Server and sample data generation method

Abstract

The invention relates to a server and a sample data generation method, and relates to the technical field of artificial intelligence. The server comprises a first communication device which is in communication connection with a terminal, and at least one processor which is connected with the first communication device and is configured to respond to a media asset recommendation request of the terminal and generate an association identifier according to a receiving timestamp and a target user identifier; obtaining current user features from a feature library according to the target user identifier; calling an initial recommendation model, and selecting target media assets from the candidate media assets according to the current user features and the current media asset features; associating the association identifier, the current user feature and the current media asset feature, and sending recommendation information to the terminal; receiving user log data of the terminal; and under the condition that a sample construction condition is met, constructing sample data for training an initial recommendation model based on the current user characteristics, the response behavior data and the current media asset characteristics associated with the association identifier. And the accuracy of the input content of the recommendation model is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a server and a method for generating sample data. Background Technology

[0002] Over-the-Top (OTT) video intelligent recommendation is based on OTT terminals, such as smart TVs, set-top boxes and other large-screen devices. It analyzes user data through recommendation models and accurately pushes personalized video content, such as TV series, movies and variety shows, to users so that the recommended content is more in line with individual user preferences.

[0003] In related technologies, the sample data used to train recommendation models cannot accurately reflect the relevant features of the recommendation scenario, thus making the content output by the recommendation model unable to well match individual user preferences. Summary of the Invention

[0004] This application provides a server and a method for generating sample data, which enables the sample data used to train a recommendation model to accurately reflect the relevant features of the recommendation scenario, thereby making the content output by the recommendation model better match individual user preferences.

[0005] In a first aspect, some embodiments provide a server, including:

[0006] The first communication device is configured to communicate with a terminal.

[0007] and at least one processor, connected to the first communication device, and configured to:

[0008] In response to a media asset recommendation request from a terminal, an associated identifier corresponding to the media asset recommendation request is generated based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request.

[0009] Based on the target user identifier, the current user features of the target user are obtained from the feature library; wherein, the feature library stores user features corresponding to different user identifiers;

[0010] Invoke the initial recommendation model and select the target media asset from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets;

[0011] Associate the association identifier, the current user characteristics, and the current media asset characteristics of the target media asset; and,

[0012] Send recommendation information to the terminal; wherein the recommendation information includes the association identifier and the media asset information of the target media asset;

[0013] Receive user log data from the terminal; wherein, the user log data includes response behavior data associated with the association identifier, and the response behavior data is the behavior data of the target user towards the target media asset;

[0014] Under the condition of sample construction, sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset.

[0015] In the above embodiments, the server generates an association identifier in response to the terminal's media asset recommendation request, and associates the association identifier with the current user characteristics and the current media asset characteristics of the target media asset. It also constructs sample data by combining this with the user's response behavior data regarding the target media asset. Since the sample data contains real-time user characteristics and real-time media asset characteristics at the inference time, as well as the corresponding real user response behavior, the sample data accurately reflects the relevant characteristics of the recommendation scenario. This allows the recommendation model, trained on this sample data, to learn a feature-behavior mapping relationship that better fits the real recommendation scenario, thereby improving the accuracy of the recommendation model and enabling it to output recommendation content that better matches individual user preferences.

[0016] Secondly, some embodiments also provide a terminal, including:

[0017] monitor;

[0018] The second communication device is configured to communicate with the server.

[0019] and at least one controller, connected to the second communication device, and configured to:

[0020] Send a media asset recommendation request to the server;

[0021] The server receives recommendation information; wherein the recommendation information includes an association identifier and media asset information of the target media asset, the association identifier is generated by the server based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request; the target media asset is selected by the server from the candidate media assets by calling the initial recommendation model, based on the current user characteristics of the target user and the current media asset characteristics of the candidate media assets;

[0022] Control the display to show the media asset information, and obtain the target user's response behavior data based on the media asset information in response to the target media asset;

[0023] Associate the association identifier with the response behavior data, and generate user log data including the response behavior data;

[0024] The user log data is sent to the server; wherein the user log data is used by the server to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset, when the sample construction conditions are met.

[0025] In the above embodiments, the terminal sends a media asset recommendation request to the server. The server selects a target media asset based on the current user characteristics and the current media asset characteristics of the candidate media assets. It also generates an association identifier based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the request. The server then sends recommendation information to the terminal, containing the association identifier and the target media asset information. The terminal displays the recommendation information. Simultaneously, the terminal records the user's response behavior data to the target media asset in real time and associates this data with the association identifier to generate user log data, which is then sent to the server. This allows the server to obtain the real user behavior data corresponding to the recommendation request. The server then trains a recommendation model based on the real-time user characteristics and real-time media asset characteristics included at the inference time, as well as the real user response behavior corresponding to these characteristics. This enables the recommendation model to output content that better matches user preferences, improving the user experience.

[0026] Thirdly, some embodiments also provide a sample data generation method applied to a server, including:

[0027] In response to a media asset recommendation request from a terminal, an associated identifier corresponding to the media asset recommendation request is generated based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request.

[0028] Based on the target user identifier, the current user features of the target user are obtained from the feature library; wherein, the feature library stores user features corresponding to different user identifiers;

[0029] Invoke the initial recommendation model and select the target media asset from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets;

[0030] Associate the association identifier, the current user characteristics, and the current media asset characteristics of the target media asset; and,

[0031] Send recommendation information to the terminal; wherein the recommendation information includes the association identifier and the media asset information of the target media asset;

[0032] Receive user log data from the terminal; wherein, the user log data includes response behavior data associated with the association identifier, and the response behavior data is the behavior data of the target user towards the target media asset;

[0033] Under the condition of sample construction, sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset.

[0034] Fourthly, some embodiments also provide a sample data generation method for use in a terminal, including:

[0035] Send a media asset recommendation request to the server;

[0036] The server receives recommendation information; wherein the recommendation information includes an association identifier and media asset information of the target media asset, the association identifier is generated by the server based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request; the target media asset is selected by the server from the candidate media assets by calling the initial recommendation model, based on the current user characteristics of the target user and the current media asset characteristics of the candidate media assets;

[0037] Control the display to show the media asset information, and obtain the target user's response behavior data based on the media asset information in response to the target media asset;

[0038] Associate the association identifier with the response behavior data, and generate user log data including the response behavior data;

[0039] The user log data is sent to the server; wherein the user log data is used by the server to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset, when the sample construction conditions are met.

[0040] Fifthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the methods provided in some embodiments of the third or fourth aspect.

[0041] In a sixth aspect, a computer program product is provided, comprising: a computer program that, when executed by a processor, implements the steps of the methods provided in some embodiments of the third or fourth aspect. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram illustrating the operational scenarios between a terminal and a control device provided in some embodiments of this application;

[0044] Figure 2 A schematic diagram of the hardware configuration of a terminal provided in some embodiments of this application;

[0045] Figure 3 This is a schematic diagram of the hardware configuration of the control device provided in some embodiments of this application;

[0046] Figure 4 A schematic diagram illustrating the software configuration of a terminal provided in some embodiments of this application;

[0047] Figure 5 A flowchart illustrating a sample data generation method provided in some embodiments of this application;

[0048] Figure 6 This is a schematic diagram of the structure of the initial recommendation model provided in some embodiments of this application;

[0049] Figure 7 This is a schematic diagram illustrating the process of constructing sample data provided in some embodiments of this application;

[0050] Figure 8 This is a flowchart illustrating the timing verification of current feature and response behavior data provided in some embodiments of this application;

[0051] Figure 9 A schematic diagram illustrating the process of selecting a target media asset from candidate media assets for some embodiments of this application;

[0052] Figure 10 A schematic flowchart illustrating a sample data generation method provided in other embodiments of this application;

[0053] Figure 11 Timing diagrams of sample data generation methods provided in some embodiments of this application;

[0054] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0055] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0056] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0057] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0058] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0059] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0060] In some embodiments of this application, the sample data generation method can be implemented through interaction between a server and a terminal. A terminal generally refers to a device with screen display and data processing capabilities. For example, terminals include, but are not limited to, smart TVs, mobile terminals, computers, monitors, advertising screens, wearable devices, virtual reality devices, and augmented reality devices.

[0061] Figure 1 This is a schematic diagram illustrating an operational scenario between a terminal and a control device provided in some embodiments of this application. For example... Figure 1 As shown, a user can operate the terminal 200 via touch operation, mobile terminal 300, and control device 100. For example, control device 100 can be a remote control, stylus, gamepad, etc.

[0062] The mobile terminal 300 can serve as a control device for human-computer interaction between the user and the terminal 200. The mobile terminal 300 can also serve as a communication device for establishing a communication connection with the terminal 200 and exchanging data. In some embodiments, the mobile terminal 300 can install software applications with the terminal 200 to establish a connection and communicate via a network communication protocol, achieving one-to-one control operations and data communication. It can also transmit audio and video content displayed on the mobile terminal 300 to the terminal 200 for synchronized display.

[0063] like Figure 1 The document also shows that terminal 200 can communicate with server 400 via various communication methods. Terminal 200 can communicate via local area network (LAN), wireless local area network (WLAN), and other networks.

