Artificial intelligence-based content recommendation method and apparatus

Abstract

The application provides an artificial intelligence-based content recommendation method and device, an electronic device, and a computer readable storage medium. The method comprises: in response to a touch operation of a content presentation interface, obtaining historical operation behaviors of a user account logged into the content presentation interface; determining an account attribute of the user account according to the historical operation behaviors; obtaining account preference behaviors conforming to the account attribute, and performing preference analysis and processing according to the account preference behaviors to obtain preference information of the user account for multiple contents in a content library; determining a content recommendation strategy according to the account attribute and the preference information, and screening out to-be-recommended contents from the content library according to the content recommendation strategy; and updating the content presentation interface according to the screened out to-be-recommended contents. Through the application, accurate content recommendation can be achieved, and the actual utilization rate of the computing resources consumed by the electronic device when performing content recommendation can be improved.

Description

Technical Field

[0001] This application relates to artificial intelligence technology and cloud technology, and in particular to an artificial intelligence-based content recommendation method, apparatus, electronic device and computer-readable storage medium. Background Technology

[0002] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence.

[0003] Content recommendation is an important application of artificial intelligence, aiming to recommend content that users are likely to be interested in. Current solutions typically recommend content to multiple user accounts based on a pre-defined, uniform strategy. However, this approach lacks specificity for different user accounts, resulting in low accuracy and low utilization of the computing resources consumed by electronic devices for content recommendation. Summary of the Invention

[0004] This application provides an artificial intelligence-based content recommendation method, apparatus, electronic device, and computer-readable storage medium, which can improve the accuracy of content recommendation and increase the actual utilization rate of computing resources consumed by electronic devices for content recommendation.

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

[0006] This application provides an artificial intelligence-based content recommendation method, including:

[0007] In response to touch operations on the content presentation interface, the historical operation behavior of the user account logged into the content presentation interface is obtained;

[0008] The account attributes of the user account are determined based on the historical operation behavior;

[0009] Obtain account preference behaviors that match the account attributes, and perform preference analysis processing based on the account preference behaviors to obtain the user account's preference information for multiple contents in the content library;

[0010] A content recommendation strategy is determined based on the account attributes and the preference information, and content to be recommended is filtered from the content library according to the content recommendation strategy.

[0011] The content presentation interface is updated based on the filtered content to be recommended.

[0012] This application provides an artificial intelligence-based content recommendation device, including:

[0013] The acquisition module is used to acquire the historical operation behavior of the user account logged into the content presentation interface in response to touch operation of the content presentation interface;

[0014] An attribute determination module is used to determine the account attributes of the user account based on the historical operation behavior.

[0015] The preference determination module is used to obtain account preference behaviors that match the account attributes, and perform preference analysis processing based on the account preference behaviors to obtain the user account's preference information for multiple contents in the content library.

[0016] The filtering module is used to determine a content recommendation strategy based on the account attributes and the preference information, and to filter out content to be recommended from the content library according to the content recommendation strategy.

[0017] The update module is used to update the content presentation interface based on the filtered content to be recommended.

[0018] This application provides an electronic device, including:

[0019] Memory, used to store executable instructions;

[0020] When the processor executes the executable instructions stored in the memory, it implements the AI-based content recommendation method provided in the embodiments of this application.

[0021] This application provides a computer-readable storage medium storing executable instructions for inducing a processor to execute and implement the AI-based content recommendation method provided in this application.

[0022] The embodiments of this application have the following beneficial effects:

[0023] Account attributes are determined based on the user's historical activity, and then preference information is determined based on the acquired account preference behaviors that match the account attributes. Furthermore, a content recommendation strategy is determined based on the account attributes and preference information, and content to be recommended is filtered from the content library according to the content recommendation strategy. In this way, suitable content recommendation strategies can be determined for different account attributes and different preference information. Through the embodiments of this application, the targeting and accuracy of content recommendations can be improved, while simultaneously increasing the actual utilization rate of computing resources consumed by electronic devices when performing content recommendations. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the architecture of an AI-based content recommendation system provided in an embodiment of this application;

[0025] Figure 2 This is a schematic diagram of the architecture of the terminal device provided in the embodiments of this application;

[0026] Figure 3A This is a flowchart illustrating the content recommendation method based on artificial intelligence provided in an embodiment of this application;

[0027] Figure 3B This is a flowchart illustrating the content recommendation method based on artificial intelligence provided in an embodiment of this application;

[0028] Figure 3C This is a flowchart illustrating the filtering process based on the content filtering strategy corresponding to the first account attribute provided in this application embodiment;

[0029] Figure 3D This is a flowchart illustrating the content recommendation method based on artificial intelligence provided in an embodiment of this application;

[0030] Figure 4 This is a flowchart illustrating the video recommendation method based on artificial intelligence provided in an embodiment of this application;

[0031] Figure 5 This is a schematic diagram of the selection interface for teams of interest provided in an embodiment of this application;

[0032] Figure 6 This is a schematic diagram of the video interface of the basketball league provided in an embodiment of this application;

[0033] Figure 7 This is a schematic diagram illustrating the use of an artificial intelligence model for predictive processing provided in an embodiment of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0036] In the following description, the terms "first," "second," and "third" are used merely to distinguish similar objects and do not represent a specific ordering of the objects. It is understood that "first," "second," and "third" may be interchanged in a specific order or sequence where permissible, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein. In the following description, the term "multiple" refers to at least two.

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0038] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0039] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0040] 1) Content: The embodiments of this application do not limit the form of the content. For example, it can be text, sound or image, or it can be multimedia content that includes at least one of the media forms of text, sound and image, such as video.

[0041] 2) User account: The account a user uses to view content. For example, for an application used to present content (such as a video application), the user account can be the account that is logged in within that application.

[0042] 3) Operation behavior: refers to the user account's operation behavior on the presented content. Here, there is no limitation on the type of operation behavior. For example, operation behavior can be the behavior of playing, liking or collecting the presented content.

[0043] 4) Account Attributes: These are determined by the user account's historical activity (i.e., historical behavior) and are abstract properties of the user account used to distinguish different user accounts. Depending on the actual application scenario, two or more account attributes can be set; for example, account attributes can include new accounts and old accounts.

[0044] 5) Account Preference Behavior: This refers to the actions performed by a user account that indicate the user's preferences. Different account attributes can correspond to different account preference behaviors. Based on account preference behaviors, the user account's preference information for multiple contents in the content library can be determined. This preference information directly reflects the user account's (or the user using the user account's) preferences.

[0045] 6) Content Recommendation Strategy: A strategy used to determine the content to be recommended in the content library. In this embodiment, different account attributes can correspond to different content recommendation strategies; for the same account attribute, different preference information can also correspond to different content recommendation strategies; different account attributes and different preference information can also share a single content recommendation strategy. Furthermore, the content recommendation strategy can include only a content filtering strategy, or it can include both a content replacement strategy and a content filtering strategy.

[0046] 7) Interactive Events: These are events involving multiple participants. For example, an interactive event could be a league match (such as a basketball league or a football league), with the participating teams being the participants. Interactive events can also be content-themed events, used to guide the uploading of content related to a specific theme (such as a person or game), with the participants being content upload accounts, such as uploaders (UPs) in video applications. An interactive event can correspond to multiple pieces of content, which can be further subdivided to correspond to different objects. It's worth noting that the same content may correspond to multiple objects; for example, a video of a match between team A and team B would correspond to both team A and team B. Furthermore, interactive events can include multiple phases; for example, a league match could include the preseason, regular season, and playoff phases.

[0047] 8) Tags: In this embodiment, the tags of the content are used to indicate the type of the content. A piece of content may include at least one tag. For example, the tags of the content may include the object corresponding to the content, or a certain stage in the interactive event. For example, for a video named "xx media predicts that team 1 (the team that enters the finals) is the champion", its tags may include champion prediction, finals (i.e., the final stage in the league event), and team 1 (team 1 is the object corresponding to the video). The tags included in the content can be obtained by manual annotation or by Natural Language Processing (NLP) analysis. For example, the title of the above video "xx media predicts that team 1 (the team that enters the finals) is the champion" includes the keyword "team 1". Therefore, "team 1" is used as the tag of the content, where the keyword can be preset.

[0048] 9) Machine Learning (ML): This field specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning typically includes techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning. In the embodiments of this application, artificial intelligence models can be constructed based on machine learning principles.

[0049] 10) Database: Similar to an electronic filing cabinet, it is a place to store electronic files, where users can perform operations such as adding, querying, updating, and deleting data. A database can also be understood as a collection of data stored together in a certain way, shareable by multiple users, with minimal redundancy, and independent of applications. In this embodiment, the content library can be constructed based on database principles, storing multiple pieces of content.

[0050] 11) Big Data: Refers to data sets that cannot be captured, managed, and processed within a certain timeframe using conventional software tools. It represents massive, rapidly growing, and diverse information assets that require new processing models to achieve stronger decision-making, insight discovery, and process optimization capabilities. Technologies applicable to big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet, and scalable storage systems. In the embodiments of this application, big data technologies can be used to process large volumes of content and to train artificial intelligence models, among other things.

[0051] This application provides an artificial intelligence-based content recommendation method, apparatus, electronic device, and computer-readable storage medium, which can improve the accuracy of content recommendation and simultaneously increase the actual utilization rate of computing resources consumed by the electronic device for content recommendation. The following describes exemplary applications of the electronic device provided in this application. The electronic device provided in this application can be implemented as various types of terminal devices or as a server.

