An interest pushing method, device and equipment and computer readable storage medium

Abstract

The application provides an interest pushing method, which comprises the following steps: obtaining user information of a first blogger recalled based on user behavior characteristics for a target user; obtaining user information of a second blogger recalled based on user interest characteristics and / or social relationship characteristics for the target user; inputting at least the user information of the first blogger and the user information of the second blogger into a ranking model to obtain ranking results of the first blogger and the second blogger, wherein the ranking results are used to reflect the interested degree of the target user to the bloggers; determining a target blogger according to the ranking results, and pushing blog information of the target blogger to the target user. Through the application, the diversity of recalled materials can be enriched, and accurate pushing to the target user can be realized.

Description

Technical Field

[0001] The present application relates to the field of computer processing technology, and in particular to an interest push method, apparatus, device, and computer-readable storage medium. Background Art

[0002] Currently, in social push notification scenarios, push notification systems typically use models to calculate content based on user characteristics, material characteristics, and blogger characteristics to push content of interest to users. There are two main approaches to improving push notification accuracy. One is to build serialized features to capture changes in users' long-term and short-term interests. However, this approach is only effective for a small number of active users and is difficult to achieve accurate push notifications for medium- and low-frequency users. The second approach is to upgrade the model structure to capture the relationships between different features during the calculation process. However, this approach increases model complexity, computation time, and computing resource consumption.

[0003] Therefore, a push method is needed that can target all frequency users, does not increase computing resource overhead, and can improve push accuracy. Summary of the Invention

[0004] The present application provides an interest push method, apparatus, device and computer-readable storage medium, which can enrich the diversity of recalled materials and achieve accurate push to target users.

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

[0006] The present application provides an interest push method, comprising: obtaining user information of a first blogger recalled for a target user based on user behavior characteristics; obtaining user information of a second blogger recalled for the target user based on user interest characteristics and / or social relationship characteristics; inputting at least the user information of the first blogger and the user information of the second blogger into a ranking model to obtain a ranking result of the first blogger and the second blogger, the ranking result being used to reflect the target user's interest level in the blogger; determining a target blogger based on the ranking result, and pushing the target blogger's blog post information to the target user.

[0007] In some possible implementations, obtaining user information of a second blogger recalled for a target user based on user interest characteristics includes: obtaining user information of a third blogger that meets preset screening conditions; obtaining a sequence of interactive behaviors of the target user with respect to the third blogger, where the interactive behavior sequence is used to indicate a sequence of interactive behaviors of the target user with respect to different third bloggers; determining user information of a fourth blogger based on the sequence of interactive behaviors of the target user with respect to the third blogger, where the fourth blogger is used to indicate a blogger among the third bloggers who has interactive behaviors with the target user; and obtaining user information of a second blogger recalled for the target user based on the user interest characteristics based on the user information of the fourth blogger, where the second blogger is at least a portion of the fourth bloggers.

[0008] In some possible implementations, based on the user information of the fourth blogger, user information of the second blogger recalled for the target user based on the user interest characteristics is obtained, including: inputting the user information of each fourth blogger into a deep learning model to obtain the blogger interest characteristics of each fourth blogger; performing weighted processing on the blogger interest characteristics of each fourth blogger to obtain a weighted processing result, and using the weighted processing result as the user interest characteristics of the target user; performing similarity calculation on the blogger interest characteristics and user interest characteristics of each fourth blogger, screening out the second blogger according to the calculation result, and determining the user information of the second blogger.

[0009] In some possible implementations, obtaining user information of third bloggers that meet preset screening conditions includes: dividing the quantity of materials published by each blogger within a preset time period by the number of exposures of each blogger within the preset time period, and multiplying the result by the interaction value of each blogger within the preset time period to obtain a popularity value of each blogger, where the popularity value is used to indicate the degree to which the materials published by each blogger in the corresponding field are liked by users; based on the preset screening conditions, screening third bloggers among each blogger that meet the preset screening conditions, and determining the user information of the third bloggers.

[0010] In some possible implementations, obtaining user information of a second blogger recalled for a target user based on social relationship characteristics includes: obtaining a user relationship sequence of the target user based on the target user's follow-up relationship; extracting user social relationship characteristics of the target user based on the user relationship sequence; the user social relationship characteristics are used to represent the social relationship between the target user and the bloggers followed by the target user, and / or the social relationship between the target user and the bloggers not followed by the target user; and obtaining the user information of the second blogger based on the user social relationship characteristics.

[0011] In some possible implementations, in response to a user social relationship feature being used to represent the social relationship between a target user and a blogger followed by the target user; obtaining user information of a second blogger based on the user social relationship feature, includes: obtaining a first blogger social relationship feature of the blogger followed by the target user; performing a similarity calculation on the user social relationship feature and the first blogger social relationship feature, screening out a second blogger based on the calculation result, and determining the user information of the second blogger.

[0012] In some possible implementations, in response to a user social relationship feature being used to represent a social relationship between a target user and a blogger that the target user does not follow; obtaining user information of a second blogger based on the user social relationship feature, includes: obtaining a second blogger social relationship feature of a blogger that the target user does not follow, the blogger that the target user does not follow having a common following relationship with the target user; performing similarity calculation on the user social relationship feature and the second blogger social relationship feature, screening out the second blogger based on the calculation result, and determining the user information of the second blogger.

[0013] In some possible implementations, in response to user social relationship characteristics used to represent the social relationship between a target user and a blogger followed by the target user and the social relationship between the target user and a blogger not followed by the target user, user information of a second blogger is obtained according to the user social relationship characteristics, including: obtaining a first blogger social relationship characteristic of a blogger followed by the target user and a second blogger social relationship characteristic of a blogger not followed by the target user, the blogger not followed by the target user having a common following relationship with the target user; performing similarity calculation on the user social relationship characteristic and the first blogger social relationship characteristic, and performing similarity calculation on the user social relationship characteristic and the second blogger social relationship characteristic, screening out the second blogger according to the calculation results, and determining the user information of the second blogger.

[0014] In some possible implementations, obtaining user information of a second blogger recalled by a target user based on user interest characteristics and social relationship characteristics includes: obtaining user information of a third blogger that meets preset screening conditions, a target user's interactive behavior sequence with respect to the third blogger, and a target user's user relationship sequence; determining user information of the second blogger recalled by the target user based on user interest characteristics based on the user information of the third blogger and the target user's interactive behavior sequence with respect to the third blogger; and determining user information of the second blogger recalled by the target user based on social relationship characteristics based on the target user's interactive behavior sequence with respect to the third blogger.

[0015] In some possible implementations, the above method also includes: inputting the user-side embedded features of the target user, the material-side embedded features, the blogger-side embedded features of the second blogger, the identity embedded features of the target user, and the identity embedded features of the second blogger into a deep learning model to obtain the identity identification features between the target user and the second blogger.

[0016] In some possible implementations, the method further includes: inputting the identity identification features between the target user and the second blogger, the user information of the first blogger, and the user information of the second blogger into a ranking model to obtain ranking results of the first blogger and the second blogger.

[0017] The present application provides an interest push device, comprising: a data acquisition module, configured to obtain user information of a first blogger recalled by a target user based on user behavior characteristics; obtain user information of a second blogger recalled by the target user based on user interest characteristics and / or social relationship characteristics; a model calculation module, configured to input at least the user information of the first blogger and the user information of the second blogger into a ranking model to obtain a ranking result of the first blogger and the second blogger, the ranking result being used to reflect the target user's interest level in the blogger; and a push module, configured to determine a target blogger based on the ranking result and push the target blogger's blog post information to the target user.

