User preference prediction model training method and related apparatus

Abstract

Embodiments of the present application disclose a user preference prediction model training method, a user preference prediction method and related devices. Each set of training sample data includes feature portrait data of a user account and preference data of the user account, the preference data including a preference label of the user for each function of an application program, and co-occurrence information between multiple functions in each function combination of the application program is obtained. An initial prediction model outputs a preference prediction result of the user account for each function of the application program according to the feature portrait data and the co-occurrence information, and the initial prediction model is trained according to the preference prediction result and the preference label to obtain a target prediction model. The target prediction model can predict the preference of the user for multiple functions of the application program by reading the portrait data of the user only once, thereby saving computing resources. The model can mine potential preferences of the user for other functions based on learned correlations between functions, so that the preference prediction is more accurate and comprehensive.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a user preference prediction model training method, a user preference prediction method, and related apparatus. Background Technology

[0002] Applications offer users a variety of functions, allowing them to obtain different user experiences. For example, music software provides users with functions such as music playback, sound effect configuration, social networking, and gaming, enabling users to configure song sound effects while listening to music, or interact with fellow music lovers, play games, and so on.

[0003] Existing solutions train a binary classification model for each function of an application, which then uses user profile data to determine whether a user likes a particular function. However, this approach requires maintaining a separate binary classification model for each function. Therefore, when identifying a user's preference for multiple functions of the application, each of these multiple binary classification models needs to read the user profile data separately, resulting in repeated readings of user profile data and a significant waste of computational resources.

[0004] Secondly, the training and application processes of multiple binary classification models are independent of each other, ignoring the correlation between multiple functions. For example, users who like to use sound effects often also have a strong preference for sound quality. However, each binary classification model is only trained and applied to user profile data corresponding to a single function, resulting in low accuracy of training and recognition of the binary classification models, which affects the prediction of user preferences. Summary of the Invention

[0005] This application provides a user preference prediction model training method, a user preference prediction method, and related apparatus to improve the accuracy of user preference function prediction.

[0006] The first aspect of this application provides a method for training a user preference prediction model, the method comprising:

[0007] Multiple sets of training sample data are obtained. Each set of training sample data includes feature profile data of a user account and preference data of that user account. The preference data includes preference tags of the user account for the functions of the application. The preference tags are used to represent the user account's preference for the functions of the application.

[0008] Obtain co-occurrence information among multiple functions in each function combination of the application. The co-occurrence information is obtained by statistically analyzing the preference data of each user account for multiple functions in the function combination.

[0009] An initial prediction model is obtained, and the multiple sets of training sample data and the co-occurrence information are input into the initial prediction model so that the initial prediction model outputs the user account's preference prediction result for each function of the application based on the feature profile data and the co-occurrence information.

[0010] The model parameters of the initial prediction model are adjusted based on the preference prediction results and the preference labels until the model training convergence condition is met, at which point the model training stops, and the target prediction model is obtained. The target prediction model is used to predict the preference of any user account for any function of the application.

[0011] A second aspect of this application provides a user preference prediction method, the method comprising:

[0012] A target prediction model is obtained, which is trained by a machine learning algorithm on multiple sets of training sample data and co-occurrence information among multiple functions in each function combination of the application; wherein, each set of training sample data includes feature profile data and user account preference data, the preference data including the user account's preference label for each function of the application; the preference label is used to represent the user account's preference for the application's functions; the co-occurrence information is used to indicate that the user account prefers each function in the function combination;

[0013] Obtain feature profile data of the target user account, and input the feature profile data of the target user account into the target prediction model so that the target prediction model outputs the target user account's preference for each function of the application based on the feature profile data of the target user account.

[0014] A third aspect of this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the methods of the first or second aspect described above.

[0015] A fourth aspect of this application provides a computer storage medium storing instructions that, when executed on a computer, cause the computer to perform the methods described in the first or second aspect.

[0016] A fifth aspect of this application provides a computer program product that, when run on a computer device, causes the computer device to perform the methods described in the first or second aspect.

