A method, apparatus, medium and equipment for determining a target object

Abstract

This invention discloses a method, apparatus, medium, and device for determining a target object, comprising: acquiring a source domain dataset and a target domain dataset, wherein each source domain data element in the source domain dataset includes a first feature, a second feature, and a label of a first user, and each target domain data element in the target domain dataset includes only the second feature of a second user; predicting the first feature of the second user based on the result of training a first machine learning model using the source domain dataset; predicting the predicted label of the second user based on the custom second feature and the first feature and label of the first user using a second machine learning model; and determining the target object based on the predicted label of the second user. This invention relates to machine learning, improving the accuracy of object prediction based on multiple features, and also enabling the prediction model to be applicable to various task scenarios.

Description

Technical Field

[0001] This invention relates to the field of machine learning, and more specifically to a method, apparatus, medium, and device for determining a target object. Background Technology

[0002] With the development of artificial intelligence, machine learning models have been widely applied across various industries, playing a crucial role in various scenarios. To meet the needs of one platform (not limited to various applications, brands, or enterprises) to provide services to users on another platform to boost user activity or promote business, a machine learning model is used to jointly predict user characteristics from both platforms. This selects user groups that meet certain predicted values ​​as the target audience for business services. The selected user groups have a high probability of meeting business metrics, thus satisfying the business needs of one platform while providing personalized services to users.

[0003] In such application scenarios, to enable machine learning models to be productized, generalized, and leverage the advantages of multi-platform user data, user features from both platforms are indispensable. However, in practice, there may be situations where user features from one platform are missing. In this case, the model is suitable for applications such as promoting the activity of existing customers, but not for applications such as acquiring new customers, because new customers, or potential users on the other platform, do not possess the user features from the first platform (since they are not yet users of the first platform). On the other hand, the lack of user features from one platform will also affect the actual utility of the model's prediction results. Summary of the Invention

[0004] To more accurately predict target objects and to make the model productable and generalizable to meet the needs of different task scenarios, this invention provides a method, apparatus, medium, and device for determining target objects. The technical solution is as follows:

[0005] In a first aspect, the present invention provides a method for determining a target object, the method comprising:

[0006] Obtain a source domain dataset and a target domain dataset. Each source domain data element in the source domain dataset includes a first feature, a second feature, and a label of a first user. Each target domain data element in the target domain dataset includes only the second feature of a second user.

[0007] The first machine learning model is trained based on the source domain dataset to obtain the first prediction model;

[0008] Based on the first prediction model, prediction is performed according to the target domain dataset to obtain the first feature of the second user;

[0009] Obtain a custom second feature, and train a second machine learning model based on the custom second feature and the first feature and label of the first user to obtain a second prediction model;

[0010] Based on the second prediction model, prediction is performed according to the custom second feature and the first feature of the second user to obtain the predicted label of the second user;

[0011] The target object is determined based on the predicted tag of the second user, so as to provide services to the target object;

[0012] Wherein, the first user is a user of both the first platform and the second platform, and the second user is a user of the second platform; the first feature represents the user's characteristics on the first platform, and the second feature represents the user's characteristics on the second platform.

[0013] In a second aspect, the present invention provides a device for determining a target object, the device comprising:

[0014] The acquisition module is used to acquire a source domain dataset and a target domain dataset. Each source domain data element in the source domain dataset includes a first feature, a second feature, and a label of a first user. Each target domain data element in the target domain dataset only includes the second feature of a second user.

[0015] The first training module is used to train the first machine learning model based on the source domain dataset to obtain the first prediction model;

[0016] The first prediction module is used to predict the first feature of the second user based on the first prediction model and the target domain dataset.

[0017] The second training module is used to obtain a custom second feature, and to train the second machine learning model based on the custom second feature and the first feature and label of the first user to obtain the second prediction model.

[0018] The second prediction module is used to make predictions based on the second prediction model, according to the custom second feature and the first feature of the second user, to obtain the predicted label of the second user.

[0019] An object determination module is used to determine a target object based on the predicted tags of the second user, so as to provide services to the target object;

[0020] Wherein, the first user is a user of both the first platform and the second platform, and the second user is a user of the second platform; the first feature represents the user's characteristics on the first platform, and the second feature represents the user's characteristics on the second platform.

[0021] Thirdly, the present invention provides a computer storage medium storing at least one instruction or at least one program, wherein the at least one instruction or at least one program is loaded and executed by a processor to implement a method for determining a target object as described in the first aspect.

[0022] Fourthly, the present invention provides a computer device comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or at least one program being loaded by the processor and executed as a method for determining a target object as described in the first aspect.

[0023] The present invention provides a method, apparatus, medium, and device for determining a target object, which has the following technical effects:

[0024] (1) The solution provided by this invention is based on transfer learning to predict the characteristics of users on other platforms, so as to combine multiple features to predict and select objects, thereby improving the accuracy of prediction and the actual utility of prediction results;

[0025] (2) The solution provided by the present invention makes the prediction model productized and generalized through feature prediction, and can be applied to a variety of task scenarios.