[0018] In some possible implementations, the data collection module is further used to obtain user information of a third blogger that meets preset screening criteria; obtain a target user's interactive behavior sequence with respect to the third blogger, where the interactive behavior sequence is used to indicate a sequence of the target user's interactive behaviors with respect to different third bloggers; determine user information of a fourth blogger based on the target user's interactive behavior sequence with respect to the third blogger, where the fourth blogger is used to indicate a blogger among the third bloggers who has interactive behaviors with the target user; and obtain user information of a second blogger recalled for the target user based on user interest characteristics based on the user information of the fourth blogger, where the second blogger is at least a portion of the fourth bloggers.

[0019] In some possible implementations, the model calculation module is further used to input the user information of each fourth blogger into the deep learning model to obtain the blogger interest characteristics of each fourth blogger; obtain a weighted processing result by weighting the blogger interest characteristics of each fourth blogger, and use the weighted processing result as the user interest characteristics of the target user; perform similarity calculation on the blogger interest characteristics and user interest characteristics of each fourth blogger, screen out the second blogger according to the calculation result, and determine the user information of the second blogger.

[0020] In some possible implementations, the apparatus further includes a popularity value calculation module; the popularity value calculation module is configured to divide the quantity of materials published by each blogger within a preset time period by the number of exposures of each blogger within the preset time period, and multiply the result by the interaction value of each blogger within the preset time period to obtain the popularity value of each blogger, where the popularity value is used to indicate the degree to which the materials published by each blogger in the corresponding field are liked by users; based on preset screening conditions, a third blogger among each blogger that meets the preset screening conditions is screened, and user information of the third blogger is determined.

[0021] In some possible implementations, the data acquisition module is further used to obtain a user relationship sequence of the target user based on the target user's follow-up relationship; extract user social relationship features of the target user based on the user relationship sequence; the user social relationship features are used to represent the social relationship between the target user and the blogger the target user follows, and / or the social relationship between the target user and the blogger the target user does not follow; and obtain user information of the second blogger based on the user social relationship features.

[0022] In some possible implementations, in response to user social relationship characteristics being used to represent the social relationship between a target user and a blogger followed by the target user, the data collection module is further used to obtain the social relationship characteristics of a first blogger of the blogger followed by the target user; a similarity calculation is performed on the user social relationship characteristics and the social relationship characteristics of the first blogger, a second blogger is screened out based on the calculation results, and the user information of the second blogger is determined.

[0023] In some possible implementations, in response to the user social relationship feature being used to represent the social relationship between the target user and a blogger that the target user does not follow; the data collection module is further used to obtain the social relationship feature of a second blogger that the target user does not follow, and the blogger that the target user does not follow has a common follow relationship with the target user; a similarity calculation is performed on the user social relationship feature and the second blogger social relationship feature, the second blogger is screened out based on the calculation result, and the user information of the second blogger is determined.

[0024] In some possible implementations, in response to user social relationship characteristics being used to represent the social relationship between a target user and a blogger followed by the target user and the social relationship between the target user and a blogger not followed by the target user; the data collection module is further used to obtain a first blogger social relationship characteristic of a blogger followed by the target user and a second blogger social relationship characteristic of a blogger not followed by the target user, where the blogger not followed by the target user has a common following relationship with the target user; similarity calculation is performed on the user social relationship characteristic and the first blogger social relationship characteristic, and similarity calculation is performed on the user social relationship characteristic and the second blogger social relationship characteristic, a second blogger is screened out according to the calculation results, and user information of the second blogger is determined.

[0025] In some possible implementations, the data collection module is further used to obtain user information of a third blogger that meets preset screening conditions, a target user's interactive behavior sequence with respect to the third blogger, and a target user's user relationship sequence; determine user information of the second blogger that the target user recalls based on user interest characteristics based on the user information of the third blogger and the target user's interactive behavior sequence with respect to the third blogger; and determine user information of the second blogger that the target user recalls based on social relationship characteristics based on the target user's interactive behavior sequence with respect to the third blogger.

[0026] In some possible implementations, the model calculation module is also used to input the user-side embedding features of the target user, the material-side embedding features, the blogger-side embedding features of the second blogger, the identity embedding features of the target user, and the identity embedding features of the second blogger into the deep learning model to obtain the identity identification features between the target user and the second blogger.

[0027] In some possible implementations, the model calculation module is further configured to input the identity identification features between the target user and the second blogger, the user information of the first blogger, and the user information of the second blogger into the ranking model to obtain ranking results for the first blogger and the second blogger.

[0028] The present application provides an electronic device, comprising: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided by the present application.

[0029] The present application provides a computer-readable storage medium storing executable instructions for implementing the method provided in the present application when the executable instructions are executed by a processor.

[0030] The present application provides a computer program product, including a computer program or instructions, for implementing the method provided in the present application when the computer program or instructions are executed by a processor.

[0031] This application has the following beneficial effects:

[0032] This application can enrich the diversity of recalled materials by recalling the second blogger. Sorting based on the existing recalled first blogger and the supplementary recalled second blogger can improve the diversity of materials in the push scenario, allowing target users to receive more different types of pushes and improve the target users' usage experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 1 is a schematic diagram of the architecture of the interest push system 100 provided in an embodiment of the present application;

[0034] Figure 2 is a structural diagram of an electronic device 400 provided in an embodiment of the present application;

[0035] Figure 3 This is a flow chart of the interest push method provided in an embodiment of the present application;

[0036] Figure 4 This is an optional flowchart of the interest push method provided in an embodiment of the present application;

[0037] Figure 5 This is an optional flowchart of the interest push method provided in an embodiment of the present application;

[0038] Figure 6 This is an optional flowchart of the interest push method provided in an embodiment of the present application;

[0039] Figure 7 This is an optional flowchart of the interest push method provided in an embodiment of the present application;

[0040] Figure 8 This is an optional flowchart of the interest push method provided in an embodiment of the present application;

[0041] Figure 9 This is a schematic diagram of a sorting model in an embodiment of the present application;

[0042] Figure 10 This is a schematic diagram of a deep learning model in an embodiment of the present application;

[0043] Figure 11 This is a flow chart of the interest push method provided in an embodiment of the present application applied to social application software. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0045] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be 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.

[0046] If similar descriptions of "first / second" appear in the application documents, the following explanation is added. In the following description, the terms "first / second / third" involved are merely to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first / second / third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0048] Currently, there are two main methods to improve the accuracy of content pushed to users in social scenarios. One method starts from the feature perspective and captures the long-term and short-term changes in users' interests by constructing serialized features, thereby more accurately recommending content that users are interested in. Serialized features usually refer to sequences of content that users have recently interacted with and clicked to view. These behaviors have a sequential relationship within them and can reflect changes in users' interests. The other method starts from the perspective of the model and upgrades the model structure to enable the model to capture the relationship between different features, improve the feature combination capability, and make the model have better fitting capabilities.

[0049] In social push scenarios, especially in the social ecosystem of Weibo push, many users' behavior sequences are not rich, so the constructed sequence features are very sparse and only useful for a small number of active users. This makes it difficult to profile low- and medium-frequency users, failing to improve their user experience, and failing to capture potential relationships between non-interactive users. Upgrading the model structure increases model complexity, requiring more computing resources. The increased computation time also affects the timeliness of push notifications and the user experience. Furthermore, in current experiments, the offline AUC (area under curve) improvement after model upgrades is minimal, while the model converges quickly during training. This suggests that the current feature complexity is low and the existing features are unable to capture user interests.

[0050] Therefore, considering the need to comprehensively reflect user interests and social relationships, user profiles can be created, which can also be understood as constructing user vectors. User interests and social relationships can be considered fine-grained features, which are more conducive to the model's matching of users and items. Furthermore, user vectors not only serve the ranking system; the user information they contain can also enrich user recall diversity, providing a better user experience and enhancing the overall social push experience.