[0017] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0018] Each set of training sample data includes user account feature profile data and the user account's preference data. The preference data includes the user's preference tags for each function of the application, and obtains the co-occurrence information among multiple functions in each function combination of the application. The co-occurrence information is obtained statistically based on the preference data of each user account for multiple functions in the function combination. Multiple sets of training sample data and this co-occurrence information are input into the initial prediction model. The initial prediction model then outputs the user account's preference prediction result for each function of the application based on the feature profile data and co-occurrence information. The initial prediction model is then trained based on the preference prediction results and preference tags to obtain the target prediction model. Compared with using multiple binary classification models to predict user preferences, using this target prediction model can predict the user's preferences for multiple functions of the application with only one reading of the user's profile data, saving computing resources. Moreover, the model can learn the correlation between multiple functions from the co-occurrence information, and thus, based on the learned correlation between functions, it can explore the user's potential preferences for other functions, making preference prediction more accurate and comprehensive. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the network framework in an embodiment of this application;

[0020] Figure 2 This is a flowchart illustrating the user preference prediction model training method in the embodiments of this application;

[0021] Figure 3 This is an exemplary schematic diagram of the co-occurrence frequency matrix corresponding to multiple functional combinations in the embodiments of this application;

[0022] Figure 4 This is a flowchart illustrating the user preference prediction method in the embodiments of this application;

[0023] Figure 5 This is an exemplary schematic diagram illustrating one display method for the function of the target prediction model to predict the target user account preferences in the client interface in this application embodiment;

[0024] Figure 6 This is a schematic diagram of a computer device in an embodiment of this application. Detailed Implementation

[0025] This application provides a user preference prediction model training method, a user preference prediction method, and related apparatus to improve the accuracy of user preference function prediction.

[0026] Please see Figure 1 The network framework in this embodiment includes:

[0027] The business server 100 and the terminal cluster; the terminal cluster may include: terminal devices 200a, 200b, 200c, ..., terminal devices 200n and other terminal devices.

[0028] The aforementioned business server 100 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud databases, cloud services, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal devices (including terminal devices 200a, 200b, 200c, ..., 200n) can be smartphones, tablets, laptops, desktop computers, PDAs, mobile internet devices (MIDs), wearable devices (such as smartwatches, smart bracelets, etc.), smart computers, smart in-vehicle systems, and other smart terminals.

[0029] The service server 100 can establish communication connections with each terminal device in the terminal cluster, and the terminal devices in the terminal cluster can also establish communication connections with each other. In other words, the service server 100 can establish communication connections with each terminal device among terminal devices 200a, 200b, 200c, ..., 200n. For example, terminal device 200a can establish a communication connection with the service server 100. Terminal devices 200a and 200b can establish a communication connection, and terminal devices 200a and 200c can also establish a communication connection. The communication connection method is not limited; it can be established directly or indirectly through wired communication or wireless communication, etc., depending on the actual application scenario. This application does not impose any restrictions on this.

[0030] It should be understood that, such as Figure 1Each terminal device in the terminal cluster shown can have an application client installed. When the application client runs on each terminal device, it can interact with the business server 100, allowing the business server 100 to receive business data from each terminal device (such as user identity data uploaded by users through the terminal device). This application client can be a music player, karaoke app, browser, social networking app, instant messaging app, live streaming app, game app, short video app, video app, shopping app, novel app, payment app, or any other application client capable of displaying text, images, audio, and video data. The specific application client can be determined based on the actual application scenario requirements and is not limited here. This application client can be a standalone client or an embedded sub-client integrated into a client (such as a music app or karaoke app), depending on the actual application scenario and is not limited here.

[0031] The following will combine Figure 1 The network framework described herein is used to describe the user preference prediction model training method in the embodiments of this application:

[0032] The user preference prediction model training method and user preference prediction method of the embodiments of this application can be applied to the recommendation scenarios of various application functions, such as the recommendation scenarios of various functions of music software. Music software can provide multiple functions, and based on the method of the embodiments of this application, the user's preference for multiple functions of music software can be predicted. Then, when the user uses music software, the user's preferred functions can be recommended on the UI interface of music software. For example, the user's preferred functions can be displayed in the form of a pop-up window when the user enters music software, or the corresponding points of each function can be displayed when the user searches for songs in music software. The user can click on the point corresponding to a function to enter that function.