[0052] See Figure 1 , Figure 1 This is a schematic diagram of the architecture of an AI-based content recommendation system 100 provided in this application embodiment. The terminal device 400 is connected to the server 200 through the network 300, and the server 200 is connected to the database 500. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0053] In some embodiments, taking the electronic device as a terminal device as an example, the AI-based content recommendation method provided in this application can be implemented by the terminal device. For example, the terminal device 400 runs a client 410, which displays a content presentation interface in the graphical interface of the terminal device 400. When the client 410 receives a touch operation on the content presentation interface, it obtains the historical operation behavior of the user account that logged into the content presentation interface (i.e., the client 410). The client 410 can obtain the historical operation behavior of the user account from local storage (such as cache) or by accessing the server 200 to obtain the historical operation behavior of the user account from the database 500. Then, the client 410 determines the account attributes of the user account based on the historical operation behavior, obtains the account preference behavior that matches the account attributes, and determines the user account's preference information for multiple contents based on the account preference behavior. The content library can be the database 500, which is used to store multiple contents (the database 500 can also store historical operation behavior); the content library can also be located locally on the client 410, using the storage resources of the terminal device 400 to store multiple contents.

[0054] Client 410 can determine a content recommendation strategy based on the obtained account attributes and preference information, and filter out content to be recommended from multiple contents in the content library according to the content recommendation strategy. It then updates the content presentation interface based on the content to be recommended, that is, presents the content to be recommended in the content presentation interface. The content recommendation strategy can be pre-stored locally on client 410 or stored in database 500. In this embodiment, client 410 can be an offline client, used to recommend content pre-stored in the content library on client 410's local machine (i.e., offline content); or it can be an online client, used to recommend content stored in database 500 (i.e., online content) to user accounts.

[0055] In some embodiments, taking the electronic device as a server as an example, the AI-based content recommendation method provided in this application can also be implemented collaboratively by the terminal device and the server. For example, when the client 410 receives a touch operation on the content presentation interface, it notifies the server 200 to retrieve the user account's historical operation behavior from the database 500. The server 200 can determine the user account's account attributes based on the historical operation behavior and send the account attributes to the client 410, so that the client 410 can determine the method for obtaining account preference behavior based on the account attributes; or, the server 200 can also directly determine the method for obtaining account preference behavior based on the account attributes and send the corresponding data (e.g., data used to construct the interface for obtaining account preference behavior) to the client 410, so that the client 410 can obtain the account preference behavior; or, the server 200 can also directly obtain account preference behavior that matches the account attributes from the database 500. Wherein, if the client 410 obtains the account preference behavior, the client 410 can send the obtained account preference behavior to the server 200.

[0056] Server 200 determines the user account's preference information based on the obtained account preference behavior, and then determines a content recommendation strategy based on the account attributes and preference information. Next, server 200 selects content to be recommended from multiple content items stored in database 500 according to the content recommendation strategy, and sends the recommended content to client 410 for presentation.

[0057] exist Figure 1 The example provided illustrates the scenario where server 200 retrieves historical operation behavior from database 500 and account preference behavior from client 410. Additionally, it shows a piece of content to be recommended (e.g., a video) presented in the content presentation interface. It is worth noting that the number of pieces of content to be recommended in this embodiment can be one or more, and there is no limitation on this.

[0058] In some embodiments, the terminal device 400 or server 200 can implement the AI-based content recommendation method provided in this application by running a computer program. For example, the computer program can be a native program or software module in an operating system; it can be a native application (APP), i.e., a program that needs to be installed in the operating system to run, such as a video application (corresponding to client 410 above); it can also be a mini-program, i.e., a program that only needs to be downloaded to a browser environment to run; or it can be a mini-program that can be embedded in any APP, such as a mini-program component embedded in a video application, wherein the mini-program component can be turned on or off by the user. In summary, the above-mentioned computer program can be any form of application, module, or plugin.

[0059] In some embodiments, server 200 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The cloud service can be a content recommendation service, which can be invoked by terminal device 400. Cloud technology refers to a hosting technology that unifies hardware, software, network, and other resources within a wide area network or local area network to achieve data computation, storage, processing, and sharing. Terminal device 400 can be a smartphone, tablet, laptop, desktop computer, smart TV, smartwatch, etc., but is not limited to these. Terminal devices and servers can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment.

[0060] Taking the example of a terminal device provided in this application embodiment, it can be understood that in the case where the electronic device is a server, Figure 2 Some parts of the structure shown (such as the user interface, presentation module, and input processing module) can be omitted. See also Figure 2 , Figure 2 This is a schematic diagram of the structure of the terminal device 400 provided in the embodiments of this application. Figure 2 The terminal 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 terminal 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to implement communication between these components. In addition to a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 440.

[0061] 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. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0062] User interface 430 includes one or more output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. 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.

[0063] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices physically located away from the processor 410.

[0064] The memory 450 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 450 described in this application embodiment is intended to include any suitable type of memory.

[0065] In some embodiments, memory 450 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0066] Operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0067] The network communication module 452 is used to reach other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0068] Presentation module 453 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 associated with user interface 430 (e.g., a display screen, a speaker, etc.).

[0069] The input processing module 454 is used to detect and translate one or more user inputs or interactions from one or more input devices 432.

[0070] In some embodiments, the apparatus provided in this application can be implemented in software. Figure 2 An AI-based content recommendation device 455, stored in memory 450, is shown. This device can be software in the form of programs and plugins, and includes the following software modules: an acquisition module 4551, an attribute determination module 4552, a preference determination module 4553, a filtering module 4554, and an update module 4555. These modules are logically connected and can therefore be arbitrarily combined or further divided according to their implemented functions. The functions of each module will be described below.

[0071] The AI-based content recommendation method provided in this application will be described by referring to exemplary applications and implementations of the electronic devices provided in the embodiments of this application.

[0072] See Figure 3A , Figure 3A This is a flowchart illustrating an AI-based content recommendation method provided in an embodiment of this application, which will be combined with... Figure 3A The steps shown are explained.

[0073] In step 101, in response to touch operations on the content presentation interface, the historical operation behavior of the user account logged into the content presentation interface is obtained.

[0074] Here, the content presentation interface can be displayed first, for example, in an application used for recommending content (such as a video application). Upon receiving a touch operation on the content presentation interface, content recommendation is performed. During the content recommendation process, the historical operation behavior of the user account logged into the content presentation interface is first retrieved. The touch operation can be a click or a long press, etc., without limitation. It is worth noting that if the content presentation interface is displayed in an application used for recommending content, the user account logged into the content presentation interface can refer to a user account that is currently logged in within that application.

[0075] Historical user activity data can be retrieved from local storage (such as the local cache of the application used for content recommendations) or a database; that is, there is no restriction on the method of obtaining historical user activity data. Each historical user activity corresponds to a piece of viewed content, that is, content that has been viewed by a user with the same account in the past.

[0076] In step 102, the account attributes of the user account are determined based on historical operation behavior.

[0077] For example, multiple account attributes can be pre-defined, with different account attributes corresponding to different ranges of historical operation behaviors. After obtaining the user account's historical operation behaviors, the range of behaviors that successfully match the obtained historical operation behaviors is determined, and the account attribute corresponding to this range is used as the user account's account attribute.

[0078] In step 103, account preference behaviors that match the account attributes are obtained, and preference analysis is performed based on the account preference behaviors to obtain the user account's preference information for multiple contents in the content library.

[0079] In this embodiment, different account attributes can correspond to different account preference behaviors, meaning the methods for obtaining account preference behaviors differ. The specific methods can be set according to the actual application scenario. For example, one account attribute might correspond to the behavior of selecting an object of interest, while another account attribute might correspond to historical operations on previously presented content. After determining the user account's attributes in step 102, account preference behaviors matching those attributes can be obtained. These preferences can then be analyzed to obtain preference information, which reflects the user account's preferences for multiple pieces of content in the content library.

[0080] In step 104, a content recommendation strategy is determined based on account attributes and preference information, and content to be recommended is selected from the content library according to the content recommendation strategy.

[0081] In this application embodiment, different account attributes can correspond to different content recommendation strategies. For the same account attribute, different determined preference information can also correspond to different content recommendation strategies, thereby improving the applicability to different situations, such as the applicability to the cold start scenario of content recommendation. The cold start scenario refers to a scenario where the user account's historical operation behavior is not obtained, or the number of historical operation behaviors obtained is small.

[0082] After determining the user account's attributes and preferences, a content recommendation strategy corresponding to both the account attributes and the preferences can be established. Then, based on this strategy, content to be recommended is filtered from multiple content items in the content library. It's worth noting that the number of content recommendation strategies determined here can be one or more. For example, if the content presentation interface includes multiple content sections, a corresponding content recommendation strategy can be determined for each section. Furthermore, there is no limit to the number of selected content items to be recommended.

[0083] This application does not limit the application method of the content recommendation strategy. For example, it can filter multiple pieces of content in the content library to obtain content to be recommended, including set tags. For example, if the set tag is an object participating in an interactive event, then the content to be recommended is the content corresponding to that object. Alternatively, it can filter multiple pieces of content in the content library according to content filtering parameters to obtain content to be recommended that meets a set quantity ratio. The content filtering parameters include at least one of content popularity and update time. That is, the purpose of the filtering process is to select several pieces of content with the highest popularity and / or the most recent update time. Popularity can be at least one of views, likes, and favorites. The quantity ratio can be the quantity ratio between content with different tags.

[0084] In some embodiments, the method further includes: responding to touch operations on the content presentation interface to obtain content recommendation strategies shared by different account attributes and different preference information; the above-mentioned filtering of content to be recommended from the content library based on the content recommendation strategy can be achieved in such a way that the content to be recommended is filtered from the content library according to the content recommendation strategy determined by the account attributes and preference information, and the shared content recommendation strategy.