[0051] The embodiments of the present application provide an interest push method, apparatus, device, and computer-readable storage medium, which can enrich the diversity of recalled materials and achieve accurate push to target users. The following describes an exemplary application of the electronic device provided by the embodiment of the present application. The electronic device provided by the embodiment of the present application can be implemented as various types of user terminals such as laptops, tablet computers, desktop computers, mobile devices (for example, mobile phones, wearable smart watches, dedicated messaging devices), and can also be implemented as a server. Below, an exemplary application when the electronic device is implemented as a server will be described.

[0052] See also Figure 1 , Figure 1This is an architectural diagram of the interest push system 100 provided in an embodiment of the present application. In order to implement the interest push method described in an embodiment of the present application, electronic devices (electronic device 400-1 and electronic device 400-2 are shown as examples) are connected to the server 200 through a network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0053] In some possible implementations, user A can publish material information A through electronic device 400-1, and user B can publish material information B through electronic device 400-2. Material information A and material information B are uploaded to server 200 via network 300. Server 200 can store material information A and material information B in database 500. In order to accurately push materials based on user interests, server 200 can recommend material information that best matches the user's interests to the user through a ranking model. The recommended material information can be displayed on graphical interface 410 of electronic device 400 (graphical interface 410-1 and graphical interface 410-2 are shown as examples).

[0054] In some embodiments, the server 200 may be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The electronic device 400 may be a smartphone, tablet computer, laptop computer, desktop computer, smartwatch, etc., but is not limited thereto. The terminal and the server may be connected directly or indirectly via wired or wireless communication, which is not limited in the embodiments of the present invention.

[0055] See also Figure 2 , Figure 2 is a structural diagram of an electronic device 400 provided in an embodiment of the present application, Figure 2 The electronic device 400 shown includes: at least one processor 410, a memory 450, at least one network interface 420 and a user interface 430. The various components in the electronic device 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 440 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 440 is not shown in FIG. Figure 2 Various buses are labeled as bus system 440 .

[0056] The processor 410 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., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0057] The user interface 430 includes one or more output devices 431 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0058] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0059] The memory 450 includes volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0060] In some embodiments, the memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplified below.

[0061] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, which are used to implement various basic services and process hardware-based tasks;

[0062] A network communication module 452 for reaching other computing devices via one or more (wired or wireless) network interfaces 420 , exemplary network interfaces 420 including Bluetooth, WiFi, and USB;

[0063] a presentation module 453 for enabling presentation of information via one or more output devices 431 (e.g., a display screen, a speaker, etc.) associated with the user interface 430 (e.g., a user interface for operating peripheral devices and displaying content and information);

[0064] The input processing module 454 is configured to detect one or more user inputs or interactions from one of the one or more input devices 432 and to translate the detected inputs or interactions.

[0065] In some embodiments, the apparatus provided in the embodiments of the present application may be implemented in software. Figure 2 The interest push device 455 stored in the memory 450 is shown. This device can be software in the form of a program or plug-in, and includes the following software modules: a data acquisition module 4551, a model calculation module 4552, and a push module 4553. These modules are logical and can be arbitrarily combined or further separated according to the functions they implement. The functions of each module will be described below.

[0066] In other embodiments, the interest pushing device provided in the embodiments of the present application can be implemented in hardware. As an example, the device provided in the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the interest pushing method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can adopt 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.

[0067] The interest push method provided in the embodiment of the present application will be explained below in combination with the exemplary application and implementation of the electronic device provided in the embodiment of the present application.

[0068] It should be noted that the interest push method described in the embodiment of the present application is to use deep models and different algorithms to recall again from other angles (such as user interests, user attention information, etc.) after the existing recall process has been implemented. It can be understood as a supplement to the existing recall materials.

[0069] See also Figure 3 , Figure 3 This is a flow chart of the interest push method provided by the embodiment of the present application. Figure 3 The steps shown are explained.

[0070] S101, obtaining user information of the first blogger recalled for the target user based on user behavior characteristics.

[0071] In some embodiments, after completing the supplementary recall, the push device needs to sort the existing recalled materials and the supplemented recalled materials together. Therefore, in the embodiment of the present application, the push device needs to obtain the existing recalled materials (that is, the user information of the first blogger recalled by the target user based on the user behavior characteristics).

[0072] S102: Obtain user information of a second blogger recalled for the target user based on user interest characteristics and / or social relationship characteristics.

[0073] In some embodiments, after obtaining the existing recall materials, the push device may further obtain supplementary recall materials (i.e., user information of the second blogger recalled by the target user based on user interest characteristics and / or social relationship characteristics). The above S102 may be implemented after S101 is executed, or before S101 is executed, and this embodiment of the application is not limited to this.

[0074] In some possible implementations, the above S102 may include S201 to S203, see Figure 4 , Figure 4 This is an optional flow chart of the interest push method provided in the embodiment of the present application. Figure 4 The steps shown are explained.

[0075] S201, obtaining user information of a third blogger that meets a preset screening condition.

[0076] In some embodiments, when executing S102 above, the push device may obtain user information of a second blogger, which is retrieved for the target user based on user interest characteristics, and user information of a second blogger, which is retrieved for the target user based on social relationship characteristics. Steps S201 to S204 involve the push device retrieving the user information of the second blogger for the target user based on user interest characteristics. During this process, the push device may first obtain user information of a third blogger, where the third blogger is a related blogger with high exposure in a different field.

[0077] In one example, consider social media platforms like Weibo, where a large number of bloggers consistently post in specific areas. If a blogger posts daily on the same topic and a large number of users interact with them, it can be assumed that they possess relevant expertise in that area and have a certain level of exposure. Therefore, calculations can be used to filter out these bloggers with high exposure and a consistent output of posts in the relevant field (i.e., obtain user information for third-party bloggers that meet the pre-set screening criteria).

[0078] In some possible implementations, the above S201 may include: dividing the quantity of materials published by each blogger within a preset time period by the number of exposures of each blogger within the preset time period, and multiplying the result by the interaction value of each blogger within the preset time period to obtain the popularity value of each blogger, where the popularity value is used to indicate the degree to which the materials published by each blogger in the corresponding field are liked by users; based on preset screening conditions, screening a third blogger among each blogger that meets the preset screening conditions, and determining the user information of the third blogger.

[0079] In some embodiments, the above expression (1) for calculating the blogger popularity value is as follows:

[0080]

[0081] In the formula, score is the popularity value of the blogger, It represents the sum of the interaction values ​​of all the blogger's fans with the blogger within 30 days. Indicates the total number of exposures of the blogger within 30 days. Indicates the number of blog posts published by the blogger in the past 30 days.

[0082] From formula (1), we can see that It can represent the overall popularity of the blogger. The higher the value, the more popular the blogger is in his or her field and the larger the user group he or she covers. The bloggers selected according to the above formula (1) can basically cover all Weibo push groups. The exposure ratio of these bloggers accounts for more than 90% of Weibo.

[0083] In one example, after calculating the popularity value of each blogger based on the above formula (1), the information and popularity value of these bloggers can be displayed in the form of a list, and further based on the preset screening conditions, the top bloggers in each field (i.e., the third blogger) can be determined. Assuming that the preset screening condition is that the popularity value of bloggers in each field is greater than 85, then the information of the top bloggers in each field can be determined based on this screening condition, as shown in Table 1. Table 1 is a schematic table of the information and popularity values ​​of some of the top bloggers in each field.