[0026] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0027] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the implementation environment of a method for determining a target object provided in an embodiment of the present invention;

[0029] Figure 2 This is a flowchart illustrating a method for determining a target object according to an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the task processing flow of a data management platform for executing a method for determining a target object, provided in an embodiment of the present invention.

[0031] Figure 4 This is a flowchart illustrating another method for determining a target object provided in an embodiment of the present invention;

[0032] Figure 5 This is a flowchart illustrating another method for determining a target object provided in an embodiment of the present invention;

[0033] Figure 6 This is a schematic diagram of the process of performing model training in a method for determining a target object provided in an embodiment of the present invention;

[0034] Figures 7(1) to (3) are schematic diagrams of the steps for performing model training tasks based on a data management platform according to the embodiments of the present invention;

[0035] Figure 8 This is a schematic diagram of the process of performing model prediction in a method for determining a target object provided in an embodiment of the present invention;

[0036] Figure 9 This is a schematic diagram of the structure of a target object determination device provided in an embodiment of the present invention.

[0037] Figure 10 This is a schematic diagram of the hardware structure of a device for implementing a method for determining a target object, provided in an embodiment of the present invention. Detailed Implementation

[0038] Artificial Intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operating / interactive systems, and mechatronics.

[0039] The solutions provided in this application involve technologies such as deep learning and transfer learning in artificial intelligence.

[0040] Deep learning (DL) is a research area within machine learning (ML) that brings it closer to its original goal—artificial intelligence. Deep learning learns the inherent patterns and hierarchical representations of sample data. The information gained during this learning process greatly aids in interpreting data such as text, images, and sound. Its ultimate goal is to enable machines to possess analytical and learning capabilities similar to humans, allowing them to recognize data such as text, images, and sound.

[0041] Transfer learning is a learning process that leverages the similarity between data, tasks, or models to apply a model learned in an old domain to a new domain. Training and test data follow different distributions, allowing the model to be transferred between different tasks.

[0042] The core of transfer learning based on deep networks lies in identifying the layers in the network that need adaptation and applying adaptive loss metrics to these layers. More and more researchers are starting to use deep networks for transfer learning. Among these methods, almost the vast majority employ neural networks to perform transfer learning on pre-trained models (such as AlexNet, Inception, GoogLeNet, ResNet, etc.).

[0043] To meet the needs of one platform (not limited to various applications, brands, and enterprises) to provide services to users on another platform to boost user activity or promote business, a machine learning model is used to jointly predict user characteristics from both platforms. This selects user groups that meet certain predicted values ​​as the target audience for business services. The selected user groups have a high probability of meeting business metrics, thus satisfying the platform's business needs while providing personalized services to users. For example, in advertising scenarios, target user groups are selected from users on social application platforms based on their characteristics. These target user groups have a higher probability of clicking / converting on advertising services, allowing for targeted advertising to promote the brand. In such applications, for the machine learning model to be productized and generalized, the user characteristics of both the advertising brand users and the social application platform users are indispensable. However, in practical applications, there may be situations where one side's user characteristics are missing. For instance, in advertising scenarios, this model is suitable for boosting the activity of existing customers but not for acquiring new customers because the user characteristics of advertising brand users are not present for new customers or potential users on the social platform (because they are not yet advertising brand users). On the other hand, the lack of user characteristics of advertised brand users will also affect the actual utility of the model's prediction results.

[0044] To more accurately predict and select target objects and to make the model productable and generalizable to meet the needs of different task scenarios, embodiments of the present invention provide a method, apparatus, medium, and device for determining target objects. The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0045] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0046] To facilitate understanding of the technical solutions and their effects described in the embodiments of the present invention, the relevant technical terms are explained in the embodiments of the present invention:

[0047] TDP: Tencent Data Partner, a business data asset management platform provided by the Advertising and Marketing Services line to its ecosystem partners (DPs). The purpose is to open up Tencent's big data and modeling capabilities to DPs, helping them to leverage their own data and technical capabilities to improve their data monetization and marketing effectiveness on Tencent's traffic.

[0048] TDW: Tencent Distribution Data Warehouse; it's a distributed data processing system based on a share-nothing architecture, featuring high availability and scalability for massive data storage and analysis. It provides users with an interface similar to a structured query language, offering petabyte-level storage and terabyte-level computing power to meet the ever-growing demands for massive data analysis and help uncover more user value.

[0049] Please see Figure 1 This is a schematic diagram illustrating the implementation environment of a method for determining a target object provided in an embodiment of this application, such as... Figure 1 As shown, the implementation environment may include at least client 01 and server 02.

