Data processing method and device, equipment and storage medium

Abstract

The embodiment of the invention discloses a data processing method and device, equipment and a storage medium, and the method comprises the steps: carrying out the layering of a business object according to the historical media interaction characteristics of the business object in a multimedia application; if the business object belongs to the pre-loss object layer, task push scores corresponding to M multimedia tasks in the multimedia application are determined according to the influence degree of the historical media interaction features on the business object belonging to the pre-loss object layer; if the business object belongs to the active object layer, predicting interaction label information corresponding to the business object under the M multimedia tasks according to historical media interaction characteristics, and determining task push scores corresponding to the M multimedia tasks according to the interaction label information corresponding to the M multimedia tasks; and pushing the multimedia tasks of which the task pushing scores are greater than or equal to a pushing score threshold in the M multimedia tasks to the business object. By adopting the method and the device, the multimedia task pushing accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a data processing method, apparatus, device, and storage medium. Background Art

[0002] With the development of Internet technology and the increasing scale of network data, a large number of multimedia applications (such as game applications, video applications, and audio applications, etc.) emerge in an endless stream, and the demands of business objects (such as users) are becoming more and more diverse and personalized. In order to increase the interest of multimedia applications, multimedia tasks related to multimedia applications (such as game tasks, video tasks, and audio tasks, etc.) will be added, and business objects can complete the multimedia tasks to obtain relevant rewards. It is found in practice that different business objects often have different multimedia task requirements. If multimedia tasks are pushed to all business objects without difference, there will be some business objects that are difficult to obtain the multimedia tasks they are interested in, resulting in low accuracy of multimedia task push. Summary of the Invention

[0003] Embodiments of this application provide a data processing method, apparatus, device, and storage medium, which can improve the accuracy of multimedia task push.

[0004] On the one hand, an embodiment of this application provides a data processing method, including:

[0005] Obtain the historical media interaction characteristics of a business object in a multimedia application, and based on the historical media interaction characteristics, layer the business object to obtain an object layering result;

[0006] If the object layering result indicates that the business object belongs to the pre-churn object layer, then determine the task push scores corresponding to M multimedia tasks in the multimedia application respectively according to the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer; M is a positive integer;

[0007] If the object layering result indicates that the business object belongs to the active object layer, then predict the interaction label information corresponding to the business object under M multimedia tasks respectively according to the historical media interaction characteristics, and determine the task push scores corresponding to M multimedia tasks respectively according to the interaction label information corresponding to M multimedia tasks;

[0008] Push to the business object the multimedia tasks among the M multimedia tasks whose task push scores are greater than or equal to the push score threshold.

[0009] On the one hand, an embodiment of this application provides a data processing apparatus, including:

[0010] A hierarchical module, configured to obtain the historical media interaction characteristics of a business object in a multimedia application, and hierarchically classify the business object according to the historical media interaction characteristics to obtain an object hierarchical result;

[0011] A first determination module, configured to, if the object hierarchical result indicates that the business object belongs to the pre-churn object layer, determine the task push scores respectively corresponding to M multimedia tasks in the multimedia application according to the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer; M is a positive integer;

[0012] A second determination module, configured to, if the object hierarchical result indicates that the business object belongs to the active object layer, predict the interaction label information respectively corresponding to the business object under M multimedia tasks according to the historical media interaction characteristics, and determine the task push scores respectively corresponding to M multimedia tasks according to the interaction label information respectively corresponding to M multimedia tasks;

[0013] A push module, configured to push to the business object the multimedia tasks among M multimedia tasks whose task push scores are greater than or equal to a push score threshold.

[0014] On the one hand, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which is suitable for being loaded and executed by a processor, so that a computer device having the processor executes the method provided by the embodiment of the present application.

[0015] On the one hand, an embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided by the embodiment of the present application.

[0016] In the embodiments of the present application, based on the object hierarchy to which a service object belongs in a multimedia application, a strategy for dynamically selecting and pushing multimedia tasks to the service object is implemented to achieve personalized pushing of multimedia tasks and improve the accuracy of multimedia task pushing. Specifically, when the service object belongs to the pre-churn object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively low, that is, the service object belongs to the pre-churn object of the media application. Therefore, personalized attribution can be performed on the service object belonging to the pre-churn object layer to obtain the influence degree of historical media interaction characteristics on the service object belonging to the pre-churn object layer, that is, the influence degree reflects which historical media interaction characteristics are the main reasons for the service object to belong to the pre-churn object layer. Pushing multimedia tasks to the service object based on the influence degree is beneficial to improving the accuracy of multimedia task pushing, reducing the possibility of service object churn, and further improving the interest and experience of the service object. When the service object belongs to the active object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively high, that is, the service object belongs to the active object of the media application. Therefore, the interaction label information corresponding to the service object under M multimedia tasks can be predicted, and the interaction label information can be used to reflect the active situation and retention situation of the service object after pushing the corresponding multimedia tasks. Based on the interaction label information corresponding to the service object under M multimedia tasks respectively, it is beneficial to improve the accuracy of multimedia task pushing, and improve the activity and retention time of the service object in the multimedia application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 is a schematic structural diagram of a data processing system provided by an embodiment of the present application;

[0019] Figure 2 is a schematic diagram of a multimedia task pushing method based on layering provided by an embodiment of the present application;

[0020] Figure 3 is a schematic flowchart of a data processing method provided by an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of a multimedia task pushing intervention period provided by an embodiment of the present application;

[0022] Figure 5 is a schematic diagram of a pre-churn object provided by an embodiment of the present application;

[0023] Figure 6 It is a schematic diagram of a mapping table from historical interaction media features to multimedia tasks provided by an embodiment of the present application;

[0024] Figure 7 It is a schematic diagram of personalized multimedia task push provided by an embodiment of the present application;

[0025] Figure 8 It is a schematic flowchart of a data processing method provided by an embodiment of the present application;

[0026] Figure 9 It is a schematic diagram of pushing a game task provided by an embodiment of the present application;

[0027] Figure 10 It is a schematic diagram of pushing a game task provided by an embodiment of the present application;

[0028] Figure 11 It is a schematic diagram of video task push provided by an embodiment of the present application;

[0029] Figure 12 It is a schematic structural diagram of a data processing device provided by an embodiment of the present application;

[0030] Figure 13 It is a schematic diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] This application relates to the field of artificial intelligence technology. Specifically, embodiments of this application can predict the feature importance of historical media interaction features through an interpretable model, improving the accuracy and efficiency of feature importance prediction. Specifically, embodiments of this application can also layer business objects according to an object hierarchical model to obtain an object layering result, which can improve the accuracy and efficiency of object layering. Specifically, embodiments of this application can also predict the interaction label information of business objects under multimedia tasks through a multi-task recognition model, improving the accuracy and efficiency of interaction label information prediction.

[0033] Among them, Artificial Intelligence (AI) uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, including theories, methods, technologies, and application systems that can perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, AI is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling machines to have functions of perception, reasoning, and decision-making. AI technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic AI technologies generally include technologies such as sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technologies, operation / interaction systems, and mechatronics. AI software technologies mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, autonomous driving, and intelligent transportation.

[0034] Specifically, this application specifically relates to machine learning under AI technology. Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of AI and the fundamental way to make computers intelligent, and its applications cover all fields of AI. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning.

[0035] Please refer to Figure 1 , Figure 1 which is a schematic structural diagram of a data processing system provided by an embodiment of this application. As Figure 1 shown, the data processing system may include a server 10 and a cluster of terminal devices. The cluster of terminal devices may include one or more terminal devices, and the number of terminal devices will not be limited here. As Figure 1 shown, it may specifically include terminal device 100a, terminal device 100b, terminal device 100c,..., terminal device 100n. As Figure 1As shown, the terminal devices 100a, 100b, 100c, …, 100n can be respectively connected to the server 10 through a network, so that each terminal device can perform data interaction with the server 10 through this network connection. Of course, the terminal devices 100a, 100b, 100c, …, 100n can communicate with each other through a direct network connection, that is, point-to-point communication can be achieved between the terminal devices; that is to say, when data interaction needs to be performed between every two terminal devices, one terminal device (i.e., the sending terminal device) can directly send the data to another terminal device (i.e., the receiving terminal).

[0036] Among them, each terminal device in the terminal device cluster can include: intelligent terminals with data processing functions such as smartphones, tablets, laptops, desktop computers, intelligent voice interaction devices, intelligent home appliances (for example, smart TVs), wearable devices, vehicle-mounted terminals, etc. It should be understood that, as Figure 1 shown, each terminal device in the terminal device cluster can be installed with an application with data processing functions. When the application runs on each terminal device, it can perform data interaction with the server 10 as shown above Figure 1 For example, the application can specifically include game applications, video applications, audio applications, etc. For ease of understanding, embodiments of the present application can select one terminal device as the target terminal device from multiple terminal devices as shown Figure 1 For example, embodiments of the present application can use the terminal device 100a as shown Figure 1 as the target terminal device. The target terminal device can be installed with an application with data processing functions. At this time, the target terminal device can achieve data interaction with the server 10 through the application.

[0037] Among them, as Figure 1 shown, the server 10 is a device that can provide background services for the applications in the terminal devices. The server 10 is an independent physical server, or can also be a server cluster or distributed system composed of multiple physical servers, or can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0038] It should be understood that based on Figure 1A data processing system among them can be applicable to a multimedia task push scenario. It can be understood that the multimedia applications in the embodiments of this application can refer to game applications, video applications, audio applications, novel applications, etc. The business object can be a user in the multimedia application, and the multimedia application can provide multimedia interaction services for the business object. For example, taking the multimedia application as a game application, the multimedia application can provide game match services, game teaming services, game task execution services, etc. for the business object. The historical media interaction characteristics of the business object for the multimedia application can include multimedia execution characteristics, multimedia preference characteristics, multimedia application login characteristics, etc. For example, when the multimedia application is a game application, the historical media interaction characteristics of the business object in the game application can refer to the game interaction characteristics such as the game login characteristics, game match characteristics, game preference characteristics, game duration characteristics, game task execution characteristics, etc. of the business object in the historical time period. When the multimedia application is a video application, the historical media interaction characteristics of the business object in the video application can refer to the video preference characteristics, video login characteristics, video viewing duration, video task execution characteristics, etc. of the business object in the historical time period. When the multimedia application is an audio application, the historical media interaction characteristics of the business object in the audio application can refer to the audio preference characteristics, audio login characteristics, audio listening duration, audio task execution characteristics, etc. of the business object in the historical time period. Specifically, the server 10 can obtain the historical multimedia interaction data of the business object in the multimedia application in the historical time period from the terminal device associated with the business object, and extract the historical media interaction characteristics of the business object in the multimedia application from the historical multimedia interaction data.

[0039] Among them, when the multimedia application is a game application, the M multimedia tasks in the game application can include game sharing tasks, game match tasks, game teaming tasks, game survival target duration tasks, etc.; when the multimedia application is a video application, the M multimedia tasks in the video application can include video viewing tasks, video sharing tasks, video comment tasks, etc.; when the multimedia application is an audio application, the M multimedia tasks in the audio application can include audio viewing tasks, audio sharing tasks, audio comment tasks, etc. Since different business objects often have different multimedia task requirements, such as some business objects like multimedia tasks with simple operations, and some business objects like multimedia tasks with challenges, etc. Therefore, the business objects can be stratified according to the historical media interaction characteristics to obtain the object stratification result of the business objects, and this object stratification result can be used to indicate that the business object belongs to the pre-churn object layer, or to indicate that the business object belongs to the active object layer.

[0040] An object belonging to the pre-churn object layer can refer to an object that is about to churn, whose activity level in the multimedia application in a future time period is less than or equal to the activity threshold. For example, an object belonging to the churn object layer will no longer log in to the multimedia application or enter the task push activity in the multimedia application in the future time period. An object belonging to the active object layer can refer to a continuously active object, whose activity level in the multimedia application in a future time period is greater than the activity threshold. For example, an object belonging to the active object layer will continue to log in to the multimedia application or continue to enter the task push activity in the multimedia application to execute multimedia tasks. Among them, the activity threshold can be 0 or other thresholds, which can be set according to specific requirements, and the embodiments of the present application do not limit this here.

[0041] Specifically, if the object stratification result indicates that the business object belongs to the pre-churn object layer, obtain the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer. It can be understood that the object stratification result is comprehensively determined according to each characteristic in the historical media interaction characteristics, and different historical media interaction characteristics have different influence degrees on the object stratification result. For example, the object stratification result is determined based on the historical media interaction characteristic T001, the historical media interaction characteristic T002, and the historical media interaction characteristic T003. The influence degree of the historical media interaction characteristic T001 on the object stratification result is 0.3, the influence degree of the historical media interaction characteristic T002 on the object stratification result is 0.4, and the influence degree of the historical media interaction characteristic T003 on the object stratification result is 0.3. Therefore, when the business object belongs to the pre-churn object layer (that is, the business object has a churn tendency), the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer can be obtained, and according to the corresponding influence degree of the historical media interaction characteristics, the object prediction result of the business object belonging to the pre-churn object can be personalized attributed to obtain the churn reason for the business object belonging to the pre-churn object layer.