[0084] Table 1

[0085] Blogger Nickname Heat value field Blogger A 98.2 internet Blogger B 96.6 physical education Blogger C 99.1 Fashion and Beauty Blogger D 92.5 Beautiful women and handsome men Blogger E 86.3 Funny and humorous

[0086] S202: Obtain an interactive behavior sequence of the target user with respect to the third blogger. The interactive behavior sequence is used to indicate the order of the target user's interactive behaviors with respect to different third bloggers.

[0087] In some embodiments, after obtaining the user information of the third blogger, the push device may further obtain a sequence of interactive behaviors between the target user and the third blogger. The interactive behavior sequence may be the order of the target user's interactive behaviors with different third bloggers within a preset time period. The interactive behaviors include liking, forwarding, commenting, clicking push, and other behaviors.

[0088] In one example, when user A (the target user) performs any of the aforementioned interactive behaviors on a third blogger's blog post, the push device retains the information of the third blogger who posted the post and generates a sequence of user A's interactive behaviors based on the order in which the interactive behaviors occurred. Table 2 exemplifies the interactive behavior sequences of users A, B, and C.

[0089] Table 2

[0090] Target users Interactive behavior sequence User A Blogger 1, Blogger 10, Blogger 1000, Blogger 9999… User B Blogger 3, Blogger 4, Blogger 100, Blogger 874... User C Blogger 6, Blogger 90, Blogger 345, Blogger 2323...

[0091] The order of bloggers in the target user's interactive behavior sequence in Table 2 represents the order in which the target user interacted. For example, user A first clicked on blogger 1's blog post, read it, and liked it. Then, he liked the blog posts of bloggers 10, 1000, 9999, and other bloggers (not shown in Table 2, but represented by ellipsis).

[0092] S203: Determine user information of a fourth blogger based on the target user's interactive behavior sequence with the third blogger. The fourth blogger is used to indicate a blogger among the third bloggers who has an interactive behavior with the target user.

[0093] In some embodiments, after the push device executes S202, it can determine the user information of the fourth blogger (such as the nickname of the fourth blogger, the field to which he belongs, etc.) according to the interactive behavior sequence.

[0094] In one example, as shown in Table 2, after obtaining the target user's interaction behavior sequence list, the blogger that appears in the interaction behavior sequence list (i.e., the blogger who interacted with the target user) is the fourth blogger. For example, the fourth bloggers corresponding to user A are blogger 1, blogger 10, blogger 1000, blogger 9999, and so on.

[0095] S204 , obtaining user information of second bloggers recalled for the target user based on user interest characteristics based on the user information of the fourth blogger, where the second bloggers are at least part of the fourth bloggers.

[0096] In some embodiments, after determining the user information of the fourth blogger corresponding to the target user, the push device can obtain the user information of the second blogger based on the user information and the user's interest characteristics.

[0097] In some possible implementations, the above S204 may include S301 to S303, see Figure 5 , Figure 5 This is an optional flow chart of the interest push method provided in the embodiment of the present application. Figure 5 The steps shown are explained.

[0098] S301: Input the user information of each fourth blogger into a deep learning model to obtain blogger interest features of each fourth blogger.

[0099] In some embodiments, after determining the user information of the fourth blogger, the push device can input the above user information into a deep learning model (for example, the deep learning model can be a model using the word2vec (word to vector) algorithm) to obtain the blogger interest characteristics of the fourth blogger output by the deep learning model.

[0100] In one example, the blogger interest features described above may represent the interest tags of a fourth blogger. For example, a blogger specializing in football may have an interest tag of sports, while a celebrity chat host may have an interest tag of entertainment. Assuming the deep learning model is a continuous bag-of-words (CBOW) model, the push device inputs the fourth blogger's user information into the CBOW model, which then uses the word2vec algorithm to determine the fourth blogger's interest features.

[0101] S302 , performing weighted processing on the blogger interest features of each fourth blogger to obtain a weighted processing result, and using the weighted processing result as the user interest feature of the target user.

[0102] In some embodiments, the push device can perform weighted processing on the blogger interest characteristics of the fourth blogger based on factors such as the target user's interaction time, interaction behavior, and interaction times with the fourth blogger, and the weighted processing result is the user interest characteristics of the target user.

[0103] In one example, the user interest characteristics of the target user can be used to represent the interest tags of the target user. For example, based on the interactive behavior of user A, the interest tags of user A are generated, and the interest tags include sports, entertainment, and finance.

[0104] S303: Calculate similarity between the blogger's interest features and the user's interest features to obtain user information of the second blogger.

[0105] In some embodiments, after obtaining the blogger's interest characteristics and the user's interest characteristics, the push device can perform similarity calculation on the blogger's interest characteristics and the user's interest characteristics. Based on the similarity calculation result, it can be determined whether to recall the fourth blogger and determine that the recalled fourth blogger is the second blogger. The above similarity calculation can be performed using cosine similarity calculation, and the calculation formula (2) is as follows:

[0106]

[0107] Where similarity is the calculated value of the cosine similarity between the blogger's interest features and the user's interest features, A represents the vector corresponding to the blogger's interest features, B represents the vector corresponding to the user's interest features, ||A|| represents the modulus of vector A, and ||B|| represents the modulus of vector B.

[0108] In some examples, the pushing device determines the user information of the second blogger based on the cosine similarity calculation result, which may include: determining the second blogger based on a preset similarity threshold, and further determining the user information of the second blogger.

[0109] In one example, assuming a preset similarity threshold of 90, the push device calculates cosine similarity between the blogger's interest characteristics and the user's interest characteristics, obtaining a calculated cosine similarity value for each fourth blogger. The top 10 fourth bloggers corresponding to cosine similarity values ​​greater than 90 are selected and recalled as second bloggers. Assuming user A typically reads blog posts about the internet, science, cars, and stocks, and follows some relevant bloggers, the user information of some second bloggers recalled based on user A's user interest characteristics is shown in Table 3. Table 3 exemplifies the information of some second bloggers recalled based on the similarity calculation results for user A.

[0110] Table 3

[0111]

[0112]

[0113] The second blogger recall is based on the premise that the target user is following the blogger and is likely to be of interest to the target user, thus supplementing the original recall. The original recall typically recalls bloggers who are close to the target user, that is, bloggers with whom the target user frequently interacts. However, some bloggers in the target user's areas of interest may only read their blog posts but rarely interact with them, and these bloggers will not pass the original recall. A similarity calculation is performed between the fourth blogger's interest characteristics and the target user's user interest characteristics. The bloggers recalled based on this similarity calculation result can be bloggers in the target user's following relationships with whom the target user is interested but has less interaction. This can enrich the push scenarios and make the recalled bloggers more in line with the target user's interests, thereby improving the target user's experience.

[0114] In some possible implementations, S102 may further include S401 to S403, see Figure 6 , Figure 6 This is an optional flow chart of the interest push method provided in the embodiment of the present application. Figure 6 The steps shown are explained.

[0115] It should be noted that steps S401 to S403 are about the push device recalling the user information of the second blogger for the target user based on the user's interest characteristics.

[0116] S401: Obtain a user relationship sequence of the target user based on the target user's attention relationship.

[0117] In some embodiments, the push device may construct an undirected graph based on the attention relationship of the target user, and randomly walk on the undirected graph to extract the user relationship sequence of the target user.

[0118] It should be noted that in actual applications, due to the large number of users on the site, the target user will likely follow a large number of bloggers. Constructing an undirected graph based on this would require excessive storage space and would be impractical with current resources. Therefore, before obtaining the target user's follower sequence, the target user's follower relationships must be screened. Considering that some bloggers with large follower counts (e.g., those with more than 10 million followers) and media and news bloggers have weaker social attributes with their followers, it is impossible to represent the social relationship between the target user and these bloggers. Therefore, these bloggers with large follower counts and media and news bloggers are removed from consideration when deriving the target user's user relationship sequence. Excluding these bloggers does not affect supplementary recall for the target user, but constructing an undirected graph for these bloggers does place higher computational performance requirements, placing significant pressure on the computing device.

