs, comics, audiobooks, novels, news, articles, and so on. Platforms can proactively publish this multimedia data; for example, they can purchase various multimedia data from copyright holders for display. Alternatively, registered users can generate various types of multimedia data using electronic devices and upload them to the platform. Other users can then search and browse this multimedia data using electronic devices, including but not limited to laptops, tablets, desktop computers, and mobile terminals.

[0045] Different business scenarios can involve various forms of multimedia data. For example, in a live streaming scenario, users can use cameras to capture video data and perform operations such as beautification and live interaction to generate live programs. In a short video scenario, users can use cameras to capture video data and perform operations such as beautification, editing, and adding special effects to generate short videos. In an information sharing and communication scenario, users can upload edited text, images, videos, etc., to generate blog posts. This multimedia data can be published on corresponding platforms, and other users can access this data on the platforms using smart devices such as terminals, or share this data content.

[0046] For clarity, the user who acquires multimedia data through smart devices can be referred to as the first user, and the user who produces this data can be referred to as the second user. For example, in a live streaming scenario, the second user is the broadcaster, and the first user is the viewer; in a self-media scenario, the second user is the editor or blogger, and the first user is the reader, and so on. Furthermore, for the platform, users can both produce this multimedia data and acquire multimedia data produced by other users; that is, a user can be both a first user and a second user.

[0047] Recommendation systems emerged to more accurately filter content from massive amounts of data and recommend it to users. By pushing content that users are likely to be interested in, recommendation systems can attract new users and increase monthly active users, thereby achieving user growth. Essentially, it recommends high-click-rate content to users and is widely used in e-commerce, search, and advertising to recommend personalized items. For example, in advertising scenarios, personalized recommendation systems can push ads to users based on their characteristics and preferences. If the user ultimately clicks and converts, the ad push is considered successful; otherwise, it fails.

[0048] Recommendation systems typically include two phases: recall and ranking. The recall phase often employs a full recall approach. For example, based on user follow data, all users followed by the first user are recalled, with only simple filtering performed on these second users, such as removing spam, advertising, or marketing accounts. Alternatively, a multi-path recall with specific restrictions can be used, where each path has the same restrictions for each user. For instance, restrictions could be based on the interaction relationship between the first and second users, such as their level of intimacy. Figure 1 This is a flowchart illustrating a user recall method in the related technology of this application. Figure 1 As shown, the user recall system first obtains the intimacy value between the first user and the second user (S1), then determines whether the intimacy value between the first user and the second user exceeds a preset threshold (S2), and retains user relationship pairs whose intimacy values ​​exceed the threshold (S3), thus obtaining the intimacy-based recall result (S4). In addition to the intimacy-based recall, a manual strategy can be added for further filtering (S5), such as manually adding restrictions on whether the first user follows the second user (S6), to obtain a recall sequence of first and second users with a follower relationship (S7), resulting in the manual strategy recall result (S8). Finally, the obtained intimacy-based recall result and the manual strategy recall result are merged to achieve recall (S9).

[0049] However, while full recall theoretically maximizes the candidate pool for recalling second users, in practice, it suffers from several drawbacks. First, an excessively large candidate pool leads to an exponential increase in computational resource costs due to the massive amount of data required for computation. Second, since some second users have no actual connection to the first user's interests, full recall introduces too many invalid second users, interfering with the subsequent ranking process and affecting the recommendation system's results. Third, simply setting recall conditions, such as the aforementioned closeness relationship between the first and second users, is too simplistic and cannot apply a specific threshold to any particular first user. For example, using a closeness threshold might result in insufficient or no second users to be recalled for some first users with low interaction and closeness, while conversely, some active first users might have an excessive number of candidate second users. This user recall method is not applicable to every user and may lead to an excessive number of invalid recalls.

[0050] To address the aforementioned issues, embodiments of this application provide a user recall method, apparatus, device, and computer-readable storage medium that can reduce computational overhead and improve recall efficiency while ensuring recall effectiveness.

[0051] The user recall method provided in this application can be implemented by various electronic devices. For example, it can be implemented by a server alone, or by a terminal and a server working together. For example, the server can execute the live streaming recommendation method described below on its own, or the terminal can send a recall request message to the server, and the server can execute the user recall method based on the received recall request message.

[0052] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, and this application embodiment does not impose specific limitations in this regard.

[0053] See Figure 2 , Figure 2 This is a schematic diagram of the architecture of the user recall system 200 provided in the embodiments of this application. In this system, the terminal (terminal 201-1 and terminal 201-2 are shown as examples) is connected to the server 203 through the network 202. The network 202 can be a wide area network or a local area network, or a combination of the two.

[0054] In some embodiments, the terminal or server 200 can implement the user recall method provided in this application embodiment by running a computer program. For example, the computer program can be a native program or software module in the operating system; for instance, it can be a native application (APP), that is, a program that needs to be installed in the operating system to run, such as a live streaming APP; it can also be a live streaming mini-program, that is, a live streaming program that only needs to be downloaded into the browser environment to run; or it can be a live streaming mini-program that can be embedded in any APP. In short, the above-mentioned computer program can be any form of application, module or plugin, and this application embodiment does not specifically limit it.

[0055] In some possible implementations, user A can publish blog post A through terminal 201-1, and user B can publish blog post B through terminal 201-2 (in this case, both user A and user B are second users). Blog posts A and B are uploaded to server 203 via network 202. Server 203 can then upload blog posts A and B to database 204. Other users can view these blog posts through terminals or other smart devices. Simultaneously, the database also stores interaction records between each user and other users, such as the time user A visits user B, the number of visits within a predetermined time period, and behavioral data such as forwarding, liking, and commenting on blog post B uploaded by user B. Alternatively, it could contain records of user B visiting user A, or records of visiting any other user (in this case, both user A and user B are first users). When implementing user recall, server 203 can filter a portion of second users as users to be recalled based on the interaction relationship between first and second users, and introduce a recall benefit calculation for recalling second users, selecting second users whose recall benefits meet preset conditions for recall.