[0042] Furthermore, according to the churn reason for the business object belonging to the pre-churn object, determine the task push scores corresponding to M multimedia tasks in the multimedia application, so as to obtain multimedia tasks that are more suitable for the business object, such as multimedia tasks that the business object is interested in or multimedia tasks with a difficulty level suitable for the business object, etc. In this way, the possibility of the business object churning can be reduced, thereby improving the interest and experience of the business object, and also increasing the retention time of the business object in the multimedia application.

[0043] Specifically, if the object stratification result indicates that the business object belongs to the active object layer, then according to the historical media interaction characteristics, the interaction label information corresponding to the business object under M multimedia tasks is predicted. Each multimedia task has one or more corresponding interaction label information. The interaction label information may include a task interaction label and a label value corresponding to the task interaction label. The task interaction label may include a task exposure label, a task completion label, a next-day retention label, a seven-day retention label, a next-day retention rate label, a seven-day retention rate label, an object online duration label, a final reward collection label, etc. after pushing the multimedia task to the business object.

[0044] Among them, taking the multimedia task M in the M multimedia tasks i as an example, the task exposure label corresponding to the multimedia task M i can be used to reflect whether the multimedia task M i is effectively displayed (effective display may refer to being displayed in the visible interface area of the terminal device, and the visible interface area is the interface area that the business object can view). The label value corresponding to the task exposure label may include 0 or 1. When the label value corresponding to the task exposure label is 0, it is used to indicate that the multimedia task M i is not effectively exposed. When the label value corresponding to the task exposure label is 1, it is used to indicate that the multimedia task M i is effectively exposed.

[0045] The task completion label corresponding to the multimedia task M i can be used to reflect whether the multimedia task M i is executed and completed by the business object. The label value corresponding to the task completion label may include 0 and 1. When the label value corresponding to the task completion label is 0, it is used to indicate that the multimedia task M i is not executed and completed by the business object. When the label value corresponding to the task completion label is 1, it is used to indicate that the multimedia task M i is executed and completed by the business object. The multimedia task M i The corresponding next-day retention label can be used to reflect whether the business object continues to enter the task push activity on the second day after pushing the multimedia task M i in the task push activity of the multimedia application. The label value corresponding to the next-day retention label may include 0 and 1. When the label value corresponding to the next-day retention label is 0, it is used to indicate that on the second day after pushing the multimedia task M i the business object does not enter the task push activity; when the label value corresponding to the next-day retention label is 1, it is used to indicate that on the second day after pushing the multimedia task M i the business object continues to enter the task push activity.

[0046] The multimedia task M iThe corresponding seven-day retention label can be used to reflect the push multimedia task M i On the seventh day after the task is pushed, the business object continues to enter the task push activity. The tag value corresponding to the seven-day retention tag can include 0 and 1. When the tag value corresponding to the seven-day retention tag is 0, it is used to indicate that the multimedia task M is pushed. i On the seventh day after the task was pushed, the business object did not enter the task push activity; when the tag value corresponding to the seven-day retention tag is 1, it is used to indicate that the multimedia task M is pushed. i On the seventh day after the task was pushed, the business object continued to participate in the task push activity. i The corresponding next-day retention rate label can be used to reflect the push multimedia task M i The number of objects that enter the task push activity on the day and the number of objects that push multimedia tasks M i The ratio of the number of objects that continue to enter the task push activity on the second day after the task is pushed. The higher the retention rate on the next day, the higher the retention intention of the business object in the multimedia application. The label value corresponding to the retention rate label on the next day is the object ratio. i The corresponding seven-day retention rate label can be used to reflect the push multimedia task M i The number of objects that enter the task push activity on the day and the number of objects that push multimedia tasks M i The ratio of the number of objects that continue to enter the task push activity on the seventh day after the last seven-day retention rate. The higher the seven-day retention rate, the higher the retention intention of the business object in the multimedia application. The label value corresponding to the next-day retention rate label is the object ratio.

[0047] Multimedia Task M i The corresponding object online duration label can be used to reflect the push multimedia task M i After that, the online time of the business object in the multimedia application or task push activity, the tag value corresponding to the object online time tag is the online time. The final reward collection tag corresponding to the multimedia task can be used to reflect the i After that, whether the business object receives the multimedia task M i Reward resources, or you will receive several multimedia tasks M i The reward resource, the tag value corresponding to the final reward collection tag may include 0 (used to indicate that no reward will be collected for multimedia task M i reward resources) or 1 (used to indicate that the multimedia task M i ), or about multimedia tasks M i The number of reward resources.

[0048] Specifically, based on the interaction tag information corresponding to each of the M multimedia tasks, the task push scores corresponding to the M multimedia tasks can be determined. For example, for multimedia task M i corresponding task exposure tags, task completion tags, next-day retention tags, seven-day retention tags, next-day retention rate tags, seven-day retention rate tags, object online duration tags, final reward collection tags, etc., determine the task push score for multimedia task M i corresponding to the task push score.

[0049] Furthermore, from the M multimedia tasks, determine the multimedia tasks whose task push scores are greater than or equal to the push score threshold, and push the multimedia tasks whose task push scores are greater than or equal to the push score threshold to the business object. In this way, it is possible to push more suitable multimedia tasks to the business object. When the business object belongs to the pre-churn object layer, the churn probability of the business object can be reduced, and the retention duration of the business object can be increased. When the business object belongs to the active object layer, various metrics of the business object in the multimedia application can be improved, such as retention duration, activity (e.g., performing more multimedia tasks), social interaction, etc.

[0050] Such as Figure 2 shown Figure 2 is a schematic diagram of a hierarchical-based multimedia task push method provided by an embodiment of the present application. As Figure 2 shown, the terminal device 20a can be any terminal device in the above-mentioned Figure 1 terminal device cluster. For example, the terminal device 20a can be the terminal device 100b in the above-mentioned Figure 1 . The terminal device 20a is installed with a multimedia application. The server 20c can be the server 10 in the above-mentioned Figure 1 . The business object 20b can be a user in the multimedia application installed on the terminal device 20a. The multimedia application installed on the terminal device 20a can provide multimedia interaction services for the business object 20b, that is, the business object 20b can perform multimedia interaction through the multimedia interaction services provided by the multimedia application. The terminal device 20a can obtain the historical media interaction data of the business object 20b in the multimedia application and send the historical media interaction data to the server 20c, and the server 20c can be the background server of the multimedia application.

[0051] The server 20c can extract features from the historical media interaction data of the service object 20b in the multimedia application to obtain the historical media interaction features of the service object 20b in the multimedia application. The historical media interaction features may include multimedia execution features, multimedia preference features, multimedia application login features, etc. Further, the server 20c can call an object hierarchical model, which can be used to hierarchically classify service objects to obtain the object hierarchical classification result of the service objects. The object hierarchical model can be a model with high complexity (such as having more model parameters), high precision, and high stability. For example, the object hierarchical model can be a deep learning model such as MLP (i.e., multi-layer perceptron) or DeepFM. In this way, the recognition accuracy of the object hierarchical classification result can be improved.

[0052] Among them, the multi-layer perceptron (MLP) is a forward-structured artificial neural network composed of multiple neuron layers, including an input layer, a hidden layer, and an output layer. The input layer is used to receive input data, the hidden layer is used to represent features through learning, and the output layer is used to generate the final prediction result. Each neuron in the hidden layer and the output layer has an activation function for introducing non-linear mapping. DeepFM is a model that shares input features between an FM network and a DNN network to jointly learn low-order features and high-order features. That is, the FM network can extract low-order features, and the DNN can extract high-order features, which can improve the model accuracy.

[0053] Specifically, the server 20c can input the historical media interaction features of the service object into the object hierarchical model. The object hierarchical model can hierarchically classify the service object according to the historical media interaction features to obtain the object hierarchical classification result of the service object. The object hierarchical classification result can be used to indicate that the service object belongs to the pre-churn object layer, or to indicate that the service object belongs to the active object layer. When the service object belongs to the pre-churn object layer, it means that the service object has a low activity level in the multimedia application in the future time period, or no longer logs in to the multimedia application, that is, the service object has a high churn tendency. When the service object belongs to the active object layer, it means that the service object has a high activity level in the multimedia application in the future time period, such as continuing to log in to the multimedia application for multimedia interaction, that is, the service object remains active in the multimedia application.

[0054] If the object stratification result indicates that the business object belongs to the pre-churn object layer, the server 20c can input the historical media interaction characteristics of the business object into an interpretable model. This interpretable model can show how decisions are made based on the entire feature space, model structure, model parameters, etc., and can also show which features have a higher impact on the final decision (i.e., which features are important), that is, it can explain how the final decision is obtained. The interpretable model integrates a feature evaluation method for evaluating the impact degree of each feature on the final decision and obtaining the feature importance of each feature. The interpretable model can include the XGBoost (eXtreme Gradient Boosting) model, the LightGBM (Light Gradient Boosting Machine) model, etc. The XGBoost model is a composite algorithm based on gradient boosting trees, which can improve efficiency and generalization ability in large-scale datasets and complex models.

[0055] The LightGBM model is a framework that implements the GBDT algorithm (i.e., an algorithm that uses weak classifiers (decision trees) for iterative training to obtain an optimal model), supports high-efficiency parallel training, and has advantages such as faster training speed, lower memory consumption, better accuracy, and support for distributed processing of massive data. Among them, the feature evaluation method can refer to SHAP (SHapley Additive exPlanations). SHAP calculates the marginal contribution of each feature to the model output and can explain the so-called "black box model" from both global and local levels. The SHAP tool is an additive explanation model, and all features are regarded as "contributors". For each prediction sample, the model generates a prediction value, and each feature of this sample will have a corresponding contribution value, which represents the impact degree of the feature on the prediction.

[0056] Specifically, the interpretable model can stratify business objects based on historical media interaction features to obtain the object stratification results of the business objects, and evaluate the influence degree of the historical media interaction features on the object stratification results of the business objects (i.e., the business objects belong to the pre-churn object layer) through the feature evaluation method in the interpretable model, so as to obtain the corresponding influence degree of the historical media interaction features. It can be understood that the number of the historical media interaction features can be one or more, and each historical media interaction feature has a corresponding influence degree. Further, the feature importance corresponding to the historical media interaction features can be determined according to the influence degree corresponding to the historical media interaction features. The influence degree corresponding to the historical media interaction features can include negative correlation influence and positive correlation influence, and the influence degree corresponding to the historical media interaction features can be directly determined as the feature importance corresponding to the historical media interaction features. Of course, it is also possible to obtain the weighted weight corresponding to each historical media interaction feature and determine the product of the weighted weight corresponding to each historical media interaction feature and the influence degree as the feature importance corresponding to the historical media interaction features.

[0057] Further, the server 20c can determine the task push scores corresponding to M multimedia tasks in the multimedia application according to the feature importance corresponding to the historical media interaction features, where M is a positive integer. It can be understood that when the business object belongs to the pre-churn object layer, personalized attribution can be performed on the fact that the business object belongs to the pre-churn object layer to obtain the important reasons for the business object to belong to the pre-churn object layer. At the same time, feature migration is performed on the feature importance corresponding to the historical media interaction features, and the feature importance corresponding to the historical media interaction features is migrated to the multimedia task push scenario as the multimedia task push feature, and then the task push scores corresponding to M multimedia tasks are determined respectively. In this way, according to the churn reasons for the business object belonging to the pre-churn object, the task push scores corresponding to M multimedia tasks in the multimedia application are determined, and it is possible to obtain multimedia tasks that are more suitable for the business object, such as multimedia tasks that the business object is interested in or multimedia tasks with a difficulty level suitable for the business object. In this way, the possibility of business object churn can be reduced, thereby improving the interest and experience of the business object, and the retention time of the business object in the multimedia application can also be provided.

[0058] If the object stratification result indicates that the business object belongs to the active object layer, the server 20c may call a multi-task recognition model, which is used to identify the interaction label information corresponding to the business object under M multimedia tasks. The interaction label information may include a task interaction label and a label value corresponding to the task interaction label. The task interaction label may include a task exposure label, a task completion label, a next-day retention label, a seven-day retention label, a next-day retention rate label, a seven-day retention rate label, an object online duration label, a final reward receiving label, etc. Since there is a certain correlation or common feature among the interaction label information, the multi-task recognition model can utilize this common knowledge to improve the learning effect of each interaction label information. The multi-task recognition model includes multiple output layers dedicated to task interaction labels and a common feature extraction layer at the bottom, and the information among multiple interaction label information is shared more efficiently, etc. The server 20c may input the historical media interaction features of the business object into the multi-task recognition model, and the multi-task recognition model may determine the interaction label information corresponding to the business object under M multimedia tasks according to the historical media interaction features.

[0059] Further, the server 20c may determine the task push scores corresponding to the M multimedia tasks according to the interaction label information corresponding to the business object under the M multimedia tasks. According to the task push scores corresponding to the M multimedia tasks, determine the multimedia tasks to be pushed to the business object from the M multimedia tasks. Specifically, the server 20c may determine, from the M multimedia tasks, the multimedia tasks whose task push scores are greater than or equal to the push score threshold as the multimedia tasks to be pushed to the business object. Send the multimedia task pushed to the business object to the terminal device 20a, and the terminal device 20a may display the multimedia task pushed to the business object in the task push activity of the multimedia application. In this way, the embodiment of the present application can adopt an adapted multimedia task push method according to the object stratification to which the business object belongs, can push more adapted multimedia tasks to the business object, improve the accuracy of multimedia task push, and further improve the user experience of the business object and the retention time and activity of the business object in the multimedia application.