[0119] In some embodiments, after filtering out the bloggers with a large number of followers and bloggers of media and news categories, the push device can construct an undirected graph using the remaining follow relationships. The push device can input the follow relationships of the target user into a deep learning model using the node2vec algorithm, and use the node2vec algorithm to randomly walk in the undirected graph to generate a user relationship sequence. The expression (3) of the node2vec algorithm is as follows:

[0120]

[0121] In the formula, represents the transition probability from node t to node x, and d tx represents the distance between node t and node x. p and q are two preset hyperparameters used to control the direction of the random walk of the node. When p > max(q, 1), the walk tends to move towards distant nodes. When p < max(q, 1), the walk tends to return to the previous node.

[0122] In an example, p = 1 and q = 0.2 are set. In this way, the random walk sequence tends to move between adjacent nearby nodes, which can focus more on the homogeneity between nodes and better reflect the social relationship between the target user and the followed bloggers.

[0123] S402. Extract the user social relationship features of the target user according to the user relationship sequence; the user social relationship features are used to represent the social relationship between the target user and the bloggers followed by the target user, and / or the social relationship between the target user and the bloggers not followed by the target user.

[0124] In some embodiments, after obtaining the user relationship sequence of the target user, the push device can input the user relationship sequence into a deep learning model and extract the user social relationship features of the target user in the deep learning model. Among them, the user social relationship features can include the user social relationship features of the social relationship between the target user and the bloggers followed by the target user, and / or the user social relationship features of the social relationship between the target user and the bloggers not followed by the target user.

[0125] In an example, the push device inputs the user relationship sequence of the target user into a deep learning model using the word2vec algorithm (such as the Skip-Gram model). The feature extraction layer of the Skip-Gram model can output the user social relationship features of the target user, and the push device can offline store the user social relationship features in a data warehouse analysis system (such as a hive table).

[0126] S403. Obtain the user information of the second blogger according to the user social relationship features.

[0127] In some possible implementations, the target user's social relationship characteristics may include user social relationship characteristics of the target user and the bloggers that the target user follows, and / or user social relationship characteristics of the target user and the bloggers that the target user does not follow. When the user social relationship characteristics represent the social relationship between the target user and the bloggers that the target user follows, the above S403 may include S501 to S502, see Figure 7 , Figure 7 This is an optional flow chart of the interest push method provided in the embodiment of the present application. Figure 7 The steps shown are explained.

[0128] S501, obtaining the first blogger social relationship feature of the blogger followed by the target user.

[0129] In some embodiments, the calculation method of the social relationship characteristics corresponding to the blogger followed by the target user (i.e., the social relationship characteristics of the first blogger) is the same as the calculation method of the user social relationship characteristics in the above embodiment, and the interest recommendation device can obtain the social relationship characteristics of the first blogger based on this.

[0130] S502 , performing similarity calculation on the social relationship characteristics of the user and the social relationship characteristics of the first blogger, screening out a second blogger based on the calculation result, and determining the user information of the second blogger.

[0131] In some embodiments, after the push device obtains the user's social relationship characteristics and the first blogger's social relationship characteristics, it can perform a similarity calculation on the user's social relationship characteristics and the first blogger's social relationship characteristics, screen out the supplementary recalled blogger (i.e., the second blogger) based on the calculation results, and determine the user information of the supplementary recalled blogger.

[0132] In one example, the push device can perform cosine similarity calculation on the user's social relationship characteristics and the first blogger's social relationship characteristics. The calculation method is as shown in formula (2). After the calculated value of the cosine similarity result of the user's social relationship characteristics and the first blogger's social relationship characteristics is calculated according to formula (2), the bloggers followed by the target user can be sorted according to the calculated value of the above cosine similarity result. According to the preset sorting screening conditions, for example, the top 10 bloggers with high rankings are selected for recall.

[0133] Supplementary recall based on the user's social relationship characteristics and the primary blogger's social relationship characteristics can be used to retrieve bloggers with the most similar social attributes to the target user. Furthermore, these bloggers are more likely to be real-life acquaintances, colleagues, classmates, etc. This enriches the secondary bloggers in the supplementary recall and increases the social attributes of the recalled bloggers.

[0134] In some possible implementations, the target user's social relationship characteristics may include user social relationship characteristics of the target user and a blogger that the target user follows, and / or user social relationship characteristics of the target user and a blogger that the target user does not follow. When the user social relationship characteristics represent the social relationship between the target user and a blogger that the target user does not follow, the above S403 may include S601 to S602, see Figure 8 , Figure 8 This is an optional flow chart of the interest push method provided in the embodiment of the present application. Figure 8 The steps shown are explained.

[0135] S601, obtaining a second blogger social relationship feature of a blogger that the target user does not follow, wherein the blogger that the target user does not follow has a common following relationship with the target user.

[0136] In some embodiments, the social relationship characteristics corresponding to bloggers that the target user does not follow (i.e., the second blogger social relationship characteristics) are calculated in the same manner as the user social relationship characteristics in the above-mentioned embodiments, and the interest recommendation device can obtain the second blogger social relationship characteristics based on this. The blogger that the target user does not follow, corresponding to the second blogger social relationship characteristics obtained by the push device, is a blogger that has a common following relationship with the target user.

[0137] In one example, user A is the target user. User A's following relationships include user B and user C, while user D's following relationships include user B and user C. It can be seen that although user A does not follow user D, they share a common following relationship with user D, namely, user B and user C. At this point, the interest recommendation device can calculate user D's social relationship features (i.e., the second blogger's social relationship features) for subsequent supplementary recall.

[0138] S602: Calculate the similarity between the user's social relationship characteristics and the second blogger's social relationship characteristics, select the second blogger based on the calculation result, and determine the user information of the second blogger.

[0139] In some embodiments, after the push device obtains the user's social relationship characteristics and the second blogger's social relationship characteristics, it can perform a similarity calculation on the user's social relationship characteristics and the second blogger's social relationship characteristics, screen out the supplementary recalled blogger (i.e., the second blogger) based on the calculation results, and determine the user information of the supplementary recalled blogger.

[0140] In one example, the push device can perform cosine similarity calculation on the user's social relationship characteristics and the second blogger's social relationship characteristics, and the calculation method is as shown in formula (2). After calculating the cosine similarity result of the user's social relationship characteristics and the second blogger's social relationship characteristics according to formula (2), the bloggers that the target user does not follow can be sorted according to the calculated value of the cosine similarity result. According to the preset threshold, these bloggers that the target user does not follow but have a common attention relationship can be filtered. For example, if the preset threshold is 80, the push device can filter out bloggers whose calculated value of the cosine similarity result is greater than 80 and recall them.

[0141] By supplementing the user's social relationship characteristics with those of the second blogger, it's possible to recruit people, colleagues, or classmates whom the target user doesn't follow but may know in real life. Since some target users don't follow many colleagues or classmates, recruiting these bloggers based on the user's social relationship characteristics and those of the second blogger can increase the target user's follower and engagement rate.

[0142] In some possible implementations, in response to user social relationship characteristics used to represent the social relationship between a target user and a blogger followed by the target user and the social relationship between the target user and a blogger not followed by the target user, user information of a second blogger is obtained according to the user social relationship characteristics, including: obtaining a first blogger social relationship characteristic of a blogger followed by the target user and a second blogger social relationship characteristic of a blogger not followed by the target user, the blogger not followed by the target user having a common following relationship with the target user; performing similarity calculation on the user social relationship characteristic and the first blogger social relationship characteristic, and performing similarity calculation on the user social relationship characteristic and the second blogger social relationship characteristic, screening out the second blogger according to the calculation results, and determining the user information of the second blogger.