[0056] See Figure 3 , Figure 3 This is a schematic diagram of the server structure provided in an embodiment of this application. Figure 3 The server 203 shown may include at least one processor 310, memory 320, at least one network interface 330, and user interface 340. The various components in device 300 are coupled together via a bus system 350. It is understood that the bus system 350 is used to implement communication between these components. In addition to a data bus, the bus system 350 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 3 The general labeled all buses as Bus System 350.

[0057] The processor 310 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0058] User interface 340 includes one or more output devices 341 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. User interface 304 also includes one or more input devices 342, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0059] The memory 320 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 320 may optionally include one or more storage devices physically located away from the processor 310.

[0060] The memory 320 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 320 described in this application embodiment is intended to include any suitable type of memory.

[0061] In some embodiments, memory 320 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0062] Operating system 321 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0063] The network communication module 322 is used to reach other computing devices via one or more (wired or wireless) network interfaces 330, such as Bluetooth, WiFi, and Universal Serial Bus (USB).

[0064] Presentation module 323 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 342 (e.g., a display screen, a speaker, etc.) associated with user interface 340;

[0065] The input processing module 324 is used to detect and translate one or more user inputs or interactions from one or more input devices 342.

[0066] In some embodiments, the user recall device provided in this application can be implemented in software. Figure 3 A user recall device 325 stored in memory 320 is shown. This device can be software in the form of programs and plug-ins, and includes the following software modules: a first determination module 3251, a prediction module 3252, a second determination module 3253, and a recall module 3254. These modules are logically linked and can therefore be arbitrarily combined or further split according to their implemented functions. The functions of each module will be described below.

[0067] In other embodiments, the user recall device provided in this application can be implemented in hardware. As an example, the user recall device provided in this application can be a processor in the form of a hardware decoding processor, which is programmed to execute the user recall method provided in this application. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0068] The user recall method provided in this application will be described below with reference to exemplary applications and implementations of the user recall device provided in the embodiments of this application.

[0069] See Figure 4 , Figure 4 This is a flowchart illustrating a user recall method provided in an embodiment of this application. The following will be combined with... Figure 4 The steps shown are explained.

[0070] S401, determine multiple user relationship pairs between the first user and multiple second users to be recalled.

[0071] A user relationship pair includes a first user and an associated second user.

[0072] In some embodiments, user relationship pairs can be used to indicate the association between a first user and a second user. The calculation process for recalling the second user can be performed on the server described above. Before recalling users, the server can first obtain multiple user relationship pairs corresponding to each first user, and determine the second users to be recalled based on these multiple user relationship pairs. That is, it first determines a set of second users to be recalled that includes multiple second users based on the user relationship pairs. The number of multiple user relationship pairs can be an empirical value, determined based on computing resources, or a manually preset threshold. This application embodiment does not specifically limit this.

[0073] In some embodiments, the determination of user relationship pairs can be based on historical interaction data between a first user and a second user. The interaction records between the first user and the second user can reflect, to some extent, whether there is a relationship between them. This allows for the filtering out of some second users who do not meet the criteria based on the relationship between the first user and the second user, thus reducing the amount of computational waste.

[0074] In some possible implementations, the user recall device determines user relationship pairs based on historical interaction data between the first user and the second user. Step S401 may include S501 to S504, see [link / reference]. Figure 5 , Figure 5 This is an optional flowchart illustrating the user recall method provided in this application embodiment. The following will be combined with... Figure 5 The steps shown are explained.

[0075] S501, obtain the historical interaction data of the first user;

[0076] In some embodiments, the historical interaction data of the first user may include interaction data between the first user and multiple other users, including users the first user has followed and users they have not followed. The historical interaction data of the first user may include various types of data. For example, taking a user recall method in a self-media business scenario as an example, the first user is a reader. The first user can browse blog posts published by bloggers they have followed or those they have not followed through devices such as terminals. Simultaneously, the first user can perform some operations on the browsed blog posts. For example, if the first user is interested in the blog post, they can perform operations such as saving, sharing, liking, or commenting; or if the first user dislikes the blog post content, they can mark it as "disliked," such as marking it to reduce recommendations for that type of blog post or clicking "dislike."

[0077] In addition, an important metric in the recommendation system is the probability that a user clicks on and navigates to the content information page recommended by the system. Therefore, in an optional embodiment, the first user's historical interaction data can also be the historical data of the first user clicking on blog posts recommended to the first user by the recommendation system.

[0078] In an optional embodiment, the historical interaction data of the first user, in addition to the historical data of the first user's actions on the blog posts published by the blogger, may also include relationship data between the first user and the blogger, such as the intimacy value between the first user and the blogger calculated using a preset algorithm model based on the aforementioned various action data. Alternatively, the historical interaction data of the first user may also comprehensively consider the aforementioned action data and relationship data. The specific types of historical interaction data of the first user to be included can be determined based on actual application requirements, and this application embodiment does not specifically limit this.

[0079] S502, Based on historical interaction data, determine the relationship information between the first user and each of the multiple third users.

[0080] Among them, the relationship information is used to indicate the strength of the interaction relationship between the first user and each third user.

[0081] It should be noted that a third user is any user who can publish blog posts. For example, a third user can be the set of all users followed by the first user, or a third user can be the set of users who have the aforementioned interactive behavior records with the first user.

[0082] Understandably, the second user mentioned above was selected from the third user based on the strength of their interaction. The strength of the interaction between the first user and the third user can, to some extent, indicate the first user's interest in the blog posts published by the third user. In other words, the higher the first user's interest in the third user, the more likely they are to click on and open the corresponding blog post recommended by the system.

[0083] In some embodiments, the relationship information between the first user and the third user can be calculated based on the various types of historical interaction data mentioned above.