[0064] Terminal 200 can provide broadcast television reception functionality; it can also be equipped with intelligent network television functionality that provides computer support, including but not limited to network television, smart television, Internet Protocol Television (IPTV), etc.

[0065] In some embodiments, such as Figure 1 As shown, a media asset recommendation request can be sent to the terminal 200 via the remote control 100 and the mobile terminal 300. The terminal 200 can send a media asset recommendation request to the server 400 via a communication device.

[0066] Figure 2 Provided for some embodiments of this application Figure 1 Hardware configuration block diagram of the terminal 200.

[0067] In some embodiments, terminal 200 may include at least one of tuner 210, communication device 220, detector 230, device interface 240, controller 250, display 260, audio output device 270, memory, power supply, and user input interface 280.

[0068] In some embodiments, detector 230 is used to acquire signals from the external environment or to interact with the outside world. For example, detector 230 includes a light receiver, a sensor for acquiring ambient light intensity; or, detector 230 includes an image acquisition device, such as a camera, which can be used to acquire external environmental scenes, user attributes, or user interaction gestures; or, detector 230 includes a sound acquisition device, such as a microphone, for receiving external sounds.

[0069] In some embodiments, the display 260 includes display function components for presenting images and driving components for driving image display. The display 260 is used to receive and display image signals output from the controller 250. For example, the display 260 can be used to display video content, image content, menu control interface components, and user control UI interfaces, etc.

[0070] In some embodiments, the communication device 220 is a component used to communicate with external devices or the server 400 according to various communication protocol types. The terminal 200 may have multiple communication devices 220 depending on the supported communication methods. For example, when the terminal 200 supports wireless network communication, it may have a communication device 220 with WiFi functionality. When the terminal 200 supports Bluetooth connection communication, it needs to have a communication device 220 with Bluetooth functionality.

[0071] The communication device 220 enables the terminal 200 to communicate with external devices or the server 400 via wireless or wired connections. Wired connections utilize data cables, interfaces, or other components to connect the terminal 200 to external devices. Wireless connections utilize wireless signals or wireless networks. The terminal 200 can establish a direct connection with external devices or indirectly through gateways, routers, or other connection devices.

[0072] In some embodiments, the controller 250 may include at least one of a central processing unit, a video processor, an audio processor, a graphics processor, and a power processor, and a first to an nth interface for input / output. The controller 250 controls the operation of the terminal and responds to user operations through various software control programs stored in memory. The controller 250 controls the overall operation of the terminal 200.

[0073] In some embodiments, the controller 250 and the tuner 210 may be located in different separate devices, that is, the tuner 210 may also be located in an external device of the main device where the controller 250 is located, such as an external set-top box.

[0074] In some embodiments, a user can input user commands through a graphical user interface (GUI) displayed on a monitor 260, and the user input interface receives the user input commands through the GUI.

[0075] In some embodiments, the audio output device 270 can be the built-in speaker of the terminal 200 or an external audio output device connected to the terminal 200. For the external audio output device connected to the terminal 200, the terminal 200 may also be provided with an external audio output terminal, through which the audio output device can be connected to the terminal 200 to output sound from the terminal 200.

[0076] In some embodiments, the user input interface 280 can be used to receive instructions from user input.

[0077] Figure 3 Provided for some embodiments of this application Figure 1 Hardware configuration block diagram of the central control device. (Example) Figure 3 As shown, the control device 100 may include: a controller 110, a communication interface 130, a user input / output interface, a memory, and a power supply.

[0078] The control device 100 is configured as a control terminal 200, capable of receiving user input operation commands and converting the operation commands into commands that the terminal 200 can recognize and respond to, thus acting as an intermediary for interaction between the user and the terminal 200.

[0079] In some embodiments, the control device 100 may be an intelligent device. For example, the control device 100 may be equipped with various applications of the control terminal 200 according to user needs.

[0080] In some embodiments, such as Figure 1 As shown, the mobile terminal 300 or other smart electronic devices can perform similar functions to the control device 100 after the application of the control terminal 200 is installed.

[0081] The controller 110 includes a processor 112, random-access memory (RAM) 113, read-only memory (ROM) 114, a communication interface 130, and a communication bus. The controller 110 is used to control the operation of the control device 100, as well as the communication and cooperation between internal components and the external and internal data processing functions.

[0082] Under the control of the controller 110, the communication interface 130 enables communication of control signals and data signals with the terminal 200. The communication interface 130 may include at least one of other near-field communication modules such as WiFi chip 131, Bluetooth module 132, and Near Field Communication (NFC) module 133.

[0083] User input / output interface 140, wherein the input interface includes at least one of other input interfaces such as microphone 141, touchpad 142, sensor 143, and button 144.

[0084] In some embodiments, the control device 100 includes at least one of a communication interface 130 and an input / output interface 140. The control device 100 is configured with the communication interface 130, such as a WiFi, Bluetooth, or NFC module, which can encode user input commands via WiFi, Bluetooth, or NFC protocols and send them to the terminal 200.

[0085] The memory 190 is used to store various operating programs, data, and applications for driving and controlling the control device 100 under the control of the controller. The memory 190 can also store various control signal instructions input by the user.

[0086] The power supply 180 is used to provide operating power support for the various components of the control device 100 under the control of the controller.

[0087] In some embodiments, to enable user interaction, terminal 200 may run an operating system. The operating system is a computer program used to manage and control the hardware and software resources of terminal 200. The operating system can (control the terminal) provide a user interface, allowing users to interact with terminal 200 and supporting the running of various applications.

[0088] It should be noted that the operating system can be a native operating system based on a specific operating platform, a third-party operating system that is deeply customized based on a specific operating platform, or an independent operating system specifically developed for the terminal.

[0089] An operating system can be divided into different modules or levels based on the functions it implements, for example... Figure 4 As shown, in some embodiments, the system is divided into four layers, from top to bottom: the Applications layer (referred to as the "Application Layer"), the Application Framework layer (referred to as the "Framework Layer"), the System Library layer, and the Kernel layer.

[0090] In some embodiments, the application layer provides services and interfaces for applications, enabling the terminal 200 to run applications and interact with the user based on the applications. The application layer may run at least one application, which may be a built-in Windows program, system settings program, or clock program of the operating system; or it may be an application developed by a third-party developer. In specific implementations, the applications in the application layer include, but are not limited to, the examples above.

[0091] The framework layer provides application programming interfaces (APIs) and a programming framework for applications. The application framework layer includes predefined functions. It acts as a central processing unit, determining the actions taken by applications within the application layer. Through the API, applications can access system resources and obtain system services during execution.

[0092] In this embodiment, the application framework layer includes a view system, managers, and content providers. The view system designs and implements the application's interface and interactions, and includes lists, grids, text boxes, and buttons. The managers include at least one of the following modules: an activity manager for interacting with all running activities in the system; a location manager for providing system services or applications with access to system location services; a package manager for retrieving various information related to application packages currently installed on the device; a notification manager for controlling the display and clearing of notification messages; and a window manager for managing icons, windows, toolbars, wallpapers, and desktop widgets on the user interface.

[0093] In some embodiments, the Activity Manager manages the lifecycle of individual applications and common navigation and back functions, such as controlling application exit, opening, and back actions. The Window Manager manages all window programs, such as obtaining the screen size, determining if a status bar is present, locking the screen, capturing the screen, and controlling changes to the display window, such as shrinking the display window, shaking the display, or distorting the display.

[0094] In some embodiments, the system runtime library layer can provide support for the framework layer. When the framework layer is used, the operating system runs the instruction library contained in the system runtime library layer, such as the C / C++ instruction library, to implement the functions to be performed by the framework layer.

[0095] In some embodiments, the kernel layer is a functional layer situated between the hardware and software of the terminal 200. The kernel layer can implement functions such as hardware abstraction, multitasking, and memory management. For example, ... Figure 4As shown, hardware drivers can be configured in the kernel layer. The kernel layer can contain at least one of the following drivers: audio driver, display driver, Bluetooth driver, camera driver, WIFI driver, USB driver, High-Definition Multimedia Interface (HDMI) driver, sensor driver (such as fingerprint sensor, temperature sensor, pressure sensor, etc.), and power driver, etc.

[0096] It should be noted that the above examples are merely a simple division of operating system functions and do not limit the specific operating system form of the terminal 200 in this application embodiment. Depending on factors such as the terminal's functions and the type of operating system, the number of layers and the specific type of layers contained in the operating system can be expressed in other forms.

[0097] The above embodiments illustrate the hardware / software architecture and functional implementation of a terminal 200. In some embodiments, the terminal 200 is configured with a user interaction application, and the display in the terminal 200 can display the interactive interface of the user interaction application; the controller in the terminal can receive control signals from the control device 100 and control the display in the terminal to display the user interaction application according to the control signals; the user can operate the terminal 200 through the control device 100 and perform user interaction operations in the user interaction application; or the user's voice can be received by the sound collector in the detector 230 to interact with the user and complete user interaction operations in the user interaction application.

[0098] Over-the-Top (OTT) video intelligent recommendation is based on OTT terminals, such as smart TVs, set-top boxes and other large-screen devices. It analyzes user data through recommendation models and accurately pushes personalized video content, such as TV series, movies and variety shows, to users so that the recommended content is more in line with individual user preferences.