[0143] In some embodiments, the push device can obtain the social relationship characteristics corresponding to the blogger followed by the target user (i.e., the social relationship characteristics of the first blogger) in accordance with S501, and obtain the social relationship characteristics corresponding to the blogger not followed by the target user (i.e., the social relationship characteristics of the second blogger) in accordance with S601. Then, the push device calculates the similarity between the user's social relationship characteristics and the first blogger's social relationship characteristics, and calculates the similarity between the user's social relationship characteristics and the second blogger's social relationship characteristics. According to the calculation results, the blogger to be supplemented and recalled (i.e., the second blogger) is screened out, and the user information of the supplemented and recalled blogger is determined.

[0144] In some embodiments, the sorting process of S502 and S602 above can be implemented using a sorting model. By inputting user-side features, material-side features, blogger-side features, user social relationship features, and blogger social relationship features (including the first blogger social relationship features and the second blogger social relationship features) into the sorting model, the user information of the second blogger can be obtained.

[0145] In one example, assuming the sorting model is a DeepFM model, see Figure 9 , Figure 9 This is a schematic diagram of the sorting model in the embodiment of this application. Figure 9 As shown, user information, material information and blogger information are input into the DeepFM model, and the user-side features, material-side features, blogger-side features, user social relationship features, and blogger social relationship features are extracted through the feature extraction layer. The above user-side features, material-side features, blogger-side features, user social relationship features, and blogger social relationship features are respectively input into the embedding layer, and user-side embedding features, material-side embedding features, blogger-side embedding features, user social relationship embedding features, and blogger social relationship embedding features can be obtained. These embedding features are finally input into the DNN side of the DeepFM model. After processing on the DNN side, the user information of the second blogger recalled according to the user social relationship features can be obtained.

[0146] It should be noted that the ranking model used in the above embodiment is pre-trained. During the recall process, user information, material information, and blogger information can be directly input into the ranking model to directly obtain the user information of the second blogger. In this way, the ranking model can recall the second blogger based on the characteristics of the user's social relationship, providing the ranking model with fine-grained features, making the ranking model more suitable.

[0147] In some possible implementations, obtaining user information of a second blogger recalled by a target user based on user interest characteristics and social relationship characteristics includes: obtaining user information of a third blogger that meets preset screening conditions, a target user's interactive behavior sequence with respect to the third blogger, and a target user's user relationship sequence; determining user information of the second blogger recalled by the target user based on user interest characteristics based on the user information of the third blogger and the target user's interactive behavior sequence with respect to the third blogger, and determining user information of the second blogger recalled by the target user based on social relationship characteristics based on the target user's interactive behavior sequence with respect to the third blogger.

[0148] In some embodiments, the push device obtains user information of a third blogger that meets preset screening criteria in accordance with S201, obtains a target user's interactive behavior sequence with the third blogger in accordance with S202, and obtains a target user's user relationship sequence in accordance with S401. The push device then determines user information of a second blogger that the target user recalls based on user interest characteristics and user information of the second blogger that the target user recalls based on social relationship characteristics. The implementation process for the push device to determine the user information of the second blogger that the target user recalls based on user interest characteristics can refer to S202-S204, and the implementation process for the push device to determine the user information of the second blogger that the target user recalls based on social relationship characteristics can refer to S402-S403.

[0149] S103: Inputting at least the user information of the first blogger and the user information of the second blogger into a ranking model to obtain ranking results of the first blogger and the second blogger. The ranking results are used to reflect the target user's interest in the blogger.

[0150] In some embodiments, after obtaining the user information of the first blogger who has been recalled and the user information of the second blogger who has been supplemented in the recall, the push device can input at least the user information of the first blogger and the user information of the second blogger into the ranking model to obtain a ranking result based on the supplemented recall, which reflects the target user's interest level in the bloggers (including the first blogger and the second blogger).

[0151] In some possible implementations, the features input into the ranking model to obtain the ranking results of the first blogger and the second blogger may also include identity identification features between the target user and the second blogger. The identity identification features are obtained by inputting the user-side features, material-side features of the target user, blogger-side features of the second blogger, identity embedding features of the target user, and identity embedding features of the second blogger into the deep learning model to obtain the identity identification features between the target user and the second blogger.

[0152] In some embodiments, after the push device obtains the user information of the second blogger, it can further obtain the identity identification features between the target user and the second blogger through a deep learning model.

[0153] In one example, assuming the deep learning model is a FM model, see Figure 10 , Figure 10 This is a schematic diagram of the deep learning model in the embodiment of this application. Figure 10As shown, the user information of the target user, the user information of the second blogger and the material information are input into the FM model, and the user-side features, material-side features, blogger-side features of the second blogger, the identity features of the target user and the identity features of the second blogger are obtained through the feature extraction layer. The above features are input into the embedding layer to obtain the user-side embedded features, material-side embedded features, blogger-side embedded features of the second blogger, the identity embedded features of the target user and the identity embedded features of the second blogger. At this time, the user-side features, material-side features, blogger-side features of the second blogger, the identity features of the target user and the identity features of the second blogger are input into the FM first order of the FM model, and the user-side embedded features, material-side embedded features, blogger-side embedded features of the second blogger, the identity embedded features of the target user and the identity embedded features of the second blogger are input into the FM second order of the FM model. The output results of the FM first order, FM second order and FM Bias are then input into the next processing layer (such as the fusion layer (concat layer)), and finally the identity identification features between the target user and the second blogger output by the FM model can be obtained.

[0154] It should be noted that the above-mentioned deep learning model is a pre-trained model. During the training process, the user information of the sample user, the user information of the sample second blogger, and the sample material information (i.e., sample data) are input into the deep learning model. Since the sample data has a large number of features, the individual features are relatively sparse, and the frequency of occurrence is low, it cannot support complex model training. Therefore, during the training process, the features extracted from the sample data (i.e., sample user-side features, sample material-side features, blogger-side features of the sample second blogger, identity features of the sample target user, and identity features of the sample second blogger) and embedded features (such as sample user-side embedded features, sample material-side embedded features, blogger-side embedded features of the sample second blogger, identity embedded features of the sample target user, and identity embedded features of the sample second blogger) can be cross-featured and introduced into different processing sides, such as the first-order FM and second-order FM in the FM model.

[0155] In one example, the calculation expression (4) of the above FM model is as follows:

[0156]

[0157] Where ω0 is the FM Bias item preset in the FM model, represents the first-order term of FM, represents the FM second-order term.

[0158] Because the sparse matrix corresponding to the identity features between the target user and the second blogger requires a large amount of memory and cannot be directly stored in the deep learning model, and because the identity features between the target user and the second blogger update frequently, they cannot be pre-assembled into the existing recall relationship features. For these reasons, after the deep model training is completed, the identity features between the target user and the second blogger can be stored in a storage system (such as a remote dictionary server (Redis)) in the form of key-value pairs for real-time updating. During the sorting process, Redis can be accessed based on the required key value to obtain the corresponding identity features between the target user and the second blogger. Finally, all the identity features between the target user and the second blogger, the user information of the first blogger, and the user information of the second blogger are input into the sorting model to obtain the sorting results.

[0159] S104: Determine a target blogger based on the ranking result, and push the target blogger's blog post information to the target user.

[0160] In some embodiments, after obtaining the ranking result output by the ranking model, the push device can determine the target blogger according to the ranking result and output the blog post information of the target blogger to the target user.

[0161] The following describes an exemplary application of the embodiments of the present application in a practical application scenario.