[0050] Specifically, the client 01 may include devices such as smartphones, desktop computers, tablets, laptops, digital assistants, smart wearable devices, monitoring devices, and voice interaction devices. It may also include software running on the device, such as web pages provided to users by service providers, or applications provided by those service providers. Specifically, the client 01 can be used to display the application interface of the management platform, input the task requirements of the first platform, display the task status list during model training and prediction, and store the final prediction results.

[0051] Specifically, server 02 may include a standalone server, a distributed server, or a server cluster consisting of multiple servers. Server 02 may include a network communication unit, a processor, and memory, etc. Specifically, server 02 can be used to run model training and prediction, configure tasks according to the task requirements of the first platform, and send prediction results to client 01.

[0052] The following describes a method for determining a target object provided by the present invention. Figure 2 This is a flowchart illustrating a method for determining a target object according to an embodiment of the present invention. This specification provides the operational steps of the method described in the embodiments or flowcharts, but based on conventional or non-inventive methods, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or server product execution, the method can be executed sequentially according to the embodiments or drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Please refer to... Figure 2 The method for determining a target object provided in the embodiments of this specification may include the following steps:

[0053] S101: Obtain the source domain dataset and the target domain dataset. Each source domain data element in the source domain dataset includes the first feature, the second feature, and the label of the first user. Each target domain data element in the target domain dataset includes only the second feature of the second user.

[0054] In the embodiments described in this specification, the first user is a user of both the first platform and the second platform, and the second user is a user of the second platform; the first feature characterizes the user's characteristics on the first platform, and the second feature characterizes the user's characteristics on the second platform.

[0055] Understandably, to meet the needs of one platform (not limited to various applications, brands, or companies) to provide services to users on another platform to boost user activity or promote business, a machine learning model is used to jointly predict user characteristics from both platforms. This selects user groups that meet certain predicted values ​​as the target audience for the services. The selected user groups have a high probability of meeting business metrics, thus satisfying the platform's business needs, while also providing personalized services to users. For example, in advertising scenarios, target user groups are selected from users on social application platforms based on user characteristics. These target user groups have a higher probability of clicking / converting from advertising services, allowing for targeted advertising of a specific type to promote the advertising brand. In such applications, for the machine learning model to be productized and generalized, the user characteristics of both the advertising brand users and the social application platform users are indispensable. However, in specific applications, there may be situations where one side's user characteristics are missing. For instance, in advertising scenarios, this model is suitable for boosting the activity of existing customers but not for acquiring new customers because the user characteristics of advertising brand users are not present for new customers, or potential users on the social platform (because they are not yet advertising brand users). On the other hand, the lack of user characteristics of advertising brand users can also affect the actual utility of the model's prediction results. To address the aforementioned problems, this specification provides a method for determining the target audience in its embodiments, enabling more accurate prediction and selection of the target group, and making the machine learning model productized and generalized to meet the needs of different task scenarios.

[0056] In the embodiments described in this specification, for example, the first platform can be the advertising brand, and the second platform can be Tencent. The advertising and marketing objectives of the first platform are mainly twofold: reactivating existing customers and acquiring new customers. Reactivating existing customers refers to extracting target users from the user base that belongs to both the advertising brand and Tencent, and then conducting precise advertising within Tencent's applications to increase the activity of these target users on the advertising brand's platform. Acquiring new customers refers to extracting users from Tencent's user base who are not yet users of the advertising brand, and then targeting these users with ads to encourage them to become users of the advertising brand, thereby increasing the advertising brand's user base or user activity. Currently, Tencent's data management platform TDP and model lab Model-Lab enable advertising brands to upload and manage custom datasets or self-created user features based on their own business objectives. These features are then used to build predictive models through the model algorithms provided by the management platform to extract target users for precise advertising. In the business scenario of reactivating existing customers, advertising brands, leveraging their accumulated data and technical capabilities, can effectively achieve their business goals through the data management platform TDP by uploading key business features and the first user, thereby improving data monetization and marketing effectiveness on Tencent's traffic. However, in the business scenario of acquiring new customers, the second user does not possess the advertising brand's key features, thus failing to leverage the advantages of the key features in the management platform's model prediction process. Therefore, the method in the embodiments of this specification is proposed to address the problem that the management platform's model algorithm is not applicable to the business scenario of acquiring new customers, in order to further scale the use of productized models and improve the business effectiveness of new customer acquisition.

[0057] In the embodiments of this specification, the source domain dataset contains features of a first user, who is a common user of both the first and second platforms. The target domain dataset contains features of a second user, who is a user only of the second platform. The problem to be solved in the source domain dataset is to identify high-click / conversion users to improve the activity of existing customers. This problem can be solved by jointly predicting the target audience using the first and second features. The problem to be solved in the target domain dataset is to identify high-click / conversion users among the second users to convert them into new customers of the first platform. Therefore, the tasks, models, and data dimensions have a certain degree of similarity in the two domains.

[0058] For example, in the execution environment provided by Tencent's data management platform, the method may further include:

[0059] S201: Obtain task requirements, which include at least model configuration requirements, audience selection requirements, and feature selection requirements.