[0033] Of course, the method of this application embodiment can also be applied to any other application besides music software, and is not limited here.

[0034] Please see Figure 2 One embodiment of the user preference prediction model training method in this application includes:

[0035] 201. Obtain multiple sets of training sample data. Each set of training sample data includes feature profile data of a user account and preference data of the user account. The preference data includes preference tags of the user account for the functions of the application. The preference tags are used to represent the user account's preference for the functions of the application.

[0036] The method of this embodiment can be applied to a computer device, which may be... Figure 1The network framework shown includes a business server 100 or various terminal devices. The computer devices can acquire multiple sets of training sample data. Each set of training sample data includes the feature profile data of the user's account and the user account's preference data. The preference data represents the user account's preference information for each function of the application. For example, it may include preference tags for each function of the application. These preference tags indicate the user account's preference for a particular function. For instance, if the user account prefers function A of the application (i.e., is interested in function A), the number "1" can represent the preference tag for function A. If the user account does not prefer function B of the application (i.e., is not interested in function B), the number "0" can represent the preference tag for function B.

[0037] It should be noted that the preference prediction and analysis of the feature profile data and preference data of the user account described in this embodiment is equivalent to the preference prediction and analysis of the user holding the user account. The same applies when mentioning user account in the following text; anything involving user account involves the user holding the user account.

[0038] User account profile data includes data describing any characteristics of a user account to construct a user account profile, such as data describing user account behavior characteristics, data describing user account attributes such as age, gender, education, and occupation, etc. Each set of training sample data includes profile data and preference data for the same user account. For example, one set of training sample data includes profile data and preference data for user account A, another set of training sample data includes profile data and preference data for user account B, and so on.

[0039] 202. Obtain co-occurrence information among multiple functions in each function combination of the application, wherein the co-occurrence information is obtained by statistically analyzing the preference data of each user account for multiple functions in the function combination;

[0040] Co-occurrence information of a feature set refers to the user account's preference for multiple features within that set. This information can be statistically derived from user account preference data for each feature within the set. For example, co-occurrence information could indicate that a user account prefers all features in the set, dislikes none, or prefers some features but dislikes others. Therefore, co-occurrence information represents the correlation between multiple features of an application. This correlation characterizes the degree of correlation between user preferences for multiple features. For instance, in music software, a user who enjoys using sound effects often has a strong preference for sound quality features; thus, there is a strong correlation between sound effects and sound quality features.

[0041] Therefore, a user account may have preferences for multiple related functions or no preferences at all. This correlation information can be used to infer the user account's potential preferences for application functions. For example, if a user account has a preference for function A of the application, and co-occurrence information indicates that function A and function B of the application are related, then it can be inferred that the user account may also have a potential preference for function B.

[0042] 203. Obtain an initial prediction model, input the multiple sets of training sample data and the co-occurrence information into the initial prediction model, so that the initial prediction model outputs the user account's preference prediction result for each function of the application based on the feature profile data and the co-occurrence information;

[0043] 204. Adjust the model parameters of the initial prediction model according to the preference prediction results and the preference labels until the convergence condition of model training is met, and then stop model training to obtain the target prediction model; the target prediction model is used to predict the preference of any user account for any function of the application.

[0044] The computer device can acquire an initial prediction model, train the initial prediction model, and use the trained model to predict user account preferences. In this embodiment, the initial prediction model can be a multi-task learning network model, such as a multi-gate mixture-of-experts (MMOE) model.

[0045] The initial prediction model outputs a prediction of a user account's preference for each function in the application, based on the correlation between multiple functions described by the user account's feature profile data and co-occurrence information. The model parameters are adjusted based on these predictions and the corresponding preference labels for each function. Through multiple rounds of iterative training and parameter updates, the model continuously learns the correspondence between the user account's feature profile and the user account's preferred functions, as well as the correlations between multiple functions, applying them to user account preference prediction. This gradually brings the model's predictions closer to, or even identical to, the actual preference labels, until the convergence condition is met, at which point training stops, yielding the target prediction model. Therefore, the target prediction model, based on the knowledge and predictive capabilities gained during training, can be used to predict the preference of any user account for any function in the application.