[0085] In this embodiment, different account attributes and different preference information can also share content recommendation strategies. When a touch operation is received on the content presentation interface, the shared content recommendation strategy can be obtained. The number of shared content recommendation strategies can be one or more. Thus, the content recommendation strategies used to filter content to be recommended include two types: one is the content recommendation strategy determined based on account attributes and preference information, and the other is the shared content recommendation strategy.

[0086] Then, based on the content recommendation strategy determined by account attributes and preference information, content to be recommended is selected from multiple contents in the content library. For ease of distinction, this selected content is named the first category of recommended content. Simultaneously, based on the shared content recommendation strategy, content to be recommended is selected from multiple contents in the content library; for ease of distinction, this selected content is named the second category of recommended content. Then, both the first and second category of recommended content are used as the recommended content to be presented in step 105. This method improves the flexibility of content recommendation.

[0087] In step 105, the content presentation interface is updated based on the selected content to be recommended.

[0088] After selecting the content to be recommended based on the content recommendation strategy, the content to be recommended can be displayed on the content presentation interface to update the content presentation interface. This application embodiment does not limit the presentation method of the content to be recommended; for example, it can be in table format or parallel format.

[0089] In some embodiments, when the determined content recommendation strategy includes content recommendation strategies corresponding to multiple content sections, the above-mentioned updating of the content presentation interface based on the selected content to be recommended can be achieved in the following way: For each content section, perform the following processing: In the content section of the content presentation interface, present the content to be recommended selected according to the content recommendation strategy corresponding to the content section.

[0090] In this embodiment of the application, the content presentation interface may include multiple content sections, and the content recommendation strategy determined in step 104 includes content recommendation strategies corresponding to each of the multiple content sections. In this case, for each content section, content to be recommended can be filtered from the content library according to the content recommendation strategy corresponding to the content section, and the content to be recommended can be presented in that content section.

[0091] It's worth noting that content recommendation strategies can be set based on the characteristics of content sections. For example, if a content section's characteristic is to only display content containing defined tags (such as an object participating in an interactive event), then the corresponding content recommendation strategy could be to filter out content from the content library that includes that tag. Furthermore, when shared content recommendation strategies exist, these strategies can correspond to one or more content sections. Through these methods, the recommended content can be displayed in sections, making it easier for users to find content of interest within sections of interest, thus further improving the effectiveness of content recommendations.

[0092] like Figure 3A As shown, this embodiment of the application determines the account attributes of a user account based on its historical operational behavior, and then determines a content recommendation strategy based on the account attributes and the account preference behaviors that match those attributes. This effectively improves the fit between the determined content recommendation strategy and the user account, thus enhancing the targeting of the user account. Through this embodiment, the accuracy of content recommendation can be effectively improved, and the computing resources consumed by electronic devices when performing content recommendation can be utilized more efficiently.

[0093] In some embodiments, see Figure 3B , Figure 3B This is a flowchart illustrating an AI-based content recommendation method provided in an embodiment of this application. Figure 3AThe step 102 shown can be updated to step 201, in which when the number of historical operation behaviors is greater than the number threshold, the user account's account attribute is determined to be the first account attribute.

[0094] Here, when the number of historical operation behaviors of a user account obtained through step 101 exceeds the quantity threshold, the user account's account attribute is determined to be the first account attribute, meaning the user account is an old account. The quantity threshold can be set according to the actual application scenario, for example, set to 100, or for example, set to 0.

[0095] exist Figure 3B middle, Figure 3A Step 103 shown can be implemented through steps 202 to 205, which will be explained in conjunction with each step.

[0096] In step 202, the user account's historical actions regarding multiple pieces of content previously presented are taken as account preference behaviors that conform to the first account attribute.

[0097] For the first account attribute, the purpose of content recommendation in this application embodiment is to improve the responsiveness and accuracy of content replacement. Here, the user account's historical operation behavior for multiple pieces of content previously presented is taken as the account preference behavior that conforms to the first account attribute.

[0098] In step 203, the viewing content corresponding to the historical operation behavior is determined from the multiple contents presented in the previous instance.

[0099] To facilitate differentiation, the content corresponding to historical user actions is named "viewed content." Among the multiple pieces of content previously presented, the viewed content corresponding to the historical user actions is identified. Here, historical user actions refer to the user account's past actions related to the multiple pieces of content previously presented. The number of viewed content items can be zero or at least one.

[0100] In step 204, the identification information of the viewed content is used as the user account's preference information.

[0101] In some embodiments, after step 201, the method further includes: determining the difference duration between the presentation time of the last presentation of multiple contents and the real-time time; when the difference duration is greater than the duration threshold, determining that the user account's preference information is empty.

[0102] Here, the difference in duration between the presentation time of the last presentation of multiple pieces of content (which could be the start or end time) and the real-time (i.e., the current time) can also be determined. When the difference in duration is greater than the duration threshold, the account's preference behavior and preference information are directly determined to be empty, i.e., prompting a complete replacement of all content presented in the last presentation to give the user a sense of novelty; when the difference in duration is less than or equal to the duration threshold, steps 202 to 204 are executed to determine the preference information. The duration threshold can be set according to the actual application scenario, such as 1 hour or 2 hours.

[0103] exist Figure 3B middle, Figure 3A Step 104 shown can be implemented through steps 205 and 207, or through steps 206 and 207, and will be explained in conjunction with each step.

[0104] In step 205, when the preference information is empty, the first content replacement strategy and the content filtering strategy corresponding to the first account attribute are used together as the determined content recommendation strategy; wherein, the first content replacement strategy is used to replace multiple pieces of content that were previously presented.

[0105] Here, the first account attribute corresponds to only one type of content filtering strategy. When the preference information is empty, it indicates that the user is not interested in any of the previously presented content. In this case, the first content replacement strategy and the content filtering strategy corresponding to the first account attribute are used together as the determined content recommendation strategy. The first content replacement strategy is used to replace all the previously presented content.

[0106] In step 206, when the preference information includes the identification information of the viewed content, the second content replacement strategy and the content filtering strategy corresponding to the first account attribute are used together as the determined content recommendation strategy; wherein, the second content replacement strategy is used to replace the previously presented viewed content.

[0107] Here, when the preference information includes the identification information of the viewed content, it is impossible to determine whether the user is interested in the content other than the viewed content presented last time. Therefore, the second content replacement strategy and the content filtering strategy corresponding to the first account attribute are used together as the determined content recommendation strategy. The second content replacement strategy is used to replace the viewed content presented last time, and the viewed content is also the content represented by the preference information.

[0108] In step 207, content to be recommended is selected from the content library according to the content recommendation strategy.

[0109] In some embodiments, the above-mentioned filtering of content to be recommended from the content library based on the content recommendation strategy can be achieved in the following manner: multiple contents in the content library are filtered according to the content filtering strategy in the content recommendation strategy to obtain candidate contents; when the content replacement strategy in the content recommendation strategy is the first content replacement strategy, the candidate contents are used as the content to be recommended; when the content replacement strategy in the content recommendation strategy is the second content replacement strategy, the candidate contents and the contents of the multiple contents presented in the previous presentation, excluding the content to be viewed, are used together as the content to be recommended.

[0110] In this embodiment, multiple contents in the content library are first filtered according to the content filtering strategy corresponding to the first account attribute. For easy differentiation, the contents obtained from this filtering process are named candidate contents. When the content replacement strategy in the content recommendation strategy is the first content replacement strategy, the candidate is used as the content to be recommended. When the content replacement strategy in the content recommendation strategy is the second content replacement strategy, the candidate contents and the contents of the multiple contents presented last time, excluding the content to be viewed, are used together as the content to be recommended. Here, if a content quantity threshold for the content to be recommended is set, the candidate contents can be further randomly selected. The randomly selected candidate contents and the contents of the multiple contents presented last time, excluding the content to be viewed, are used together as the content to be recommended, so that the number of the content to be recommended is equal to the content quantity threshold.

[0111] like Figure 3B As shown, this application embodiment implements a strategy of replacing all or part of the content based on the user account's historical operation behavior for multiple content items presented previously. This can effectively improve the responsiveness and accuracy of content replacement and bring a sense of novelty to the user.

[0112] In some embodiments, see Figure 3C , Figure 3C This is a flowchart illustrating how a content filtering strategy based on the first account attribute is used to filter multiple pieces of content in a content library, as provided in this application embodiment. Figure 3C The steps shown are explained.

[0113] In step 301, among multiple contents in the content library, the contents other than those viewed corresponding to historical operation behaviors are identified as unviewed contents.

[0114] Here is an example of a content filtering strategy corresponding to the first account attribute. First, among the multiple pieces of content included in the content library, all content except for content viewed based on historical actions is considered unviewed content. It is worth noting that historical actions here are not limited to the user account's historical actions regarding the last presented content, but rather refer to all historical actions of the user account.

[0115] In some embodiments, all unread content can be filtered according to the filtering parameters, and step 302 can be performed based on the unread content obtained from the filtering process.

[0116] In step 302, for each unread content, the content features of the unread content, the account features of the user account, and the operation features are combined into a test sample; wherein, the operation features correspond to multiple historical operation behaviors of the user account.

[0117] Here, the content characteristics of unviewed content include, but are not limited to, the tags, duration, number of views, and update time of the unviewed content; the account characteristics of the user account include, but are not limited to, the account attributes of the user account, as well as other attributes such as age, gender, region, and occupation; the operation characteristics can be determined based on multiple historical operation behaviors of the user account, for example, based on all historical operation behaviors of the user account within a time limit (such as one week) before the real time. The operation characteristics include, but are not limited to, the tags of the viewed content corresponding to the historical operation behavior, and the terminal device model when the historical operation behavior was executed. In addition, the operation characteristics may also include the user account's login time distribution, etc.