[0099] In related technologies, the sample data used to train recommendation models cannot accurately reflect the relevant features of the recommendation scenario, thus making the content output by the recommendation model unable to well match individual user preferences.

[0100] For example, in related technologies, the sample data used to train recommendation models does not use real-time user features and real-time media asset features corresponding to the inference time, but rather user features and media asset features from historical time corresponding to the inference time. This makes the sample data unable to accurately reflect the relevant features of the recommendation scenario, resulting in low accuracy of the trained recommendation model, and consequently, the content output by the recommendation model cannot well match individual user preferences.

[0101] Based on this, in some embodiments, a sample data generation method is provided. This sample data generation method can be implemented by a server.

[0102] In an exemplary embodiment, taking the application of the sample data generation method to a processor in a server as an example, the server includes a first communication device and at least one processor. The communication connection between the first communication device and the terminal can be a wired communication connection or a wireless communication connection, and at least one processor is connected to the first communication device.

[0103] The first communication device can be a hardware component or a combined hardware and software module in a server used for data interaction with the terminal. It supports wired communication (such as Ethernet) or wireless communication (such as Wi-Fi, 5G) and is capable of sending and receiving data. The terminal refers to a device used by the user that has video playback and network communication functions, such as smart TVs, set-top boxes, tablets, and other OTT terminal devices.

[0104] like Figure 5 As shown, Figure 5 A method for generating sample data for a processor in a server is provided, which may include the following steps:

[0105] S501, in response to the media asset recommendation request from the terminal, generates an association identifier corresponding to the media asset recommendation request based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request.

[0106] The media asset recommendation request is a signal or data packet sent by the terminal to the server requesting the server to recommend media assets. This request contains relevant information needed to implement the recommendation, such as recommended media assets like TV series, movies, and variety shows. The receiving timestamp is the specific time the server receives the media asset recommendation request from the terminal, usually expressed with millisecond precision, such as "2025-11-01 19:30:00.123". The target user identifier is information used to uniquely identify the target user, such as the user device ID or user registration account ID, for example, "TV-1234567890ABCDEF". The association identifier is a unique identifier generated by the server to associate the media asset recommendation request, current user characteristics, current media asset characteristics of the target media, and user response behavior data, ensuring the traceability of the correspondence between the data.

[0107] For example, the system can monitor request data received from the terminal by the first communication device in real time. When a media asset recommendation request conforming to a preset format is detected, the operation of generating an association identifier is triggered. For instance, after the server's processor receives a media asset recommendation request containing a user ID and scene identifier from the smart TV via listening port 8080, it generates an association identifier corresponding to the media asset recommendation request based on the receiving timestamp of the request and the target user identifier carried in the request. For example, the receiving timestamp and the target user identifier can be directly concatenated to obtain the association identifier, or the receiving timestamp and the target user identifier can be converted and then concatenated to obtain the association identifier.

[0108] In some optional implementations, the receiving timestamp of the media asset recommendation request, the target user identifier carried in the media asset recommendation request, and a random number can be concatenated to obtain the association identifier corresponding to the media asset recommendation request.

[0109] The random number is an irregular sequence of numbers generated by the server processor to ensure that the associated identifier remains unique under the same user ID and the same receiving timestamp, such as "001122" and "987654".

[0110] For example, the processor can obtain the received timestamp and convert it into a preset format, such as the string "20251102101530789" after removing the delimiter; then obtain the target user identifier, such as "TV-987654", and then generate a random number of a preset number of bits, such as a 6-bit or 8-bit random number, through a random number generation algorithm, such as a pseudo-random number generator or a cryptographically secure random number generator.

[0111] Furthermore, the three elements are concatenated into a string-based association identifier according to a preset order such as "target user identifier + receiving timestamp + random number" or "receiving timestamp + target user identifier + random number"; alternatively, a globally unique association identifier can be generated through a distributed ID generator, such as a generator based on the snowflake algorithm.

[0112] For example, if the target user identifier is “TV-987654”, the receiving timestamp is “20251102101530789”, and the random number is “654321”, the associated identifier obtained by concatenating them in sequence is “TV-98765420251102101530789654321”.

[0113] For example, the target user identifier carried in the media asset recommendation request sent by the terminal is "USER-5678". The server receives the request with a timestamp of "2025-11-03 14:20:15.333". The processor converts the timestamp to "20251103142015333", generates an 8-bit random number "87654321" using an encrypted secure random number generator, and then concatenates them in the order of "target user identifier + timestamp + random number" to obtain the associated identifier "USER-56782025110314201533387654321". This identifier can uniquely correspond to this media asset recommendation request.

[0114] In the above embodiments, an association identifier is generated by concatenating the received timestamp, the target user identifier, and a random number. The received timestamp and the target user identifier are used to ensure that the association identifier corresponds to the specific recommendation request and user. The random number is used to avoid the problem of duplicate association identifiers when the same user sends multiple recommendation requests at the same time, thereby ensuring the uniqueness of the association identifier. This provides a reliable identifier foundation for the subsequent association of current user features, target media asset features, and user response behavior data, making the associated current user features, target media asset features, and user response behavior data more accurate, and thus improving the accuracy of sample data construction.

[0115] S502, based on the target user identifier, obtain the current user characteristics of the target user from the feature library.

[0116] The feature database is a data storage system used to store user features corresponding to different user identifiers and media asset features of various media assets. It can be implemented using a database and supports fast querying and data updates. The feature database stores user features corresponding to different user identifiers.

[0117] Current user characteristics refer to the relevant characteristic information of the target user when the server receives a media asset recommendation request, including the user's age, membership level, recent viewing types, viewing duration, current network status, and viewing progress. For example, "30 years old, gold member, recently prefers science fiction series, current network is Wi-Fi, and viewing progress is 30% of the movie 'A'".

[0118] For example, based on the target user identifier, the corresponding user feature data can be retrieved from the feature library through a query interface provided by the feature library. The query interface can be a Structured Query Language (SQL) query statement, an Application Programming Interface (API) interface for a non-relational database, etc.; for instance, the processor can send a query command to the feature library to obtain the target user's current user features such as age, membership level, and recent movie viewing preferences.

[0119] S503, invoke the initial recommendation model, and select the target media asset from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets.

[0120] The initial recommendation model is an algorithmic model used to select target media assets from candidate media assets based on user characteristics and media asset characteristics. It can be configured according to different scenarios and columns, such as collaborative filtering-based models and deep learning ranking models.

[0121] Candidate media assets are the collection of all media assets stored on the server that are available for recommendation, such as all movies, TV series, variety shows, and other content in the server's video library.

[0122] Current media asset characteristics refer to the relevant feature information of candidate media assets when the server receives a media asset recommendation request, including director, actors, genre, duration, real-time popularity, number of clicks in the past hour, and number of current online viewers. For example, a science fiction movie has a real-time popularity of 8.5 points and 5,000 clicks in the past hour.

[0123] The target media asset is the media asset that the initial recommendation model selects from the candidate media assets that is suitable for recommendation to the target user, based on the current user characteristics and the current media asset characteristics of the candidate media assets.

[0124] For example, an initial recommendation model can be started via a model call interface (such as a RESTful API or a Remote Procedure Call (RPC) interface). The current user characteristics and the current characteristics of candidate media assets are input into the model. The model calculates the matching degree between the user characteristics and the characteristics of each candidate media asset, such as cosine similarity or the score of the deep learning model, and selects a predetermined number of media assets with the highest matching degree as target media assets. For instance, by inputting the user's characteristics of "science fiction preference 0.8, gold membership level" and the candidate media asset characteristics such as "type and popularity" into the initial recommendation model, the model outputs the top 5 science fiction movies as target media assets.

[0125] In some alternative implementations, see [link to relevant documentation]. Figure 6 , Figure 6A schematic diagram of an initial recommendation model is provided, wherein the initial recommendation model includes:

[0126] The input layer receives the current user features and the current media asset features of the candidate media assets, forming an original feature set;

[0127] The embedding layer transforms various raw features into multi-dimensional vectors, thereby digitizing the features;

[0128] The fusion layer normalizes the vectors, eliminates scale differences, and then concatenates them into a global feature vector.

[0129] Hidden layers use multi-layer neural networks to mine non-linear relationships between features and output deep features rich in matching patterns.

[0130] The output layer calculates the matching score of candidate media assets based on deep features, sorts the target media assets by score, and generates recommendation reasons as needed.

[0131] S504, associates the associated identifier, current user characteristics, and current media asset characteristics of the target media asset; and sends recommendation information to the terminal.

[0132] The recommendation information includes the associated identifier and the media asset information of the target media asset. The recommendation information is the data sent by the server to the terminal, which includes the associated identifier and the media asset information of the target media asset. The media asset information includes the media asset name, cover image, introduction, playback link and other content.

[0133] For example, the three elements can be stored in a temporary database in the form of key-value pairs, with the association identifier as the key and the current user characteristic and current media asset characteristic as the value; alternatively, the three elements can be assembled into a data record according to a preset data format and written into a dedicated association data table. For example, the corresponding user characteristic "age 30, science fiction preference 0.8" and media asset characteristic "Movie B, science fiction genre, popularity 8.5" can be stored with the association identifier as the key.