[0046] The conditions for model training convergence can be that the number of model iterations reaches a preset number, or that the loss value of the loss function constructed by the model training meets a preset value or the loss value tends to be stable. This embodiment does not limit the conditions for model training convergence.

[0047] In this embodiment, each set of training sample data includes user account feature profile data and user account preference data. The preference data includes the user's preference tags for each function of the application, and co-occurrence information among multiple functions in each function combination of the application is obtained. The co-occurrence information is statistically obtained based on the preference data of each user account for multiple functions in the function combination. Multiple sets of training sample data and the co-occurrence information are input into the initial prediction model. The initial prediction model then outputs the user account's preference prediction result for each function of the application based on the feature profile data and the co-occurrence information. The initial prediction model is then trained based on the preference prediction results and preference tags to obtain the target prediction model. Compared with using multiple binary classification models to predict user preferences, using this target prediction model can predict the user's preferences for multiple functions of the application by reading the user's profile data only once, saving computing resources. Moreover, the model can learn the correlation between multiple functions from the co-occurrence information, and thus, based on the learned correlation between functions, it can mine the user's potential preferences for other functions, making preference prediction more accurate and comprehensive.

[0048] based on Figure 2In a preferred embodiment of the illustrated example, the user account feature profile data in each set of training sample data includes the user account's access path data for the application's functions, the user account's operational behavior data for the application, and the user account's personal information data. Therefore, when the model outputs a preference prediction result for each function of the application based on this set of training sample data, the initial prediction model can extract behavioral features from the access path data and operational behavior data, extract attribute features from the personal information data, and output the user account's preference prediction result for each function of the application based on the behavioral features, the attribute features, and co-occurrence information.

[0049] The access methods for application functions refer to the ways and means by which a user account accesses the application's functions. For example, the music listening function in a music app can be accessed in multiple ways, such as clicking the listen button at the bottom of the homepage, clicking a playlist, searching for songs, or clicking on recommended songs on the homepage. Access method usage data indicates how a user account uses each access method for application functions, reflecting user preferences when using the application.

[0050] User account behavior data includes data on any action a user takes while using the application, such as clicking on various controls or buttons, or accessing different functions. For example, for music software, user account behavior data could include data on listening to music, searching for songs, or accessing various functions. This data can reflect user preferences and habits when using the application and can serve as a reference for predicting user preferences.

[0051] User account personal information data can include personal attributes such as age, gender, education, and occupation. This data is also related to user preferences. For example, for music software, older users may prefer its social features, while younger users may prefer its built-in mini-games; music creators may prefer the music software's sheet music function. User account personal information data is also related to users' preferences for application functions, and therefore can serve as a reference for predicting user preferences.

[0052] The network structure used in the initial prediction model to extract behavioral and attribute features of user accounts from the feature profile data of user accounts can be a convolutional neural network, a recurrent neural network, or a Transformer network structure, etc., and is not limited here.

[0053] After extracting the behavioral and attribute features of user accounts, the prediction results of user preferences for multiple functions of the application can be output by combining co-occurrence information. Specifically, the neuron weights used to predict the user account's preference for each function of the application can be set in the initial prediction model based on the co-occurrence information, and the prediction results of the user account's preference for each function of the application can be output according to the set neuron weights and the behavioral and attribute features.

[0054] For example, the co-occurrence information of this feature combination can be represented as the co-occurrence frequency. The co-occurrence frequency can be calculated by counting the number m of user accounts whose preference labels for each feature in the feature combination are the same across n user accounts' preference data, and then calculating m / n as the co-occurrence frequency of that feature combination. For example, a feature combination might be feature 1 and feature 2 of a music software. In the preference data of 100 user accounts, the number of user accounts whose preference labels for features 1 and 2 are the same across all 100 user accounts (e.g., all with a label of "1" indicating a preference for both features 1 and 2) is counted. Assuming there are 60 such user accounts, meaning all 60 user accounts prefer and are interested in both features, the co-occurrence frequency of this feature combination is 60%. The co-occurrence frequencies of multiple feature combinations can be used to construct a co-occurrence frequency matrix.