[0118] For each unread content, its content characteristics, the user account's account characteristics, and operational characteristics are combined to form a test sample. This means there is a one-to-one relationship between unread content and test samples. It's worth noting that the types of features in the test sample are not limited to this; more types of features can be added based on the actual application scenario.

[0119] In step 303, an artificial intelligence model is invoked to perform prediction processing on the sample to be tested, and the content score of the unread content corresponding to the sample to be tested is obtained.

[0120] This application does not limit the type of artificial intelligence model, such as an extreme gradient boosting (XGBoost) model, a logistic regression model, a Naive Bayes model, a decision tree model, or a neural network model. For each test sample, the artificial intelligence model is invoked to perform prediction processing on the test sample, obtaining a content score for the unread content corresponding to the test sample. The higher the content score, the greater the probability that the user is interested in the unread content.

[0121] In some embodiments, the artificial intelligence model can be trained before step 303. For example, for multiple sample contents that have been presented, a labeled content score is determined for each sample content. When a sample content corresponds to a certain historical operation, i.e., it belongs to viewed content, the labeled content score of the sample content is determined to be 1; when a sample content does not correspond to any historical operation, i.e., it belongs to unviewed content, the labeled content score of the sample content is determined to be 0. Of course, the method of determining the labeled content score is not limited to this. For each sample content, a corresponding test sample is constructed, and the constructed test sample is predicted according to the artificial intelligence model to obtain the content score of the sample content. For ease of distinction, the content score obtained here is named the content score to be compared. Then, based on the difference between the content score to be compared and the labeled content score of the sample content, the weight parameters of the artificial intelligence model are updated (i.e., the artificial intelligence model is trained). The weight parameters can be updated by minimizing the objective function of the artificial intelligence model. The type of objective function is not limited here.

[0122] In step 304, multiple unread contents are filtered based on content rating.

[0123] For example, unread content is selected as candidate content in descending order of content rating, until the number of selected candidate content equals the content quantity threshold.

[0124] In some embodiments, the types of artificial intelligence models include multiple types. After step 303, the method further includes: fusing the content scores obtained by calling multiple artificial intelligence models to obtain a fused content score of the unread content corresponding to the test sample; the above-mentioned filtering of multiple unread contents based on content scores can be achieved in this way: filtering of multiple unread contents based on fused content scores.

[0125] In this embodiment, various types of artificial intelligence models can be used. For example, for each test sample, multiple artificial intelligence models are used to predict the test sample, resulting in multiple content scores. These multiple content scores are then fused to obtain a fused content score for the test sample. The fusion process can be such as summation or weighted summation, and is not limited thereto. After obtaining the fused content score for each unread content, the unread content is filtered based on the fused content score. This method improves the accuracy of the obtained fused content score, thereby further enhancing the accuracy of filtering the unread content.

[0126] In some embodiments, after step 301, the method further includes: for each unread content, calling an artificial intelligence model to predict the content features of the unread content and obtaining a content score for the unread content.

[0127] In this embodiment of the application, an artificial intelligence model can also be invoked to directly predict the content features of unread content, obtain a content score for the unread content, and then execute step 304. During the training phase of the artificial intelligence model, multiple sample contents that have been presented and the labeled content score for each sample content are obtained. The labeling method for the labeled content score is similar to that described above.

[0128] For each sample content, the content features of the sample content are predicted using an artificial intelligence model to obtain a content score. For ease of differentiation, this content score is named the "comparison score." Then, based on the difference between the comparison score and the labeled content score, the weight parameters of the artificial intelligence model are updated, thus training the model. This method improves the efficiency of model training and application by only requiring the acquisition of content features. It is worth noting that for each unread content, multiple artificial intelligence models can be used to predict its content features separately, resulting in multiple content scores. These scores can then be fused to obtain a final content score for the unread content.

[0129] In some embodiments, after step 301, the method further includes: constructing global features of the first user account based on the content features of multiple viewed content corresponding to the first user account; wherein the first user account is the user account to be recommended content; determining the similarity between the first user account and multiple second user accounts in terms of global features; and determining the overlapping content between multiple viewed content corresponding to the second user account and multiple unviewed content corresponding to the first user account that satisfies the first similarity condition, as candidate content obtained through filtering.

[0130] Here, for ease of distinction, the user account to be recommended content is named the first user account, and other user accounts different from the first user account are named the second user accounts. In this embodiment, a global feature of the first user account can be constructed based on the content features of multiple viewed content corresponding to the first user account. For example, the global feature of the first user account can be obtained by concatenating or weighting the content features of the multiple viewed content corresponding to the first user account. Similarly, for each second user account, a global feature of the second user account can be constructed based on the content features of the multiple viewed content corresponding to the second user account.

[0131] For each second user account, the similarity between the global features of the first user account and the global features of the second user account is determined. This similarity can be cosine similarity or other similarity metrics. Then, second user accounts whose similarity satisfies a first similarity condition are identified. This first similarity condition includes, for example, a similarity greater than a set first similarity threshold, or the highest number of similarities (e.g., one). The overlapping content (i.e., identical content) between multiple viewed content items corresponding to the identified second user accounts and multiple unviewed content items corresponding to the first user accounts is used as candidate content, thus achieving the filtering process for multiple contents in the content library. This method represents user accounts through viewed content and performs filtering based on the similarity between user accounts, improving the flexibility of the filtering process.

[0132] In some embodiments, after step 301, the method further includes: for each piece of content in the content library, identifying multiple user accounts that have performed historical operations on the content, and constructing global features of the content based on the account characteristics of the multiple user accounts; for each piece of content viewed by the first user account, determining the similarity between the viewed content and each unviewed content in terms of global features; and using unviewed content whose similarity satisfies the second similarity condition as candidate content obtained through filtering; wherein, the first user account is the user account to be recommended for the content.

[0133] In this application embodiment, the global features of each content in the content library can be determined. For example, for a certain content, multiple user accounts that have performed historical operations on the content can be identified, and the account features of the identified multiple user accounts can be concatenated or weighted to obtain the global features of the content.

[0134] For ease of distinction, the user account for which content recommendation is to be performed is also named the first user account. Here, multiple viewed content items corresponding to the first user account can be traversed. For each viewed content item, the similarity between the viewed content and each unviewed content item corresponding to the first user account is determined on a global feature basis. Unviewed content items whose similarity meets a second similarity condition are selected as candidate content for filtering. This second similarity condition includes items with similarity values ​​greater than a set second similarity threshold, or items with the highest similarity values ​​(e.g., one). This method represents content using user accounts and performs filtering based on the similarity between content items, thus improving the flexibility of the filtering process from another perspective.

[0135] like Figure 3C As shown, this application embodiment determines the content score of unread content based on the principle of artificial intelligence, and then filters multiple unread contents based on the content score, which can improve the accuracy of the filtering process.

[0136] In some embodiments, see Figure 3D , Figure 3D This is a flowchart illustrating an AI-based content recommendation method provided in an embodiment of this application. Figure 3A The step 102 shown can be updated to step 401, in which when the number of historical operation behaviors is less than or equal to the number threshold, the user account's account attribute is determined to be the second account attribute.

[0137] In this embodiment, account attributes may include two categories: first account attributes and second account attributes. The first account attribute corresponds to the old account, and the second account attribute corresponds to the new account. When the number of historical operation behaviors of the user account obtained through step 101 is less than or equal to the number threshold, the user account's account attribute is determined to be the second account attribute.

[0138] exist Figure 3D middle, Figure 3A Step 103 shown can be implemented through steps 402 to 404, which will be explained in conjunction with each step.

[0139] In step 402, the identification information of multiple objects is presented; wherein, each object corresponds to multiple contents.

[0140] For user accounts with a secondary account attribute, the limited number of historical activity records makes it impossible to effectively analyze preference information based on these records. Therefore, multiple object identifiers can be proactively presented, for example, within the content presentation interface. Each object corresponds to multiple pieces of content. The object identifiers are used to easily distinguish between different objects and can be at least one of the object's name and icon; however, the form of the identifiers is not limited to these.

[0141] For example, when objects participate in interactive events, if the interactive event is a league competition, the identification information of multiple teams (objects) participating in the league competition can be displayed. The identification information of the teams can be at least one of their names and team logos. If the interactive event is a content-themed event, the identification information of multiple content-uploading accounts (objects) participating in the content-themed event can be displayed. The identification information of the content-uploading accounts can be at least one of their names and avatars.

[0142] In step 403, the user account's selection behavior of identifying information of any object is taken as the account preference behavior that conforms to the second account attribute.

[0143] Here, users can actively select the identifier information of a specific object to be presented based on their own preferences using their user account. For electronic devices, the user account's selection of the identifier information of any presented object is considered an account preference behavior that conforms to the second account attribute.

[0144] In some embodiments, after step 402, the method further includes: when no user account selection behavior for the identification information of the presented object is obtained (i.e. no account preference behavior is obtained), the identification information of any object is used as preference information.

[0145] In step 404, the identification information of the object selected by the selected behavior is used as the user account's preference information.

[0146] In this embodiment, the second account attribute can correspond to multiple types of content filtering strategies, and different types of content filtering strategies target different objects. Each type of content filtering strategy can have one or more components. Multiple types of content filtering strategies can be set according to the characteristics of the objects targeted in the actual application scenario. This embodiment does not limit this. For example, if a certain type of content filtering strategy targets object A, then this type of content filtering strategy can be used to filter out content corresponding to object A in the content library, so that the content corresponding to object A accounts for 80% of the final multiple recommended contents, and the remaining 20% ​​is content corresponding to other objects filtered from the content library. This satisfies the user's interests and preferences by presenting the content corresponding to object A, which has a larger proportion.