[0162] When the above interest push method is applied to the social application software (application, APP) of the user terminal (such as Weibo), assuming that the deep learning model is the DeepFM model and the sorting model is the DeepFM model, the implementation process of the above interest push method applied to Weibo can be found in Figure 11 , Figure 11 This is a flow chart of the interest push method provided by the embodiment of the present application applied to social application software. Figure 11 The steps shown are explained.

[0163] S701 , the push device obtains user information of a first blogger recalled for user A (ie, target user) based on user behavior characteristics.

[0164] S702 , the push device inputs the user information of the fourth blogger whose popularity value meets the preset screening condition and interacts with the user A into the DeepFM model to obtain the blogger interest feature of the fourth blogger.

[0165] S703 , the push device performs weighted processing on the blogger interest features of the fourth blogger obtained in S702 to obtain user interest features of user A.

[0166] S704: The push device performs similarity calculation on the blogger interest feature and the user interest feature to obtain user information of the second blogger recalled based on the user interest feature of user A.

[0167] S705: The push device obtains the user relationship sequence of user A based on the following relationship of user A.

[0168] S706: The push device obtains user social relationship characteristics of user A according to the user relationship sequence of user A.

[0169] S707, the push device calculates the similarity between the user social relationship characteristics of user A and the first blogger social relationship characteristics of the blogger followed by user A, so as to obtain user information of a part of the second blogger recalled based on the social relationship characteristics of user A.

[0170] S708, the push device calculates the similarity between the user social relationship characteristics of user A and the second blogger social relationship characteristics of the blogger that user A does not follow, so as to obtain user information of another part of the second blogger recalled based on the social relationship characteristics of user A.

[0171] S709, the push device inputs the user-side embedded features of user A, the material-side embedded features, the blogger-side embedded features of the second blogger, the identity embedded features of user A, and the identity embedded features of the second blogger into the DeepFM model to obtain the identity identification features between user A and the second blogger.

[0172] S710 , the push device inputs the identity identification features between user A and the second blogger obtained in S709 , the user information of the first blogger obtained in S701 , and the user information of the second blogger into the DeepFM model to obtain a ranking result of the first blogger and the second blogger.

[0173] S711, the pushing device pushes the target blogger's blog information to user A according to the sorting result.

[0174] At this point, the above interest push method is completed.

[0175] The following continues to describe the exemplary structure of the interest push device 455 provided in the embodiment of the present application implemented as a software module. In some embodiments, such as Figure 2As shown, the software modules stored in the interest pushing device 455 of the memory 450 may include: a data acquisition module 4551, used to obtain user information of the first blogger recalled for the target user based on user behavior characteristics; obtain user information of the second blogger recalled for the target user based on user interest characteristics and / or social relationship characteristics; a model calculation module 4552, used to input at least the user information of the first blogger and the user information of the second blogger into the sorting model to obtain the sorting results of the first blogger and the second blogger, and the sorting results are used to reflect the target user's interest level in the blogger; a push module 4553, used to determine the target blogger according to the sorting results, and push the target blogger's blog information to the target user.

[0176] In some possible implementations, the data acquisition module 4551 is further used to obtain user information of a third blogger that meets preset screening conditions; obtain a target user's interactive behavior sequence with respect to the third blogger, where the interactive behavior sequence is used to indicate the order of the target user's interactive behaviors with respect to different third bloggers; determine user information of a fourth blogger based on the target user's interactive behavior sequence with respect to the third blogger, where the fourth blogger is used to indicate a blogger among the third bloggers who has interactive behaviors with the target user; and obtain user information of a second blogger recalled for the target user based on user interest characteristics based on the user information of the fourth blogger, where the second blogger is at least a portion of the fourth bloggers.

[0177] In some possible implementations, the model calculation module 4552 is further used to input the user information of each fourth blogger into the deep learning model to obtain the blogger interest characteristics of each fourth blogger; obtain a weighted processing result by weighting the blogger interest characteristics of each fourth blogger, and use the weighted processing result as the user interest characteristics of the target user; calculate the similarity between the blogger interest characteristics and the user interest characteristics, screen out the second blogger according to the calculation result, and determine the user information of the second blogger.

[0178] In some possible implementations, the apparatus further includes a popularity value calculation module; the popularity value calculation module is configured to divide the quantity of materials published by each blogger within a preset time period by the number of exposures of each blogger within the preset time period, and multiply the result by the interaction value of each blogger within the preset time period to obtain the popularity value of each blogger, where the popularity value is used to indicate the degree to which the materials published by each blogger in the corresponding field are liked by users; based on preset screening conditions, a third blogger among each blogger that meets the preset screening conditions is screened, and user information of the third blogger is determined.

[0179] In some possible implementations, the data acquisition module 4551 is also used to obtain a user relationship sequence of the target user based on the target user's follow-up relationship; extract user social relationship features of the target user based on the user relationship sequence; the user social relationship features are used to represent the social relationship between the target user and the blogger the target user follows, and / or the social relationship between the target user and the blogger the target user does not follow; and obtain user information of the second blogger based on the user social relationship features.

[0180] In some possible implementations, in response to the user social relationship characteristics being used to represent the social relationship between the target user and the blogger the target user follows; the data acquisition module 4551 is further used to obtain the first blogger social relationship characteristics of the blogger the target user follows; perform similarity calculation on the user social relationship characteristics and the first blogger social relationship characteristics, screen out the second blogger based on the calculation results, and determine the user information of the second blogger.

[0181] In some possible implementations, in response to the user social relationship characteristics being used to represent the social relationship between the target user and a blogger that the target user does not follow; the data acquisition module 4551 is further used to obtain the social relationship characteristics of a second blogger that the target user does not follow, and the blogger that the target user does not follow has a common follow-up relationship with the target user; a similarity calculation is performed on the user social relationship characteristics and the second blogger social relationship characteristics, the second blogger is screened out based on the calculation results, and the user information of the second blogger is determined.

[0182] In some possible implementations, in response to user social relationship characteristics being used to represent the social relationship between a target user and a blogger followed by the target user and the social relationship between the target user and a blogger not followed by the target user; the data acquisition module 4551 is further used to obtain a first blogger social relationship characteristic of a blogger followed by the target user and a second blogger social relationship characteristic of a blogger not followed by the target user, where the blogger not followed by the target user has a common following relationship with the target user; similarity calculation is performed on the user social relationship characteristic and the first blogger social relationship characteristic, and similarity calculation is performed on the user social relationship characteristic and the second blogger social relationship characteristic, a second blogger is screened out according to the calculation results, and user information of the second blogger is determined.

[0183] In some possible implementations, the data acquisition module 4551 is further used to obtain user information of a third blogger that meets preset screening conditions, a target user's interactive behavior sequence with respect to the third blogger, and a target user's user relationship sequence; based on the user information of the third blogger and the target user's interactive behavior sequence with respect to the third blogger, determine the user information of the second blogger that the target user recalls based on user interest characteristics; based on the target user's interactive behavior sequence with respect to the third blogger, determine the user information of the second blogger that the target user recalls based on social relationship characteristics.

[0184] In some possible implementations, the model calculation module 4552 is also used to input the user-side embedded features of the target user, the material-side embedded features, the blogger-side embedded features of the second blogger, the identity embedded features of the target user, and the identity embedded features of the second blogger into the deep learning model to obtain the identity identification features between the target user and the second blogger.

[0185] In some possible implementations, the model calculation module 4552 is further configured to input the identity identification features between the target user and the second blogger, the user information of the first blogger, and the user information of the second blogger into the ranking model to obtain ranking results for the first blogger and the second blogger.

[0186] The present application provides a computer program product or computer program, which 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 interest push method described above in the present application.