[0084] Understandably, each type of historical interaction data can establish a correlation on a specific dimension, and the correlation strength of that type of interaction data can be calculated based on the specific type of historical interaction data. For example, a user recall device calculates the correlation strength between the first user and the third user based on the first user's reposting records of posts by the third user; a user recall device calculates the correlation strength between the first user and the third user based on the first user's liking records of posts by the third user, and so on. Generally, the more historical interaction data between the first user and the third user, the stronger the correlation between them, that is, the higher the correlation strength between the first user and the third user. However, in general, the longer the time span, the more interaction records the first user and the third user accumulate. It is possible that even if the interaction frequency between the first user and the third user is not high, the correlation calculated directly based on historical interactions may be high due to the accumulation over a long period of time. Therefore, when calculating the correlation between the first user and the third user based on the first user's historical interaction data, the time decay factor can be taken into account to reduce potential errors. For example, a user recall device can calculate the association between a first user and a third user based on whether the first user is newly following the third user, using this as a form of interaction data. Alternatively, the user recall device can directly add a time decay parameter when calculating the association to reduce the impact of time span.

[0085] For example, taking the historical interaction data of the first user, including the first user's forwarding records, like records, and comment records on the blog posts published by the third user, as an example, with the addition of a time decay parameter, in the case of the first user u interacting with the third user b within a predetermined time window k days, t i The timing, the decay value of the like association strength from the current value is shown in expression (1):

[0086]

[0087] Wherein, likP(u,b,t) i ) indicates that the first user u interacts with the third user b at time t. i The timing, the strength of the like association, and its current decay value. Attenuation parameter, The value is a constant, as described in the embodiments of this application. The value can be 0.012, k is an integer greater than or equal to 1, and i is a positive integer less than or equal to k.

[0088] Based on the above expression (1), after summing over the time window, we can obtain the overall like association strength likAll(u,b) of the first user u for the third user b within the time window k days, as shown in expression (2):

[0089]

[0090] Based on the same method, the forwarding association strength fwdAll(u,b) of the first user u to the third user b as a whole within a time window of k days and the comment association strength cmtAll(u,b) of the first user u to the third user b as a whole within a time window of k days can be obtained respectively, as shown in the following expressions (3) and (4):

[0091]

[0092]

[0093] Furthermore, in the recommendation system, whether the first user clicks to open recommended content has a stronger correlation than actions such as liking, forwarding, and commenting. Also, typically, the number of clicks by the first user on recommended content within the same time window is much smaller than the interactions of liking, commenting, and forwarding. Therefore, simply using methods to calculate the correlation strength of these interactions and introducing a time decay parameter may result in a smaller degree of separation for the third user. Therefore, in this embodiment, when calculating the click management strength of the first user on the third user's recommended blog posts based on the first user's click records, in addition to introducing a time decay parameter to calculate the absolute click correlation strength of the first user u on the third user b within a preset time window k days, a relative click correlation strength of the first user u on the third user b can also be calculated within a longer time window kl than k. This relative click strength of the first user u on the third user is compared to the click strength of other third users. Finally, the obtained absolute click strength and relative click correlation strength of the first user are used together to calculate the comprehensive click strength of the first user u on the third user b.

[0094] For example, within a time window k, the absolute click association strength of the first user u to the third user b, clickAbs(u,b), can be expressed as in expression (5):

[0095]

[0096] After obtaining the absolute click association strength of the first user u to the third user b within the time window k, the relative click association strength clickRel(u,b) of the first user u to the third user b within the time window kl, which is longer than the time window k, can be calculated as shown in expression (6):

[0097]

[0098] Where kl is an integer and the value of kl is greater than k, and e is a constant.

[0099] After obtaining the relative and absolute click strengths of the first user u to the third user b, the combined click strength of the first user u to the third user b, clickScore(u,b), can be expressed as in expression (7):

[0100] clickScore(u,b)=clickAbs(u,b)×clickRel(u,b) (7)

[0101] In this embodiment, the user recall device comprehensively considers the short-term absolute click intensity and long-term relative click intensity of the first user towards the third user within different time windows. This reduces the situation where the click behavior of the first user towards the third user is relatively sparse compared to other interactive behaviors, due to the smaller magnitude of the first user's clicks on the third user's recommended blog posts compared to interactive information such as forwarding and liking. Consequently, the calculation of the click behavior of the first user and the click intensity of the third user are less separated from other third users, thereby improving the accuracy of calculating the association between the first user and the third user.

[0102] In this embodiment of the application, after the user recall device calculates the correlation strength of various data based on the above-mentioned interaction data between the first user and the third user, it can determine the interaction relationship strength between the first user and each third user based on a comprehensive consideration of the correlation strength of the various data.

[0103] In some embodiments, when calculating the interaction strength between the first user and each third user, the user recall device may also introduce relationship data obtained based on various interaction data between the first user and the third user, such as whether the first user follows the third user, or whether the first user has newly followed the third user based on the time the first user followed the third user. Alternatively, it may obtain the intimacy value between the first user and the second user using a preset intimacy algorithm model based on the interaction data between the first user and the second user. The intimacy algorithm model can be pre-trained or can be other existing models; this embodiment does not specifically limit its application.

[0104] For example, the intimacy between the first user u and the third user b can be represented as: qmd(u,b).

[0105] Whether the first user u has recently become a new user of interest to the third user b can be determined by expression (8):

[0106]

[0107] After obtaining the association strengths of the interaction behaviors between the first user u and the third user b (likAll(u,b), commenting, forwarding, and clicks), as well as the association strengths of intimacy (qmd(u,b) and whether the user recently followed newFlw(u,b)), the user recall device inputs this association strength data into a regression model to calculate the weights of each data point. For example, using a logistic regression (LR) model, the feature vector containing the above association strength data is denoted as x, and the parameter vector is denoted as θ. The parameters are trained based on the historical data of the first user u clicking on the recommended blog posts of the third user b. Historical data samples of the first user u clicking on the recommended information of the third user b are labeled as positive samples 1, and historical data samples of the information of the first user u not clicking on the recommended information of the third user b are labeled as negative samples 0. The value of the parameter vector θ is calculated through logistic regression. The interaction strength between the first user u and the third user b is determined based on the calculated parameter vector θ.