[0060] Further, please refer to Figure 3 , Figure 3 is a schematic flowchart of a data processing method provided by an embodiment of the present application. As Figure 3 shown, this method may be executed by any terminal device in Figure 1 , or may be executed by the server 10 in Figure 1 , or may also be executed by Figure 1The terminal device and the server in [it] jointly execute, and the devices used to execute the data processing method in this application can be collectively referred to as computer devices. Among them, the data processing method may include but is not limited to the following steps:

[0061] S101. Obtain the historical media interaction characteristics of the service object in the multimedia application, and layer the service object according to the historical media interaction characteristics to obtain an object layering result.

[0062] The computer device can layer the service object, determine the task push method adapted to the service object according to the object layering result, and determine the multimedia task to be pushed to the service object by using the task push method adapted to the service object. In this way, it is possible to push a multimedia task more adapted to the service object for the service object, improve the accuracy of multimedia task pushing, and further improve the user experience of the service object as well as the retention time and activity of the service object in the multimedia application. Specifically, the computer device can obtain the historical media interaction data of the service object in the multimedia application during the historical time period. The historical media interaction data may include media execution data, media viewing data, multimedia application login data, multimedia task execution data, etc. Taking the multimedia application as a game application as an example, the historical media interaction data of the service object in the game application may include game application login data, game execution data, game teaming data, game sharing data, game task execution data, etc.

[0063] A computer device can extract historical media interaction features of a business object for a multimedia application from the historical media interaction data of the business object. The historical media interaction features can include multimedia execution features, multimedia preference features, multimedia application login features, etc. Taking the multimedia application as a game application as an example, the historical media interaction features of the business object in the game application can refer to game interaction features such as the game application login feature, game session feature, game preference feature, game duration feature, game task execution feature, etc. Since different business objects often have different multimedia task requirements, such as some business objects like multimedia tasks with simple operations, and some business objects like multimedia tasks with challenges, etc. At the same time, due to different retention tendencies of different business objects, the retention intervention space and retention intervention difficulty of different business objects are different. Therefore, business objects can be stratified, and different retention intervention methods are adopted for objects in different strata. Specifically, the computer device can perform feature analysis on the historical media interaction features, detect the churn tendency of the business object in the future time period, and stratify the business object according to the churn tendency of the business object in the future time period to obtain the object stratification result. The object stratification result can be used to indicate that the business object belongs to the pre-churn object layer, or to indicate that the business object belongs to the active object layer. An object belonging to the pre-churn object layer can refer to an object that is about to churn, that is, has a higher churn tendency; an object belonging to the active object layer can refer to a continuously active object, that is, has a lower churn tendency.

[0064] Optionally, the specific method for the computer device to stratify the business object according to the historical media interaction features to obtain the object stratification result can include: calling an object stratification model, and predicting the predicted activity of the business object in the future time period according to the historical media interaction features. If the predicted activity is less than or equal to the activity threshold, an object stratification result for indicating that the business object belongs to the pre-churn object layer is generated. If the predicted activity is greater than the activity threshold, an object stratification result for indicating that the business object belongs to the active object layer is generated.

[0065] Specifically, the computer device can input the historical media interaction characteristics of the business object into the object hierarchical model, call the object hierarchical model, and predict the predicted activity of the business object in the future time period according to the historical media interaction characteristics. The predicted activity can be used to reflect the activity of the business object in the multimedia application in the future time period, that is, to predict the churn trend of the business object. If the predicted activity is less than or equal to the activity threshold, an object hierarchical result indicating that the business object belongs to the pre-churn object layer is generated, that is, the business object has a high churn tendency, indicating that the business object has a low interest in the multimedia application. If the predicted activity is greater than the activity threshold, an object hierarchical result indicating that the business object belongs to the active object layer is generated, that is, the business object has a low churn tendency, indicating that the business object has a high interest in the multimedia application. Among them, the object hierarchical model can refer to a model with high complexity (such as more model parameters), high precision, and high stability. For example, the object hierarchical model can be a deep learning model such as MLP (i.e., multi-layer perceptron) or DeepFM. In this way, the recognition accuracy of the object hierarchical result can be improved.

[0066] Optionally, the historical media interaction characteristics include the media interaction characteristics of the business object in the first historical time period and the media interaction characteristics of the business object in the second historical time period, and the first historical time period is before the second historical time period. For example, the second historical time period can be the historical time period in the current active week, and the second historical time period can be the time period in the previous week of the current active week. The specific method for the computer device to call the object hierarchical model and predict the predicted activity of the business object in the future time period according to the historical media interaction characteristics can include: identifying the first activity of the business object in the first historical time period according to the media interaction characteristics of the business object in the first historical time period. Identifying the second activity of the business object in the second historical time period according to the media interaction characteristics of the business object in the second historical time period. Determining the activity change trend of the business object according to the first activity and the second activity, and predicting the predicted activity of the business object in the future time period according to the activity change trend.

[0067] Specifically, the computer device can identify the first activity level of the business object in the first historical time period based on the media interaction characteristics of the business object in the first historical time period. For example, taking the multimedia application as a game application, the computer device can obtain the game execution characteristics, game preference characteristics, game duration characteristics, and game task execution characteristics of the business object in the game application within the first historical time period, and determine the game activity level of the business object in the first historical time period as the first activity level. For example, the more game execution times, the more game preference characteristics, the longer the game duration, and the more game tasks executed by the business object in the first historical time period, the higher the game activity level of the business object in the first historical time period. Similarly, the computer device can identify the second activity level of the business object in the second historical time period based on the media interaction characteristics of the business object in the second historical time period.

[0068] The computer device can analyze the first activity level and the second activity level to generate the activity change trend of the business object. The activity change trend can be a gradually decreasing trend, a gradually increasing trend, a stable trend, or a fluctuating trend, etc. The computer device can predict the predicted activity level of the business object in the future time period through the object stratification model based on this activity change trend.

[0069] Optionally, the object stratification model can be pre-trained. The training process of the object stratification model can include: the computer device can obtain the third sample media interaction characteristics of the sample object in the multimedia application, and the stratification result label corresponding to the sample object. The stratification result label corresponding to the sample object can be determined manually according to the churn trend of the sample object in the prediction time period. The sample object can refer to an object with historical media interaction data in the multimedia application. If the sample object no longer logs in to the multimedia application or has a relatively low probability of using the multimedia application in the prediction time period, etc., the stratification result label corresponding to the sample object can be determined to reflect that the business object belongs to the pre-churn object layer. If the sample object has a relatively high probability of continuing to log in to the multimedia application or using the multimedia application in the prediction time period, etc., the stratification result label corresponding to the sample object can be determined to reflect that the business object belongs to the active object layer.

[0070] For example, a computer device can extract historical media interaction features belonging to the T1 cycle, historical media interaction features belonging to the T2 cycle, and historical media interaction features belonging to the T3 cycle from the historical media interaction data of a sample object. The historical media interaction features belonging to the T1 cycle and the historical media interaction features belonging to the T2 cycle are determined as the third sample media interaction features of the sample object in the multimedia application. At the same time, according to the historical media interaction features of the T3 cycle, the stratified result label corresponding to the sample object is determined, that is, if the historical media interaction features of the T3 cycle indicate the loss of the business object, the stratified result label is determined to reflect that the business object belongs to the pre-loss object layer. If the historical media interaction features of the T3 cycle indicate the activity of the business object, the stratified result label is determined to reflect that the business object belongs to the active object layer. The T1 cycle precedes the T2 cycle, the T2 cycle precedes the T3 cycle, and the T1 cycle, T2 cycle, and T3 cycle can be adjacent cycles.

[0071] As Figure 4 shown Figure 4 is a schematic diagram of a multimedia task push intervention cycle provided by an embodiment of the present application. As Figure 4 shown, in the training stage of training the initial object stratification model, an object belonging to the pre-loss object layer refers to an object that is active in the T1 cycle, in a pre-loss state (still belonging to the active state) in the T2 cycle, and in a lost state (such as no longer logging in to the multimedia application) in the T3 cycle. In the embodiment of the present application, the initial object stratification model can be trained to obtain the object stratification model. In the application stage of the object stratification model (i.e., the model application stage), it is possible to predict in the T1 cycle that the business object belongs to the pre-loss object layer. The cycle after the T1 cycle is determined as the multimedia task push intervention period (including the T2 cycle and the T3 cycle). According to the influence degree of the historical media interaction features on the business object belonging to the pre-loss object layer, the task push scores corresponding to M multimedia tasks in the multimedia application are determined, and then more accurate multimedia tasks are pushed for the business object according to the task push scores. This can improve the retention willingness of the business object and thus reduce the probability of the business object's loss.

[0072] As Figure 5 shown Figure 5 is a schematic diagram of a pre-loss object provided by an embodiment of the present application. As Figure 5As shown, the business object in 50a is an active object in the T1 cycle, the business object in 50b is an active object in the T2 cycle, and the business object in 50c is an active object in the T3 cycle. Since the business object in 50d is active in the T1 cycle and lost in the T3 cycle, the business object in 50d is a pre-lost object and belongs to the pre-lost object layer. The business object in 50e is an object that is active in the T1 cycle, the T2 cycle, and the T3 cycle, and the business object in 50f is a return object.

[0073] Furthermore, the computer device can call the initial object stratification model, predict the stratification result corresponding to the sample object according to the third sample media interaction feature, and obtain the predicted stratification result output by the initial object stratification model. The initial object stratification model can be any one of deep learning models such as MLP (i.e., multi-layer perceptron) and DeepFM. The computer device can determine the model loss of the initial object stratification model according to the difference between the predicted stratification result output by the initial object stratification model and the stratification result label, and adjust the model parameters of the initial object stratification model according to the model loss of the initial object stratification model to obtain the adjusted initial object stratification model. The computer device can detect whether the adjusted initial object stratification model meets the convergence condition, and the convergence condition can refer to that the model loss is less than the loss threshold, or the number of model training times reaches the target number. If the adjusted initial object stratification model does not meet the convergence condition, continue to train the adjusted initial object stratification model; if the adjusted initial object stratification model meets the convergence condition, the adjusted initial object stratification model reaches the convergence state, and the adjusted initial object stratification model is determined as the object stratification model.

[0074] S102, if the object stratification result indicates that the business object belongs to the pre-lost object layer, determine the task push scores corresponding to the M multimedia tasks in the multimedia application according to the influence degree of the historical media interaction feature on the business object belonging to the pre-lost object layer.

[0075] Specifically, the number of historical media interaction features is one or more, and each historical media interaction feature has a certain impact on the object stratification result of the business object (i.e., provides a certain contribution to the object stratification result of the business object). When the business object belongs to the pre-churn object layer, it indicates that the probability of the business object using the multimedia application in the future time period is relatively low, that is, the business object belongs to the pre-churn object of the multimedia application. Therefore, personalized attribution can be performed on the business object belonging to the pre-churn object layer to obtain the influence degree of the historical media interaction features on the business object belonging to the pre-churn object layer, that is, the influence degree reflects which historical media interaction features are the main reasons for the business object to belong to the pre-churn object layer. Pushing multimedia tasks to the business object based on the influence degree is beneficial to improving the accuracy of multimedia task pushing, reducing the possibility of business object churn, and then improving the interest and experience of the business object.

[0076] When the number of historical media interaction features is multiple, the influence degree of each historical media interaction feature on the object stratification result of the business object belonging to the pre-churn object is different. For example, the influence degree of some historical media interaction features on the object stratification result of the business object belonging to the pre-churn object is relatively low, and the influence degree of some historical media interaction features on the object stratification result of the business object belonging to the pre-churn object is relatively high. Further, the computer device can determine the task push scores corresponding to the M multimedia tasks in the multimedia application according to the influence degree of the historical media interaction features on the business object belonging to the pre-churn object layer. It can be seen that the embodiments of the present application can determine the task push scores corresponding to the M multimedia tasks respectively through the influence degree of the historical media interaction features that cause the business object to belong to the pre-churn object layer (i.e., the feature importance of the churn reasons that cause the business object to pre-churn). In this way, it is possible to obtain multimedia tasks that are more adapted to the business object, such as multimedia tasks that the business object is interested in or multimedia tasks with a difficulty level suitable for the business object. In this way, the possibility of business object churn can be reduced, the interest and experience of the business object can be improved, and the retention time of the business object in the multimedia application can also be increased.

[0077] Optionally, the specific method for the computer device to determine the task push scores corresponding to the M multimedia tasks in the multimedia application according to the influence degree of the historical media interaction features on the business object belonging to the pre-churn object layer may include: calling an interpretable model to predict the influence degree of the historical media interaction features on the business object belonging to the pre-churn object layer. Determining the feature importance of the historical media interaction features according to the influence degree, and determining the task push scores corresponding to the M multimedia tasks respectively according to the historical media interaction features and the feature importance of the historical media interaction features.