[0187] The embodiment of the present application provides a computer-readable storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will execute the interest push method provided by the embodiment of the present application, for example, Figure 3 The interest push method shown.

[0188] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or various devices including one or any combination of the above memories.

[0189] In some embodiments, executable instructions may be in 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 stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0190] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0191] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0192] To sum up, the embodiment of the present application can enrich the diversity of recalled materials by recalling the second blogger. Sorting based on the existing recalled first blogger and the supplementary recalled second blogger can improve the diversity of materials in the push scenario, allowing target users to receive more different types of pushes, thereby improving the target users' usage experience.

[0193] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A method for pushing interest, characterized in that: The method comprises: Obtain user information of the first blogger recalled for the target user based on user behavior characteristics; Obtaining user information of a second blogger recalled for the target user based on user interest characteristics and / or social relationship characteristics; Inputting at least the user information of the first blogger and the user information of the second blogger into a ranking model to obtain ranking results of the first blogger and the second blogger, wherein the ranking results are used to reflect the target user's interest in the blogger; Determine a target blogger based on the ranking result, and push the target blogger's blog information to the target user; The obtaining of user information of a second blogger recalled by the target user based on user interest characteristics includes: obtaining user information of a third blogger that meets a preset screening condition; obtaining a sequence of interactive behaviors of the target user with respect to the third blogger, wherein the interactive behavior sequence is used to indicate a sequence of interactive behaviors of the target user with respect to different third bloggers; determining user information of a fourth blogger based on the sequence of interactive behaviors of the target user with respect to the third blogger, wherein the fourth blogger is used to indicate a blogger among the third bloggers who has interactive behaviors with the target user; and obtaining user information of a second blogger recalled by the target user based on user interest characteristics based on the user information of the fourth blogger, wherein the second blogger is at least a part of the fourth bloggers.

2. The interest push method according to claim 1, characterized in that: The step of obtaining, based on the user information of the fourth blogger, the user information of the second blogger recalled for the target user based on the user interest characteristics, includes: Inputting the user information of each of the fourth bloggers into a deep learning model to obtain blogger interest features of each of the fourth bloggers; performing weighted processing on the blogger interest features of each of the fourth bloggers to obtain a weighted processing result, and using the weighted processing result as the user interest feature of the target user; A similarity calculation is performed between the blogger interest feature of each of the fourth bloggers and the user interest feature, the second blogger is screened out according to the calculation result, and the user information of the second blogger is determined.

3. The interest push method according to claim 1, characterized in that: The obtaining of user information of third bloggers meeting the preset screening conditions includes: Divide the number of materials published by each blogger in a preset time period by the number of exposures of each blogger in the preset time period, and multiply the result by the interaction value of each blogger in the preset time period to obtain the popularity value of each blogger. The popularity value is used to indicate the degree to which the materials published by each blogger in the corresponding field are popular among users; Based on the preset screening condition, the third blogger that meets the preset screening condition among each blogger is screened, and user information of the third blogger is determined.

4. The interest push method according to claim 1, characterized in that: The obtaining of user information of a second blogger recalled by the target user based on the social relationship feature includes: Based on the attention relationship of the target user, obtaining a user relationship sequence of the target user; Extracting user social relationship features of the target user based on the user relationship sequence; the user social relationship features are used to represent the social relationship between the target user and the bloggers followed by the target user, and / or the social relationship between the target user and the bloggers not followed by the target user; The user information of the second blogger is obtained according to the user social relationship characteristics.

5. The interest push method according to claim 4, characterized in that: In response to the user social relationship feature being used to represent the social relationship between the target user and a blogger followed by the target user, obtaining user information of the second blogger according to the user social relationship feature includes: Obtaining first blogger social relationship characteristics of bloggers followed by the target user; A similarity calculation is performed on the user social relationship feature and the first blogger social relationship feature, the second blogger is screened out according to the calculation result, and the user information of the second blogger is determined.

6. The interest push method according to claim 4, characterized in that: In response to the user social relationship feature being used to represent a social relationship between the target user and a blogger that the target user does not follow, obtaining user information of the second blogger according to the user social relationship feature includes: Obtaining a second blogger social relationship feature of a blogger that the target user does not follow, where the blogger that the target user does not follow has a common following relationship with the target user; A similarity calculation is performed on the user social relationship feature and the second blogger social relationship feature, the second blogger is screened out according to the calculation result, and the user information of the second blogger is determined.

7. The interest push method according to claim 4, characterized in that: In response to the user social relationship feature being used to represent a social relationship between the target user and a blogger followed by the target user and a social relationship between the target user and a blogger not followed by the target user, obtaining user information of the second blogger according to the user social relationship feature includes: Obtaining a first blogger social relationship feature of a blogger followed by the target user and a second blogger social relationship feature of a blogger not followed by the target user, wherein the blogger not followed by the target user has a common following relationship with the target user; A similarity calculation is performed on the user social relationship feature and the first blogger social relationship feature, and a similarity calculation is performed on the user social relationship feature and the second blogger social relationship feature, the second blogger is screened out according to the calculation result, and the user information of the second blogger is determined.

8. The interest push method according to claim 1, characterized in that: The obtaining of user information of a second blogger recalled for the target user based on user interest characteristics and social relationship characteristics includes: Obtaining user information of a third blogger that meets preset screening conditions, a sequence of interactive behaviors of the target user with respect to the third blogger, and a sequence of user relationships of the target user; Based on the user information of the third blogger and the target user's interactive behavior sequence with respect to the third blogger, the user information of the second blogger recalled by the target user based on the user interest characteristics is determined; and based on the target user's interactive behavior sequence with respect to the third blogger, the user information of the second blogger recalled by the target user based on the social relationship characteristics is determined.

9. The interest push method according to claim 1, characterized in that: The method further comprises: The user-side embedded features, material-side embedded features of the target user, the blogger-side embedded features of the second blogger, the identity embedded features of the target user, and the identity embedded features of the second blogger are input into a deep learning model to obtain the identity identification features between the target user and the second blogger; the identity identification features are used to characterize the association relationship between the target user and the second blogger.

10. The interest push method according to claim 9, characterized in that: The method further includes: inputting the identity identification features between the target user and the second blogger, the user information of the first blogger, and the user information of the second blogger into the ranking model to obtain ranking results of the first blogger and the second blogger.

11. An interest pushing device, characterized in that: The device comprises: A data collection module is configured to obtain user information of a first blogger recalled for a target user based on user behavior characteristics; and obtain user information of a second blogger recalled for the target user based on user interest characteristics and / or social relationship characteristics; a model calculation module, configured to input at least the user information of the first blogger and the user information of the second blogger into a ranking model to obtain ranking results of the first blogger and the second blogger, wherein the ranking results are used to reflect the target user's interest in the blogger; A push module is used to determine a target blogger according to the ranking result and push the blog post information of the target blogger to the target user; The data collection module is further used to obtain user information of third bloggers that meet preset screening conditions; obtain a target user's interactive behavior sequence with respect to the third blogger, where the interactive behavior sequence is used to indicate the order of the target user's interactive behaviors with respect to different third bloggers; determine user information of a fourth blogger based on the target user's interactive behavior sequence with respect to the third blogger, where the fourth blogger is used to indicate a blogger among the third bloggers who has interactive behaviors with the target user; and obtain user information of a second blogger recalled for the target user based on user interest characteristics based on the user information of the fourth blogger, where the second blogger is at least a portion of the fourth bloggers.

12. An electronic device, characterized in that: The electronic device comprises: a memory for storing executable instructions; A processor, configured to implement the method according to any one of claims 1 to 10 when executing the executable instructions or computer program stored in the memory.

13. A computer-readable storage medium storing executable instructions or a computer program, characterized in that: When the executable instructions are executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images