[0108] For example, the interaction strength recallScore(u,b) between the first user u and the third user b can be expressed as in expression (9):

[0109] recallScore(u,b)=sigmoid(θ·x) (9)

[0110] In some embodiments, the logistic regression model described above can be updated based on the various interaction data obtained daily, and the model parameters can be updated with the latest interaction data to improve the accuracy of the strength of the interaction relationship between the first user u and the third user b.

[0111] The interaction strength between the first user and each third user is calculated using the above method, and the second user to be recalled is selected from the third users based on the interaction strength.

[0112] S503, based on the relationship information, the third user whose interaction relationship strength meets the second preset condition is identified as multiple second users;

[0113] The second preset condition can be a preset fixed threshold for the strength of the interaction relationship. For example, after the user recall device calculates the interaction relationship strength between the first user and each third user according to the above-mentioned method, it determines all third users whose interaction relationship meets the threshold as second users to be recalled. Alternatively, the second preset condition can also be a preset user number threshold. For example, if the user number threshold is 500, after the user recall device calculates the interaction relationship strength between the first user and each third user according to the above-mentioned method, it sorts the third users based on the interaction relationship strength and selects the top 500 third users as second users to be recalled.

[0114] In summary, the second preset condition is to select multiple third users from the third users to be identified as the second users to be recalled. The specific number of third users to be identified can be selected based on actual application requirements, and this application embodiment does not specifically limit this.

[0115] S504, associate the first user with each of the multiple second users to obtain multiple user relationship pairs.

[0116] In some embodiments, the user recall device associates a first user with multiple second users, which can be achieved by individually tagging each second user for the first user. Alternatively, the multiple second users obtained above can be placed in the same sequence list, and the first user can be associated with the corresponding sequence list. Alternatively, any other method can be used to associate the first user with each second user, thereby obtaining multiple user relationship pairs.

[0117] After obtaining multiple user relationship pairs between the first user and the second user through the above method, the user recall method further includes:

[0118] S402, iterate through multiple user relationship pairs and predict the click probability value of the first user clicking the second user and the number of blog posts published by the second user;

[0119] Understandably, the ultimate goal of a recommendation system is to increase the probability of a user clicking on recommended content. By predicting the probability of a first user clicking on a second user's post and the probability of the second user publishing a post, we can, to some extent, indicate the probability that if a post that the second user is currently likely to publish is recommended to the first user, the first user will click on and open the post. This allows us to determine the recall benefit of recalling the second user, select the second user with higher benefit for recall, and improve recall efficiency.

[0120] In some embodiments, the probability of a first user clicking on a second user's click can be predicted through the following steps.

[0121] S4021, Obtain the first user data of the first user and the second user data of the second user;

[0122] It should be noted that the first user data can be feature data reflecting the first user's operational behavior, while the second user data can be feature data of the blog posts published by the second user. For example, taking a self-media scenario as an example, the first user can be a registered user of the self-media platform, and the second user can be a registered blogger on the self-media platform. Within a preset time window, such as a week, the first user's first user data includes, but is not limited to: the number of active days of the first user on the platform that week, the average number of times the first user opened the platform per day that week, the average daily online time of the first user that week, the number of blog posts viewed by the first user that week, the number of times the first user clicked on recommended blog posts that week, the number of bloggers followed by the first user, the total number of likes of the first user that week, the total number of comments of the first user that week, the total number of reposts of the first user that week, the number of new bloggers followed by the first user that week, the number of bloggers privately messaged by the first user that week (whether followed or not), and the number of mutual follows between the first user and other users. Correspondingly, the characteristics of the second user include, but are not limited to: the number of blog posts published by the second user in that week, the number of followers of the second user, the number of active fans of the second user, the number of likes received by the second user in that week, the number of comments received by the second user in that week, the number of reposts received by the second user in that week, and the number of times the second user was recalled and recommended to any first user in that week.

[0123] S4022, Extract the interaction feature data between the first user and the second user based on the first historical data and the second historical data;

[0124] Understandably, interaction feature data refers to the characteristic data of the interaction behavior between the first user and the second user. For example, taking a time window of one month as an example, interaction feature data may include, but is not limited to: the number of times the first user viewed the second user's blog posts in that month, the number of times the first user commented on the second user's blog posts in that month, the number of times the first user liked the second user's blog posts in that month, the number of times the first user forwarded the second user's blog posts in that month, the number of times the first user clicked on the second user's recommended information in that month, the number of private messages between the first and second users in that month, and whether the first and second users follow each other, etc.

[0125] S4023. Based on the first user data, the second user data, and the interaction feature data between the first user and the second user, the click probability prediction model is used to obtain the click probability value of the first user clicking the second user.

[0126] In some embodiments, after obtaining the first user data, the second user data, and the interaction feature data between the first user and the second user, the user recall device can use these data to train a click probability prediction model to predict the probability of the first user clicking on the second user based on the trained model.

[0127] Understandably, when constructing training samples based on the historical moments when the first user clicked on the second user's posts, for each user relationship pair, regardless of how many times the second user recommended blog posts to the first user that day, as long as the first user clicked on any of those recommendations, it can be considered that the first user clicked on the second user's post. Therefore, when constructing training samples, click records for that day can be recorded as click samples, and otherwise as non-click samples. For example, click samples can be recorded as 1, and non-click samples can be recorded as 0.

[0128] For example, when training a click prediction model, the user recall device uses a feature classification algorithm to take the records of whether the first user clicked on the second user's click within a preset time window as training samples. The obtained first user data, second user data, and interaction feature data between the first and second users are used as input features for model training, resulting in a trained click probability prediction model. The feature classification algorithm can be any existing algorithm model, such as logistic regression. A loss function is added to reduce data bias; for example, the loss function could be binary cross-entropy loss (BCE Loss).

[0129] In some embodiments, after the user recall device obtains the trained click probability prediction model through the above training process, it can also optimize the model using the newly acquired data every day, and at the same time update the model every preset time period.