[0134] Then, the associated identifier and target media asset information are encapsulated according to a preset communication protocol (such as Hypertext Transfer Protocol (HTTP) or WebSocket), and sent to the terminal through the first communication device. For example, the recommendation information is encapsulated as JSON data and sent to the smart TV via a Hypertext Transfer Protocol POST (HTTP POST) request.

[0135] In some optional implementations, the association identifier is added to a preset field of the target media asset's media asset information to obtain recommendation information, which is then sent to the terminal.

[0136] Among them, the preset field is a predefined field in the media asset information used to store the association identifier. This field can be a specific attribute in the media asset information data structure. For example, in JSON format media asset information, the preset field name is "associate_id".

[0137] For example, the server's processor can first parse the media asset information data of the target media asset, such as JSON format data, find a preset field, such as "associate_id", and then assign the generated association identifier to the preset field, completing the fusion of the association identifier and the media asset information. The media asset information containing the association identifier is then encapsulated according to the communication protocol supported by the terminal and sent to the terminal through a first communication device, ensuring that the terminal can correctly receive and parse the association identifier and the media asset information. The communication protocol includes, but is not limited to, HTTP, WebSocket, etc.

[0138] In the above embodiments, the association identifier is added to a preset field of the media asset information of the target media asset and sent to the terminal. This enables the terminal to easily extract the association identifier from the received media asset information and associate it with the user's response behavior data for the media asset to generate user log data. This simplifies the binding process between the association identifier and the response behavior data on the terminal side, ensures the accuracy of the association identifier in the user log data, and improves the efficiency of sample data construction.

[0139] S505 receives user log data from the terminal.

[0140] User log data consists of behavioral data of target users on the recommended target media assets, recorded on the terminal. This data is associated with an association identifier and used by the server to build training samples. User log data includes response behavior data associated with the association identifier. Response behavior data is the behavioral data of target users on the target media assets, that is, the specific behavioral records generated by target users on the terminal for the target media assets, such as clicking to play, viewing duration, adding to favorites, skipping, and not clicking after exposure, for example, "clicked to play and watched for 25 minutes" or "not clicked after exposure".

[0141] For example, the first communication device can receive log data sent by the terminal in real time. The server's processor parses and verifies the received data, extracting response behavior data associated with the association identifier. For instance, after the first communication device receives log data from the smart TV containing the association identifier and the "click to play" action, the processor parses out the response behavior data.

[0142] S506, Under the condition of meeting the sample construction conditions, sample data for training the initial recommendation model is constructed based on the current user characteristics associated with the association identifier, response behavior data, and current media asset characteristics of the target media asset.

[0143] The sample construction conditions are the preset conditions that the server must meet to start the sample data construction process, ensuring the feasibility and efficiency of sample construction. The sample data, used to train the initial recommendation model, consists of current user features, response behavior data, and the current media asset features of the target media asset.

[0144] In some optional implementations, the sample construction condition can be that the server's current remaining computing resources are greater than or equal to the computing resources required to construct the sample data, or that the current time is within a preset time period.

[0145] The remaining computing resources refer to the computing resources of the server that are not occupied at a certain moment, including the remaining CPU utilization, remaining memory capacity, and remaining disk I / O bandwidth, such as 60% remaining CPU utilization and 8GB remaining memory.

[0146] The computing resources required to build sample data are the computing resources that the server needs to occupy during the process of building sample data. They are estimated in advance based on factors such as the size of the sample data and the complexity of feature processing. For example, building a batch of sample data requires 30% CPU utilization and 4GB of memory.

[0147] The preset time period is a time period that the server has pre-set to be suitable for building sample data. It is usually a time period with low server load and low business request volume, such as 2:00-4:00 am.

[0148] For example, system monitoring tools, such as the `top` command on Linux or Task Manager on Windows, can be used to obtain real-time information on server computing resources, such as CPU utilization and memory usage. This allows for the calculation of remaining computing resources and the acquisition of a pre-estimated amount of computing resources needed to build sample data. The two are then compared. For instance, if the current remaining CPU utilization is 50% and the required CPU utilization is 30%; and the remaining memory is 10GB while the required memory is 5GB, it can be determined that the remaining computing resources exceed the required computing resources.

[0149] For example, the current system time can also be obtained and compared with the start and end times of a preset time period to determine whether the current time is within that range. For instance, if the preset time period is 02:00-04:00 and the current system time is 03:30, it can be determined that the current time is within the preset time period; if the current system time is 10:00, it can be determined that it is not within the preset time period.

[0150] If the remaining computing resources are greater than or equal to the required computing resources, or if the current time is within a preset time period, the processor can initiate the sample data construction process. If neither condition is met, the process will not start until the conditions are met. For example, if the remaining computing resources are sufficient, sample data construction can be initiated even if the time period is not specified; if the remaining computing resources are insufficient, but the time period is specified and the server load is low, sample data construction can also be initiated.

[0151] For example, suppose the server's preset sample construction conditions include that the remaining computing resources are greater than or equal to the required computing resources and the current time is between 01:00 and 03:00. At one moment, the server detects through system monitoring tools that the current CPU utilization is 70%, assuming the resources required to build the sample data are 35%, and the remaining memory is 12GB, while the memory required to build the sample data is 6GB. That is, the current remaining computing resources are greater than the required computing resources, satisfying the first sample construction condition, and the processor starts the sample data construction process. At another moment, the server's current remaining CPU utilization is 25%, less than the required 35%, but the current time is 02:15, which is within the preset time period, and the server's business request volume is extremely low at this time, so the processor can also start the sample data construction process.

[0152] In the above embodiments, using whether the server has sufficient remaining computing resources or whether the current time is within a preset time period as a condition for sample construction ensures that the sample data construction process does not consume too many server resources, thus avoiding affecting the server's performance in processing core business such as online media asset recommendation requests. At the same time, selecting a time period with low server load to construct sample data can improve the efficiency and stability of sample construction, enabling sample data to be generated in a timely manner for model training. This provides stable sample data support for the continuous optimization of the initial recommendation model without affecting the normal operation of the server.

[0153] For example, when the sample construction conditions are met, the server's processor can extract the current user features and current media asset features corresponding to the associated identifier from the associated storage, combine them with the response behavior data in the user log data, and assemble them into sample data according to a preset sample format (such as feature vector + label). The sample format can be a combination of feature vectors and labels. For example, the user feature vector, the media asset feature vector, and the positive label corresponding to "click to play" can be combined to form positive sample data.

[0154] For example, suppose a user opens a media asset viewing app on a smart TV (terminal). The media asset viewing app sends a media asset recommendation request to the cloud server, carrying the target user identifier "TV-987654" and the scene identifier "Homepage Recommendation". The server's first communication device receives the request via Wi-Fi, with a reception timestamp of "2025-11-02 10:15:30.789". The processor responds to the request and generates an associated identifier "TV-987654_20251102101530789" using the "target user identifier + reception timestamp" method.

[0155] Then, the processor retrieves the current user characteristics from the feature database based on the target user identifier "TV-987654": "Age 28, Platinum membership level, watched 3 comedy variety shows in the last 3 days, current network Wi-Fi".

[0156] Furthermore, the processor calls the initial recommendation model RankModel-V3 corresponding to the homepage recommendation, inputting the current user characteristics and the current media asset characteristics of all variety shows in the candidate media asset library into the model. For example, "Variety C" is a comedy, with a real-time popularity of 9.2 and 8,000 clicks in the last hour; "Variety D" is a comedy, with a real-time popularity of 8.8 and 6,000 clicks in the last hour, etc. The model calculates the three variety shows with the highest matching degree as the target media assets.

[0157] The processor stores the association identifier, current user characteristics, and current media asset characteristics of the three target media assets in the database as key-value pairs. Then, it encapsulates the association identifier and media asset information of the target media assets (such as name, cover image, and playback link) into recommendation information and sends it to the smart TV through the first communication device.

[0158] After receiving the information, the smart TV displays the variety show information. The user clicks on "Variety C" and watches it for 30 minutes. The terminal generates user log data based on the association identifier and the response behavior data, and sends it to the server.

[0159] After receiving the data, the server detects that the remaining computing resources are greater than the resources required to build the sample, thus determining that the sample construction conditions are met. It then extracts the corresponding current user features and the current media asset features of the target media asset from the database, and combines them with the response behavior data of "click to watch for 30 minutes" to construct positive sample data for training the initial recommendation model.

[0160] In the above embodiments, the server generates an association identifier in response to the terminal's media asset recommendation request, and associates the association identifier with the current user characteristics and the current media asset characteristics of the target media asset. It also constructs sample data by combining this with the user's response behavior data regarding the target media asset. Since the sample data contains real-time user characteristics and real-time media asset characteristics at the inference time, as well as the corresponding real user response behavior, the sample data accurately reflects the relevant characteristics of the recommendation scenario. This allows the recommendation model, trained on this sample data, to learn a feature-behavior mapping relationship that better fits the real recommendation scenario, thereby improving the accuracy of the recommendation model and enabling it to output recommendation content that better matches individual user preferences.