[0060] Specifically, Figure 3 The task processing flow of the data management platform is shown, such as... Figure 3 As shown, the modeling interface provided by the management platform can meet the model configuration requirements, prediction configuration requirements, audience selection requirements, feature selection requirements, and trigger model training and prediction requirements of the first platform. This specification does not limit the types of requirements that the modeling interface can meet; it can be specifically designed according to business processes and needs. For example, to meet the user object prediction requirements of the first platform, the audience creation interface of the modeling interface allows the first platform to create user audiences through file uploads, historical interactive activities, geographical location, keyword selection, etc., and to perform filtering, editing, authorization, and deletion operations on the created user audiences in the audience management interface. Figure 3 In this model, one-sided features represent the first feature, two-sided features represent the second feature, one-sided feature embedding represents the model, feature vector represents the model, one-sided feature embedding predicts the model, representing the first machine learning model and the first prediction model, and the super model represents the second machine learning model and the second prediction model.

[0061] S203: Create a first machine learning model and a second machine learning model according to the model configuration requirements.

[0062] For example, the model configuration requirements include the model algorithm and hyperparameters customized by the first platform according to the task type and business objectives. In machine learning, hyperparameters are parameters whose values ​​are set before the learning process begins, such as the number of latent factors in matrix factorization, the number of hidden layers in a deep neural network, and the number of clusters in k-means clustering.

[0063] In the embodiments of this specification, the first machine learning model is used to predict the first feature based on the second feature, so as to complete the transfer learning from the first feature of the first user to the first feature of the second user; the second machine learning model is used to make joint prediction based on the first feature and the second feature of the second user. At the same time, it can be considered that in the application scenario of reactivating old customers, the second machine learning model can also directly make prediction based on the first feature and the second feature of the first user. Therefore, the second machine learning model can be generalized and productized.

[0064] For example, the first machine learning model can use a deep learning network (DNN) model structure, and the second machine learning model can use a deep learning network or combine a deep learning network with model algorithms such as LR (Logistic Regression) or XGboost (Extreme Gradient Boosting).

[0065] S205: Determine the first user set and the second user set based on the population selection requirements.

[0066] For example, a first user set and a second user set are determined based on the user groups created and managed by the first platform in the modeling interface. The data format of the first user set and the second user set is user identity identifier + tag. The data type of the user identity identifier can be QQ number, mobile phone number, user terminal device number, MAC address, or its encrypted format, etc. The tag depends on the specific business objectives of the first platform. Based on the tags, the first user set can be divided into a positive example set and a negative example set, and the corresponding second user set can be called a candidate set.

[0067] S207: Based on the selected features, determine the custom second feature.

[0068] It is understood that the custom second feature is selected by the first platform from all user profile feature dimensions provided by the second platform for use in training and prediction of the second machine learning model.

[0069] In the embodiments described in this specification, when in the modeling environment provided by the data management platform, obtaining the source domain dataset may include the following steps:

[0070] S1011: Determine the identity identifier and tag of each first user in the first user set based on the first user set.

[0071] In one feasible implementation, for the purpose of protecting user information and controlling usage rights, the identity of the first user is encrypted using MD5 (Message Digest Algorithm MD5, a cryptographic hash function).

[0072] S1013: Obtain user behavior data uploaded through the dataset interface.

[0073] Specifically, refer to Figure 3 As shown, the dataset's application programming interface (API) allows the first platform to directly upload its own created first feature or user behavior data of the first user. This user behavior data refers to user actions occurring in scenarios such as mobile applications, websites, and offline activities, and can be composed of user identifiers, behavior types, time, and parameters. Furthermore, the dataset API also provides an interface for registering first features to add user profile feature tags not present in the data management platform.

[0074] S1015: Perform feature standardization management and feature matching based on the user behavior data to generate the first feature of each first user.

[0075] Specifically, the generated first feature may be stored in the database or not stored, depending on the authorization permissions or resource level of the first platform. The usage permission of the first feature may be restricted to a single task or to multiple tasks within a single project.

[0076] S1017: Obtain the second feature of each first user from the data warehouse based on the identity identifier of each first user.

[0077] For example, when Tencent acts as both the second platform and the data management platform provider, the user's second characteristics are stored in the data warehouse or a dedicated feature warehouse of the data management platform. The user profile characteristics of the first user within Tencent are obtained based on the matching of identity identifiers. These second characteristics can be all the second characteristics of the first user, encompassing all dimensions of the user profile characteristics within Tencent. If the data management platform is a third platform, the method by which the data management platform obtains the second characteristics of the first user can refer to the method used to obtain the first user's first characteristics, and will not be elaborated further here.

[0078] S1019: Use the identity identifier, first feature, second feature, and label of each first user as the source domain dataset.

[0079] It is understood that the source domain dataset is used as training samples to train the first machine learning model and the second machine learning model, and can also be used as test samples for the first machine learning model to obtain the first feature of the predicted first user.