[0130] For example, the first user's daily new data may include, but is not limited to: the number of times the first user opened the platform on that day, the first user's online time on that day, the first user's click on recommended blog post information on that day, the number of bloggers the first user newly followed on that day, the number of likes the first user received on that day, the number of comments the first user made on that day, the number of blog posts the first user forwarded on that day, and the number of bloggers the first user sent private messages to on that day.

[0131] See Figure 6 As shown, Figure 6This is a flowchart illustrating the process of predicting the probability of a first user clicking on a second user's click in an embodiment of this application. The training samples for the click probability prediction model include historical data of the first user clicking on the second user's clicks. The input sample features for model training include first user feature data, second user feature data, and interaction feature data between the first and second users. A click probability prediction model is trained based on the logistic regression algorithm using the training samples and sample feature data to predict the probability of the first user clicking on the second user's click. Simultaneously, the model is optimized based on new sample feature data acquired daily, and every week, newly acquired sample data from the previous week is added to retrain the model to ensure the accuracy of the prediction model.

[0132] In some embodiments, predicting the number of blog posts published by a second user can be achieved through the following steps.

[0133] S4024, respectively obtain the record data of blog posts published by each second user within a preset time period;

[0134] Understandably, the preset time period can be selected based on the actual application scenario requirements. For example, it can be to obtain the record data of the second user's blog posts within the past week, or the record data within a month, or record data of any other time length.

[0135] In some embodiments, the record data of the second user's blog posts may be a record of blog post data uploaded to the server by the second user through a smart device such as a terminal. For example, this record data may include the content of all blog posts published by the second user, the publication time of each blog post, etc.

[0136] S4025, Extract the posting characteristics of each second user based on the recorded data;

[0137] It should be noted that the second user's posting characteristics are extracted based on the second user's posting records and are used to indicate the second user's posting habits, preferences, and other characteristic data. For example, the second user's posting characteristics may include, but are not limited to, the number of times the second user has posted in the past 3 days, the number of times the second user has posted in the past 3 days, the number of times the second user has posted in the past 5 days, the number of times the second user has posted in the past 5 days, the number of times the second user has posted in the past 7 days, the number of times the second user has posted in the past 7 days, the number of times the second user has posted in the past 10 days, the number of times the second user has posted in the past 10 days, the number of times the second user has posted in the past 30 days, and the number of times the second user has posted in the past 30 days.

[0138] S4026, Input the posting characteristics of each second user into the posting quantity prediction model to obtain the predicted number of posts published by each second user.

[0139] In some embodiments, the post count prediction model can be trained based on the post count feature data of the second user obtained above, so as to predict the number of posts published by the second user according to the trained model.

[0140] For example, when training a post count prediction model, training samples can be constructed based on the historical post count of the second user. For instance, the average daily post count of the second user within a preset time window can be used as the training sample, or the actual daily post count of the second user within a preset time window can be used as the training template.

[0141] The preset time window can be determined based on actual application needs. For example, the preset time window could be one month. In this case, the total number of blog posts published by the second user in the past month can be counted, and the average number of blog posts per day by the second user in the past month can be calculated as a training sample. Alternatively, it can be based on the actual number of blog posts by the second user per day in the past month. The choice of the size of the preset time window does not limit the scope of protection of this application.

[0142] In some embodiments, a linear regression algorithm can be used to train the blog post count prediction model. For example, the average daily blog post count of the second user within a preset time window can be used as a training sample, and the blog post characteristic data of the second user obtained above can be used as sample features to train the model, thereby obtaining the trained blog post count prediction model.

[0143] In some embodiments, when training a model using a linear regression algorithm, a loss function can be added to reduce data bias during the model training process. For example, the loss function could be the mean squared error loss (MSE Loss).

[0144] Understandably, the output of the post count prediction model is the predicted post count for the second user, which is a specific predicted value. Therefore, when the predicted post count is greater than or equal to 1, the model's output value is the predicted post count for the second user, while when the predicted value is less than 1, the model's output value can be approximated as the posting probability for the second user.

[0145] For example, the blog post count prediction model predicts that the second user's blog post count for the next day is 0.6 based on the second user's historical blog post data. In this case, it can be approximately assumed that the second user has a 60% probability of posting a blog post the next day.

[0146] In one embodiment, after the user recall device obtains the trained post count prediction model through the above training process, it can further optimize the model based on the post count of the newly acquired second user each day, and can also update the model at preset intervals.

[0147] See Figure 7 As shown, Figure 7 This is a schematic diagram illustrating the process of predicting the number of posts by a second user in this embodiment of the application. The training samples for the post count prediction model include the average number of posts by the second user per day over the past 30 days. The input sample feature data for model training includes the post characteristics of the second user. A post count prediction model is trained based on a linear regression algorithm using the training samples and sample feature data. This model is then used to predict the number of posts the second user will publish the following day. Simultaneously, the model is optimized based on new sample feature data acquired daily. Furthermore, every week, the number of posts by the second user acquired daily over the past week is used as new sample feature data to retrain the model, ensuring the accuracy of the prediction model.

[0148] It is understandable that the execution order of predicting the probability of the first user clicking on the second user and predicting the number of posts by the second user is not limited. The user recall device may first predict the probability of the first user clicking on the second user and then predict the number of posts by the second user, or the user recall device may first predict the number of posts by the second user and then predict the probability of the first user clicking on the second user, or the two prediction processes may be performed simultaneously. This application embodiment does not specifically limit this.

[0149] Understandably, recalling the second user for the first user aims to recommend the blog posts published by the second user to the first user so that the first user can click to view the recommended posts. Therefore, to avoid recommending too many blog posts to a first user on the same day, a daily limit on the number of blog posts recommended to a first user can be set to reduce the possibility of invalid recommendations. Based on this, before performing step S404 to recall the second user for the first user, the user recall method further includes:

[0150] S4041, the predicted number of blog posts published by the second user who has been recalled as the first user;

[0151] S4042, when the total number of predicted posts of all recalled second users exceeds the first threshold, the recall of second users for the corresponding first user is stopped.