[0147] Once it is determined that the user account's account attribute is a secondary account attribute and the preference information is determined, a content filtering strategy can be selected from multiple content filtering strategies corresponding to the secondary account attribute to determine the content filtering strategy for the object represented by the preference information (i.e. the selected object), which can then be used as a content recommendation strategy to filter out the content to be recommended.

[0148] exist Figure 3D middle, Figure 3A Step 104 shown can be implemented through steps 405 and 407, or steps 406 and 407, which will be explained in conjunction with each step.

[0149] In step 405, when the selected object has participated in the latest stage of the interactive event, the first content filtering strategy corresponding to the second account attribute in the latest stage is used as the determined content recommendation strategy.

[0150] Here, multiple objects can participate in interactive events, and each interactive event corresponds to multiple pieces of content, including the content specific to each object. There are no restrictions on the type of interactive event; it can be, for example, a league event or a video-themed event.

[0151] In one embodiment of this application, the interactive event includes multiple stages, such as a league event including a preseason stage, a regular season stage, and a playoff stage. For each stage, the content filtering strategy corresponding to the second account attribute in that stage includes a first content filtering strategy for objects participating in that stage and a second content filtering strategy for objects not participating in that stage. That is, based on whether the object participates in that stage, there are two types of content filtering strategies: the first content filtering strategy is one type of content filtering strategy, and the second content filtering strategy is another type of content filtering strategy.

[0152] When the object represented by the preference information (i.e. the selected object) has participated in the latest stage of the interactive event, the first content filtering strategy corresponding to the second account attribute in the latest stage will be used as the determined content recommendation strategy.

[0153] In some embodiments, when the interaction event includes multiple stages, after step 402, the method further includes: when no selection behavior of the user account regarding the identification information of the presented object is obtained, using the identification information of any object participating in the latest stage as preference information. This improves the accuracy and applicability of the obtained preference information.

[0154] In step 406, if the selected object does not participate in the latest stage, the second content filtering strategy corresponding to the second account attribute in the latest stage is used as the determined content recommendation strategy.

[0155] When the object represented by the preference information (i.e. the selected object) does not participate in the latest stage of the interactive event, the second content filtering strategy corresponding to the second account attribute in the latest stage will be used as the determined content recommendation strategy.

[0156] In step 407, content to be recommended is selected from the content library according to the content recommendation strategy.

[0157] For example, content to be recommended can be selected from multiple pieces of content in the content library that correspond to interactive events, based on a content recommendation strategy.

[0158] In some embodiments, when the determined content recommendation strategy is a first content filtering strategy, the above-mentioned filtering of content to be recommended from the content library according to the content recommendation strategy can be achieved in the following manner: performing at least one of the following processes: filtering multiple contents corresponding to the latest stage in the content library and multiple contents corresponding to the selected object according to filtering parameters to obtain content to be recommended that meets a first quantity ratio; wherein, the first quantity ratio is the quantity ratio between the content corresponding to the latest stage and the content corresponding to other stages; filtering multiple contents corresponding to the object participating in the latest stage and multiple contents corresponding to the object not participating in the latest stage in the content library according to filtering parameters to obtain content to be recommended that meets a second quantity ratio; wherein, the second quantity ratio is the quantity ratio between the content corresponding to the object participating in the latest stage and the content corresponding to the object not participating in the latest stage; wherein, the filtering parameters include at least one of popularity and update time.

[0159] Here, two examples of the first content filtering strategy are provided. The first example involves filtering multiple pieces of content corresponding to the latest stage (i.e., tags including multiple pieces of content from the latest stage, and the same applies below) and multiple pieces of content corresponding to the selected object based on filtering parameters. This yields content to be recommended that meets a first quantity ratio, where the first quantity ratio is the ratio between the content corresponding to the latest stage and the content corresponding to other stages (where the number of pieces of content corresponding to the latest stage can be greater than the number of pieces of content corresponding to other stages), such as a first quantity ratio of 6:4. In this way, some content corresponding to the latest stage can be presented, as well as some content corresponding to other stages and the selected object, improving the presentation effect and rationality. It is worth noting that this application embodiment can also limit the number of content to be recommended obtained through filtering, such as limiting the threshold for the number of content to be recommended to 10, and the same applies below.

[0160] The second example involves filtering multiple pieces of content corresponding to objects participating in the latest phase and multiple pieces of content corresponding to objects not participating in the latest phase based on filtering parameters. This yields recommended content that meets a second quantity ratio, where the second quantity ratio is the ratio between the content corresponding to objects participating in the latest phase and the content corresponding to objects not participating in the latest phase (where the number of pieces of content corresponding to objects participating in the latest phase can be greater than the number of pieces of content corresponding to objects not participating in the latest phase), for example, a second quantity ratio of 8:2. The first content filtering strategies in the above two examples can be applied individually or simultaneously. Furthermore, the first content filtering strategy is not limited to the above examples and can be specifically set according to the actual application scenario.

[0161] In some embodiments, when the determined content recommendation strategy is the second content filtering strategy, the above-mentioned filtering of content to be recommended from the content library according to the content recommendation strategy can be achieved in the following manner: performing at least one of the following processes: filtering multiple contents corresponding to the latest stage in the content library and multiple contents corresponding to the selected object according to the filtering parameters to obtain content to be recommended that meets the third quantity ratio; wherein, the third quantity ratio is the quantity ratio between the content corresponding to the latest stage and the content corresponding to the selected object; filtering multiple contents corresponding to the object participating in the latest stage in the content library and multiple contents corresponding to the selected object according to the filtering parameters to obtain content to be recommended that meets the fourth quantity ratio; wherein, the fourth quantity ratio is the quantity ratio between the content corresponding to the object participating in the latest stage and the content corresponding to the selected object; wherein, the filtering parameters include at least one of popularity and update time.

[0162] Here are two examples of the second content filtering strategy. The first example filters multiple pieces of content corresponding to the latest stage and multiple pieces of content corresponding to the selected object based on filtering parameters, resulting in recommended content that meets a third quantity ratio. This third quantity ratio is the ratio of the content corresponding to the latest stage to the content corresponding to the selected object, such as 6:4. This satisfies the user's potential preference for the latest stage, i.e., preferences not reflected in the selection behavior. The second example filters multiple pieces of content corresponding to the object participating in the latest stage and multiple pieces of content corresponding to the selected object based on filtering parameters, resulting in recommended content that meets a fourth quantity ratio. This fourth quantity ratio is the ratio of the content corresponding to the object participating in the latest stage to the content corresponding to the selected object, such as 6:4. This satisfies the user's potential preference for the object participating in the latest stage. Similarly, the second content filtering strategies in the above two examples can be applied individually or simultaneously. The second content filtering strategy is not limited to the above examples and can be specifically set according to the actual application scenario.

[0163] like Figure 3D As shown, this application embodiment determines whether to apply a first content filtering strategy or a second content filtering strategy based on whether the object participates in the latest stage of the interactive event, which can fully meet the user's content viewing needs in the latest stage of the interactive event.

[0164] The following will describe an exemplary application of the embodiments of this application in a real-world application scenario. For ease of understanding, the example given is a video (i.e., content mentioned above) recommendation scenario during the finals preparation period (i.e., the finals preparation stage, corresponding to the latest stage mentioned above) of a basketball league (corresponding to the interactive event above). The finals preparation period refers to the period between the election of the East and West division champions of the basketball league and the start of the first game of the finals. The embodiments of this application provide, as follows... Figure 4 The diagram shown illustrates an AI-based video recommendation method, which will be presented step-by-step. Figure 4 Please provide an explanation.

[0165] Step 1: Receive touch operations from user accounts for the content display interface of the basketball league, which is the video interface.

[0166] Step 2: Determine whether the user using the user account is a new user. If the user is a new user (i.e., a new account, corresponding to the second account attribute above), proceed to steps 3 and 4; if the user is an old user (i.e., an old account, corresponding to the first account attribute above), proceed to step 9.

[0167] Step 3: Present the selection interface for interested teams in the video interface, for example, in the form of a window. This application embodiment provides, for example... Figure 5 The diagram shows the selection interface 51 for teams of interest. The selection interface 51 includes identification information for each team in the basketball league (such as team name) and a selection box for each team. Figure 5 (Using selection box 52 as an example), it's worth noting that the team corresponds to the object mentioned above. Figure 5 The example uses 6 teams, but the number of teams participating in a basketball league in actual applications is not limited to this. When a user account's selection behavior for any team is obtained (such as checking the selection box corresponding to any team and clicking the "OK" button), the team selected by the selection behavior is taken as the user's team of interest, which corresponds to the selected object mentioned above; when no user account's selection behavior for a team is obtained, for example, when the selection interface 51 for the team of interest is closed (no user account selection behavior was obtained during the presentation of the selection interface 51 for the team of interest), either of the two teams that enter the finals will be taken as the user's team of interest.

[0168] Step 4: Determine if the user's interested team is one of the two teams that have entered the finals. If yes, proceed to steps 5 and 6; otherwise, proceed to steps 7 and 8.

[0169] Step 5: Based on the popular section filtering strategy 1 for new users, obtain the videos displayed in the popular section of the basketball league video interface.

[0170] Step 6: Based on the new user's highlights section filtering strategy 1, obtain the videos displayed in the highlights section of the basketball league's video interface. The new user's popular section filtering strategy 1 and highlights section filtering strategy 1 correspond to the first content filtering strategy for the second account attribute mentioned above in the latest stage.

[0171] Step 7: Based on the new user's popular section filtering strategy 2, obtain the videos displayed in the popular section of the basketball league video interface.