[0161] In some optional implementations, feature collection switches can be configured according to business scenarios. By specifying the "scenario identifier and collection status" through the configuration file, different feature data can be collected for different scenarios to construct sample data. For example, for the "homepage recommendation" scenario, full collection can be enabled, for the "topic page" scenario, incremental collection can be enabled, and for the "test environment" scenario, collection can be disabled to avoid invalid data consuming resources.

[0162] In some alternative implementations, to improve the reliability of the constructed sample data, the current user characteristics, current media asset characteristics, and response behavior data used to construct the sample data can be subject to time-series verification. If the verification passes, the sample data can then be constructed based on the current user characteristics, current media asset characteristics, and response behavior data.

[0163] See Figure 7 , Figure 7 A flowchart for constructing sample data is provided, which includes the following steps:

[0164] S701, obtain the first timestamp and the second timestamp.

[0165] The first timestamp is the timestamp at which the current feature is acquired, i.e., the specific time record when the server acquires the current feature. Its precision can be consistent with the received timestamp, for example, "2025-11-05 16:40:22.555". The current feature can include both current user features and current media asset features. The second timestamp is the timestamp recorded in the user log data for the generation of response behavior data, i.e., the specific time record of the target user generating response behavior data, as recorded by the terminal, for example, "2025-11-05 16:40:30.123".

[0166] For example, when retrieving current user features and current media asset features from the feature library, the time when the retrieval operation is completed can be recorded as the first timestamp; after receiving user log data sent by the terminal, the second timestamp recording the time when the response behavior occurred can be parsed from the log data. For example, if the processor completes feature retrieval at 16:40:22.555, the first timestamp is "2025-11-05 16:40:22.555"; if the user click behavior is parsed from the user log data at 16:40:30.123, the second timestamp is "2025-11-05 16:40:30.123".

[0167] S702, based on the first timestamp and the second timestamp, verifies the temporal sequence between the current feature and response behavior data.

[0168] Among them, the time sequence verification is used to verify whether the acquisition time of the current feature is earlier than the occurrence time of the user's response behavior, to ensure that the current feature is the feature on which the user generates the response behavior, and to avoid the distortion of sample data caused by the reversal of the time sequence.

[0169] For example, the first and second timestamps can be converted into timestamp values, such as Unix timestamps, and then their values ​​can be compared to determine whether the acquisition time of the current feature is earlier than the occurrence time of the response behavior. For instance, if the Unix timestamp corresponding to the first timestamp is 1751762422555 and the Unix timestamp corresponding to the second timestamp is 1751762430123, the comparison shows that the value of the first timestamp is less than the value of the second timestamp, and the timing verification passes.

[0170] S703, if the timing verification passes, determine the time difference between the first timestamp and the second timestamp.

[0171] The time difference refers to the time interval between the first timestamp and the second timestamp, calculated in milliseconds or seconds. For example, "7568 milliseconds" means 7.568 seconds.

[0172] For example, the difference between the second timestamp and the first timestamp can be calculated to obtain the time difference. For instance, subtracting the first timestamp 1751762430123 from the second timestamp 1751762422555 gives a time difference of 7568 milliseconds.

[0173] S704, when the time difference is less than the time threshold, construct sample data for training the initial recommendation model based on the current user characteristics associated with the association identifier, response behavior data, and the current media asset characteristics of the target media asset.

[0174] The time threshold is a preset critical time interval used to determine the strength of the correlation between the current feature and the response behavior data. It can be configured according to business scenario requirements. For example, it can be set to 30 seconds or 60 seconds. No specific limit is made on the time threshold here.

[0175] For example, the calculated time difference can be compared with a preset time threshold. If the time difference is less than the time threshold, the current feature and response behavior data are considered to have a strong correlation, and sample data is constructed based on the current user feature, response behavior data, and current media asset feature associated with the correlation identifier. If the time difference is greater than or equal to the time threshold, the correlation is considered to be weak, and the sample data is not constructed or is marked as a low-quality sample. For example, if the time threshold is 30 seconds (30,000 milliseconds), and the time difference of 7,568 milliseconds is less than 30,000 milliseconds, sample data is constructed.

[0176] In the above embodiments, by obtaining the first and second timestamps and performing time sequence verification, it is ensured that the current features used to construct the sample data are the features at the time when the user generates a response behavior, thus avoiding sample data distortion caused by the reversal of time sequence. At the same time, by setting a time threshold and judging whether the time difference is less than the time threshold, feature data with strong correlation to response behavior is filtered out, further improving the quality of sample data. This enables the initial recommendation model after training to more accurately learn the real correlation between user features, media asset features and user behavior, thereby improving the recommendation accuracy of the recommendation model.

[0177] In some alternative implementations, see [link to relevant documentation]. Figure 8 , Figure 8 A flowchart illustrating the temporal verification of current feature and response behavior data is provided, specifically including the following steps:

[0178] S801, compare the first timestamp and the second timestamp.

[0179] For example, the first and second timestamps can be converted into time strings of the same format for character comparison. For instance, they can be converted into strings in the format "YYYY-MM-DDHH:MM:SS.sss" or into Unix timestamps, which are numerical in form, and then their sizes can be compared. For example, the first timestamp "2025-11-07 09:10:05.222" corresponds to the Unix timestamp 1751833805222, and the second timestamp "2025-11-07 09:10:10.777" corresponds to the Unix timestamp 1751833810777. The timestamps can be compared by comparing the size of the two values.

[0180] S802. When the first timestamp is less than the second timestamp, it is determined that the timing check between the current feature and the response behavior data passes.

[0181] Exemplarily, when the value corresponding to the first timestamp is less than the value corresponding to the second timestamp, it indicates that the acquisition time of the current feature is earlier than the occurrence time of the user's response behavior, and the current feature is the basis for the user to generate the response behavior. Therefore, it is determined that the timing check passes. For example, if 1751833805222 <1751833>810777, then it is determined that the timing check passes.

[0182] S803. When the first timestamp is not less than the second timestamp, it is determined that the timing check between the current feature and the response behavior data fails.

[0183] Exemplarily, when the value corresponding to the first timestamp is greater than or equal to the value corresponding to the second timestamp, it indicates that the acquisition time of the current feature is later than or equal to the occurrence time of the user's response behavior, and the current feature cannot be the basis for the user to generate the response behavior. Therefore, it is determined that the timing check fails. For example, if the Unix timestamp corresponding to the first timestamp is 1751833815333 and the Unix timestamp corresponding to the second timestamp is 1751833810777, and 1751833815333 > 1751833810777, then it is determined that the timing check fails.

[0184] Exemplarily, assume that the first timestamp is "2025-11-08 15:30:20.444" and the second timestamp is "2025-11-08 15:30:18.666". Convert the two timestamps to Unix timestamps, which are 1751895020444 and 1751895018666 respectively. By comparison, it is found that 1751895020444 is greater than 1751895018666, that is, the first timestamp is not less than the second timestamp. Therefore, it is determined that the timing check between the current feature and the response behavior data fails, and sample data is not constructed based on this set of data. If the second timestamp is "2025-11-08 15:30:25.888" and the corresponding Unix timestamp is 1751895025888, and by comparison, 1751895020444 is less than 1751895025888, the timing check passes, and then sample data can be constructed based on this set of data.

[0185] In the above embodiments, by directly comparing the magnitudes of the first timestamp and the second timestamp to determine whether the timing passes, it is possible to quickly screen out the feature data and response behavior data with reasonable time order, avoid using the invalid data with reversed time order for sample construction, achieve the time logic consistency between features and behaviors in the sample data, and further improve the reliability of the sample data.

[0186] In some optional implementations, the media asset recommendation request may also include scene identifiers and column identifiers. During the process of selecting target media assets from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets, the initial recommendation model can be determined based on the column identifier, which can improve the targeting of the model recommendation. Furthermore, the scene identifier can also be used as input data for model recommendation, making the input data more dimensional, thereby enabling the recommendation model to obtain more dimensional data features and improve the accuracy of the model output data.

[0187] For example, see Figure 9 , Figure 9 A flowchart illustrating the process of selecting a target media asset from candidate media assets is provided, which includes the following steps:

[0188] S901, Based on the column identification, select the initial recommendation model corresponding to the column identification from different candidate recommendation models.

[0189] The scenario identifier is used to identify the business scenario corresponding to the media asset recommendation request, such as "homepage recommendation", "children's zone recommendation", "movie channel recommendation", "special event recommendation", etc. Different scenarios correspond to different recommendation needs and user groups.

[0190] The column identifier is used to identify the specific column corresponding to the media asset recommendation request. A business scenario can contain multiple columns, such as "Popular TV Series", "You May Like", and "New Releases" in the "Homepage Recommendation" scenario, which correspond to different media asset recommendation ranges and recommendation logics.

[0191] Different candidate recommendation models are multiple recommendation models pre-deployed on the server, each corresponding to a different category. Each model is optimized for the recommendation needs of the corresponding category. For example, the "Popular Dramas" category corresponds to a recommendation model based on popularity ranking, while the "You May Like" category corresponds to a recommendation model based on user collaborative filtering.