[0078] Specifically, the computer device can obtain an interpretable model. The interpretable model can refer to giving an explanation of how to make a decision based on the entire feature space, model structure, model parameters, etc., and at the same time can give which features have a higher impact on the final decision (that is, which features are important), that is, it can explain how to obtain the final decision. The interpretable model integrates a feature evaluation method for evaluating the impact degree of each feature on the final decision to obtain the feature importance of each feature. The interpretable model can include the XGBoost model, the LightGBM model, etc. Input the historical media interaction features of the business object into the interpretable model, and through the interpretable model, extract the features of the historical media interaction features of the business object, and predict the impact degree of the historical media interaction features on the business object belonging to the pre-churn object layer.

[0079] It can be understood that the interpretable model can stratify the business object according to the historical media interaction features to obtain the object stratification result of the business object, and through the feature evaluation method in the interpretable model, evaluate the impact degree of the historical media interaction features on the object stratification result of the business object (that is, the business object belongs to the pre-churn object layer) to obtain the corresponding impact degree of the historical media interaction features. Specifically, the computer device can predict the prediction probability of the business object belonging to the active object layer according to the interpretable model, and generate the object stratification result of the business object according to the prediction probability. For example, when the prediction probability is less than or equal to 0.5, it is determined that the business object does not belong to the active object layer and belongs to the pre-churn object layer; when the prediction probability is greater than 0.5, it is determined that the business object belongs to the active object layer. The impact degree corresponding to each historical media interaction feature can include the positive correlation impact degree and the negative correlation impact degree. When the prediction probability is less than or equal to 0.5, the prediction probability is jointly affected by the historical media interaction features with positive correlation impact degree and the historical media interaction features with negative correlation impact degree.

[0080] Taking the historical media interaction features of a business object, including historical media interaction feature T001, historical media interaction feature T002, and historical media interaction feature T003 as an example, if based on these historical media interaction features T001, T002, and T003, the predicted probability that the business object belongs to the active object layer is 0.3, at this time, the business object belongs to the pre-churn object layer. When predicting the influence degree of each historical media interaction feature on the business object belonging to the pre-churn object layer through an interpretable model, the influence degree (i.e., contribution degree) of historical media interaction feature T001 on the business object belonging to the active object layer is +0.2, the influence degree of historical media interaction feature T002 on the business object belonging to the active object layer is -0.2, and the influence degree of historical media interaction feature T003 on the business object belonging to the active object layer is +0.3. The predicted probability of 0.3 that the business object belongs to the active object layer is the sum of the influence degrees corresponding to historical media interaction features T001, T002, and T003 respectively. It can be seen that historical media interaction features T001 and T003 have a positive correlation influence degree on the probability of the business object belonging to the active object layer. However, historical media interaction feature T002 has a positive correlation influence degree on the probability of the business object belonging to the active object layer.

[0081] This means that historical media interaction features T001 and T003 can make the business object belong to the active object layer, while historical media interaction feature T002 will not make the business object belong to the active object layer, that is, historical media interaction feature T002 will cause the object to churn. Therefore, the influence degree corresponding to the historical media interaction feature can be directly determined as the feature importance (i.e., shapley value) corresponding to the historical media interaction feature. Among them, shapley is an additive feature attribution that can effectively express the prediction result of the model as an additive binary linear function. In this way, important historical media interaction features that can make the business object belong to the active object layer (i.e., features that keep the business object record in the multimedia) can be determined from the historical media interaction features, which is beneficial to improving the accuracy of multimedia task push, reducing the possibility of business object churn, and further improving the interest and experience of the business object.

[0082] Of course, it is also possible to obtain the weighted weight corresponding to each historical media interaction feature, and determine the product of the weighted weight corresponding to each historical media interaction feature and the influence degree as the feature importance corresponding to the historical media interaction feature. Further, the computer device can perform feature migration on the feature importance corresponding to the historical media interaction feature, and migrate the feature importance corresponding to the historical media interaction feature to the multimedia task push scenario as a multimedia task push feature, and then determine the task push scores corresponding to M multimedia tasks.

[0083] Optionally, the interpretable model may be pre-trained. The training process of the interpretable model may include: obtaining an initial interpretable model, the first sample media interaction features of the sample object in the multimedia application, and obtaining the labeled hierarchical result of the sample object. Invoke the initial interpretable model, and based on the first sample media interaction features, perform hierarchical classification on the sample object to obtain the predicted hierarchical result of the sample object. According to the labeled hierarchical result and the predicted hierarchical result, determine the model loss of the initial interpretable model. According to the model loss of the initial interpretable model, train the initial interpretable model until the initial interpretable model meets the convergence condition to obtain the interpretable model. Among them, the AUC of the interpretable model can be 86.65%. AUC is an evaluation index for binary classification models. The AUC value is a probability value. When randomly selecting a positive sample and a negative sample, the probability that the current classification algorithm ranks this positive sample in front of the negative sample based on the calculated score value is the AUC value. The larger the AUC value, the more likely the current classification algorithm is to rank the positive sample in front of the negative sample, and thus can better classify.

[0084] Specifically, the computer device may have an initial interpretable model, which can be used to output the object hierarchical result and determine the influence degree of each input feature on the object hierarchical result. The initial interpretable model is in a non-convergent state. Among them, the initial interpretable model may be an XGBoost model, a LightGBM model, etc. The computer device may obtain the first sample media interaction features of the sample object in the multimedia application. The sample object may refer to an object with historical media interaction data in the multimedia application. The first sample media interaction features may be the media interaction features of the sample object in the historical time period in the multimedia application. At the same time, the computer device may obtain the labeled hierarchical result of the sample object. The labeled hierarchical result may be determined manually or may be determined by a guiding model (such as a teacher model) with high complexity and high accuracy and in a convergent state based on the soft label obtained by performing object hierarchical classification according to the first sample media interaction features.

[0085] Further, the computer device may invoke the initial interpretable model, stratify the sample object according to the first sample media interaction feature, and obtain the predicted stratification result of the sample object. According to the labeled stratification result and the predicted stratification result, the model loss of the initial interpretable model is determined, and whether the initial interpretable model meets the convergence condition is detected according to the model loss. The convergence condition means that the model loss is less than the loss threshold, or the number of model training times reaches the target number. If the initial interpretable model does not meet the convergence condition, the model parameters of the initial interpretable model are adjusted according to the model loss of the initial interpretable model to obtain the adjusted initial interpretable model. The computer device may detect whether the adjusted initial interpretable model meets the convergence condition. If the adjusted initial interpretable model meets the convergence condition, the adjusted initial interpretable model is determined as the interpretable model. If the adjusted initial interpretable model does not meet the convergence condition, the adjusted initial interpretable model is continuously trained until the convergence condition is met to obtain the interpretable model.

[0086] Optionally, the labeled stratification result of the initial interpretable model regarding the sample object may be determined by the object stratification model. Since the model structure of the initial interpretable model itself is relatively simple and the number of model parameters is small, the accuracy of the initial interpretable model is low. Therefore, in the embodiments of the present application, a high-precision and high-complexity object stratification model is first trained, and the AUC of the object stratification model may be 89.53%. The object stratification model may be a Teacher model, and the initial interpretable model is a Student model. By using the object stratification model to generate more soft labels to guide the training of the lightweight initial interpretable model, the model accuracy of the trained interpretable model can be improved. It can be understood that through the Teacher-Student deep learning model framework, soft labels are generated by the Teacher model with higher accuracy (i.e., the object stratification model). The soft label refers to a probability value given by the Teacher model with higher accuracy for stratifying each sample object, which can convert the discrete label containing only two types of [0,1] into a continuous probability value, and more accurately guide the Student model (i.e., the initial interpretable model) to learn. In the Teacher-Student model, the training goal of the Student model is to maximize the similarity with the soft labels generated by the Teacher model, thereby improving the model accuracy of the interpretable model.

[0087] Specifically, the computer device can obtain an object stratification model for stratifying an object. The object stratification model is a model with a model complexity higher than that of the initial interpretable model and in a converged state. Invoke the object stratification model, and stratify the sample object according to the first sample media interaction feature to obtain the stratification result output by the object stratification model, and determine the stratification result output by the object stratification model as the labeled stratification result of the sample object. In this way, the model accuracy of the trained interpretable model can be improved.

[0088] Optionally, the specific manner in which the computer device determines the task push scores corresponding to the M multimedia tasks according to the historical media interaction features and the feature importance of the historical media interaction features may include: performing feature splitting on the historical media interaction features to obtain Q atomic features corresponding to the historical media interaction features; Q is an integer greater than 1. Determine the feature importance corresponding to each of the Q atomic features according to the feature importance of the historical media interaction features. For the multimedia task M i among the M multimedia tasks, perform task splitting to obtain P atomic tasks corresponding to the multimedia task M i ; P is an integer greater than 1, and i is a positive integer less than or equal to M. Determine the task push score corresponding to the multimedia task M i according to the feature importance corresponding to the P atomic tasks and the Q atomic features respectively.

[0089] Specifically, the computer device can perform feature splitting on the historical media interaction features, split the historical media interaction features of the high-order cross feature type into atomic features of the first-order atomic feature type, and obtain Q atomic features corresponding to the historical media interaction features. The number of the historical media interaction features is one or more. When the number of the historical media interaction features is multiple, feature splitting can be performed on each of the multiple historical media interaction features to obtain the atomic features corresponding to each historical media interaction feature. The Q atomic features corresponding to the historical media interaction features refer to the set of atomic features corresponding to all historical media interaction features. Further, the computer device can determine the feature importance corresponding to each of the Q atomic features according to the feature importance of the historical media interaction features.

[0090] In the process of determining the feature importance corresponding to each of the Q atomic features, the computer device can obtain the contribution degree of the historical media interaction features to each atomic feature. Taking the number of historical media interaction features as one example, the computer device can obtain the ratio between 1 and the number of the Q atomic features (i.e., 1 / Q) as the contribution degree of the historical media interaction features to each atomic feature. The computer device can obtain the product of the contribution degree of the Q atomic features and the feature importance of the historical media interaction features to obtain the feature importance corresponding to each of the Q atomic features. It can be understood that the computer device can evenly distribute the feature importance of the historical media interaction features to the atomic features corresponding to the historical media interaction features to obtain the feature importance of each atomic feature (i.e., the feature importance of the historical media interaction features multiplied by 1 / Q).

[0091] Of course, when the number of historical media interaction features is multiple, the computer device can obtain the ratio between 1 and the number of atomic features corresponding to each historical media interaction feature as the contribution degree of the corresponding historical media interaction feature to each of its atomic features. Obtain the product of the feature importance of each historical media interaction feature and the contribution degree of each of its atomic features to obtain the feature importance of the corresponding historical media interaction feature. It can be understood that when the contribution degree of each historical media interaction feature to each of its atomic features is the same, the feature importance of each historical media interaction feature can be evenly distributed to each of its atomic features to obtain the feature importance corresponding to each of its atomic features.

[0092] Of course, when the atomic features corresponding to multiple historical media interaction features are the same, the feature importance of the same atomic feature can be the sum of the feature importance of the atomic features corresponding to the multiple historical media interaction features. For example, the historical media interaction feature T001 includes the atomic feature "island", and the feature importance of the atomic feature "island" included in the historical media interaction feature T001 is 0.1. The historical media interaction feature T002 also includes the atomic feature "island", and the feature importance of the atomic feature "island" included in the historical media interaction feature T002 is 0.3. Then the feature importance of the atomic feature "island" is 0.4.

[0093] For example, taking a multimedia application as a game application, the historical media interaction features may include game mode preference features. If the game mode preference feature of a business object is "Top 3 in Classic Island", that is, the business object likes to carry out game activities in the "Top 3 in Classic Island" game mode, and the feature importance of "Top 3 in Classic Island" is 0.9. The computer device can split the feature of "Top 3 in Classic Island" into 3 atomic features, namely the atomic feature "Classic", the atomic feature "Island", and the atomic feature "Top 3". Further, the computer device can obtain the feature of "Top 3 in Classic Island" and calculate the contribution degrees corresponding to the 3 atomic features of the atomic feature "Classic", the atomic feature "Island", and the atomic feature "Top 3" respectively. Among them, the contribution degree mapping table of the feature of "Top 3 in Classic Island" to the 3 atomic features is W1 = [0.33, 0.33, 0.33], that is, the feature of "Top 3 in Classic Island" has the same contribution degree (i.e., 0.33) to the 3 atomic features of the atomic feature "Classic", the atomic feature "Island", and the atomic feature "Top 3". In this way, the computer device can evenly distribute the feature importance of "Top 3 in Classic Island" to the 3 atomic features of the atomic feature "Classic", the atomic feature "Island", and the atomic feature "Top 3". In this way, the feature importance of the atomic feature "Classic", the atomic feature "Island", and the atomic feature "Top 3" is all 0.3.

[0094] The computer device can also split each of the M multimedia tasks to obtain P atomic tasks corresponding to each multimedia task. When the multimedia tasks are different, the value of P can be the same or different. Specifically, taking the multimedia task M among the M multimedia tasks i as an example, the computer device can split the multimedia task M i to obtain P atomic tasks corresponding to the multimedia task M i where i is a positive integer less than M. When the value of i is different, the value of P can be the same or different, that is, the number of atomic tasks corresponding to each multimedia task can be the same or different. For example, if the multimedia task M i is "Classic Mode Island Top", then the computer device splits the "Classic Mode Island Top" to obtain 3 atomic tasks, namely the atomic task "Classic", the atomic task "Island", and the atomic task "Top".