[0172] Step 8: Based on the new user's highlights section filtering strategy 2, obtain the videos displayed in the highlights section of the basketball league's video interface. The new user's popular section filtering strategy 2 and highlights section filtering strategy 2 correspond to the second content filtering strategy for the second account attribute mentioned above at the latest stage.

[0173] Step 9: When the user account is an existing user, first determine whether the user account has accessed the basketball league video interface within a set time period before the current time (corresponding to the duration threshold above). The set time period can be specifically set according to the actual application scenario, such as 1 hour or 2 hours. If the user account has accessed the basketball league video interface within the set time period before the current time and has not triggered (triggered the execution history behavior corresponding to the above) any video in the basketball league video interface, then the all-content replacement mechanism is activated (corresponding to the first content replacement strategy above), that is, steps 10 to 11 are executed to completely replace the videos presented in the previously accessed popular sections and highlights sections; if the user account has accessed the basketball league video interface within the set time period before the current time and triggered a video in the basketball league video interface, then step 12 is executed.

[0174] Step 10: Based on the popular section filtering strategy of old users, obtain the videos displayed in the popular section of the basketball league video interface.

[0175] Step 11: Based on the filtering strategy for the highlights section of returning users, obtain the videos displayed in the highlights section of the basketball league's video interface. The filtering strategies for popular sections and highlights sections of returning users correspond to the content filtering strategies for the first account attribute mentioned above.

[0176] Step 12: Enable the partial content replacement mechanism (corresponding to the second content replacement strategy above). That is, for the videos presented in the video interface of the basketball league visited last time, keep the untriggered videos unchanged, and replace the triggered videos (corresponding to the viewed videos above) with a new random video from the section to which the triggered video belongs (such as the popular section, the highlights section, or other content sections). (This can be obtained by filtering according to the corresponding filtering strategy for old users). In this way, the videos presented in the video interface of the basketball league visited this time can be obtained.

[0177] Step 13: Based on the filtering strategy for other content sections, obtain the videos presented in the other content sections of the basketball league video interface. It is worth noting that other content sections can also be included in the scope of the full content replacement mechanism, that is, enabling the full content replacement mechanism can refer to executing steps 10, 11, and 13. Of course, other content sections can also be included in the partial content replacement mechanism.

[0178] Through the above steps, videos (i.e., content to be recommended) can be obtained and presented in popular sections, highlight sections, and other content sections, and then recommended to users. This application provides the following embodiments: Figure 6 The diagram shown is a schematic of the basketball league's video interface 61, illustrating videos 1 to 4 in the "Popular" section, videos 5 to 8 in the "Highlights" section, and videos 9 to 12 in other content sections. It is worth noting that... Figure 6 The presentation of the popular sections, highlights sections, and other content sections does not constitute a limitation on the embodiments of this application. For example, the three sections can be presented in a side-by-side structure, and the size of the area occupied by each section can be adjusted according to the actual application scenario.

[0179] Next, we will explain the various content recommendation strategies involved in the above steps.

[0180] 1) Popular section selection strategy for new users.

[0181] When a user is identified as a new user, and their team of interest is one of the two teams that made it to the finals, the videos to be presented in the popular sections are determined according to the new user's popular section filtering strategy 1. For example, among all videos tagged with the team of interest (i.e., all videos corresponding to the selected object), the top 100 videos by play count are identified. Here, the play count ranking in this embodiment refers to sorting by play count from highest to lowest. Simultaneously, among all videos tagged with the finals (i.e., all videos corresponding to the latest stage), the top 50 videos by update time are identified. Here, the update time ranking in this embodiment refers to sorting by update time from newest to oldest.

[0182] Randomly select from the initial 150 videos until N videos are obtained. Of these N videos, 60% must be tagged with "Finals," meaning videos corresponding to the Finals. These 60% of videos may also include tags such as champion predictions, player statements, media commentary, team highlights, or player highlights, etc., without restriction. The remaining 40% of videos must be tagged without "Finals," meaning videos not corresponding to the Finals (i.e., videos corresponding to other stages). These 40% of videos may include tags such as team highlights or player highlights, without restriction.

[0183] The N mentioned above is an integer greater than 1 and less than 150. For ease of understanding, the following example uses N=10, meaning each section of the basketball league video interface needs to display 10 videos. Taking the two teams that entered the finals as Team 1 and Team 2, and the user's interested team as Team 1, the following example shows the 10 videos displayed in the popular section as determined by the new user's popular section filtering strategy 1:

[0184]

[0185] 2) New user highlight section filtering strategy 1.

[0186] When a user is identified as a new user, and their team of interest is one of the two teams that made it to the finals, the videos to be presented in the highlights section are determined according to the new user's highlight selection strategy 1. For ease of differentiation, either of the two teams that made it to the finals is designated as the "Finals Team" (i.e., the team participating in the latest stage), and the other team that did not make it to the finals is designated as the "Non-Finals Team" (i.e., the team not participating in the latest stage). For example, among all videos tagged with both "Finals Team" and "Highlights" (using team highlights as an example), the top 100 videos with an update time within one month prior to the current time and the highest view count are selected, and 8 videos are randomly selected from these 100. Simultaneously, among all videos tagged with both "Non-Finals Team" and "Team Highlights," the top 100 videos with the highest view count are selected, and 2 videos are randomly selected from these 100.

[0187] In other words, of the videos presented in the Highlights section, 80% are team highlights videos from the Finals teams, and the remaining 20% ​​are team highlights videos from teams that did not participate in the Finals. Taking the two teams that reached the Finals as Team 1 and Team 2, and assuming the user is interested in Team 1, the following example shows the 10 videos presented in the Highlights section based on the new user's Highlights section filtering strategy 1:

[0188]

[0189] 3) Popular section selection strategy for new users 2.

[0190] When a user is identified as a new user, and their interested teams are not from the Finals, the videos to be presented in the popular sections are determined according to the new user's popular section filtering strategy 2. For example, among all videos tagged with Finals teams, the top 100 videos by view count are selected. Simultaneously, among all videos tagged with Finals, the top 50 videos by update time are selected. Then, 6 videos are randomly selected from the selected 150 videos, requiring that all 6 videos have the tag "Finals". For each of these 6 videos, the tags can also include championship prediction, player comments, media commentary, team highlights, or player highlights, etc., which are not required.

[0191] Additionally, among all videos tagged with teams of interest, the top 200 videos by view count are identified. From these 200 videos, four are randomly selected. For each of these four videos, tags such as team highlights or player highlights are optional and not mandatory.

[0192] In other words, among the videos presented in the popular section, 60% are videos related to the finals, and the remaining 40% are videos related to teams the user is interested in. Taking the two teams that entered the finals as Team 1 and Team 2, and the user's interested team as Team 7, as an example, the following shows the 10 videos presented in the popular section determined by the new user's popular section filtering strategy 2:

[0193]

[0194] 4) New user highlight section filtering strategy 2.

[0195] When a user is identified as a new user, and their interested teams are not from the Finals, the videos to be presented in the Highlights section are determined according to the new user's Highlights section filtering strategy 2. For example, among all videos tagged with both Finals teams and Highlights (using team highlights as an example), the top 100 videos with an update time within one month prior to the current time and the highest number of views are selected. Simultaneously, among all videos tagged with Finals, the top 50 videos with the highest update time are selected. Then, from these 150 selected videos, 6 videos are randomly chosen, requiring that these 6 videos have tags that simultaneously include both Finals teams and team highlights.

[0196] In addition, among all videos tagged with both the team of interest and highlights (referring to team highlights or player highlights), the top 100 videos by view count were identified, and four videos were randomly selected from these 100 videos.

[0197] In other words, of the many videos presented in the Highlights section, 60% are videos of the teams that made it to the Finals, and the remaining 40% are videos of teams the user is interested in. Taking the two teams that made it to the Finals as Team 1 and Team 2, and the user's team of interest as Team 7, as an example, the following shows the 10 videos presented in the Highlights section determined by the new user's Highlights section filtering strategy 2:

[0198]

[0199] 5) Popular section filtering strategy for existing users.

[0200] Once a user is identified as a returning user, the videos to be displayed in the popular sections are determined based on the returning user's preferred section filtering strategy. For example, videos already accessed by the user's account are removed, and from the remaining unviewed videos, the top 1000 videos by view count are selected. Figure 7 As shown, for each of the 1000 unviewed videos, the video features (i.e., content features), user features, interaction features (i.e., operation features), and video freshness features of the unviewed video are input into an artificial intelligence model (such as a machine learning model) to obtain the video score of the unviewed video.

[0201] Then, among the top 50 unviewed videos in the video rating ranking, 10 are randomly selected as the videos to be presented in the popular section. Here, the video rating ranking refers to sorting the videos in descending order of their ratings. The higher the video rating of an unviewed video, the greater the probability that users are interested in that unviewed video.

[0202] The video features of unviewed videos include, but are not limited to, the tags, duration, and number of views. User features may include whether the user is a returning user, as well as basic attributes such as age, gender, region, and occupation. Interaction features include, but are not limited to, the tags of videos triggered by the user account within the past week, the model of the terminal device used, the distribution of login times, and the teams the user is interested in. The video freshness features of unviewed videos include the update time of the unviewed videos; here, the video freshness features of unviewed videos can also be considered as part of the video features. Of course, the above features are just examples, and the content of each feature can be adjusted according to the actual application scenario. In addition, the types of features input to the artificial intelligence model can also be adjusted, for example, by adding more types of features.

[0203] This application does not limit the type of artificial intelligence model, such as a logistic regression model, a Naive Bayes model, a decision tree model, or a neural network model. Before using the artificial intelligence model, it can be trained based on the video features, user features, interaction features, and video freshness features of the sample videos, where sample videos refer to videos that have already been presented.