[0192] For example, a mapping table between column identifiers and candidate recommendation models can be pre-established, such as column 1 corresponding to model A and column 2 corresponding to model B. After obtaining the column identifiers in the media asset recommendation request, this mapping table can be queried to select the corresponding initial recommendation model. For instance, if the column identifier is "You May Like," and the corresponding recommendation model obtained from the mapping table is a collaborative filtering model, then this collaborative filtering model can be used as the initial recommendation model for media asset recommendation.

[0193] S902, invoke the initial recommendation model, match the target feature with the current media asset features of the candidate media assets, and select the target media asset from the candidate media assets based on the matching results.

[0194] The target features are the feature set used to input the initial recommendation model for media asset matching, including current user features and scene identifiers. Scene identifiers enable the model to make accurate recommendations based on specific scenes.

[0195] For example, the target feature and the current media asset features of the candidate media assets can be input into the selected initial recommendation model. The initial recommendation model calculates the matching score between the target feature and each candidate media asset feature through a preset matching algorithm, and then sorts them from high to low according to the score, selecting the top-ranked candidate media assets as the target media assets.

[0196] For example, the target features are user preferences for science fiction and homepage recommendations. The current media asset features of the candidate media assets include the type and popularity of each movie. The model calculates that movies with higher popularity in the science fiction genre have higher matching scores, and the top 10 movies with the highest scores are selected as target media assets.

[0197] Accordingly, the above S506 may include the following steps:

[0198] Based on the current user characteristics associated with the association identifier, response behavior data, current media asset characteristics of the target media asset, and scene identifier, sample data is constructed for training the initial recommendation model.

[0199] When constructing sample data, in addition to associating current user characteristics, response behavior data, and current media asset characteristics, scene identifiers can also be added as one of the features to the sample data. This allows the sample data to contain scene information, which can be used to train the model's ability to make recommendations based on the scene. For example, the sample data may include "Scene identifier: Children's Zone Recommendation", "User characteristics: Age 5, Preference for animation", "Media asset characteristics: 'Animated Film E', Animation type, Suitable for children aged 3-6", and "Response behavior: Watched for 15 minutes".

[0200] For example, suppose the media asset recommendation request sent by the terminal includes the scene identifier "Homepage Recommendation" and the column identifier "New Movie Express". The processor queries the mapping table between column identifiers and candidate recommendation models and finds that the initial recommendation model corresponding to the "New Movie Express" column is a dedicated model for new movie recommendations. This dedicated model focuses on the media asset's release time and recent popularity. Then, it obtains the target user's current user characteristics "age 25, likes romance movies", combines these characteristics with the scene identifier "Homepage Recommendation" to form the target feature, and inputs it into the dedicated model for new movie recommendations. Simultaneously, it inputs the current media asset characteristics of romance movies released in the past month from the candidate media assets, such as "Movie I", romance genre, release date 2025-10-20, and real-time popularity 8.6. The model matches and scores the target feature and each candidate media asset feature, selecting the top 5 new movies as the target media assets.

[0201] Furthermore, the server receives user log data sent by the terminal, such as when a user clicks to watch "Movie I". Based on the current user characteristics associated with the association identifier, the response behavior data of clicking to watch, the current media asset characteristics of "Movie I", and the scene identifier homepage recommendation, the processor constructs sample data to train the initial recommendation model corresponding to the "New Movie Express" section.

[0202] In the above embodiments, the corresponding initial recommendation model is selected based on the column identification, enabling the recommendation model to adapt to the recommendation needs of specific columns and improving the targeting of media asset recommendations. Simultaneously, incorporating scene identifiers into target features for media asset matching allows the model to make recommendations based on business scenarios, further improving recommendation accuracy. Furthermore, adding scene identifiers when constructing sample data ensures that the training samples contain scene information, allowing the initial recommendation model to learn the correlation between user characteristics, media asset features, and user behavior in different scenarios during training. This enables subsequent recommendations to output media assets that better suit user needs based on different scenarios, improving the recommendation model's scene adaptability and recommendation effectiveness.

[0203] In some alternative implementations, in order to ensure that the training samples can comprehensively cover whether users are interested in or not interested in the recommended target media assets, positive and negative sample data can be constructed based on response behavior data.

[0204] For example, if the response behavior data is the first response behavior data, then positive sample data for training the initial recommendation model is constructed based on the current user characteristics associated with the association identifier and the current media asset characteristics of the target media asset; wherein, the first response behavior data includes behavior data that characterizes the target user’s attention to the target media asset, that is, the user’s behavior records that generate positive feedback to the recommended target media asset, such as clicking to play, watching for a duration exceeding a preset threshold, such as 10 minutes, adding to favorites, adding to favorites, sharing, etc.

[0205] Positive sample data is used to train the initial recommendation model. It represents the sample data that users are interested in media assets. It consists of current user features, current media asset features of the target media asset, and positive behavior labels. It enables the model to learn which feature combinations will lead users to pay attention to media assets.

[0206] If the response behavior data is the second response behavior data, then negative sample data for training the initial recommendation model is constructed based on the current user characteristics associated with the association identifier and the current media asset characteristics of the target media asset. The second response behavior data includes behavior data that represents the target user's lack of attention to the target media asset, that is, the user's behavior records of generating negative feedback or no effective feedback to the recommended target media asset, such as not clicking after exposure, watching for less than the preset threshold after clicking, exiting within 1 minute and not due to network problems, skipping, etc.

[0207] Negative sample data is used to train the initial recommendation model. It represents the sample data that indicates that users are not interested in media assets. It consists of current user features, current media asset features of the target media asset, and negative behavior labels. It enables the model to learn which feature combinations will cause users to disregard media assets.

[0208] For example, rules for judging first and second response behavior data can be pre-defined. For instance, a viewing duration of 10 minutes or more is considered first response behavior data, while no clicks are made after exposure, making it second response behavior data. Upon receiving user log data, the type of response behavior data can be determined based on these rules. For example, if the user log data records a viewing duration of 15 minutes, it meets the rules for first response behavior data and is therefore classified as first response behavior data; if it records no clicks within 30 seconds of exposure, it meets the rules for second response behavior data and is therefore classified as second response behavior data.

[0209] For example, if the response behavior data is the first response behavior data, then the current user characteristics associated with the association identifier and the current media asset characteristics of the target media asset can be processed using feature engineering, such as feature normalization and feature encoding. The processed features are then combined with a positive behavior label, such as "1," to form positive sample data. For instance, the current user characteristics are "age 35, likes action movies," the target media asset characteristics are "Movie N," action type, popularity 9.5, and the positive behavior label is "1," which are then combined to form positive sample data.

[0210] For example, if the response behavior data is the second response behavior data, the current user characteristics associated with the association identifier and the current media asset characteristics of the target media asset can be processed by feature engineering, and then combined with a negative behavior label, such as "0", to form negative sample data. For example, the current user characteristics are "age 35 years old, likes action movies", the target media asset characteristics are "TV series H", historical romance genre, popularity 8.0, and the negative behavior label is "0", which are combined to form negative sample data.

[0211] For example, suppose the server recommends the media asset "Animated Film G" to the terminal. The user log data from the terminal shows the response behavior data as "clicked to play and watched for 40 minutes." The server's processor, based on a preset rule (assuming a viewing time greater than or equal to 10 minutes), identifies this as the first response behavior data. Then, the server's processor extracts the associated current user characteristics ("age 22, membership level silver, recently clicked 3 superhero movies") and the target media asset's current media asset characteristics ("superhero type, real-time popularity 9.8, 10,000 clicks in the last hour"), performs feature normalization, and combines it with the positive label "1" to construct positive sample data. If the terminal's response behavior data is "no clicks within 1 minute of exposure," the processor identifies this as the second response behavior data, extracts the corresponding current user characteristics and the target media asset's current media asset characteristics, and combines them with the negative label "0" to construct negative sample data.

[0212] In the above embodiments, by constructing positive and negative sample data according to the type of user response behavior data, the training samples can comprehensively cover the situations where users are interested in and uninterested in media assets. This allows the initial recommendation model to learn user behavioral preferences under different feature combinations, clarify which feature combinations will lead users to pay attention to media assets and which will lead users to not pay attention, thereby optimizing the model's recommendation decision logic, improving the model's accuracy in judging user preferences, and ultimately outputting recommendation content that is more in line with individual user preferences.

[0213] In some optional implementations, a sample data generation method is provided. This sample data generation method can be implemented by a terminal.

[0214] In an exemplary embodiment, taking the application of the sample data generation method to a controller in a terminal as an example, the terminal includes a display and a second communication device, as well as at least one controller; the second communication device is configured to communicate with a server; and at least one controller is connected to the second communication device.

[0215] The second communication device can be a hardware component or a combined hardware and software module in the terminal used for data interaction with the server. It supports wired communication (such as Ethernet) or wireless communication (such as Wi-Fi, 5G) and can send and receive data. The display is a hardware device in the terminal used to display media information, such as the screen of a smart TV or the display of a tablet computer, capable of displaying images, text, and other content. The controller is the core component in the terminal used to control the operation of various components, perform data processing, and issue commands, such as the main control chip of a smart TV.

[0216] See Figure 10 , Figure 10A method for generating sample data for a controller used in a terminal is provided, which may include the following steps:

[0217] S1001, send a media asset recommendation request to the server.