[0095] Further, the computer device performs feature conversion on the feature importance of Q atomic features, and converts the Q atomic features and the feature importance of the Q atomic features into task scores corresponding to the P atomic tasks respectively. Further, according to the task scores corresponding to the P atomic tasks, the multimedia task M iThe corresponding task push score is obtained in this way, and each multimedia task among the M multimedia tasks respectively corresponds to a task push score.

[0096] Optionally, the computer device determines the multimedia task M according to P atomic tasks and the feature importance respectively corresponding to Q atomic features i The specific manner of determining the corresponding task push score may include: screening out sub-features that match the task features respectively corresponding to the P atomic tasks from the Q atomic features to obtain associated sub-features. According to the conversion ratio between the feature importance and the task score, and the feature importance corresponding to the associated sub-features, determine the task scores respectively corresponding to the P atomic tasks. Sum up the task scores respectively corresponding to the P atomic tasks to obtain the task push score corresponding to the multimedia task M i The corresponding task push score.

[0097] Specifically, the computer device can screen out sub-features that match the task features respectively corresponding to the P atomic tasks from the Q atomic features to obtain associated sub-features. It can be understood that the computer device can screen out sub-features that match the task features of each atomic task among the P atomic tasks from the Q atomic features to obtain associated sub-features. The computer device can determine the conversion ratio between the feature importance and the task score according to the matching degree between the atomic feature and the task feature of the atomic task. This conversion ratio can be obtained from the mapping table between the feature importance and the task score. This mapping table can be W2 = I, and I can be values such as 0.1, 0.3, 1, 2, 3, etc., which can be determined according to the matching degree between the atomic feature and the task feature of the atomic task. For example, the higher the matching degree between the atomic feature and the task feature of the atomic task, the higher the conversion ratio between the feature importance and the task score; of course, the lower the matching degree between the atomic feature and the task feature of the atomic task, the lower the conversion ratio between the feature importance and the task score. According to the conversion ratio between the feature importance and the task score, and the feature importance corresponding to the associated sub-features, determine the task scores respectively corresponding to the P atomic tasks.

[0098] For example, if the Q atomic features include atomic features "classic", "island", "Top3", "desert", "forest", "popular", etc., and the P atomic tasks include atomic tasks "classic", "island", "Top". The computer device can determine, from the Q atomic features, the atomic feature that matches the task feature of the atomic task "classic", that is, the atomic feature "classic". According to the matching degree between the atomic feature "classic" and the task feature of the atomic task "classic", determine the conversion ratio for converting the feature importance of the atomic feature "classic" into the task score of the atomic task "classic", and use it as the conversion ratio corresponding to the atomic task "classic". The computer device can use the product of the feature importance of the atomic feature "classic" and the conversion ratio corresponding to the atomic task "classic" as the task score of the atomic task "classic". For example, since the matching degree between the atomic feature "classic" and the task feature of the atomic task "classic" is 100%, the conversion ratio 1 is used as the conversion ratio corresponding to the atomic task "classic". At this time, the feature importance of the atomic feature "classic" can be determined as the task score of the atomic task "classic".

[0099] Similarly, the computer device can determine, from the Q atomic features, the atomic feature that matches the task feature of the atomic task "island", that is, the atomic feature "island". According to the matching degree between the atomic feature "island" and the task feature of the atomic task "island", determine the conversion ratio for converting the feature importance of the atomic feature "island" into the task score of the atomic task "island", and use it as the conversion ratio corresponding to the atomic task "island". Use the product of the feature importance of the atomic feature "island" and the conversion ratio corresponding to the atomic task "island" as the task score of the atomic task "island". For example, since the matching degree between the atomic feature "island" and the task feature of the atomic task "island" is 100%, the conversion ratio 1 is used as the conversion ratio corresponding to the atomic task "island". At this time, the feature importance of the atomic feature "island" can be determined as the task score of the atomic task "island".

[0100] Similarly, the computer device can determine, from the Q atomic features, the atomic feature that matches the task feature of the atomic task "Top", i.e., the atomic feature "Top3", and use the product of the feature importance of the atomic feature "Top3" and the conversion ratio corresponding to the atomic task "Top" as the task score of the atomic task "Top". Since the matching degree between the atomic feature "Top3" and the task feature of the atomic task "Island" is less than 100%, a value can be determined between 0 and 1 as the conversion ratio corresponding to the atomic task "Top". At this time, the product of the feature importance of the atomic feature "Top3" and the conversion ratio corresponding to the atomic task "Top" can be determined as the task score of the atomic task "Top".

[0101] Further, when the computer device obtains the task scores corresponding to the P atomic tasks respectively, it can obtain the contribution degree of the multimedia task M i , to the P atomic tasks. Among them, since the multimedia task M i is a combination of the P atomic tasks, theoretically completing the multimedia task M i also means completing the P atomic tasks. Therefore, the contribution degree of the multimedia task M i to the P atomic tasks is all 1, and the mapping matrix of the contribution degree of the multimedia task M i to the P atomic tasks is W3 = [1, 1, 1] T . Therefore, the computer device can sum up the task scores corresponding to the P atomic tasks respectively to obtain the task push score corresponding to the multimedia task M i . For example, if the P atomic tasks include the atomic task "Classic", the atomic task "Island", and the atomic task "Top", and the task score of each atomic task is 0.3, then the task push score corresponding to the multimedia task M i = 1 * 0.3 + 1 * 0.3 + 1 * 0.3 = 0.9.

[0102] As Figure 6 shown, Figure 6 is a schematic diagram of a mapping table from historical interactive media features to multimedia tasks provided by an embodiment of the present application. As Figure 6As shown, the computer device can generate a mapping table 60a for converting historical media interaction features into atomic features of historical media interaction features. The mapping table 60a from historical media interaction features to atomic features records the contribution degree W1 of historical media interaction features to atomic features. At the same time, the computer device can obtain a mapping table 60b for converting atomic tasks into multimedia tasks. The mapping table 60b from atomic tasks to multimedia tasks records the contribution degree W2 of multimedia tasks to atomic tasks. The computer device can obtain the product between the mapping table 60a and the mapping table 60b, and then obtain a mapping table 60c from historical media interaction features to multimedia tasks. Through the mapping table 60c from historical media interaction features to multimedia tasks, the historical media interaction features can be quickly converted into task features of multimedia tasks.

[0103] Specifically, for the business objects in the pre-churn object layer, during the task push activity, the online duration of the business objects is relatively increased by 7.19%, the average number of online days per day is relatively increased by 6.38%, and the average exposure completion rate (UV) is relatively increased by 6.04%. The next-day retention rate of the business objects is relatively increased by 2.78%, the 3-day retention rate is relatively increased by 7.58%, and the weekly retention rate is relatively increased by 8.48%.

[0104] S103, if the object stratification result indicates that the business object belongs to the active object layer, then according to the historical media interaction features, predict the interaction label information corresponding to the business object under M multimedia tasks respectively. According to the interaction label information corresponding to M multimedia tasks respectively, determine the task push scores corresponding to M multimedia tasks respectively.

[0105] Specifically, if the object stratification result indicates that the business object belongs to the active object layer, the computer device can predict the interaction label information corresponding to the business object under M multimedia tasks respectively according to the historical media interaction features. Among them, each multimedia task has one or more corresponding interaction label information. The interaction label information can include task interaction labels and label values corresponding to the task interaction labels. The task interaction labels can include task exposure labels, task completion labels, next-day retention labels, seven-day retention labels, next-day retention rate labels, seven-day retention rate labels, object online duration labels, final reward receipt labels, etc. after pushing the multimedia task to the business object. Further, according to the interaction label information corresponding to M multimedia tasks respectively, comprehensively determine the task push scores corresponding to M multimedia tasks respectively. In this way, more suitable multimedia tasks can be pushed for the business object. When the business object belongs to the pre-churn object layer, the churn probability of the business object can be reduced and the retention duration of the business object can be increased. When the business object belongs to the active object layer, various indicators of the business object in the multimedia application can be improved, such as retention duration, activity (such as executing more multimedia tasks), social and other indicators.

[0106] Optionally, the specific manner in which the computer device predicts the interaction label information corresponding to the service object under M multimedia tasks according to the historical media interaction characteristics may include: invoking a multi-task recognition model to extract, from the historical media interaction characteristics, the associated media characteristics associated with the multimedia task M among the M multimedia tasks; i is a positive integer less than or equal to M. According to the associated media characteristics, label prediction is performed on the multimedia task M i to obtain the interaction label information of the service object under the multimedia task M i i

[0107] Specifically, the computer device may invoke a multi-task recognition model, which can be used to predict the interaction label information corresponding to the service object under M multimedia tasks according to the historical media interaction characteristics of the service object. The computer device may input the historical media interaction characteristics of the service object into the multi-task recognition model. At this time, the historical media interaction characteristics of the service object may include the task interaction characteristics of the service object in the task push activity of the multimedia application. The computer device may invoke a multi-task recognition model to extract, from the historical media interaction characteristics, the associated media characteristics associated with the multimedia task M among the M multimedia tasks i

[0108] Among them, the interaction label information under each multimedia task includes multiple task interaction labels and the label values corresponding to each task interaction label respectively. The multiple task interaction labels under each multimedia task may include the task interaction label and the label value corresponding to the task interaction label. The task interaction label may include any of the labels such as task exposure label, task completion label, next-day retention label, seven-day retention label, next-day retention rate label, seven-day retention rate label, object online duration label, final reward collection label, etc. There is a certain correlation or common feature among the task interaction labels. Therefore, the multi-task recognition model can utilize this common knowledge to improve the learning effect of each task interaction label. The multi-task recognition model includes multiple output layers dedicated to task interaction labels and a common feature extraction layer at the bottom layer, and the information between multiple interaction label information is shared more efficiently, etc

[0109] Among them, the multi-task recognition model may include a shared network layer (i.e., the network layer shared by each task interaction label) and an exclusive network layer dedicated to each task interaction label. Through the shared network layer in the multi-task recognition model, from the associated media characteristics associated with the multimedia task M i extract the features associated with the multimedia task M i ​​​Fusion media features associated with multiple task interaction tags under. Further, the computer device can extract, from the fusion media features, exclusive media features associated with each task interaction tag through an exclusive network layer dedicated to each task interaction tag in the multi-task recognition model, and then predict the tag value of each task interaction tag based on the exclusive media features of each task interaction tag. The multimedia task M i The multiple task interaction tags under, and the tag value corresponding to each task interaction tag, are determined as the interaction tag information of the business object in the multimedia task M i under.

[0110] Optionally, the number of pieces of interaction tag information corresponding to each multimedia task is multiple, and the interaction tag information may include a task interaction tag and the tag value corresponding to the task interaction tag. Taking the multimedia task M i in the M multimedia tasks as an example, the interaction tag information corresponding to the multimedia task M i may include a task interaction tag and the tag values respectively corresponding to the task interaction tag. When calculating the task push score corresponding to the multimedia task M i , a target task interaction tag can be determined from the task interaction tags corresponding to the multimedia task M i , and the tag value corresponding to the target task interaction tag is determined as the task push score corresponding to the multimedia task M i . For example, the task interaction tags corresponding to the multimedia task M i may include a task exposure tag, a task completion tag, a next-day retention tag, a seven-day retention tag, a next-day retention rate tag, a seven-day retention rate tag, an object online duration tag, a final reward collection tag, etc. The tag value corresponding to the seven-day retention rate tag (i.e., the retention rate) can be determined as the task push score corresponding to the multimedia task M i .

[0111] Optionally, the interaction tag information corresponding to the multimedia task M i in the M multimedia tasks includes the tag values respectively corresponding to multiple task interaction tags. When there are multiple task interaction tags corresponding to the multimedia task M i , the task push score corresponding to the multimedia task M i can also be comprehensively determined according to the tag values respectively corresponding to the multiple task interaction tags corresponding to the multimedia task M i . The specific manner in which the computer device determines the task push scores respectively corresponding to the M multimedia tasks according to the interaction tag information respectively corresponding to the M multimedia tasks may include: summing the tag values respectively corresponding to the multiple task interaction tags corresponding to the multimedia task M i to obtain a total tag value. The total tag value is determined as the multimedia task M iThe corresponding task push score.

[0112] Of course, the computer device can also score each task interaction label according to the scoring rules corresponding to each task interaction label and the label value corresponding to each task interaction label, and obtain the label score corresponding to each task interaction label. For the multimedia task M i Sum the label scores corresponding to the multiple task interaction labels respectively corresponding to it to obtain the multimedia task M i The corresponding task push score.