[0055] Suppose an application has multiple functions, including functions A, B, C, and D. These functions can be combined in pairs: "function A + function D" and "function B + function D". The co-occurrence frequency (COF) of the "function A + function D" combination is 0.6, and the COF of the "function B + function D" combination is 0.1. A COF matrix of these functions can then be constructed, as follows: Figure 3As shown, taking the first row of the co-occurrence frequency matrix as an example, the model learns the correlation of the combination of "function A + function D" based on the co-occurrence frequencies in this row. That is, the model inputs the co-occurrence frequencies into the last layer of the network (such as a fully connected layer). This is equivalent to setting the weights of the neurons used to predict the user's preference for functions B and C to 0, while setting the weights of the neurons used to predict the user's preference for functions A and D to 0.6. Then, combining the behavioral and attribute features extracted from the previous layers of the network structure (such as convolutional layers), and based on the set neuron weights, the model predicts the user's preference for functions A, B, C, and D, and outputs the preference prediction result. Similarly, the model can learn the correlation between multiple functions in other function combinations based on other co-occurrence frequencies in the co-occurrence frequency matrix.

[0056] Therefore, by setting neuron weights based on the co-occurrence frequency matrix, functional correlations can be introduced into the model, enabling the model to learn the correlations between multiple functions and make user preference predictions based on this. This can more comprehensively predict users' potential functional preferences, improve the accuracy of user preference predictions, and thus enhance the user's experience with the application.

[0057] based on Figure 2 In another preferred embodiment of the illustrated example, when acquiring user preference data, for each function of the application, behavioral data of multiple user accounts for that function can be acquired, and it can be determined whether the behavioral data of each user account for that function contains data corresponding to the preferred behavior. If the user account's behavioral data for that function contains data corresponding to the preferred behavior, then the user account's preference label for that function is determined to be a first label, which indicates that the user account prefers that function. If the user account's behavioral data for that function does not contain data corresponding to the preferred behavior, then the user account's preference label for that function is determined to be a second label, which indicates that the user account does not prefer that function.

[0058] For example, computer devices can collect user usage data of applications to construct multiple sets of training sample data, such as collecting user behavior log data to determine whether users have shown preferred behavior towards each function of the application. Therefore, for each function of the application, the computer device can acquire behavioral data from multiple users regarding that function, and determine whether each user's behavioral data for that function contains data corresponding to a preferred behavior. If the user's behavioral data for that function contains data corresponding to a preferred behavior, then the user's preference label for that function is determined to be the first label; if the user's behavioral data for that function does not contain data corresponding to a preferred behavior, then the user's preference label for that function is determined to be the second label.

[0059] This preference behavior characterizes actions taken by a user out of interest in an application's functionality. Specifically, a predefined condition can be used; if a user's behavior meets this condition, it is considered a preference behavior; otherwise, it is not. For example, a music app might offer a badge feature that generates matching badges based on the user's listening habits and preferences. The predefined preference behavior for this feature could be clicking the "Wear Badge" button. When the behavioral data includes this action, it's determined that the user has a preference for this badge feature. Alternatively, the predefined preference behavior could be upgrading badges or changing badge skins. Therefore, if the behavioral data includes upgrading badges or changing badge skins, it's determined that the user has a preference for this badge feature. This preference behavior can be configured according to the specific functionality of the application and is not limited here.

[0060] Therefore, in this way, the preferences of each user for multiple functions of the application can be statistically analyzed, thereby forming the preference data of each user. For example, an example of the storage structure of preference data can be: User ID | Preference label of function 1, Preference label of function 2, ..., Preference label of function n.