[0152] The first threshold can be an empirical value or it can be calculated based on actual application requirements.

[0153] For example, for each recalled second user, the sum of the predicted post counts can approximate the number of recommendable posts associated with the corresponding first user. For instance, the recalled second users for the first user include users B, C, and D. User B's predicted post count is T(B) = 0.5, user C's is T(C) = 1.5, and user D's is T(D) = 2. Then, for the first user, on the day following the predicted post counts of the recalled second users, the predicted number of recommended posts for the first user is D(A) = T(B) + T(C) + T(D) = 4. If the first threshold is 3, and D(A) is greater than the first threshold, then recalling the first user is no longer performed.

[0154] After obtaining the predicted click probability of the first user clicking on the second user's post and the predicted number of posts published by the second user using the method in step S402 above, step S403 is executed. Step S403 includes:

[0155] S403, Based on the predicted click probability value and the number of blog posts, determine the target user relationship pairs among multiple user relationship pairs whose recall revenue meets the first preset condition;

[0156] Understandably, recall benefit can be represented as the probability that, after a first user recalls a blog post from a second user, that second user's post is recommended to the first user, and the first user clicks to view the recommended post. A higher recall benefit indicates a greater likelihood that the first user will click to view the second user's post, meaning a higher probability of a valid recommendation. Conversely, a lower recall benefit indicates a higher probability of an invalid recommendation. Therefore, selecting user relationship pairs with higher recall benefits from user relationship pairs can improve the probability of valid recommendations.

[0157] In one embodiment, recalling target user relationship pairs that meet preset conditions for revenue from multiple user relationship pairs can be achieved through the following steps:

[0158] Step S4031: Based on the predicted click probability value of the first user clicking on the second user and the predicted number of blog posts published by the second user, determine the recall revenue of the first user to recall the second user.

[0159] Understandably, step S402 above can yield the predicted probability value of the first user clicking on the second user's post, as well as the predicted value of the number of posts published by the second user. Furthermore, considering that the number of recommendations for a single first user on a given day should not be excessive, a maximum daily recommendation limit parameter for the first user can be introduced when calculating the recall revenue for the second user's posts, thereby optimizing the recall revenue calculation.

[0160] For example, the predicted number of posts by the second user currently recalled by the first user A can be calculated as the proportion of the first user A's maximum number of recommendations on that day, denoted as Q(A). The predicted click probability of the first user A clicking on the second user B to be recalled can be denoted as P(AB), and the predicted number of posts by the second user can be denoted as T(B). At this time, the recall revenue L(A) for recalling the blog posts corresponding to the second user B for the first user A can be as shown in expression (10):

[0161] L(A)=Q(A)×P(AB) / T(B) (10)

[0162] Using expression (10), the recall revenue for recalling each blog post corresponding to the second user for the first user can be calculated. Thus, the following steps are performed to determine...

[0163] Step S4032: Determine the target user relationship pair that meets the first preset condition from multiple user relationship pairs.

[0164] The first preset condition can be a preset recall revenue threshold, which can be an empirical value or a value calculated based on actual application.

[0165] From multiple user relationship pairs, target user relationship pairs that meet the first preset condition are identified. The recall revenue corresponding to each second user's blog post recalled by the first user can be calculated using the above expression (10). Then, the recall revenue is sorted, and the blog posts of the second user corresponding to the first preset condition with recall revenue greater than the first preset condition are selected for recall. Alternatively, a preset recall strategy can be used to automatically calculate the blog posts corresponding to the second user with the largest recall revenue. For example, the existing greedy strategy method can be used to compare the recall revenue of the blog posts of the second user and select the user relationship pairs corresponding to the second user with the largest recall revenue.

[0166] S404, based on the second user information in the target user relationship pair, recall the corresponding second user's blog post information for the first user.

[0167] In some embodiments, see Figure 8 As shown, Figure 8 This is an optional flowchart illustrating a user recall method provided in an embodiment of this application, based on... Figure 8 After step S404, the following can also be executed:

[0168] S801, determine whether the total number of predicted posts from the second users recalled by each first user is greater than the second threshold;

[0169] Understandably, recommendation systems have certain computational resource limitations. Beyond a certain number of recommended posts, the required computational resources increase exponentially, leading to excessive computational costs. Furthermore, excessive recommendations also result in more invalid recommendations. Therefore, setting a maximum limit on the total number of recommendations can reduce the waste of computational resources. This second threshold can be an empirical value or determined based on current computational resources; this embodiment does not impose specific limitations on it.

[0170] S802, when the total predicted number of posts by the second user corresponding to each first user is greater than the second threshold, the user recall ends.

[0171] Otherwise, continue with the user recall method described above.

[0172] For example, the predicted number of posts from the second users recalled by the first user A can be denoted as D(A). Then, the total number of predicted posts from the second users recalled by each first user can be denoted as... Where i is greater than or equal to n, and both i and n are integers greater than or equal to 1. The second threshold can be denoted as C. When When the value of is less than or equal to C, the above user recall method can continue to be executed to recall the blog post corresponding to the second user for the first user. When the value of is greater than C, it means that the predicted number of posts by the second user who has been recalled has exceeded the upper limit of the number of recommendations, and the user recall method ends at this time.

[0173] In some optional embodiments, the above method may further include:

[0174] S8021, sort multiple second users according to the predicted click probability of the first user clicking the second user to generate a recall sequence;

[0175] S8022: After recalling the second user for the first user, the corresponding second user is removed from the pending recall sequence and added to the already recalled sequence.

[0176] Understandably, generating a recall sequence based on click probability allows for the direct selection of the second user with the highest click probability during user recall calculations, improving the efficiency of calculating recall revenue using a greedy algorithm. Simultaneously, adding the recalled second user to the already recalled sequence facilitates the calculation of the predicted total number of posts from that user. Furthermore, when recommending posts, the system can directly select posts from the second user's corresponding posts in the already recalled sequence, further improving the efficiency of the post recommendation process.