[0218] For example, the controller can generate a media asset recommendation request containing information such as the target user identifier, scene identifier, and column identifier based on user actions, such as the user opening a media asset viewing app and entering the homepage recommendation page, or preset trigger conditions, such as sending recommendation requests at regular intervals. This request is then sent to the server via a second communication device. For instance, when a user opens the homepage of a media asset viewing app on a smart TV, the controller generates a media asset recommendation request containing the user device ID "TV-112233", the scene identifier "Homepage Recommendation", and the column identifier "Popular Variety Shows", and sends it to the server via Wi-Fi.

[0219] S1002, Receive recommendation information sent by the server.

[0220] The recommendation information includes an association identifier and the media asset information of the target media asset. The association identifier is generated by the server based on the receiving timestamp in the media asset recommendation request and the target user identifier carried in the media asset recommendation request. The target media asset is the media asset selected by the server from the candidate media assets based on the current user characteristics of the target user and the current media asset characteristics of the candidate media assets when the initial recommendation model is called.

[0221] For example, the second communication device monitors the data sent by the server in real time. Upon receiving recommendation information, it transmits it to the controller. The controller parses the recommendation information and extracts the association identifier and media asset information of the target media assets. For instance, if the second communication device receives recommendation information in JSON format from the server, the controller parses out the association identifier "TV-112233_20251109134050222_567890" and the names, cover images, and playback links of three popular variety shows.

[0222] S1003, control the display to show media asset information, and obtain the target user's response behavior data based on the media asset information.

[0223] For example, the controller converts the parsed target media asset information into signals recognizable by the display, such as video or image signals, and sends them to the display. The controller then controls the display to show the cover image, name, and other information of the media asset according to a preset layout, such as a grid or list layout. For instance, the controller can control a smart TV to display the cover images and names of three popular variety shows in a grid layout for the user to view and select. The controller can monitor the user's response behavior on the display in real time regarding the target media asset, such as clicking, watching, and saving, and obtain the target user's response behavior data.

[0224] S1004, associate the association identifier with the response behavior data, and generate user log data including the response behavior data.

[0225] For example, the controller can record response behavior data and the corresponding occurrence time, then associate the response behavior data with the associated identifier, and generate user log data according to a preset log format, including fields such as the associated identifier, behavior type, behavior time, and media asset ID. For instance, if a user clicks on a variety show and watches it for 20 minutes, the controller records the response behavior data "click to play, watching time 20 minutes" and the associated identifier "TV-112233_20251109134050222_567890", generating user log data containing this information.

[0226] S1005, Send user log data to the server; the user log data is used by the server to construct sample data for training the initial recommendation model based on the current user characteristics associated with the association identifier, response behavior data, and the current media asset characteristics of the target media asset, provided that the sample construction conditions are met.

[0227] For example, the controller sends the generated user log data to the server via a second communication device according to a preset transmission protocol, such as HTTP or TCP, ensuring that the server can receive it in a timely manner and use it for sample data construction. For instance, the controller sends the user log data to the server via an HTTP POST request.

[0228] For example, when a user opens a media asset viewing app on a smart TV, the controller detects that the user has entered the "Movie Channel" page and generates a media asset recommendation request containing the target user identifier "TV-445566", the scene identifier "Movie Channel Recommendation", and the column identifier "High-scoring Movies". This request is sent to the server via a second communication device (Wi-Fi module). After processing, the server returns recommendation information, which is received by the second communication device and transmitted to the controller. The controller parses out the associated identifier "TV-445566_20251110162030888_987654" and media asset information for five high-scoring movies, including "Movie L" and "Movie M". This information includes, for example, the title, cover image, synopsis, and playback link. The controller controls the smart TV's display to show this movie information in a list layout. If the user clicks on "Movie L" and watches it for one hour, the controller records this response behavior data, associates it with the associated identifier to generate user log data, and sends it to the server via the second communication device for the server to build sample data.

[0229] In the above embodiments, the terminal provides users with personalized media asset recommendation services by sending media asset recommendation requests to the server, receiving and displaying recommendation information; at the same time, the terminal records the user's response behavior data to the target media asset in real time, and generates user log data by associating it with the association identifier and sending it to the server, so that the server can obtain the real user behavior data corresponding to the recommendation request, and the recommendation model can output content that is more in line with user preferences, thereby improving the user experience.

[0230] In some alternative implementations, see [link to relevant documentation]. Figure 11 , Figure 11 A timeline diagram of a sample data generation method is provided, which specifically includes the following steps:

[0231] S1101, the terminal sends a media asset recommendation request to the server.

[0232] S1102, the server responds to the terminal's media asset recommendation request by generating an association identifier corresponding to the media asset recommendation request based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request.

[0233] S1103, the server retrieves the current user characteristics of the target user from the feature database based on the target user identifier.

[0234] S1104, the server calls the initial recommendation model and selects the target media asset from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets.

[0235] S1105, the server associates the associated identifier, the current user characteristics, and the current media asset characteristics of the target media asset.

[0236] S1106, the server sends recommendation information to the terminal.

[0237] S1107, the terminal receives recommendation information sent by the server.

[0238] S1108, the terminal control display shows media asset information and obtains response behavior data of the target user based on the media asset information.

[0239] S1109, the terminal associates the associated identifier with the response behavior data and generates user log data including the response behavior data.

[0240] S1110, the terminal sends user log data to the server.

[0241] S1111, the server receives user log data from the terminal.

[0242] S1112, Under the condition of meeting the sample construction conditions, the server constructs sample data for training the initial recommendation model based on the current user characteristics associated with the association identifier, response behavior data, and the current media asset characteristics of the target media asset.

[0243] The specific implementation methods of S1101-S1112 are the same as those in the above embodiments, and will not be repeated here.

[0244] Based on the same inventive concept, some embodiments also provide a sample data generation apparatus for implementing the sample data generation method described above. This apparatus is applied to a server, and the solution provided by this apparatus is similar to the implementation scheme described in the above method. Therefore, the specific limitations in one or more sample data generation apparatus embodiments provided below can be found in the limitations of the sample data generation method described above, and will not be repeated here.

[0245] In one exemplary embodiment, a sample data generation apparatus is provided, applied to a server, comprising:

[0246] The generation module 10 is used to respond to the media asset recommendation request from the terminal and generate the associated identifier corresponding to the media asset recommendation request based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request.

[0247] The acquisition module 20 is used to acquire the current user features of the target user from the feature library based on the target user identifier; wherein, the feature library stores user features corresponding to different user identifiers;

[0248] Selection module 30 is used to call the initial recommendation model and select the target media asset from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets;

[0249] The first sending module 40 is used to associate the association identifier, the current user characteristics, and the current media asset characteristics of the target media asset; and to send recommendation information to the terminal; wherein the recommendation information includes the association identifier and the media asset information of the target media asset;

[0250] The first receiving module 50 is used to receive user log data from the terminal; wherein, the user log data includes response behavior data associated with the association identifier, and the response behavior data is the behavior data of the target user towards the target media asset;

[0251] The construction module 60 is used to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, response behavior data, and the current media asset features of the target media asset, provided that the sample construction conditions are met.

[0252] In the above embodiments, the server generates an association identifier in response to the terminal's media asset recommendation request, and associates the association identifier with the current user characteristics and the current media asset characteristics of the target media asset. It also constructs sample data by combining this with the user's response behavior data regarding the target media asset. Since the sample data contains real-time user characteristics and real-time media asset characteristics at the inference time, as well as the corresponding real user response behavior, the sample data accurately reflects the relevant characteristics of the recommendation scenario. This allows the recommendation model, trained on this sample data, to learn a feature-behavior mapping relationship that better fits the real recommendation scenario, thereby improving the accuracy of the recommendation model and enabling it to output recommendation content that better matches individual user preferences.

[0253] In one exemplary embodiment, the generation module 10 is specifically used for:

[0254] The associated identifier corresponding to the media asset recommendation request is obtained by concatenating the receiving timestamp of the media asset recommendation request, the target user identifier carried in the media asset recommendation request, and the random number.

[0255] In an exemplary embodiment, the first sending module 40 is specifically used for:

[0256] The association identifier is added to the preset field of the target media asset's media asset information to obtain the recommendation information, and then the recommendation information is sent to the terminal.

[0257] In one exemplary embodiment, the construction module 60 specifically includes:

[0258] The acquisition unit is used to acquire a first timestamp and a second timestamp; wherein, the first timestamp is the timestamp for acquiring the current feature, and the second timestamp is the timestamp for generating response behavior data recorded in the user log data; the current feature includes the current user feature and the current media asset feature;

[0259] The verification unit is used to verify the temporal sequence between the current feature and response behavior data based on the first timestamp and the second timestamp.

[0260] The determining unit is used to determine the time difference between the first timestamp and the second timestamp if the timing verification passes.

[0261] The building unit is used to construct sample data for training the initial recommendation model based on the current user characteristics associated with the associated identifier, response behavior data, and the current media asset characteristics of the target media asset, when the time difference is less than the time threshold.

[0262] In one exemplary embodiment, the verification unit is specifically used for:

[0263] Compare the first timestamp and the second timestamp; if the first timestamp is less than the second timestamp, the time sequence check between the current feature and the response behavior data is passed; if the first timestamp is not less than the second timestamp, the time sequence check between the current feature and the response behavior data is failed.