[0080] In the embodiments described in this specification, when in the execution environment provided by the data management platform, obtaining the target domain dataset may include the following steps:

[0081] S1012: Determine the identity identifier of each second user in the second user set based on the second user set.

[0082] In one feasible implementation, the identity identifiers of the target domain population are encrypted using MD5 for the purpose of protecting user information and controlling usage rights.

[0083] S1014: Obtain the second feature of each second user from the data warehouse based on the identity identifier of each second user.

[0084] For example, when Tencent acts as both the second platform provider and the data management platform provider, the user's second characteristics are stored in the data warehouse of the data management platform or a dedicated feature warehouse. The second user's user profile characteristics within Tencent are obtained based on the matching of identity identifiers. These second characteristics can be all of the second user's second characteristics, encompassing all dimensions of the user profile characteristics within Tencent. If the data management platform is a third platform, the method by which the data management platform obtains the second user's second characteristics can refer to the method used to obtain the first user's first characteristics, which will not be elaborated upon here.

[0085] S1016: Use the identity identifiers and second features of each second user as the target domain dataset.

[0086] It is understood that the target domain dataset is used as test data for the prediction input of the first machine learning model, and the first feature of the second user is transferred to the second machine learning model for prediction.

[0087] S103: Train the first machine learning model based on the source domain dataset to obtain the first prediction model.

[0088] In the embodiments described in this specification, the lack of the second user's first feature prevents the use of the general model of the data management platform for predicting and selecting user objects. It also prevents the leverage of the first feature in model prediction, affecting the actual utility of the prediction results. Considering the similarity in tasks, models, and data dimensions between the source and target domain datasets, transfer learning can be used to obtain the second user's first feature. Transfer learning involves transferring the first user's first feature as known knowledge from the source domain dataset to the target domain dataset. Specifically, a mapping relationship between the user's second feature and the first feature is established through the first prediction model, and the second feature is used to predict the first feature.

[0089] Specifically, refer to Figure 4 and Figure 5 ,exist Figure 4 In this embodiment, a first feature is represented by a one-sided feature, a second feature is represented by a two-sided feature, a one-sided feature embedding represents a model, a feature vector represents a model, a one-sided feature embedding predicts a model, representing a first machine learning model and a first prediction model, a super model represents a second machine learning model and a second prediction model, a DP user represents a first user, a non-DP user / targeted audience candidate set represents a second user, and DP represents a first platform. Step 103 may further include the following steps:

[0090] S301: Train the third machine learning model using the first features and labels of the first user to obtain a feature vector representation model.

[0091] It is understandable that the first user's first feature data is large in volume and has a lot of information redundancy. Therefore, it is more efficient to extract the first feature and represent it in the form of an embedding vector than to use the first feature directly.

[0092] In one feasible implementation, in response to task requirements, a deep learning model is constructed, taking the first feature of the first user as data input and business objectives such as clicks / conversions as label input, and the deep learning model is trained to obtain a feature vector representation model.

[0093] S303: Obtain the first feature embedding vector of the first user according to the feature vector representation model, and update the source domain dataset.

[0094] In one feasible implementation, the output of the last 8-dimensional hidden layer of the deep learning model is used as the embedding vector output, representing the first feature embedding vector of the first user. Furthermore, the first feature embedding vector can be concatenated with the model input to generate data in tfrecord (a binary file format) format to accelerate data reading and preprocessing.

[0095] S305: Train the first machine learning model based on the second feature, label and first feature embedding vector of the first user to obtain the first prediction model.

[0096] In one feasible implementation, a deep learning model is constructed to predict the embedding vector representation of the first feature using the user's second feature. The deep learning model is trained by taking the second feature of the first user, the embedding vector of the first feature, and user samples (including user identifier and label) as input to obtain a first prediction model.

[0097] S307: Based on the first prediction model, make a prediction according to the second feature of the first user to obtain the predicted first feature embedding vector of the first user.

[0098] In another feasible implementation, the output of step S305 is directly used as the first user's predicted first feature embedding vector, and the prediction result and input data are concatenated to generate data in tfrecord format.

[0099] S309: By comparing the first feature embedding vector of the first user with the predicted first feature embedding vector, the first prediction model is optimized to obtain the optimized first prediction model.

[0100] S311: Based on the optimized first prediction model, predict according to the second feature of the first user to obtain the target prediction first feature embedding vector of the first user.

[0101] Understandably, by comparing the difference between the first user's first feature embedding vector and the predicted first feature embedding vector, the training effect of the first prediction model can be evaluated, and then the model structure, parameters, functions, etc., of the first prediction model can be optimized and adjusted. At the same time, based on the optimized first prediction model, prediction can be performed again to obtain the first user's target predicted first feature embedding vector, which can then be used to better train the second machine learning model.

[0102] S105: Based on the first prediction model, make a prediction according to the target domain dataset to obtain the first feature of the second user.