[0177] In this embodiment, by identifying associated user pairs and excluding second users who are not associated with the first user, the computational load is greatly reduced while maintaining effectiveness. Simultaneously, based on predictions of the first user's click probability on the second user and the number of blog posts published by the second user, the recall benefit of recalling the second user for the first user is comprehensively considered. The second user is recalled using a first preset condition that satisfies the recall benefit, ensuring that the first user has a reasonable amount of recall materials while reducing the problem of invalid calculations caused by excessive recall volume, thus improving recall efficiency.

[0178] The following will describe an exemplary application of the embodiments of this application in a real-world application scenario.

[0179] When the above user recall method is applied to social applications (APPs), such as Weibo and other social platforms, the second user can be a blogger who can upload and publish blog posts, and the first user can be any registered user on the platform. The number of multiple second users can be M, where M is an integer greater than 0. For the implementation process of the above user recall method applied to the Weibo platform, please refer to [link to relevant documentation]. Figure 9 As shown, Figure 9 This is a flowchart illustrating the user recall method provided in this application embodiment applied to social application software. The following will combine... Figure 9 The steps shown are explained.

[0180] S901, the user recall device obtains the first user's historical interaction data of forwarding, liking and commenting;

[0181] S902, the user recall device calculates the correlation strength between the first user and the third user's forwarding, liking, and commenting based on the first user's historical interaction data.

[0182] S903, the user recall device obtains the historical interaction data of the first user clicking on the third user;

[0183] S904, The user recall device calculates the click intensity of the first user on the third user based on the historical interaction data of the clicks;

[0184] S905, The user recall device obtains the affinity value between the first user and the third user;

[0185] S906, The user recall device determines whether the first user is a newly concerned third user;

[0186] S907, The user recall device obtains the historical data of the first user clicking on the third user within a preset time window;

[0187] S908, the user recall device uses the historical data of the first user clicking on the third user as training samples, and uses the association strength of the first user's forwarding, liking and commenting with the third user, the click strength of the first user on the third user, the intimacy value between the first user and the third user, and whether the first user is a newly followed third user as feature vector x. The parameter θ is calculated by the logistic regression algorithm, and the interaction strength between the first user and the third user is determined based on the parameter vector θ.

[0188] S909, the user recall device sorts the third users based on the strength of the interaction relationship between the first user and the third user and selects the top M third users as the second users to be recalled to generate a recall sequence.

[0189] S910, the user recall device predicts the probability of the first user clicking on the second user's click based on a click probability prediction model;

[0190] S911, the user recall device predicts the number of posts of the second user based on the post count prediction model;

[0191] S912, the user recall device determines whether to recall the first user based on whether the predicted number of posts of the second user who has been recalled by the first user is greater than the first threshold; if not, proceed to step S913; if yes, proceed to step S914.

[0192] S913, terminate the recall of the second user for the first user;

[0193] S914, the user recall device determines the target user relationship pair to be recalled based on the predicted probability of the first user clicking on the second user and the number of posts by the second user, using a greedy strategy.

[0194] S915, the user recall device determines whether the total number of predicted posts of the second users recalled by each first user is greater than the second threshold C. If yes, proceed to step S917; otherwise, proceed to step S916.

[0195] S916, the user recall device continues to select a second user from the recall sequence for the first user and adds the recalled second user to the already recalled sequence;

[0196] S917, recall ended.

[0197] It is understandable that the execution order of the above steps S901 to S902, S903 to S904, S905, S906 and S907 is not limited.

[0198] This completes the user recall method described above.

[0199] The following continues to describe the exemplary structure of the user recall device 325 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 3 As shown, the software modules stored in the user recall device 325 in the memory 320 may include: a first determining module 3251, used to determine multiple user relationship pairs between a first user and multiple second users to be recalled, wherein a user relationship pair includes a first user and an associated second user; a prediction module 3252, used to traverse multiple user relationship pairs and predict the click probability value of the first user clicking on the second user and the number of blog posts published by the second user; a second determining module 3253, used to determine the target user relationship pairs among the multiple user relationship pairs whose recall benefits meet the first preset condition based on the predicted click probability value and the number of blog posts; and a recall module 3254, used to recall the corresponding blog post information of the second user for the first user based on the second user information in the target user relationship pair.

[0200] In some possible implementations, the first determining module 3251 includes: a first acquiring module, used to acquire historical interaction data of a first user; a third determining module, used to determine relationship information between the first user and each of a plurality of third users based on the historical interaction data, wherein the relationship information is used to indicate the strength of the interaction relationship between the first user and each of the third users; a fourth determining module, used to determine the third users whose interaction relationship strength meets a second preset condition as a plurality of second users based on the relationship information; and an association module, used to associate the first user with each of the plurality of second users to obtain a plurality of user relationship pairs.

[0201] In some possible implementations, the prediction module 3252 is further configured to acquire first user data of the first user and second user data of the second user; extract interaction feature data between the first user and the second user based on the first user data and the second user data; and obtain the click probability value of the first user clicking the second user based on the first user data, the second user data, and the interaction feature data using a pre-trained click probability prediction model.

[0202] In some possible implementations, the prediction module 3252 is further configured to acquire the record data of blog posts published by each second user within a preset time period; extract the blog posting features of each second user based on the record data; input the blog posting features of each second user into the blog posting number prediction model to obtain the predicted number of blog posts published by each second user.

[0203] In some possible implementations, the user recall device further includes: a fifth determining module, configured to determine the predicted number of blog posts published by the recalled second user who is the first user; and a termination module, configured to terminate the recall of blog posts corresponding to the second user for the corresponding first user when the total number of predicted blog posts of all recalled second users is greater than a first threshold.

[0204] In some possible implementations, the second determining module 3253 is further configured to determine the recall benefit of the first user recalling the second user based on the predicted click probability value of the first user clicking the second user and the predicted number of blog posts published by the second user; and to determine the user relationship pair that meets the first preset condition from multiple user relationship pairs as the target user relationship pair.