[0204] To facilitate understanding, we will use the XGBoost model as an example to illustrate its training process. The XGBoost model fuses the results of multiple sub-classifiers, considering the complexity of each sub-classifier, and optimizes both complexity and classification results simultaneously. Its objective function formula is as follows:

[0205]

[0206] in, For the first i The expected output of each sample video (i.e.) (corresponding to the labeled video rating above) and predicted output (i.e. The loss function between the video ratings to be compared (as mentioned above) is, for example, if the user account triggers the first... i The nth sample video will be the nth i The expected output of each sample video is labeled as 1; if the user account does not trigger the first... i The nth sample video will be the nth i The expected output annotation for each sample video is 0; however, the actual annotation method is not limited to this. Additionally, For the first k The complexity of a subclassifier w These are the weight parameters that need to be trained in the subclassifier.

[0207] During training, the objective function is iterated multiple times, the 1st... t The objective function of the wheel can be expressed as:

[0208]

[0209] Performing a second-order Taylor expansion on the above formula, we obtain:

[0210]

[0211] in, g The first derivative, h This is the second derivative. For the XGBoost model, the weight parameters of the subclassifiers can be effectively trained by minimizing the objective function in each iteration.

[0212] It is worth noting that the embodiments of this application can also use multiple artificial intelligence models and fuse the video scores obtained by the multiple artificial intelligence models to filter unviewed videos based on the fused video score. The fusion process can include summation or weighted summation. For example, for a given unviewed video, video scores are obtained using an XGBoost model, a Naive Bayes model, and a neural network model, respectively. Then, the three video scores of the unviewed video are fused to obtain the fused video score of the unviewed video.

[0213] 6) Filtering strategy for the highlights section of old users.

[0214] Once a user is identified as a returning user, the videos to be presented in the highlights section are determined based on the selection strategy for returning users. Here, it is assumed that returning users have already watched team highlights from previous stages and are familiar with the events of previous matches; therefore, team highlights from previous rounds will not be recommended again to returning users.

[0215] For example, among all videos tagged with both Finals teams and player highlights, the top 100 videos by view count can be identified. These 100 videos are then randomly selected until six videos with tags including highlights of the player's current season and four videos with tags including highlights of different stages of the player's career are obtained. Here, highlights of the player's current season and highlights of different stages of the player's career are subordinate concepts to player highlights.

[0216] 7) Other content section filtering strategies.

[0217] In other content sections, there is no distinction between new and returning users. For example, from all videos that a user's account has not triggered and whose tags include gossip (or off-field information), the top 100 videos by view count can be identified, and 3 videos can be randomly selected from these 100. Simultaneously, from all videos whose tags include both non-finalist teams and player highlights, the top 100 videos by view count can be identified, and 5 videos can be randomly selected from these 100. Furthermore, from all videos whose tags include both finalist teams and team highlights, the top 100 videos by view count can be identified, and 2 videos can be randomly selected from these 100. This ultimately results in 10 videos being presented in other content sections.

[0218] The embodiments of this application can achieve at least the following technical effects: 1) By guiding users to select teams of interest and applying corresponding video recommendation strategies (content recommendation strategies) for video recommendation, the cold start problem of video recommendation for new users can be effectively solved; 2) The recommendation effect can be effectively improved from multiple aspects such as information richness and video replacement responsiveness, thereby increasing user stickiness and the time users spend on the basketball league video interface (stay time), achieving a "personalized" recommendation effect; 3) The embodiments of this application are decoupled from the platform used to present the video, and can be applied to any platform that needs to recommend basketball league-related videos, thus having high reusability.

[0219] The following continues to describe the exemplary structure of the AI-based video recommendation device 455 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 2 As shown, the software modules stored in the AI-based video recommendation device 455 in the memory 450 may include: an acquisition module 4551, used to acquire the historical operation behavior of the user account logged into the content presentation interface in response to a touch operation on the content presentation interface; an attribute determination module 4552, used to determine the account attributes of the user account based on the historical operation behavior; a preference determination module 4553, used to acquire account preference behaviors that match the account attributes, and perform preference analysis processing based on the account preference behaviors to obtain the user account's preference information for multiple contents in the content library; a filtering module 4554, used to determine a content recommendation strategy based on the account attributes and preference information, and filter out content to be recommended from the content library according to the content recommendation strategy; and an update module 4555, used to update the content presentation interface based on the filtered content to be recommended.

[0220] In some embodiments, the preference determination module 4553 is further configured to: when the account attribute is a first account attribute, take the user account's historical operation behavior for multiple previously presented contents as the account preference behavior that conforms to the first account attribute; determine the viewing content corresponding to the historical operation behavior among the multiple previously presented contents; and take the identification information of the viewing content as the user account's preference information; wherein the number of historical operation behaviors corresponding to the first account attribute is greater than a number threshold.

[0221] In some embodiments, the filtering module 4554 is further configured to: when the preference information is empty, use the first content replacement strategy and the content filtering strategy corresponding to the first account attribute together as the determined content recommendation strategy; wherein the first content replacement strategy is used to replace multiple contents presented previously; when the preference information includes the identification information of the viewed content, use the second content replacement strategy and the content filtering strategy corresponding to the first account attribute together as the determined content recommendation strategy; wherein the second content replacement strategy is used to replace the viewed content presented previously.

[0222] In some embodiments, the filtering module 4554 is further configured to: filter multiple contents in the content library according to the content filtering strategy in the content recommendation strategy to obtain candidate contents; when the content replacement strategy in the content recommendation strategy is the first content replacement strategy, use the candidate contents as the content to be recommended; when the content replacement strategy in the content recommendation strategy is the second content replacement strategy, use the candidate contents and the contents of the multiple contents presented last time, excluding the content to be viewed, as the content to be recommended.

[0223] In some embodiments, the filtering module 4554 is further configured to: identify, from multiple contents in the content library, contents other than those viewed corresponding to historical operation behaviors, as unviewed contents; for each unviewed content, combine the content features of the unviewed content, the account features of the user account, and the operation features into a test sample; wherein the operation features correspond to multiple historical operation behaviors of the user account; call an artificial intelligence model to perform prediction processing on the test sample to obtain a content score for the unviewed content corresponding to the test sample; and filter multiple unviewed contents based on the content score.

[0224] In some embodiments, the types of artificial intelligence models include multiple types; the filtering module 4554 is further configured to: perform fusion processing on the content scores obtained by calling multiple artificial intelligence models respectively to obtain a fused content score of the unread content corresponding to the test sample; and perform filtering processing on multiple unread contents based on the content scores, including: filtering processing on multiple unread contents based on the fused content scores.

[0225] In some embodiments, the preference determination module 4553 is further configured to: determine the difference duration between the presentation time of the last presentation of multiple contents and the real-time time; when the difference duration is greater than the duration threshold, determine that the user account's preference information is empty.

[0226] In some embodiments, the preference determination module 4553 is further configured to: present the identification information of multiple objects when the account attribute is a second account attribute; wherein each object corresponds to multiple contents; take the user account's selection behavior for the identification information of any object as the account preference behavior that conforms to the second account attribute; take the identification information of the object selected by the selected behavior as the user account's preference information; wherein the number of historical operation behaviors corresponding to the second account attribute is less than or equal to the number threshold.

[0227] In some embodiments, the second account attribute corresponds to multiple types of content filtering strategies, and different types of content filtering strategies target different objects; the filtering module 4554 is further configured to: determine, among the multiple types of content filtering strategies corresponding to the second account attribute, a content filtering strategy for the selected object, as a content recommendation strategy.

[0228] In some embodiments, the interactive event involving multiple objects includes multiple stages; the content filtering strategy corresponding to the second account attribute in each stage includes a first content filtering strategy for objects participating in the stage and a second content filtering strategy for objects not participating in the stage; the filtering module 4554 is further configured to: when the selected object has participated in the latest stage of the interactive event, use the first content filtering strategy corresponding to the second account attribute in the latest stage as the determined content recommendation strategy; when the selected object has not participated in the latest stage, use the second content filtering strategy corresponding to the second account attribute in the latest stage as the determined content recommendation strategy.

[0229] In some embodiments, the filtering module 4554 is further configured to: when the determined content recommendation strategy is a first content filtering strategy, perform at least one of the following processes: filter multiple contents corresponding to the latest stage in the content library and multiple contents corresponding to the selected object according to the filtering parameters to obtain content to be recommended that meets a first quantity ratio; wherein, the first quantity ratio is the quantity ratio between the content corresponding to the latest stage and the content corresponding to other stages; filter multiple contents corresponding to the object participating in the latest stage and multiple contents corresponding to the object not participating in the latest stage in the content library according to the filtering parameters to obtain content to be recommended that meets a second quantity ratio; wherein, the second quantity ratio is the quantity ratio between the content corresponding to the object participating in the latest stage and the content corresponding to the object not participating in the latest stage; wherein, the filtering parameters include at least one of popularity and update time.

[0230] In some embodiments, the filtering module 4554 is further configured to: when the determined content recommendation strategy is a second content filtering strategy, perform at least one of the following processes: filter multiple contents corresponding to the latest stage in the content library and multiple contents corresponding to the selected object according to the filtering parameters to obtain content to be recommended that meets a third quantity ratio; wherein, the third quantity ratio is the quantity ratio between the content corresponding to the latest stage and the content corresponding to the selected object; filter multiple contents corresponding to the object participating in the latest stage in the content library and multiple contents corresponding to the selected object according to the filtering parameters to obtain content to be recommended that meets a fourth quantity ratio; wherein, the fourth quantity ratio is the quantity ratio between the content corresponding to the object participating in the latest stage and the content corresponding to the selected object; wherein, the filtering parameters include at least one of popularity and update time.