[0113] Optionally, the multi-task recognition model can be pre-trained. The training process of the multi-task recognition model can include: obtaining the second sample media interaction features of the sample object in the multimedia application. The sample object can refer to an object with historical media interaction data in the multimedia application. The second sample media interaction features can include the task interaction features of the sample object in the task push activities of the multimedia application during the historical time period. The computer device can also obtain the labeled interaction label information corresponding to the sample object under M multimedia tasks, call the initial multi-task recognition model, and predict the predicted interaction label information corresponding to the sample object under M multimedia tasks according to the second sample media interaction features. Determine the model loss of the initial multi-task recognition model according to the labeled interaction label information and the predicted interaction label information. Train the initial multi-task recognition model according to the model loss of the initial multi-task recognition model until the initial multi-task recognition model meets the convergence condition to obtain the multi-task recognition model. Among them, the convergence condition can refer to that the model loss is less than the loss threshold, or the number of model training times reaches the target number.

[0114] Among them, when obtaining the second sample media interaction feature, the task interaction data of the sample object can be extracted from the task interaction data of all objects in the task push activity of the multimedia application, and the second sample media interaction feature of the business object can be extracted from the task interaction data of the sample object. Since the multimedia task space is different in different periods, the access caliber of the delivery object is also different, so the computer device can refine the task delivery feature of the task push activity of the multimedia application, and extract the task interaction data of the sample object according to the task delivery feature. Since the exposure object coverage level of different tasks is quite different, and the task exposure coverage object UV (i.e., the number of visitors) is very different, and the activities that different objects have participated in in history are quite different, ranging from a few to dozens, there is a cold start problem for some multimedia tasks and objects. In one embodiment, the computer device can statistically analyze the sample online period and the coverage object level, truncate according to the sampling k cycles (such as 2 cycles) of a single task coverage object, and reverse according to the exposure time, so as to cover all multimedia tasks and objects, and have sufficient sample data to extract sufficient second sample media interaction features.

[0115] Among them, the multi-task recognition model can output the label interaction information of the business object under each multimedia task, and the label interaction information can include the task interaction label and the label value of the task interaction label. The number of label interaction information under each multimedia task can be multiple. The task interaction label can include a task exposure label, a task completion label (which can be represented by is_complete_label), a next-day retention label, a seven-day retention label, a next-day retention rate label, a seven-day retention rate label, a T-day online time label, a T+1-day online time label (which can be represented by onlinetime_2day_nom), a T+6-day online time label (which can be represented by onlinetime_7day_nom), a final reward collection label, a task score (which can be represented by complete_value_label), a normalized task score (which can be represented by complete_value_nom), etc. Among them, the normalized task score can refer to the task score obtained by normalization processing, and its value is in the interval [0,1]. The specific types and meanings of the task interaction labels can be seen in the following Table 1.

[0116]

[0117]

[0118] Table 1

[0119] Specifically, the embodiments of the present application adopt a multi-task recognition model, which can improve the prediction accuracy of the interaction label information of business objects under M multimedia tasks. Table 2 below can show the differences between the multi-task recognition model and the single-task recognition model.

[0120]

[0121] Table 2

[0122] Specifically, for the objects in the active object layer, during the task push activity, the average daily online duration of the business object increased by 8.43% relative to the baseline, the average number of days the business object was online per day increased by 6.78% relative to the baseline, the average exposure task completion rate (UV) increased by 4.24% relative to the baseline. The average exposure task reward collection rate (UV) increased by 3.25% relative to the baseline, the next-day retention rate increased by 6.89% relatively, the three-day retention rate increased by 4.76% relatively, and the weekly retention rate increased by 4.45% relatively.

[0123] S104. Push the multimedia tasks among the M multimedia tasks to the business object, where the task push score of the multimedia task is greater than or equal to the push score threshold.

[0124] Specifically, the computer device can screen out the multimedia tasks among the M multimedia tasks whose task push scores are greater than or equal to the push score threshold, and push the screened multimedia tasks to the business object. In this way, more suitable multimedia tasks can be pushed to the business object. When the business object belongs to the pre-churn object layer, the churn probability of the business object can be reduced, and the retention duration of the business object can be increased. When the business object belongs to the active object layer, various indicators of the business object in the multimedia application can be improved, such as the retention duration, activity (such as executing more multimedia tasks), social interaction, etc. Of course, the computer device can extract the target number of multimedia tasks from the M multimedia tasks according to the task push scores corresponding to the M multimedia tasks as the multimedia tasks pushed to the business object. For example, the top 5 multimedia tasks with the highest push rankings among the M multimedia tasks are used as the multimedia tasks pushed to the business object.

[0125] In the embodiments of the present application, when the business object belongs to the active object layer, the interaction label information of the business object under M multimedia tasks predicted by the multi-task recognition model is used to determine the task push scores corresponding to the M multimedia tasks respectively. Then, when pushing multimedia tasks for the business object according to the task push scores, the in-game time during the activity period of the business object in the multimedia application is increased by 8.43% compared with the baseline, the next-day retention rate is increased by 6.89% compared with the baseline, the 3-day retention rate is increased by 4.76% relatively, and the 7-day retention rate is increased by 4.45% relatively. It can be seen that for business objects belonging to the active object layer, their activity can be improved and their online duration can be maintained. When the business object belongs to the pre-churn object layer, based on the historical media interaction characteristics predicted by the interpretable model, and aiming at the influence degree of the business object belonging to the pre-churn object layer, when determining the task push scores corresponding to the M multimedia tasks respectively, the online duration of the business object in the multimedia application is increased by 7.19% relatively compared with the control group, and the average single-day online duration of each exposed business object is increased by 7.19% relatively. It can be seen that for business objects belonging to the pre-churn object layer, their retention willingness can be improved and the churn probability of the business object can be reduced.

[0126] As Figure 7 shown, Figure 7 is a schematic diagram of personalized multimedia task push provided by the embodiments of the present application. As Figure 7 shown, in the natural state, the highly active object A01 at time T remains in the retained state in the multimedia application at time T + 1, and the pre-churn object A02 at time T is in the churn state at time T + 1, such as no longer logging in to the multimedia application. When using the personalized task push technology of the embodiments of the present application, at time T, personalized multimedia tasks can be obtained for the highly active object A01 and the pre-churn object A02 respectively. For example, a multimedia task with a higher complexity is pushed for the highly active object A01 to increase the fun of task execution, and a multimedia task with a lower complexity and more task reward resources is pushed for the pre-churn object A02 to mainly improve the retention tendency of the pre-churn object A02. When the highly active object A01 and the pre-churn object A02 complete their respective multimedia tasks, they can obtain the task reward resources corresponding to the multimedia tasks. At time T + 1, the activity of the highly active object A01 can be improved, and the retention tendency of the pre-churn object A02 can be improved.

[0127] In the embodiments of the present application, based on the object stratification to which a service object belongs in a multimedia application, a strategy for pushing multimedia tasks to the service object is dynamically selected to implement personalized pushing of multimedia tasks and improve the accuracy of multimedia task pushing. Specifically, when the service object belongs to the pre-churn object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively low, that is, the service object belongs to the pre-churn object of the media application. Therefore, personalized attribution can be performed on the fact that the service object belongs to the pre-churn object layer to obtain the influence degree of historical media interaction characteristics on the service object belonging to the pre-churn object layer, that is, the influence degree reflects which historical media interaction characteristics are the main reasons for the service object to belong to the pre-churn object layer. Pushing multimedia tasks to the service object based on the influence degree is beneficial to improving the accuracy of multimedia task pushing, reducing the possibility of service object churn, and further improving the interest and experience of the service object. When the service object belongs to the active object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively high, that is, the service object belongs to the active object of the media application. Therefore, the interaction label information corresponding to the service object under M multimedia tasks can be predicted, and the interaction label information can be used to reflect the active situation and retention situation of the service object after pushing the corresponding multimedia tasks. Based on the interaction label information corresponding to the service object under M multimedia tasks respectively, it is beneficial to improve the accuracy of multimedia task pushing, and improve the activity and retention time of the service object in the multimedia application.

[0128] Further, please refer to Figure 8 , Figure 8 which is a schematic flowchart of a data processing method provided by an embodiment of the present application. As Figure 8 shown, this method can be executed by any terminal device in Figure 1 , or can be executed by server 10 in Figure 1 , or can also be jointly executed by the terminal device and the server in Figure 1 . The devices used to execute this data processing method in the present application can be collectively referred to as computer devices. Among them, this data processing method may include but is not limited to the following steps:

[0129] S201, obtain the historical media interaction characteristics of the service object in the multimedia application, and stratify the service object according to the historical media interaction characteristics to obtain an object stratification result.

[0130] S202, if the object stratification result indicates that the service object belongs to the pre-churn object layer, determine the task push scores corresponding to M multimedia tasks in the multimedia application according to the influence degree of the historical media interaction characteristics on the service object belonging to the pre-churn object layer.

[0131] S203. If the object stratification result indicates that the service object belongs to the active object layer, then according to the historical media interaction characteristics, predict the interaction label information corresponding to the service object under M multimedia tasks respectively, and determine the task push scores corresponding to the M multimedia tasks respectively according to the interaction label information corresponding to the M multimedia tasks respectively.

[0132] Specifically, for the content of steps S201 - S203 in the embodiments of the present application, reference can be made to the content of steps S101 - S103 above, and the embodiments of the present application will not be elaborated herein.

[0133] S204. In response to the trigger operation of the service object for the game task, display the game task interface.

[0134] S205. In the game task interface, display the pushed game task and the executable times of the service object for the pushed game task.

[0135] Specifically, the multimedia application can be a game application, and the multimedia task can be a game task in the game application. When the multimedia task is a game task, the computer device can determine, from M game tasks, the game tasks whose task push scores are greater than or equal to the push score threshold as the pushed game tasks. When the computer device detects the trigger operation of the service object for the game task, the computer device can, in response to the trigger operation of the service object for the game task, display the game task interface, and in the game task interface, display the pushed game task and the executable times of the service object for the pushed game task. The executable times of the service object for the pushed game task refer to the maximum number of times the service object can execute the pushed game task. For example, when the executable times is 10 times, the service object can execute the pushed game task at most 10 times. In a feasible manner, when the service object executes a pushed game task once, the computer device can issue task reward resources for the pushed game task to the service object, such as game props, game skins, game mode experience permissions, etc.

[0136] Optionally, when calculating the number of executable times of a push game task for a business object, the computer device may obtain the task push score of the push game task, the maximum number of executable times associated with the push game task, and the minimum number of executable times associated with the push game task. The computer device may determine the difference in the number of times between the maximum number of executable times and the minimum number of executable times, obtain the product of the difference in the number of times and the task push score, and sum the product and the minimum number of executable times to obtain the number of executable times of the business object for the push game task. For example, the push game task is a game teaming task, the maximum number of executable times of the game teaming task is 20 times, the maximum number of executable times of the game teaming task is 3 times, and the task push score of the game teaming task is 0.5. At this time, the number of executable times of the business object for the push game task = 0.5 * (20 - 10) + 10 = 15, that is, the number of executable times = task push score * (the maximum number of executable times of the multimedia task - the minimum number of executable times of the multimedia task) + the minimum number of executable times of the multimedia task. In this way, the number of executable times of the push game task can be determined for the business object in a personalized manner to implement customizing the quantity of different task reward resources for the business object.

[0137] As Figure 9 shown, Figure 9 is a schematic diagram of a push game task provided by an embodiment of the present application. As Figure 9 shown, taking the multimedia application as a game application and the multimedia task as a game task as an example, the computer device may display the push game task to the business object in the game task interface, such as the game task "Complete a classic battle in a team with friends" and the game task "Survive for 25 minutes in any game mode". At the same time, the computer device may display the number of executable times of the business object for the game task. For example, the number of executable times of the game task "Complete a classic battle in a team with friends" is 20 times, and each time the game task "Complete a classic battle in a team with friends" is executed, 5 lucky values can be obtained. For example, the number of executable times of the game task "Survive for 25 minutes in any game mode" is 25 times, and each time the game task "Survive for 25 minutes in any game mode" is executed, 5 lucky values can also be obtained. The computer device may also display the task reward resources that the business object can finally obtain in the game task interface, such as a battle suit. The current lucky value of the business object is 45. When the lucky value of the business object reaches 100, the task reward resources (such as a battle suit) displayed in the game task interface can be obtained.

[0138] As Figure 10 shown, Figure 10 is a schematic diagram of a push game task provided by an embodiment of the present application. As Figure 10As shown, taking the multimedia application as the game application and the multimedia task as the game task as an example, the computer device can display, in the game task interface, the game tasks pushed to the service object, such as game task 100a and game task 100b. The game task 100a is used to indicate participating in 4 rounds in any game mode, and the game task 100b is used to indicate eliminating 5 virtual objects in the mission game mode. At the same time, the computer device can display the executable times corresponding to the game task 100a and the game task 100b respectively. For example, the executable times corresponding to the game task 100a are 4 times, and the executable times of the game task 100b are 5 times. After completing the executable times corresponding to the game task 100a, the task reward resources corresponding to the game task 100a (i.e., the battle suit) can be obtained. Similarly, after completing the executable times corresponding to the game task 100b, the task reward resources corresponding to the game task 100b (i.e., the battle vehicle) can be obtained.