[0205] In some possible implementations, the user recall device 3254 further includes: a sixth determining module, configured to determine whether the total number of predicted posts by the second users recalled by each first user is greater than a second threshold; and a termination recall module, configured to terminate the user recall when the total number of predicted posts is greater than the second threshold; otherwise, to continue executing the user recall method.

[0206] Figure 10 This is a schematic diagram of a computer device structure provided in an embodiment of this application. The computer device 110 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 111 (e.g., one or more processors) and a memory 112, and one or more storage media 113 (e.g., one or more mass storage devices) for storing application programs 121 or data 122. The memory 112 and storage media 113 can be temporary or persistent storage. The program stored in the storage media 113 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the computer device. Furthermore, the CPU 111 may be configured to communicate with the storage media 113 and execute the series of instruction operations in the storage media 113 on the computer device 110.

[0207] Computer device 110 may also include one or more power supplies 114, one or more wired or wireless network interfaces 115, one or more input / output interfaces 116, and / or one or more operating systems 117, such as Windows Server™, Mac OS X™, Uni™, Linux™, FreeBSD™, etc.

[0208] This application provides a computer program product or computer program that includes 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 user recall method described in this application.

[0209] This application provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to execute the user recall method provided in this application. For example, ... Figure 4 The user recall method is shown.

[0210] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0211] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0212] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0213] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0214] In summary, this application embodiment, by identifying associated user relationship pairs and excluding second users who are not associated with the first user, significantly reduces computational load while maintaining effectiveness. Furthermore, based on predictions of the probability of the first user clicking on the second user's click and the number of blog posts published by the second user, it comprehensively considers the recall benefits for the first user in recalling the second user. It uses constraints that satisfy the first preset condition for recall benefits to recall the second user, ensuring that the first user has a reasonable amount of recall materials while reducing invalid calculations caused by excessive recall volume, thus improving recall efficiency.

[0215] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A user recall method, characterized in that, The method includes: Multiple user relationship pairs are determined between a first user and multiple second users to be recalled. Each user relationship pair includes the first user and an associated second user. The user relationship pairs are determined based on historical interaction data between the first user and the second user. Iterate through the multiple user relationship pairs to predict the click probability value of the first user clicking the second user and the number of blog posts published by the second user; Based on the predicted click probability value and the number of blog posts, target user relationship pairs among the multiple user relationship pairs whose recall revenue meets the first preset condition are determined; the recall revenue is used to indicate the probability that the first user will view the recommended blog posts published by the second user. Based on the second user information in the target user relationship pair, the blog post information of the corresponding second user is recalled for the first user.

2. The method according to claim 1, characterized in that, The determination of multiple user relationship pairs between the first user and multiple second users to be recalled includes: Obtain the historical interaction data of the first user; Based on the historical interaction data, the relationship information between the first user and each of the multiple third users is determined, and the relationship information is used to indicate the strength of the interaction relationship between the first user and each of the third users. Based on the relationship information, the third user whose interaction relationship strength meets the second preset condition is identified as one of the multiple second users; The first user is associated with each of the plurality of second users to obtain the plurality of user relationship pairs.

3. The method according to claim 2, characterized in that, The prediction of the probability value of the first user clicking the second user's click includes: Obtain the first user data of the first user and the second user data of the second user; Based on the first user data and the second user data, extract the interaction feature data between the first user and the second user; Based on the first user data, the second user data, and the interaction feature data, a pre-trained click probability prediction model is used to obtain the click probability value of the first user clicking the second user.

4. The method according to claim 2, characterized in that, The prediction of the number of blog posts published by the second user includes: Obtain the record data of blog posts published by each of the second users within a preset time period; Extract the posting characteristics of each second user based on the recorded data; Input the posting characteristics of each second user into the posting quantity prediction model to obtain the predicted number of posts published by each second user.

5. The method according to claim 4, characterized in that, Before recalling the blog post information of the corresponding second user for the first user based on the second user information in the target user relationship pair, the method further includes: The predicted number of blog posts published by the second user who has been recalled and identified as the first user; When the total number of predicted posts from all recalled second users exceeds the first threshold, the recall of posts corresponding to the second user for the first user is suspended.

6. The method according to claim 5, characterized in that, The step of determining the target user relationship pairs among the multiple user relationship pairs whose recall revenue meets the first preset condition based on the predicted click probability value and the number of blog posts includes: Based on the predicted click probability value of the first user clicking on the second user and the predicted number of blog posts published by the second user, the recall revenue of the first user recalling the second user is determined; From the plurality of user relationship pairs, the user relationship pairs that satisfy the first preset condition are determined as the target user relationship pairs.

7. The method according to claim 4, characterized in that, The method further includes: Determine whether the total number of predicted posts from each of the second users recalled by the first user is greater than a second threshold. When the total number of predicted posts exceeds the second threshold, the user recall process ends; otherwise, the user recall method continues.

8. A user recall device, characterized in that, The device includes: The first determining module is used to determine multiple user relationship pairs between a first user and multiple second users to be recalled. Each user relationship pair includes the first user and an associated second user. The user relationship pair is determined based on historical interaction data between the first user and the second user. The prediction module is used to traverse the multiple user relationship pairs and predict the click probability value of the first user clicking the second user and the number of blog posts published by the second user. The second determining module is used to determine, based on the predicted click probability value and the number of blog posts, the target user relationship pairs among the multiple user relationship pairs whose recall benefits meet the first preset condition; the recall benefits are used to indicate the probability that the first user will view the recommended blog posts published by the second user. The recall module is used to recall the blog post information of the corresponding second user for the first user based on the second user information in the target user relationship pair.

9. A user recall device, characterized in that, The user recall device includes: Memory, used to store executable instructions; A processor, when executing executable instructions or computer programs stored in the memory, implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing executable instructions or a computer program, characterized in that, When the executable instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images