[0103] Understandably, the first prediction model is trained using the second and first features of the first user, thus establishing a mapping relationship between the second and first features. Considering that the feature dimensions of the first and second users are similar, and the problems to be solved in the source and target domain datasets are also similar, the theory of transfer learning can be used to apply the first prediction model to predict the first feature of the second user.

[0104] In the embodiments of this specification, compared to using the second feature alone for prediction, using the first feature and the second feature together for prediction results in a target object that is more consistent with the service object characteristics targeted by the business of the first platform. While leveraging the data advantages of the first feature, it also makes the prediction model more generalizable and productizable.

[0105] In one feasible implementation, as described in steps S301-S311 of the embodiments of this specification, the first feature is represented in the form of an embedding vector, and step S105 may include:

[0106] S501: Based on the optimized first prediction model, a prediction is made according to the second feature of the second user to obtain the predicted first feature embedding vector of the second user.

[0107] Specifically, the second feature of the second user, or candidate user, is used as input to predict the first feature embedding vector of the second user, and the output and input data are concatenated to obtain data in tfrecord format.

[0108] S107: Obtain a custom second feature, and train a second machine learning model based on the custom second feature and the first feature and label of the first user to obtain a second prediction model.

[0109] In the embodiments of this specification, by way of example, the custom second feature may be selected by the first platform from all the second features used by the second platform, and the use of the custom second feature of the first platform makes the selected target object more in line with the business objectives of the first platform.

[0110] In one feasible implementation, as described in steps S301-S311 of the embodiments of this specification, the first feature is represented by an embedding vector, and step S107 may include:

[0111] S701: Train the second machine learning model based on the first user's label, the target prediction first feature embedding vector, and the custom second feature to obtain the second prediction model.

[0112] Specifically, the second machine learning model can use a deep learning network model architecture or combine deep learning networks with model algorithms such as LR (Logistic Regression) or XGboost (Extreme Gradient Boosting).

[0113] For example, such as Figure 6 As shown, in the execution environment provided by Tencent's data management platform, the acquired task requirements may also include model training, and steps S103 and S107 may further include the following steps:

[0114] S601: In response to the model training requirement in the task, configure the model training task and generate a training task status list, wherein the model training task includes a first machine learning model training task and a second machine learning model training task.

[0115] In one feasible implementation, task configuration is performed for model training. The process of configuring tasks mainly includes the acquisition and processing of feature data and the triggering of downstream tasks in the data management platform.

[0116] In another feasible implementation, the first feature of the first user is extracted to obtain the representation of the first feature embedding vector. When configuring the model training task, the training task of the feature vector representation model and the first prediction task of the first prediction model can also be configured. The first prediction task of the first prediction model is used to predict the first feature of the first user to optimize the first prediction model.

[0117] S603: Execute the model training task to obtain the first prediction model and the second prediction model, and update the training task status list.

[0118] In one feasible implementation, as shown in Figures 7(1), 7(2) and 7(3), the training tasks of the feature vector representation model, the training tasks of the first prediction model and the first prediction task of the first prediction model are executed respectively. After the model training is completed, the task execution results are stored and the status is updated for query.

[0119] S109: Based on the second prediction model, make a prediction according to the custom second feature and the first feature of the second user to obtain the predicted label of the second user.

[0120] In the embodiments of this specification, the second prediction model is trained using the first feature of the first user and a custom second feature, constructing a mapping relationship between the first feature, the second feature, and the label. Considering that the feature dimensions of the first user and the second user are similar, and the problems to be solved in the source domain dataset and the target domain dataset are also similar, the theory of transfer learning can be used to transfer the second prediction model to predict the label of the second user. Simultaneously, the second prediction model can be applied to various business scenarios, realizing the productization and generalization of the model.

[0121] In the embodiments provided by the present invention, as described in steps S301-S311 of the embodiments of this specification, the first feature is represented by an embedding vector, and step S109 may include the following steps:

[0122] S901: Based on the second prediction model, prediction is performed according to the custom second feature and the prediction first feature embedding vector of the second user to obtain the prediction label of the second user. The prediction label of the second user represents the predicted value of the second user for the target business indicator.

[0123] For example, such as Figure 8 As shown, in the execution environment provided by Tencent's data management platform, the acquired task requirements may also include model prediction, and steps S105 and S109 may further include the following steps:

[0124] S801: In response to the model prediction in the task requirements, configure the model prediction task and generate a prediction task status list, wherein the model prediction task includes a first machine learning model prediction task and a second machine learning model prediction task.

[0125] In one feasible implementation, task configuration is performed for model prediction. The process of configuring tasks mainly includes the acquisition and processing of feature data and the triggering of downstream tasks in the data management platform.

[0126] S803: Execute the model prediction task to obtain the first feature of the second user and the predicted label of the second user, and update the prediction task status list.

[0127] In one feasible implementation, the prediction tasks of the first prediction model and the second prediction model are executed respectively. After the model prediction is completed, the task execution results are stored and the status is updated for querying.