[0139] Optionally, the embodiments of the present application can be applied to the video task push scenario. The multimedia application can be a video application. The M multimedia tasks in the multimedia application can be video tasks in the video application, and the historical media interaction feature can be the historical video interaction feature of the service object in the video application. Through the personalized push task solution in the embodiments of the present application, when the service object belongs to the pre-churn object in the video application, the personalized attribution can be performed on the pre-churn object layer to obtain the influence degree of the historical video interaction feature on the pre-churn object layer of the service object, that is, the influence degree reflects which historical video interaction features are the main reasons for the service object to belong to the pre-churn object layer. Based on the influence degree, video tasks are pushed to the service object, which is beneficial to improving the accuracy of video task push, reducing the possibility of service object churn, and then improving the interest and experience of the service object. When the service object belongs to the active object in the video application, by predicting the interaction label information corresponding to the service object under each of the M video tasks respectively, the interaction label information can be used to reflect the active situation and retention situation of the service object after pushing the corresponding video task. Therefore, based on the interaction label information corresponding to the service object under each of the M video tasks respectively, it is beneficial to improve the accuracy of video task push and improve the activity and retention time of the service object in the video application.

[0140] As Figure 11 shown, Figure 11 is a schematic diagram of a video task push provided by an embodiment of the present application. As Figure 11As shown, the terminal device 110a may be installed with a video application. The service object 110b may be a user in the video application installed on the terminal device 110a, and the server 110c may be the background server of the video application installed on the terminal device 110a. When it is necessary to push a video task to the service object 110b, the terminal device 110a may send the historical video interaction features of the service object 110b in the video application to the server 110c. The server 110c may input the historical video interaction features into an object hierarchical model, and through the object hierarchical model, layer the service object 110b to obtain the object hierarchical result of the service object 110b. If the object hierarchical result of the service object 110b indicates that the service object 110b belongs to the pre-churn object layer, then through an interpretable model, predict the influence degree of the historical video interaction features on the service object 110b belonging to the pre-churn object layer. Further, according to the influence degree of the historical video interaction features on the service object 110b belonging to the pre-churn object layer, determine the task push scores corresponding to the M video tasks in the video application respectively.

[0141] If the object hierarchical result of the service object 110b indicates that the service object 110b belongs to the active object layer, then through a multi-task recognition model, predict the interaction label information of the service object under the M video tasks. Further, according to the interaction label information of the service object under the M video tasks, determine the task push scores corresponding to the M video tasks in the video application respectively. The server 110c may screen out the video tasks with task push scores greater than or equal to the push score threshold from the M video tasks as the video tasks to be pushed to the service object, and return the video tasks to be pushed to the service object to the terminal device 110a.

[0142] The terminal device 110a may display the video tasks pushed to the service object and the executable times of the service object for the video tasks in the video task interface 110d. For example, the video task pushed to the service object is "Watch a video to receive a membership reward" and the corresponding executable times are 8 times. At the same time, the terminal device 110a may also display in the video task interface 110d the reward resources (i.e., a one-week video membership) that can be obtained after the video tasks with the executable times are completed. The service object 110b may complete the video tasks pushed to the service object by triggering the "Go to complete" control in the video task interface 110d. As Figure 11 shown, after the service object 110b has executed the video tasks 8 times, the terminal device 110a may display a game task interface 110e. The game task interface 110e displays a control for receiving the task reward resources (i.e., a one-week video membership). The service object 110b may receive the task reward resources (i.e., a one-week video membership) by triggering the control.

[0143] In the embodiments of the present application, based on the object hierarchy to which a service object belongs in a multimedia application, a strategy for dynamically selecting multimedia tasks to be pushed to the service object is adopted to achieve personalized pushing of multimedia tasks and improve the accuracy of multimedia task pushing. Specifically, when the service object belongs to the pre-churn object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively low, that is, the service object belongs to the pre-churn object of the media application. Therefore, personalized attribution can be performed on the service object belonging to the pre-churn object layer to obtain the influence degree of historical media interaction features on the service object belonging to the pre-churn object layer, that is, the influence degree reflects which historical media interaction features are the main reasons for the service object to belong to the pre-churn object layer. Pushing multimedia tasks to the service object based on the influence degree is beneficial to improving the accuracy of multimedia task pushing, reducing the possibility of service object churn, and further improving the interest and experience of the service object. When the service object belongs to the active object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively high, that is, the service object belongs to the active object of the media application. Therefore, the interaction label information corresponding to the service object under M multimedia tasks can be predicted, and the interaction label information can be used to reflect the active situation and retention situation of the service object after pushing the corresponding multimedia tasks. Based on the interaction label information corresponding to the service object under M multimedia tasks respectively, it is beneficial to improve the accuracy of multimedia task pushing, and improve the activity and retention time of the service object in the multimedia application.

[0144] Further, please refer to Figure 12 , Figure 12 which is a schematic structural diagram of a data processing device provided by an embodiment of the present application. The data processing device may be a computer program (including program code) running in a computer device. For example, the data processing device is an application software; the data processing device may be used to execute the corresponding steps in the method provided by the embodiments of the present application. As Figure 12 shown, the data processing device may be any blockchain node in the blockchain network. The data processing device may include: a layering module 11, a first determination module 12, a second determination module 13, and a pushing module 14.

[0145] The layering module 11 is configured to obtain the historical media interaction features of the service object in the multimedia application, and layer the service object according to the historical media interaction features to obtain an object layering result;

[0146] The first determination module 12 is configured to, if the object layering result indicates that the service object belongs to the pre-churn object layer, determine the task push scores corresponding to M multimedia tasks in the multimedia application respectively according to the influence degree of the historical media interaction features on the service object belonging to the pre-churn object layer; M is a positive integer;

[0147] The second determination module 13 is configured to, if the object stratification result indicates that the service object belongs to the active object layer, predict the interaction label information corresponding to the service object under each of the M multimedia tasks according to the historical media interaction characteristics, and determine the task push scores corresponding to the M multimedia tasks respectively according to the interaction label information corresponding to the M multimedia tasks respectively;

[0148] The push module 14 is configured to push to the service object the multimedia tasks among the M multimedia tasks for which the task push score is greater than or equal to the push score threshold.

[0149] Among them, the first determination module 12 is specifically configured to:

[0150] Call an interpretable model to predict the influence degree of the historical media interaction characteristics on the service object belonging to the pre-churn object layer;

[0151] Determine the feature importance of the historical media interaction characteristics according to the influence degree;

[0152] Determine the task push scores corresponding to the M multimedia tasks respectively according to the historical media interaction characteristics and the feature importance of the historical media interaction characteristics.

[0153] Among them, the first determination module 12 is also specifically configured to include:

[0154] Perform feature splitting on the historical media interaction characteristics to obtain Q atomic features corresponding to the historical media interaction characteristics; Q is an integer greater than 1;

[0155] Determine the feature importance corresponding to each of the Q atomic features according to the feature importance of the historical media interaction characteristics;

[0156] For the multimedia task Mi among the M multimedia tasks i Perform task splitting to obtain Pi atomic tasks corresponding to the multimedia task Mi i ; Pi is an integer greater than 1, and i is a positive integer less than or equal to M;

[0157] Determine the task push score corresponding to the multimedia task Mi according to the Pi atomic tasks and the feature importance corresponding to each of the Q atomic features i respectively.

[0158] Among them, the first determination module 12 is also specifically configured to include:

[0159] Screen out the sub-features that match the task features corresponding to the Pi atomic tasks respectively from the Q atomic features to obtain associated sub-features;

[0160] Determine the task scores corresponding to the Pi atomic tasks respectively according to the conversion ratio between the feature importance and the task score and the feature importance corresponding to the associated sub-features;

[0161] Sum the task scores corresponding to P atomic tasks to obtain the multimedia task M i The corresponding task push score.

[0162] Among them, the second determination module 13 is specifically used for:

[0163] Call the multi-task recognition model to extract, from the historical media interaction features, the associated media features associated with the multimedia task M among the M multimedia tasks; i is a positive integer less than or equal to M; i The associated media features; i is a positive integer less than or equal to M;

[0164] According to the associated media features, perform label prediction on the multimedia task M i to obtain the interaction label information of the service object in the multimedia task M i under.

[0165] Among them, the interaction label information corresponding to the multimedia task M among the M multimedia tasks i includes the label values corresponding to multiple task interaction labels respectively; i is a positive integer less than or equal to M; the second determination module 13 is specifically used for:

[0166] Sum the label values corresponding to the multiple task interaction labels corresponding to the multimedia task M i to obtain the total label value;

[0167] Determine the total label value as the task push score corresponding to the multimedia task M i under.

[0168] Among them, the layering module 11 is specifically used for:

[0169] Call the object layering model to predict the predicted activity of the service object in the future time period according to the historical media interaction features;

[0170] If the predicted activity is less than or equal to the activity threshold, generate an object layering result indicating that the service object belongs to the pre-churn object layer;

[0171] If the predicted activity is greater than the activity threshold, generate an object layering result indicating that the service object belongs to the active object layer.

[0172] Among them, the historical media interaction features include the media interaction features of the service object in the first historical time period and the media interaction features of the service object in the second historical time period, and the first historical time period is before the second historical time period; the layering module 11 is specifically used for:

[0173] Identify the first activity of the service object in the first historical time period according to the media interaction features of the service object in the first historical time period;

[0174] Identify the second activity level of the business object within the second historical time period according to the media interaction characteristics of the business object in the second historical time period;

[0175] Determine the activity change trend of the business object according to the first activity level and the second activity level, and predict the predicted activity level of the business object within the future time period according to the activity change trend.

[0176] Among them, the first determination module 12 is specifically further used for:

[0177] Obtain the initial interpretable model, the first sample media interaction characteristics of the sample object in the multimedia application, and obtain the labeled stratification result of the sample object;

[0178] Call the initial interpretable model, and stratify the sample object according to the first sample media interaction characteristics to obtain the predicted stratification result of the sample object;

[0179] Determine the model loss of the initial interpretable model according to the labeled stratification result and the predicted stratification result;

[0180] Train the initial interpretable model according to the model loss of the initial interpretable model until the initial interpretable model meets the convergence condition to obtain the interpretable model.

[0181] Among them, the first determination module 12 is specifically further used for:

[0182] Obtain the object stratification model for stratifying the object; the object stratification model is a model with a model complexity higher than that of the initial interpretable model and in a converged state;

[0183] Call the object stratification model, and stratify the sample object according to the first sample media interaction characteristics to obtain the stratification result output by the object stratification model;

[0184] Determine the stratification result output by the object stratification model as the labeled stratification result of the sample object.

[0185] Among them, the second determination module 13 is specifically further used for:

[0186] Obtain the second sample media interaction characteristics of the sample object in the multimedia application, and the labeled interaction label information corresponding to the sample object under M multimedia tasks respectively;

[0187] Call the initial multi-task recognition model, and predict the predicted interaction label information corresponding to the business object under M multimedia tasks respectively according to the second sample media interaction characteristics;

[0188] Determine the model loss of the initial multi-task recognition model according to the labeled interaction label information and the predicted interaction label information;

[0189] Train the initial multi-task recognition model according to the model loss of the initial multi-task recognition model until the initial multi-task recognition model meets the convergence condition, and obtain the multi-task recognition model.

[0190] Wherein, the multimedia application is a game application, and the M multimedia tasks are M game tasks in the game application; the push module 14 is specifically further configured to:

[0191] In response to a trigger operation of the service object for the game task, display the game task interface;

[0192] In the game task interface, display the pushed game task and the executable times of the service object for the pushed game task; the pushed game task is a game task in the M game tasks, and the task push score of the game task is greater than or equal to the push score threshold.

[0193] The push module 14 is specifically further configured to:

[0194] Obtain the task push score of the pushed game task, the maximum executable times associated with the pushed game task, and the minimum executable times associated with the pushed game task;

[0195] Determine the difference in times between the maximum executable times and the minimum executable times;

[0196] Obtain the product of the difference in times and the task push score, and sum the product and the minimum executable times to obtain the executable times of the service object for the pushed game task.

[0197] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of the overall module or unit that includes the function of the module or unit. According to an embodiment of the present application, Figure 12Each module in the data processing device shown can be separately or wholly combined into one or several units to form, or a certain one (or some) of the units can be further split into at least two smaller sub-units in terms of function, and the same operations can be achieved without affecting the realization of the technical effects of the embodiments of this application. The above modules are divided based on logical functions. In practical applications, the function of one module can also be realized by at least two units, or the functions of at least two modules are realized by one unit. In other embodiments of this application, the data processing device can also include other units. In practical applications, these functions can also be assisted by other units and can be realized by the cooperation of at least two units.

[0198] According to an embodiment of this application, it can be achieved by running a computer program (including program code) that can execute the respective steps involved in the corresponding method shown in Figure 3 on a general computer device such as a computer including processing elements and storage elements such as a central processing unit (CPU), a random access storage medium (RAM), and a read-only storage medium (ROM), to construct a data processing device as shown in Figure 12 and to implement a data processing method of the embodiments of this application. The above computer program can be recorded on, for example, a computer-readable recording medium, loaded into the above computer device through the computer-readable recording medium, and run therein.