[0061] Furthermore, in some optional approaches, for each function of the application, the number of positive samples with the preference label of the first label for that function can be counted. Since the number of interested users may differ between different functions—for example, many users may be interested in function 1, while few may be interested in function 2—this difference can lead to excessively large discrepancies between the associations learned during model training and user features, resulting in prediction bias. To mitigate the prediction bias caused by the large differences in the number of users exhibiting preference behavior across functions, positive samples can be preprocessed based on their quantity. If the number of positive samples is greater than a first preset number, undersampling is performed to reduce the number of positive samples for that function; if the number of positive samples is less than a second preset number, oversampling is performed to increase the number of positive samples for that function. The first preset number is greater than or equal to the second preset number. The positive samples obtained through undersampling or oversampling are then used as the preference data for that function.

[0062] Therefore, by applying different sampling methods to positive samples based on the number of positive samples, the amount of preference data among multiple functions can be balanced, improving the model training accuracy and prediction precision, and avoiding poor model training performance and prediction bias caused by excessive differences in the number of positive samples among multiple functions.

[0063] Since each set of training sample data corresponds to a user account, processing a set of training sample data allows the model to output a preference prediction result for that user account. This preference prediction result includes the user account's preference prediction for each function of the application. Similarly, processing multiple sets of training sample data separately allows the model to output multiple user preference prediction results for multiple functions of the application. Therefore, based on... Figure 2 In another preferred embodiment of the illustrated example, when adjusting the model parameters of the initial prediction model based on preference prediction results and preference labels, for each function of the application, a loss function for that function can be constructed based on the preference prediction results of multiple user accounts for that function obtained from multiple sets of training sample data and multiple preference labels for that function in the multiple sets of training sample data. When constructing the loss function corresponding to each function, the model parameters of the initial prediction model can be further adjusted based on the loss values ​​of the multiple loss functions for the multiple functions of the application.

[0064] For example, the loss function Loss for each function of the application can be expressed as follows:

[0065]

[0066] Where N is the number of training samples, i represents the i-th sample, and p(y i ) refers to the model's prediction of user preferences for a feature of the application, y i Refers to the preference tag corresponding to this feature.

[0067] This formula represents the loss function for a single function. Therefore, the loss function for each function of the application can be constructed based on this formula, resulting in multiple loss functions for multiple functions. The adjustment of model parameters is then based on the loss values ​​of these multiple loss functions for the application's multiple functions. As the model is trained and the parameters are adjusted, the loss values ​​of these multiple loss functions for the application's multiple functions gradually decrease, indicating that the model's predictions gradually approach and converge with the actual results, signifying that the model is continuously converging.

[0068] Therefore, by constructing loss functions corresponding to multiple functions and adjusting model parameters based on these loss functions, the training of the model for predicting preferences for each function can converge accurately, laying the foundation for the model to accurately predict users' preferences for various functions and improving the model's preference prediction ability.

[0069] After training to obtain the target prediction model, this model can be applied to predict any user's preference for any feature of an application. Therefore, embodiments of this application also propose application scenarios and processes for this target prediction model. Please refer to [link / reference] for details. Figure 4 One embodiment of the user preference prediction method in this application includes:

[0070] 401. Obtain a target prediction model, wherein the target prediction model is trained by a machine learning algorithm on multiple sets of training sample data and co-occurrence information among multiple functions in each function combination of the application; wherein each set of training sample data includes feature profile data and preference data of the user account, the preference data including preference labels of the user account for each function of the application; the preference labels are used to represent the user account's preference for the functions of the application; the co-occurrence information is used to indicate that the user account prefers each function in the function combination;

[0071] The method of this embodiment can be applied to a computer device, which may be... Figure 1 The network framework shown includes the service server 100 or various terminal devices. This target prediction model can be based on the aforementioned... Figure 2 The user preference prediction model training method shown in the embodiment and its various preferred embodiments is obtained.

[0072] The user feature profile data and the user preference data in each training sample data set have been described in detail above, and will not be repeated here.

[0073] 402. Obtain the feature profile data of the target user account, and input the feature profile data of the target user account into the target prediction model so that the target prediction model outputs the target user account's preference for each function of the application based on the feature profile data of the target user account;

[0074] The target user is the user for whom preference prediction needs to be performed. Feature profile data of the target user can be obtained and input into the target prediction model. The target prediction model then extracts user profile features such as behavioral features and attribute features from the feature profile data of the target user. Based on the user profile features such as behavioral features and attribute features, as well as the co-occurrence information representing the correlation between multiple functions of the application, the model outputs the target user's preference for each function of the application. This preference can be the degree of interest.