[0231] In some embodiments, the AI-based video recommendation device 455 further includes: a sharing module, used to acquire content recommendation strategies shared by different account attributes and different preference information; and a filtering module 4554, used to: filter out content to be recommended from the content library according to the content recommendation strategies determined by the account attributes and preference information and the shared content recommendation strategies.

[0232] In some embodiments, the update module 4555 is further configured to: when the determined content recommendation strategy includes content recommendation strategies corresponding to multiple content sections respectively, perform the following processing for each content section: in the content section of the content presentation interface, present the content to be recommended filtered according to the content recommendation strategy corresponding to the content section.

[0233] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the artificial intelligence-based video recommendation method described above in this application.

[0234] This application provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to perform the method provided in this application, for example... Figure 3A , Figure 3B and Figure 3D The video recommendation method based on artificial intelligence is shown.

[0235] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0236] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0237] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0238] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0239] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method for content recommendation based on artificial intelligence, characterized by, The method comprises: in response to a touch operation of a content presentation interface, obtaining a historical operation behavior of a user account logged into the content presentation interface; determining an account attribute of the user account according to the historical operation behavior; obtaining an account preference behavior conforming to the account attribute, and performing preference analysis processing according to the account preference behavior to obtain preference information of the user account for multiple contents in a content library; determining a content recommendation strategy according to the account attribute and the preference information, and performing screening processing on the multiple contents in the content library according to a content screening strategy in the content recommendation strategy to obtain candidate contents; when a content replacement strategy in the content recommendation strategy is a first content replacement strategy, taking the candidate contents as to-be-recommended contents; wherein the first content replacement strategy is used to replace multiple contents presented last time; when the content replacement strategy in the content recommendation strategy is a second content replacement strategy, taking the candidate contents and contents other than reading contents in the multiple contents presented last time together as to-be-recommended contents; wherein the second content replacement strategy is used to replace the reading contents presented last time; updating the content presentation interface according to the screened to-be-recommended contents.

2. The method of claim 1, wherein, When the account attribute is a first account attribute, the obtaining of the account preference behavior conforming to the account attribute and the performing of the preference analysis processing according to the account preference behavior to obtain the preference information of the user account for the multiple contents in the content library comprises: taking historical operation behaviors of the user account for the multiple contents presented last time as the account preference behavior conforming to the first account attribute; determining reading contents corresponding to the historical operation behaviors in the multiple contents presented last time; taking identification information of the reading contents as the preference information of the user account; wherein the number of the historical operation behaviors corresponding to the first account attribute is greater than a quantity threshold.

3. The method of claim 2, wherein, The determination of the content recommendation strategy according to the account attribute and the preference information comprises: when the preference information is empty, taking a first content replacement strategy and a content screening strategy corresponding to the first account attribute together as the determined content recommendation strategy; when the preference information includes the identification information of the reading contents, taking a second content replacement strategy and the content screening strategy corresponding to the first account attribute together as the determined content recommendation strategy.

4. The method of claim 1, wherein, The screening processing of the multiple contents in the content library according to the content screening strategy in the content recommendation strategy comprises: in the multiple contents in the content library, determining contents other than the reading contents corresponding to the historical operation behaviors as unread contents; for each of the unread contents, combining a content feature of the unread content, an account feature of the user account, and an operation feature corresponding to multiple historical operation behaviors of the user account into a test sample; calling an artificial intelligence model to perform prediction processing on the test sample to obtain a content score of the unread content corresponding to the test sample; Filter a plurality of the unread contents according to the content scores.

5. The method of claim 4, wherein, The types of the artificial intelligence models include multiple types; after the artificial intelligence model performs the prediction processing on the to-be-tested sample to obtain the content score of the unread content corresponding to the to-be-tested sample, the method further includes: Fusing the content scores obtained by calling the multiple types of artificial intelligence models respectively to obtain a fused content score of the unread content corresponding to the to-be-tested sample; The filtering processing of the plurality of the unread contents according to the content scores includes: Filtering a plurality of the unread contents according to the fused content score.

6. The method of claim 2, wherein, The method further includes: Determining a difference duration between a presentation time of a last time of presenting a plurality of contents and a real-time time; When the difference duration is greater than a duration threshold, determining that the preference information of the user account is empty.

7. The method of claim 1, wherein, When the account attribute is a second account attribute, the method further includes: Presenting identification information of a plurality of objects; each of the objects corresponds to a plurality of contents; Selecting behavior of the user account for the identification information of any one of the objects as account preference behavior conforming to the second account attribute; Identification information of the object selected by the selecting behavior is used as the preference information of the user account. The number of the historical operation behaviors corresponding to the second account attribute is less than or equal to a quantity threshold.

8. The method of claim 7, wherein, The second account attribute corresponds to a plurality of types of content filtering strategies, and different types of content filtering strategies are for different objects. The determining of the content recommendation strategy according to the account attribute and the preference information includes: In the plurality of types of content filtering strategies corresponding to the second account attribute, a content filtering strategy for the selected object is determined as the content recommendation strategy.

9. The method of claim 8, wherein, The interactive event in which the plurality of objects participate includes a plurality of stages; the content filtering strategy of the second account attribute corresponding to each of the stages includes a first content filtering strategy for an object participating in the stage and a second content filtering strategy for an object not participating in the stage; The determining of the content recommendation strategy in the plurality of types of content filtering strategies corresponding to the second account attribute includes: When the selected object has participated in a latest stage of the interactive event, the first content filtering strategy of the second account attribute corresponding to the latest stage is determined as the determined content recommendation strategy; When the selected object has not participated in the latest stage, the second content filtering strategy of the second account attribute corresponding to the latest stage is determined as the determined content recommendation strategy.

10. The method of claim 9, wherein, When the determined content recommendation strategy is the first content filtering strategy, the method further includes: Performing at least one of the following processing: screening the contents corresponding to the selected object and the contents corresponding to the latest stage according to the screening parameter to obtain to-be-recommended contents satisfying a first quantity ratio; wherein the first quantity ratio is a quantity ratio between the contents corresponding to the latest stage and the contents corresponding to other stages; screening the contents corresponding to the selected object and the contents corresponding to the latest stage according to the screening parameter to obtain to-be-recommended contents satisfying a first quantity ratio; wherein the first quantity ratio is a quantity ratio between the contents corresponding to the latest stage and the contents corresponding to other stages; wherein the screening parameter comprises at least one of a hotness and an update time.

11. The method of claim 9, wherein, When the determined content recommendation strategy is the second content screening strategy, the method further comprises: performing at least one of the following processing: screening the contents corresponding to the selected object and the contents corresponding to the latest stage according to the screening parameter to obtain to-be-recommended contents satisfying a third quantity ratio; wherein the third quantity ratio is a quantity ratio between the contents corresponding to the latest stage and the contents corresponding to the selected object; screening the contents corresponding to the selected object and the contents corresponding to the latest stage according to the screening parameter to obtain to-be-recommended contents satisfying a first quantity ratio; wherein the first quantity ratio is a quantity ratio between the contents corresponding to the latest stage and the contents corresponding to other stages; wherein the screening parameter comprises at least one of a hotness and an update time.

12. The method according to any one of claims 1 to 11, characterized in that, The method further comprises: obtaining a content recommendation strategy shared by different account attributes and different preference information; The screening of to-be-recommended contents from the content library according to the content recommendation strategy comprises: screening to-be-recommended contents from the content library according to the content recommendation strategy determined by the account attribute and the preference information, and the shared content recommendation strategy.

13. The method according to any one of claims 1 to 11, characterized in that, When the determined content recommendation strategy comprises content recommendation strategies corresponding to a plurality of content blocks respectively, the updating of the content presentation interface according to the screened to-be-recommended contents comprises: performing the following processing for each content block: presenting to-be-recommended contents screened according to the content recommendation strategy corresponding to the content block in the content block of the content presentation interface. 14.A content recommendation apparatus based on artificial intelligence, characterized by, The device comprises: an acquisition module configured to acquire historical operation behaviors of a user account logging into a content presentation interface in response to a touch operation of the content presentation interface; an attribute determination module configured to determine an account attribute of the user account according to the historical operation behaviors; a preference determination module configured to acquire account preference behaviors conforming to the account attribute, and perform preference analysis processing according to the account preference behaviors to obtain preference information of the user account on a plurality of contents in a content library; The screening module is configured to determine a content recommendation strategy according to the account attribute and the preference information, and screen a plurality of contents in the content library according to a content screening strategy in the content recommendation strategy to obtain candidate contents; when a content replacement strategy in the content recommendation strategy is a first content replacement strategy, the candidate contents are taken as to-be-recommended contents; the first content replacement strategy is configured to replace a plurality of contents presented last time; when the content replacement strategy in the content recommendation strategy is a second content replacement strategy, the candidate contents and contents other than reading contents in the plurality of contents presented last time are taken together as to-be-recommended contents; the second content replacement strategy is configured to replace the reading contents presented last time. The updating module is configured to update the content presentation interface according to the screened to-be-recommended contents.

15. An electronic device, comprising: The memory is configured to store executable instructions. The processor is configured to execute the executable instructions stored in the memory to implement the artificial intelligence-based content recommendation method in any one of claims 1 to 13. The memory is configured to store executable instructions.

16. A computer-readable storage medium, characterized in that, The computer program or instructions are configured to be executed by the processor to implement the artificial intelligence-based content recommendation method in any one of claims 1 to 13.

17. A computer program product comprising computer programs or instructions, characterized in that, ​

Images