[0128] S111: Determine the target object based on the predicted tag of the second user, so as to provide services to the target object.

[0129] In one feasible implementation, the predicted tag can be a rating of a second user based on a preset business objective. For example, in a business scenario where advertising is delivered to a target audience, the preset business objective includes, but is not limited to, purchase metrics and click / conversion metrics. Second users whose predicted tags meet a preset threshold are selected as target audiences, and corresponding business services are provided. For example, the preset threshold is used to filter second users; it can be used to select second users whose ratings fall within a certain range, or it can be used to sort second users based on their ratings and select a certain number of second users as target audiences based on the sorting. The preset business objective and the preset threshold can be designed according to specific business scenarios, and this embodiment does not limit this.

[0130] It is understood that the solutions provided in the embodiments of this specification are based on transfer learning to predict user profile features, and combine multiple features to predict and select objects, thereby improving the accuracy of prediction and the practical utility of the prediction results. At the same time, the solutions provided in the embodiments of this specification supplement the prediction model through feature prediction, making the prediction model productized and generalized, and applicable to a variety of task scenarios.

[0131] This invention also provides a device for determining a target object, as shown in FIG9. The device may include:

[0132] The acquisition module 910 is used to acquire a source domain dataset and a target domain dataset. Each source domain data element in the source domain dataset includes a first feature, a second feature, and a label of a first user. Each target domain data element in the target domain dataset only includes the second feature of a second user.

[0133] The first training module 920 is used to train the first machine learning model based on the source domain dataset to obtain the first prediction model;

[0134] The first prediction module 930 is used to make predictions based on the first prediction model and the target domain dataset to obtain the first feature of the second user.

[0135] The second training module 940 is used to obtain a custom second feature, and to train a second machine learning model based on the custom second feature and the first feature and label of the first user to obtain a second prediction model.

[0136] The second prediction module 950 is used to make predictions based on the second prediction model, according to the custom second feature and the first feature of the second user, to obtain the predicted label of the second user.

[0137] The object determination module 960 is used to determine the target object based on the predicted tag of the second user, so as to provide services to the target object;

[0138] Wherein, the first user is a user of both the first platform and the second platform, and the second user is a user of the second platform; the first feature represents the user's characteristics on the first platform, and the second feature represents the user's characteristics on the second platform.

[0139] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0140] This invention provides a computer device including a processor and a memory. The memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement a method for determining a target object as provided in the above method embodiments.

[0141] Figure 10 A schematic diagram of a hardware structure for implementing the method provided in the embodiments of this application is shown. This device may participate in or include the apparatus or system provided in the embodiments of this application. Figure 10 As shown, device 10 may include one or more processors 1002 (shown as 1002a, 1002b, ..., 1002n in the figure) 1002 (processor 1002 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1004 for storing data, and a transmission device 1006 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 10 The structure shown is for illustrative purposes only and does not limit the structure of the electronic device described above. For example, device 10 may also include a... Figure 10 The more or fewer components shown, or having the same Figure 10 The different configurations shown.

[0142] It should be noted that the aforementioned one or more processors 1002 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be wholly or partially integrated into any other element within device 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0143] The memory 1004 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of this application. The processor 1002 executes various functional applications and data processing by running the software programs and modules stored in the memory 1004, thereby realizing the above-mentioned method for determining a target object. The memory 1004 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1004 may further include memory remotely located relative to the processor 1002, and these remote memories can be connected to the device 10 via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0144] The transmission device 1006 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of device 10. In one example, the transmission device 1006 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 1006 may be a radio frequency (RF) module used for wireless communication with the Internet.

[0145] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows a user to interact with the user interface of device 10 (or a mobile device).

[0146] This invention also provides a computer storage medium, which can be located in a server to store at least one instruction or at least one program related to implementing a method for determining a target object in the method embodiments. The at least one instruction or the at least one program is loaded and executed by the processor to implement the method for determining a target object provided in the above method embodiments.

[0147] Optionally, in this embodiment, the storage medium may be located at at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0148] As can be seen from the embodiments of the target object determination method, apparatus, medium and device provided by the present invention above,

[0149] (1) The solution provided by the present invention is based on transfer learning to predict user profile features, and combines multiple features to predict and select objects, thereby improving the accuracy of prediction and the actual utility of prediction results;

[0150] (2) The solution provided by the present invention makes the prediction model productized and generalized through feature prediction, and can be applied to a variety of task scenarios.