[0075] The target prediction model is trained by machine learning algorithms on multiple sets of training sample data. During the training process, the model continuously learns the correspondence between user profile features such as user behavior features and user attribute features and preference labels. When the model training converges, it learns an accurate representation of the correspondence, and then can accurately predict the user's preference for application functions based on the learned knowledge and prediction ability.

[0076] Therefore, in a preferred embodiment, the target prediction model, based on the predictive capabilities obtained through model training, can output a preference score for each function of the application for the target user account according to the feature profile data of the target user account, and sort the multiple preference scores of the target user account for multiple functions of the application, outputting the sorting result. This sorting result can be output to the downstream profile platform for business users to call, representing the functions that the user prefers and the functions that the user does not prefer. For example, the data output example of this sorting result can be: <User ID|TOP1 Function Name_Corresponding Preference Score, TOP2 Function Name_Corresponding Preference Score, ...>.

[0077] After determining the user's preferred functions, those functions can be displayed on the interface, such as... Figure 5 As shown, for example, when a user uses a music software client and performs an in-app search, such as searching for functions, music, or other information, multiple functions predicted above can be displayed to recommend to the user. These functions can include user preferences such as "new sound effects," "simplified mode," and "radar mode," allowing the user to quickly access the desired function without searching, thus improving the user experience of the client.

[0078] Therefore, in this embodiment, compared to using multiple binary classification models to predict user preferences, using this target prediction model can predict a user's preferences for multiple functions of the application by reading the user's profile data only once, saving computing resources. At the same time, by combining the co-occurrence information used to characterize the correlation between multiple functions, the prediction can uncover the user's potential preferences for other functions, making the preference prediction more accurate and comprehensive.

[0079] The computer device in the embodiments of this application is described below. Please refer to [link / reference]. Figure 6 One embodiment of the computer device in this application includes:

[0080] The computer device 600 may include one or more central processing units (CPUs) 601 and a memory 605, in which one or more applications or data are stored.

[0081] The memory 605 can be volatile or persistent storage. The program stored in the memory 605 can include one or more modules, each module including a series of instruction operations on the computer device. Furthermore, the central processing unit 601 can be configured to communicate with the memory 605 and execute the series of instruction operations stored in the memory 605 on the computer device 600.

[0082] The computer device 600 may also include one or more power supplies 602, one or more wired or wireless network interfaces 603, one or more input / output interfaces 604, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0083] The central processing unit 601 can perform the aforementioned... Figure 2 The operations performed by the computer device in the illustrated embodiments and their various preferred embodiments are not described in detail here.

[0084] This application also provides a computer storage medium, one embodiment of which includes: the computer storage medium storing instructions, which, when executed on a computer, cause the computer to perform the aforementioned... Figure 2 The operations performed by the computer device in the illustrated embodiments and their various preferred embodiments.

[0085] This application also provides a computer program product, one embodiment of which includes: when the computer program product is run on a computer device, causing the computer device to perform the aforementioned... Figure 2 The operations performed by the computer device in the illustrated embodiments and their various preferred embodiments.

[0086] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0087] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0088] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0089] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0090] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for training a user preference prediction model, characterized in that, The method includes: Multiple sets of training sample data are obtained. Each set of training sample data includes feature profile data of a user account and preference data of that user account. The preference data includes preference tags of the user account for the functions of the application. The preference tags are used to represent the user account's preference for the functions of the application. Obtain co-occurrence information among multiple functions in each function combination of the application. The co-occurrence information is obtained by statistically analyzing the preference data of each user account for multiple functions in the function combination. An initial prediction model is obtained by inputting the multiple sets of training sample data and the co-occurrence information into the initial prediction model, so that the initial prediction model outputs the user account's preference prediction result for each function of the application based on the feature profile data and the co-occurrence information; wherein, the initial prediction model sets the neuron weights for predicting the user account's preference for each function of the application based on the co-occurrence information, and outputs the preference prediction result based on the set neuron weights and the user account's behavioral and attribute features extracted from the feature profile data; the co-occurrence information includes a co-occurrence frequency matrix composed of the co-occurrence frequencies of multiple function combinations, and the neuron weights are set according to the co-occurrence frequency matrix; The model parameters of the initial prediction model are adjusted based on the preference prediction results and the preference labels until the model training convergence condition is met, at which point the model training stops, and the target prediction model is obtained. The target prediction model is used to predict the preference of any user account for any function of the application.