[0264] In an exemplary embodiment, the media asset recommendation request further includes scene identifiers and column identifiers; the selection module 30 is specifically used for:

[0265] Based on the column identification, select the initial recommendation model corresponding to the column identification from different candidate recommendation models; call the initial recommendation model to match the target features and the current media features of the candidate media assets, and select the target media asset from the candidate media assets based on the matching results;

[0266] Target features include current user characteristics and scene identifiers; Module 60 is specifically used for:

[0267] Based on the current user characteristics associated with the association identifier, response behavior data, current media asset characteristics of the target media asset, and scene identifier, sample data is constructed for training the initial recommendation model.

[0268] In one exemplary embodiment, the construction module 60 is specifically used for:

[0269] If the response behavior data is the first response behavior data, then positive sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier and the current media asset features of the target media asset; wherein, the first response behavior data includes behavior data representing the target user's attention to the target media asset; if the response behavior data is the second response behavior data, then negative sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier and the current media asset features of the target media asset; wherein, the second response behavior data includes behavior data representing the target user's non-attention to the target media asset.

[0270] In one exemplary embodiment, the sample construction conditions include any of the following:

[0271] The server's current remaining computing resources are greater than or equal to the computing resources required to build the sample data; the current time is within a preset time period.

[0272] Based on the same inventive concept, some embodiments also provide a sample data generation apparatus for implementing the sample data generation method described above. This apparatus is applied to a terminal, and the solution provided by this apparatus is similar to the solution described in the above method. Therefore, the specific limitations in one or more sample data generation apparatus embodiments provided below can be found in the limitations of the sample data generation method described above, and will not be repeated here.

[0273] In one exemplary embodiment, a sample data generation apparatus is provided, applied to a terminal, comprising:

[0274] The second sending module 70 is used to send a media asset recommendation request to the server;

[0275] The second receiving module 80 is used to receive recommendation information sent by the server. The recommendation information includes an association identifier and media asset information of the target media asset. The association identifier is generated by the server based on the receiving timestamp in the media asset recommendation request and the target user identifier carried in the media asset recommendation request. The target media asset is the media asset selected by the server from the candidate media assets based on the current user characteristics of the target user and the current media asset characteristics of the candidate media assets when the server calls the initial recommendation model.

[0276] Display module 90 is used to control the display to show media asset information and to acquire the target user's response behavior data based on the media asset information.

[0277] The generation module 100 is used to associate the association identifier with the response behavior data and generate user log data including the response behavior data;

[0278] The second sending module 70 is used to send user log data to the server; wherein, the user log data is used by the server to construct sample data for training the initial recommendation model based on the current user characteristics associated with the association identifier, response behavior data and the current media asset characteristics of the target media asset, when the sample construction conditions are met.

[0279] In the above embodiments, the terminal sends a media asset recommendation request to the server and receives recommendation information containing an association identifier and target media asset information. The association identifier is generated by the server based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request. The terminal displays the recommendation information, providing users with personalized media asset recommendation services. At the same time, the terminal records the user's response behavior data to the target media asset in real time and associates it with the association identifier to generate user log data, which is then sent to the server. This allows the server to obtain the real user behavior data corresponding to the recommendation request, enabling the recommendation model to output content that better matches user preferences and improve the user experience.

[0280] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 12As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores PCB product-related data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a sample data generation method.

[0281] Those skilled in the art will understand that Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0282] In one alternative embodiment, Figure 12 The computer device shown can be the aforementioned server.

[0283] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0284] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0285] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0286] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0287] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A server, characterized in that, include: The first communication device is configured to communicate with a terminal. and at least one processor, connected to the first communication device, and configured to: In response to a media asset recommendation request from a terminal, an associated identifier corresponding to the media asset recommendation request is generated based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request. Based on the target user identifier, the current user features of the target user are obtained from the feature library; wherein, the feature library stores user features corresponding to different user identifiers; The initial recommendation model is invoked, and the target media asset is selected from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets; Associate the association identifier, the current user characteristics, and the current media asset characteristics of the target media asset; and, Send recommendation information to the terminal; wherein the recommendation information includes the association identifier and the media asset information of the target media asset; Receive user log data from the terminal; wherein, the user log data includes response behavior data associated with the association identifier, and the response behavior data is the behavior data of the target user towards the target media asset; Under the condition of sample construction, sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset.

2. The server according to claim 1, characterized in that, When the processor generates an association identifier corresponding to the media asset recommendation request based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request, it is configured as follows: The associated identifier corresponding to the media asset recommendation request is obtained by concatenating the receiving timestamp of the media asset recommendation request, the target user identifier carried in the media asset recommendation request, and the random number.

3. The server according to claim 1, characterized in that, When the processor executes the command to send recommendation information to the terminal, it is configured to: The associated identifier is added to a preset field in the media asset information of the target media asset to obtain recommendation information, and the recommendation information is sent to the terminal.

4. The server according to any one of claims 1-3, characterized in that, When the processor executes the process to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset, it is configured to: Obtain a first timestamp and a second timestamp; wherein, the first timestamp is the timestamp for obtaining the current feature, and the second timestamp is the timestamp recorded in the user log data for generating the response behavior data; the current feature includes the current user feature and the current media asset feature of the target media asset; The temporal sequence between the current feature and the response behavior data is verified based on the first timestamp and the second timestamp; If the timing verification passes, the time difference between the first timestamp and the second timestamp is determined; If the time difference is less than a time threshold, sample data for training the initial recommendation model is constructed based on the current user characteristics associated with the association identifier, the response behavior data, and the current media asset characteristics of the target media asset.

5. The server according to claim 4, characterized in that, When the processor performs a time-sequence verification based on the first timestamp and the second timestamp, it is configured to: Compare the first timestamp and the second timestamp; If the first timestamp is less than the second timestamp, the temporal verification between the current feature and the response behavior data is determined to be successful; If the first timestamp is not less than the second timestamp, it is determined that the temporal sequence check between the current feature and the response behavior data fails.

6. The server according to any one of claims 1-3, characterized in that, The media asset recommendation request also includes scene identifiers and column identifiers; when the processor executes the call to the initial recommendation model and selects the target media asset from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets, it is configured as follows: Based on the column identifier, select an initial recommendation model corresponding to the column identifier from different candidate recommendation models; The initial recommendation model is invoked to match the target features with the current media asset features of the candidate media assets, and the target media asset is selected from the candidate media assets based on the matching results; wherein, the target features include the current user features and the scene identifier; When the processor executes the process to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset, it is configured to: Based on the current user characteristics associated with the association identifier, the response behavior data, the current media asset characteristics of the target media asset, and the scene identifier, sample data for training the initial recommendation model is constructed.

7. The server according to any one of claims 1-3, characterized in that, When the processor executes the process to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset, it is configured to: If the response behavior data is the first response behavior data, then positive sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier and the current media asset features of the target media asset; wherein, the first response behavior data includes behavior data characterizing the target user's attention to the target media asset; If the response behavior data is the second response behavior data, then negative sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier and the current media asset features of the target media asset; wherein, the second response behavior data includes behavior data characterizing the target user's lack of interest in the target media asset.

8. The server according to any one of claims 1-3, characterized in that, The sample construction conditions include any of the following: The server's current remaining computing resources are greater than or equal to the computing resources required to construct the sample data; The current time is within the preset time period.

9. A terminal, characterized in that, include: monitor; The second communication device is configured to communicate with the server. and at least one controller, connected to the second communication device, and configured to: Send a media asset recommendation request to the server; The server receives recommendation information; wherein the recommendation information includes an association identifier and media asset information of the target media asset, the association identifier is generated by the server based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request; the target media asset is selected by the server from the candidate media assets by calling the initial recommendation model, based on the current user characteristics of the target user and the current media asset characteristics of the candidate media assets; Control the display to show the media asset information, and obtain the target user's response behavior data based on the media asset information in response to the target media asset; Associate the association identifier with the response behavior data, and generate user log data including the response behavior data; The user log data is sent to the server; wherein the user log data is used by the server to construct sample data for training the initial recommendation model based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset, when the sample construction conditions are met.

10. A method for generating sample data, characterized in that, Applied to servers, including: In response to a media asset recommendation request from a terminal, an associated identifier corresponding to the media asset recommendation request is generated based on the receiving timestamp of the media asset recommendation request and the target user identifier carried in the media asset recommendation request. Based on the target user identifier, the current user features of the target user are obtained from the feature library; wherein, the feature library stores user features corresponding to different user identifiers; The initial recommendation model is invoked, and the target media asset is selected from the candidate media assets based on the current user characteristics and the current media asset characteristics of the candidate media assets; Associate the association identifier, the current user characteristics, and the current media asset characteristics of the target media asset; and, Send recommendation information to the terminal; wherein the recommendation information includes the association identifier and the media asset information of the target media asset; Receive user log data from the terminal; wherein, the user log data includes response behavior data associated with the association identifier, and the response behavior data is the behavior data of the target user towards the target media asset; Under the condition of sample construction, sample data for training the initial recommendation model is constructed based on the current user features associated with the association identifier, the response behavior data, and the current media asset features of the target media asset.

Images