[0151] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0152] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0153] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0154] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for determining a target object, characterized in that, The method includes: Obtain a source domain dataset and a target domain dataset. Each source domain data element in the source domain dataset includes a first feature, a second feature, and a label of a first user. Each target domain data element in the target domain dataset includes only the second feature of a second user. The first machine learning model is trained based on the source domain dataset to obtain the first prediction model; Based on the first prediction model, prediction is performed according to the target domain dataset to obtain the first feature of the second user; Obtain a custom second feature, and train a second machine learning model based on the custom second feature and the first feature and label of the first user to obtain a second prediction model; Based on the second prediction model, prediction is performed according to the custom second feature and the first feature of the second user to obtain the predicted label of the second user; The target object is determined based on the predicted tag of the second user, so as to provide services to the target object; Wherein, the first user is a user shared by the first platform and the second platform, and the second user is a user of the second platform; the first feature represents the user's characteristics on the first platform, and the second feature represents the user's characteristics on the second platform; the custom second feature is selected by the first platform from all the second features used by the second platform, so that the selected target object conforms to the business objectives of the first platform.

2. The method for determining a target object according to claim 1, characterized in that, The method further includes: Obtain task requirements, which include at least model configuration requirements, audience selection requirements, and feature selection requirements; Create a first machine learning model and a second machine learning model according to the model configuration requirements; Based on the aforementioned user selection requirements, a first user set and a second user set are determined; Based on the requirements of the aforementioned features, determine the custom second feature.

3. The method for determining a target object according to claim 2, characterized in that, The acquisition of the source domain dataset includes: Determine the identity identifier and tag of each first user in the first user set based on the first user set; Retrieve user behavior data uploaded via the dataset interface; Based on the user behavior data, feature standardization management and feature matching are performed to generate the first feature of each first user; The second characteristics of each first user are obtained from the data warehouse based on the identity identifier of each first user; The identity identifier, first feature, second feature, and label of each first user are used as the source domain dataset.

4. The method for determining a target object according to claim 2, characterized in that, The acquisition of the target domain dataset includes: Determine the identity identifier of each second user in the second user set based on the second user set; The second characteristics of each second user are obtained from the data warehouse based on the identity identifier of each second user; The identity identifiers and second features of each second user are used as the target domain dataset.

5. The method for determining a target object according to claim 1, characterized in that, The method further includes: The third machine learning model is trained using the first features and labels of the first user to obtain a feature vector representation model; The first feature embedding vector of the first user is obtained according to the feature vector representation model, and the source domain dataset is updated.

6. The method for determining a target object according to claim 5, characterized in that, The step of training the first machine learning model based on the source domain dataset to obtain the first prediction model and the step of training the second machine learning model based on the custom second feature and the first feature and label of the first user to obtain the second prediction model include: The first machine learning model is trained based on the second feature, label, and first feature embedding vector of the first user to obtain the first prediction model; Based on the first prediction model, a prediction is made according to the second feature of the first user to obtain the predicted first feature embedding vector of the first user. By comparing the first feature embedding vector of the first user with the predicted first feature embedding vector, the first prediction model is optimized to obtain the optimized first prediction model. Based on the optimized first prediction model, the first user's second feature is used to make a prediction, and the first user's target prediction first feature embedding vector is obtained. The second machine learning model is trained based on the first user's label, the target prediction first feature embedding vector, and the custom second feature to obtain the second prediction model.

7. The method for determining a target object according to claim 6, characterized in that, The process of predicting the first feature of the second user based on the first prediction model and the target domain dataset, and predicting the predicted label of the second user based on the second prediction model, the custom second feature, and the first feature of the second user, includes: Based on the optimized first prediction model, prediction is made according to the second feature of the second user to obtain the predicted first feature embedding vector of the second user. Based on the second prediction model, prediction is performed according to the custom second feature and the prediction first feature embedding vector of the second user to obtain the prediction label of the second user. The prediction label of the second user represents the predicted value of the second user for the target business indicator.

8. A device for determining a target object, characterized in that, The device includes: The acquisition module is used to acquire a source domain dataset and a target domain dataset. Each source domain data element in the source domain dataset includes a first feature, a second feature, and a label of a first user. Each target domain data element in the target domain dataset only includes the second feature of a second user. The first training module is used to train the first machine learning model based on the source domain dataset to obtain the first prediction model; The first prediction module is used to predict the first feature of the second user based on the first prediction model and the target domain dataset. The second training module is used to obtain a custom second feature, and to train the second machine learning model based on the custom second feature and the first feature and label of the first user to obtain the second prediction model. The second prediction module is used to make predictions based on the second prediction model, according to the custom second feature and the first feature of the second user, to obtain the predicted label of the second user. An object determination module is used to determine a target object based on the predicted tags of the second user, so as to provide services to the target object; Wherein, the first user is a user shared by the first platform and the second platform, and the second user is a user of the second platform; the first feature represents the user's characteristics on the first platform, and the second feature represents the user's characteristics on the second platform; the custom second feature is selected by the first platform from all the second features used by the second platform, so that the selected target object conforms to the business objectives of the first platform.

9. A computer storage medium, characterized in that, The computer storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor to implement a method for determining a target object as described in any one of claims 1 to 7.

10. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or at least one program being loaded by the processor and executed as a method for determining a target object as described in any one of claims 1 to 7.

Images