2. The method according to claim 1, characterized in that, The feature profile data includes data on the user account's access to the application's functions, data on the user account's operational behavior in the application, and personal information data of the user account. The initial prediction model outputs the user account's preference prediction results for each function of the application based on the feature profile data and the co-occurrence information, including: The initial prediction model extracts behavioral features of user accounts from the path usage data and the operation behavior data, extracts attribute features of user accounts from the personal information data, and outputs the user account's preference prediction results for each function of the application based on the behavioral features, the attribute features, and the co-occurrence information.

3. The method according to claim 1, characterized in that, The steps for obtaining the preference data include: For each function of the application, obtain behavioral data of multiple user accounts for that function; Determine whether the behavioral data for each user account for this function includes data corresponding to preferred behaviors; If a user account's behavioral data for this function includes data corresponding to the preferred behavior, then the user account's preference tag for this function is determined to be the first tag; the first tag is used to indicate that the user account prefers this function. If the user account's behavior data for this function does not include the data corresponding to the preferred behavior, then the user account's preference label for this function is determined to be the second label; the second label is used to indicate that the user account does not prefer this function.

4. The method according to claim 3, characterized in that, The method further includes: For each function of the application, count the number of positive samples whose preference label for that function is the first label; If the number of positive samples is greater than a first preset number, then the positive samples are undersampled; if the number of positive samples is less than a second preset number, then the positive samples are oversampled; the first preset number is greater than or equal to the second preset number. The positive samples obtained from the undersampling or oversampling processes are used as the preference data for this function.

5. The method according to any one of claims 1 to 4, characterized in that, The step of adjusting the model parameters of the initial prediction model based on the preference prediction result and the preference label includes: For each function of the application, a loss function for that function is constructed based on the preference prediction results of multiple user accounts for that function obtained from the multiple sets of training sample data and the multiple preference labels for that function in the multiple sets of training sample data. The model parameters of the initial prediction model are adjusted based on the loss values ​​of multiple loss functions for multiple functions of the application.

6. A user preference prediction method, characterized in that, The method includes: A target prediction model is obtained, which is trained by a machine learning algorithm on multiple sets of training sample data and co-occurrence information among multiple functions in each function combination of an application. Each set of training sample data includes feature profile data and user account preference data. The preference data includes preference labels for each function of the application by the user account. The preference labels represent the user account's preference for the application's functions. The co-occurrence information indicates that the user account prefers each function in the function combination. During training, the target prediction model sets neuron weights based on the co-occurrence information to predict the user account's preference for each function of the application. It outputs preference prediction results based on the set neuron weights and the user account's behavioral and attribute features extracted from the feature profile data. The model parameters of the target prediction model are adjusted based on the preference prediction results and the preference labels. The co-occurrence information includes a co-occurrence frequency matrix composed of the co-occurrence frequencies of multiple function combinations, and the neuron weights are set according to the co-occurrence frequency matrix. Obtain feature profile data of the target user account, and input the feature profile data of the target user account into the target prediction model so that the target prediction model outputs the target user account's preference for each function of the application based on the feature profile data of the target user account.

7. The method according to claim 6, characterized in that, The target prediction model outputs the target user account's preference for each function of the application based on the target user account's feature profile data, including: The target prediction model outputs the target user account's preference score for each function of the application based on the feature profile data of the target user account, sorts the multiple preference scores of the target user account for multiple functions of the application, and outputs the sorting result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

9. A computer storage medium, characterized in that, The computer storage medium stores instructions that, when executed on the computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images