[0199] In the embodiments of the present application, based on the object layer to which the service object belongs in the multimedia application, a strategy for dynamically selecting and pushing multimedia tasks to the service object is implemented to achieve personalized pushing of multimedia tasks and improve the accuracy of multimedia task pushing. Specifically, when the service object belongs to the pre-churn object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively low, that is, the service object belongs to the pre-churn object of the media application. Therefore, personalized attribution can be performed on the service object belonging to the pre-churn object layer to obtain the influence degree of historical media interaction characteristics on the service object belonging to the pre-churn object layer, that is, the influence degree reflects which historical media interaction characteristics are the main reasons for the service object to belong to the pre-churn object layer. Pushing multimedia tasks to the service object based on the influence degree is beneficial to improving the accuracy of multimedia task pushing, reducing the possibility of service object churn, and further improving the interest and experience of the service object. When the service object belongs to the active object layer, it indicates that the probability of the service object using the multimedia application in the future time period is relatively high, that is, the service object belongs to the active object of the media application. Therefore, the interaction label information corresponding to the service object under M multimedia tasks can be predicted, and the interaction label information can be used to reflect the active situation and retention situation of the service object after pushing the corresponding multimedia tasks. Based on the interaction label information corresponding to the service object under M multimedia tasks, it is beneficial to improve the accuracy of multimedia task pushing, and improve the activity and retention time of the service object in the multimedia application.

[0200] Further, please refer to Figure 13 , Figure 13 which is a schematic diagram of a computer device provided by an embodiment of the present application. As Figure 13 shown, the computer device 3000 may be the terminal device or server corresponding to the above Figure 2 embodiment. The computer device 3000 may include: at least one processor 3001, such as a CPU, at least one network interface 3004, a user interface 3003, a memory 3005, and at least one communication bus 3002. Among them, the communication bus 3002 is used to realize the connection and communication between these components. Among them, the user interface 3003 may include a display screen (Display) and a keyboard (Keyboard). The network interface 3004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 3005 may be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The storage 3005 may optionally also be at least one storage device located far from the aforementioned processor 3001. As Figure 13 shown, the memory 3005, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a computer program control application.

[0201] exist Figure 13 In the computer device 3000 shown, the network interface 3004 is mainly used for the second node device to communicate with the target relay server and the target oracle server; the user interface 3003 is mainly used to provide an input interface for the user; and the processor 3001 can be used to call the computer program control application stored in the memory 3005 to achieve:

[0202] Acquire historical media interaction features of business objects in multimedia applications, stratify business objects according to the historical media interaction features, and obtain object stratification results;

[0203] If the object stratification result indicates that the business object belongs to the pre-churn object layer, then according to the historical media interaction characteristics and the influence degree of the business object belonging to the pre-churn object layer, the task push scores corresponding to the M multimedia tasks in the multimedia application are determined; M is a positive integer;

[0204] If the object stratification result indicates that the business object belongs to the active object layer, then the interaction label information corresponding to the business object under M multimedia tasks is predicted according to the historical media interaction characteristics, and the task push scores corresponding to the M multimedia tasks are determined according to the interaction label information corresponding to the M multimedia tasks;

[0205] Push the multimedia tasks whose task push score is greater than or equal to the push score threshold among the M multimedia tasks pushed to the business object.

[0206] It should be understood that the computer device 3000 described in the embodiment of the present application can also execute the above Figure 8 In the description of a data processing method in the corresponding embodiment, the computer device 3000 described in the embodiment of the present application can also execute the above Figure 12 The description of the data processing device in the corresponding embodiments will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated here.

[0207] In addition, it should be pointed out here that: the embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the data processing device mentioned above, and the computer program includes program instructions. When the processor executes the program instructions, it can execute the above-mentioned Figure 3 or Figure 8For the description of the data processing method in the corresponding embodiment, therefore, it will not be elaborated here. In addition, the beneficial effects of adopting the same method will not be elaborated either. For the technical details not disclosed in the embodiment of the computer-readable storage medium involved in this application, please refer to the description of the method embodiment of this application. As an example, the program instructions can be deployed to be executed on one computing device, or on multiple computing devices located at one location, or on multiple computing devices distributed at multiple locations and interconnected through a communication network. The multiple computing devices distributed at multiple locations and interconnected through a communication network can form a blockchain system.

[0208] On the one hand, this application provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device can execute the description of a data processing method in the foregoing Figure 3 or Figure 8 corresponding embodiment, which will not be elaborated here. In addition, the beneficial effects of adopting the same method will not be elaborated either.

[0209] It should be noted that the relevant data collection and processing in this application book should strictly comply with the requirements of relevant national laws and regulations when applied in practice, obtain the informed consent or separate consent (or have a legal basis) of the personal information subject, and carry out subsequent data use and processing behaviors within the scope authorized by laws and regulations and the personal information subject. For example, when this application obtains the historical media interaction characteristics (such as multimedia execution characteristics, multimedia preference characteristics, multimedia application login characteristics, etc.) of business objects and sample objects for multimedia applications respectively, it is necessary to obtain the informed consent or separate consent of the corresponding business objects or sample objects.

[0210] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0211] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A data processing method, characterized in that, Including: Obtain the historical media interaction characteristics of a business object in a multimedia application, and based on the historical media interaction characteristics, layer the business object to obtain an object layering result; If the object layering result indicates that the business object belongs to the pre-churn object layer, then according to the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer, determine the task push scores respectively corresponding to M multimedia tasks in the multimedia application; M is a positive integer; If the object layering result indicates that the business object belongs to the active object layer, then according to the historical media interaction characteristics, predict the interaction label information respectively corresponding to the business object under the M multimedia tasks, and based on the interaction label information respectively corresponding to the M multimedia tasks, determine the task push scores respectively corresponding to the M multimedia tasks; Push the multimedia tasks in the M multimedia tasks whose task push scores are greater than or equal to the push score threshold to the business object.

2. The method according to claim 1, wherein The determining the task push scores respectively corresponding to the M multimedia tasks in the multimedia application according to the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer includes: Call an interpretable model to predict the influence degree of the historical media interaction characteristics on the business object belonging to the pre-churn object layer; Determine the feature importance of the historical media interaction characteristics according to the influence degree; Based on the historical media interaction characteristics and the feature importance of the historical media interaction characteristics, determine the task push scores respectively corresponding to the M multimedia tasks.

3. The method according to claim 2, wherein The determining the task push scores respectively corresponding to the M multimedia tasks based on the historical media interaction characteristics and the feature importance of the historical media interaction characteristics includes: Perform feature splitting on the historical media interaction characteristics to obtain Q atomic features corresponding to the historical media interaction characteristics; Q is an integer greater than 1; Determine the feature importance respectively corresponding to the Q atomic features according to the feature importance of the historical media interaction characteristics; For the multimedia task M among the M multimedia tasks i perform task splitting to obtain P atomic tasks corresponding to the multimedia task M i where P is an integer greater than 1 and i is a positive integer less than or equal to M Determine the multimedia task M according to the P atomic tasks and the feature importance degrees respectively corresponding to the Q atomic features i The corresponding task push score value.

4. The method according to claim 3, characterized in that Determining the multimedia task M according to the feature importance degrees respectively corresponding to the P atomic tasks and the Q atomic features i The corresponding task push score includes: Select sub-features that match the task features respectively corresponding to P atomic tasks from the Q atomic features to obtain associated sub-features; Determine the task scores respectively corresponding to the P atomic tasks according to the conversion ratio between the feature importance and the task score and the feature importance corresponding to the associated sub-features; Sum the task scores corresponding to the P atomic tasks respectively to obtain the multimedia task M i The corresponding task push score.

5. The method according to claim 1, characterized in that The predicting the interaction label information respectively corresponding to the business object under the M multimedia tasks according to the historical media interaction characteristics includes: Invoke the multi-task recognition model to extract, from the historical media interaction features, the associated media features associated with the multimedia task M among the M multimedia tasks; i is a positive integer less than or equal to M; i associated associated media features; i is a positive integer less than or equal to M; Based on the associated media features, for the multimedia task M i perform label prediction to obtain the interaction label information of the service object in the multimedia task M i below.

6. The method according to claim 1, wherein, The multimedia task M among the M multimedia tasks i The corresponding interactive label information includes label values respectively corresponding to multiple task interactive labels; i is a positive integer less than or equal to M; The determining the task push scores respectively corresponding to the M multimedia tasks according to the interaction label information respectively corresponding to the M multimedia tasks includes: For the multimedia task M i Sum the tag values corresponding to the respective multiple task interaction tags, and obtain the total tag value; Determine the total value of the label as the multimedia task M i The corresponding task push score value.

7. The method according to claim 1, wherein The layering the business object based on the historical media interaction characteristics to obtain an object layering result includes: Call an object layering model and predict the predicted activity of the business object in a future time period according to the historical media interaction characteristics; If the predicted activity level is less than or equal to the activity threshold, an object stratification result is generated to indicate that the business object belongs to the pre-churn object layer; If the predicted activity level is greater than the activity threshold, an object stratification result is generated to indicate that the business object belongs to the active object layer.

8. The method according to claim 7, wherein The historical media interaction features include the media interaction features of the business object in the first historical time period and the media interaction features of the business object in the second historical time period, where the first historical time period is before the second historical time period; Invoking the object stratification model to predict the predicted activity level of the business object in the future time period according to the historical media interaction features includes: Identifying the first activity level of the business object in the first historical time period according to the media interaction features of the business object in the first historical time period; Identifying the second activity level of the business object in the second historical time period according to the media interaction features of the business object in the second historical time period; Determining the activity change trend of the business object according to the first activity level and the second activity level, and predicting the predicted activity level of the business object in the future time period according to the activity change trend.

9. The method according to claim 2, characterized in that, The method further includes: Obtaining an initial interpretable model, the first sample media interaction features of the sample object in the multimedia application, and obtaining the labeled stratification result of the sample object; Invoking the initial interpretable model to stratify the sample object according to the first sample media interaction features to obtain the predicted stratification result of the sample object; Determining the model loss of the initial interpretable model according to the labeled stratification result and the predicted stratification result; Training the initial interpretable model according to the model loss of the initial interpretable model until the initial interpretable model meets the convergence condition to obtain the interpretable model.

10. The method according to claim 9, wherein The obtaining the labeled stratification result of the sample object includes: Obtaining an object stratification model for stratifying the object; the object stratification model is a model with a model complexity higher than that of the initial interpretable model and in a converged state; Invoking the object stratification model to stratify the sample object according to the first sample media interaction features to obtain the stratification result output by the object stratification model; Determining the stratification result output by the object stratification model as the labeled stratification result of the sample object.

11. The method according to claim 5, characterized in that, The method further includes: Obtaining the second sample media interaction features of the sample object in the multimedia application and the labeled interaction label information corresponding to the sample object under the M multimedia tasks respectively; Invoking the initial multi-task recognition model to predict the predicted interaction label information corresponding to the business object under the M multimedia tasks respectively according to the second sample media interaction features; Determining the model loss of the initial multi-task recognition model according to the labeled interaction label information and the predicted interaction label information; Train the initial multi-task recognition model according to the model loss of the initial multi-task recognition model until the initial multi-task recognition model meets the convergence condition, and obtain the multi-task recognition model.

12. The method according to claim 1, characterized in that The multimedia application is a game application, and the M multimedia tasks are M game tasks in the game application; Among the steps of pushing the M multimedia tasks to the service object, pushing the multimedia tasks with a push score greater than or equal to the push score threshold includes: In response to a trigger operation of the service object for a game task, display a game task interface; In the game task interface, display the pushed game tasks and the number of executable times of the service object for the pushed game tasks; the pushed game tasks are the game tasks in the M game tasks with a task push score greater than or equal to the push score threshold.

13. The method according to claim 12, wherein, The method further includes: Obtain the task push score of the pushed game task, the maximum number of executable times associated with the pushed game task, and the minimum number of executable times associated with the pushed game task; Determine the difference in the number of times between the maximum number of executable times and the minimum number of executable times; Obtain the product of the difference in the number of times and the task push score, and sum the product and the minimum number of executable times to obtain the number of executable times of the service object for the pushed game task.

14. A data processing device, characterized in that, It includes: A layering module for obtaining the historical media interaction characteristics of a service object in a multimedia application, and layering the service object according to the historical media interaction characteristics to obtain an object layering result; A first determination module for, if the object layering result indicates that the service object belongs to the pre-churn object layer, determining the task push scores corresponding to the M multimedia tasks in the multimedia application respectively according to the influence degree of the historical media interaction characteristics on the service object belonging to the pre-churn object layer; M is a positive integer; A second determination module for, if the object layering result indicates that the service object belongs to the active object layer, predicting the interaction label information corresponding to the service object under the M multimedia tasks respectively according to the historical media interaction characteristics, and determining the task push scores corresponding to the M multimedia tasks respectively according to the interaction label information corresponding to the M multimedia tasks; A push module for pushing the multimedia tasks with a task push score greater than or equal to the push score threshold among the M multimedia tasks to the service object.

15. A computer device, characterized in that, It includes: A processor and a memory; The processor is connected to the memory. Among them, the memory is used to store a computer program, and the processor is used to call the computer program so that the computer device executes the method according to any one of claims 1-13.

16. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and the computer program is suitable for being loaded and executed by a processor so that a computer device with the processor executes the method according to any one of claims 1-13.

17. A computer program product or a computer program, characterized in that, The computer program product or computer program includes computer instructions stored in a computer-readable storage medium, and the computer instructions are adapted to be read and executed by a processor so that a computer device having the processor executes the method according to any one of claims 1-13.

Images