A display method and device of a target object, an electronic device, and a storage medium

Abstract

The application provides a target object display method and device, electronic equipment and storage medium. The method collects input features, which include user historical behavior data, user attribute data, item attribute data and context data. The input features are processed by a pre-trained multi-service multi-task learning model to predict evaluation scores of each target object in an application software, the target object including a service category and a sub-service category under the service category. Each service category is sorted according to the evaluation scores, and each sub-service category under the service category is sorted. Each service category and each sub-service category are displayed on a corresponding display interface of the application software according to the sorting results. The input features are predicted by the pre-trained multi-service multi-task learning model, which can ensure that the target objects with high click rates are displayed on the corresponding interface of the application software, and the recommendation accuracy of the corresponding target objects is enhanced.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to a method, apparatus, electronic device, and storage medium for displaying a target object. Background Technology

[0002] In mobile apps, recommendation features are typically located on the homepage, directly impacting user engagement and retention. Recommendations are also often found on semi-display pages.

[0003] In the recommendation field, multi-objective systems have become the industry mainstream, and various business scenarios of major companies are basically built on the framework of multi-objective systems. However, in existing multi-objective models, it is difficult to optimize multiple objectives in multiple business scenarios at the same time, and negative transfer and seesaw phenomena often occur. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, electronic device, and storage medium for displaying target objects. The aim is to process input features using a multi-service, multi-task learning model to predict evaluation scores for each target object in the application software. Then, based on the evaluation scores, the target objects are sorted, and based on the sorting results, each target object is displayed on the corresponding display interface of the application software. This ensures that the target objects displayed in the application software are business categories or sub-business categories with high user click-through rates, facilitating user maintenance of the application software.

[0005] To achieve the above objectives, a first aspect of this application provides a method for displaying a target object, the method comprising:

[0006] The input features are collected, including user historical behavior data, user attribute data, item attribute data, and contextual data;

[0007] The input features are processed by a pre-trained multi-service multi-task learning model to predict the evaluation scores of each target object in the application software. The target objects include business categories and sub-business categories under the business categories in the application software.

[0008] The business categories are sorted according to the evaluation scores, and the sub-business categories under each business category are sorted.

[0009] Based on the sorting results, each of the business categories and each of the sub-business categories are displayed on the corresponding display interface of the application software.

[0010] In some embodiments, after training the multi-service, multi-task learning model, the method further includes:

[0011] Retrieve the first ranking results for each target object that matches the search keywords;

[0012] The target objects are predicted by a pre-trained multi-service multi-task learning model to obtain the scoring results of each target object;

[0013] Based on the scoring results, the first ranking result is corrected to obtain the second ranking result;

[0014] Based on the second sorting result, each of the target objects is displayed on the corresponding display interface of the application software.

[0015] In some embodiments, ranking each of the business categories and ranking the sub-business categories under each of the business categories according to the evaluation scores includes:

[0016] The evaluation scores corresponding to each of the aforementioned business categories are sorted in descending order of numerical value to obtain a first sorted list;

[0017] Alternatively, the evaluation scores corresponding to each of the aforementioned business categories can be sorted in ascending order of numerical value to obtain a second sorting list;

[0018] The evaluation scores corresponding to each of the aforementioned sub-business categories are sorted in descending order of numerical value to obtain a third sorting list;

[0019] Alternatively, the evaluation scores corresponding to each of the sub-business categories can be sorted in ascending order of numerical value to obtain a fourth sorting list.

[0020] In some embodiments, displaying each of the business categories and each of the sub-business categories on the corresponding display interface of the application software according to the sorting result includes:

[0021] Select the top N business categories from the first sorting list and display them on the corresponding display interface;

[0022] Alternatively, select the last N business categories from the second sorting list and display them on the corresponding display interface, where N is a positive integer;

[0023] Select the top M sub-business categories from the third sorting list and display them on the corresponding display interface;

[0024] Alternatively, the last M sub-business categories in the fourth sorting list can be selected and displayed on the corresponding display interface, where M is a positive integer.

[0025] In some embodiments, pre-training the multi-service, multi-task learning model includes:

[0026] A training sample set is constructed based on users' historical behavior data;

[0027] The training sample set is used to train the multi-service multi-task learning model.

[0028] Determine the loss function;

[0029] Based on the loss function, the parameters of the multi-service multi-task learning model are updated through backpropagation;

[0030] Determine whether the training of the multi-service multi-task learning model meets the termination condition. If the termination condition is met, output the multi-service multi-task learning model. If the termination condition is not met, continue training.

[0031] In some embodiments, determining the loss function includes:

[0032] Determine the weights for each business function and each task.

[0033] Obtain the sample space for each task under each business;

[0034] Calculate the loss function for each task under each business;

[0035] The loss function is calculated based on the weights corresponding to each business, the weights corresponding to each task, the sample space of each task under each business, and the loss function of each task under each business.

[0036] In some embodiments, the loss function is calculated based on the weights corresponding to each business, the weights corresponding to each task, the sample space of each task under each business, and the loss function of each task under each business, using the following formula:

[0037]

[0038] in,

[0039] In the formula, L N,K (θ N,K ,θ s ) represents the loss function of the multi-service, multi-task learning model, θ N,K Let θ represent the characteristics of N business processes and K tasks, where N represents the total number of business processes and K represents the total number of tasks. n,k Let θ represent the characteristics of the k-th task under the n-th business. s Represents shared features, ω n ω represents the weight of business n. kL represents the weight of task k. n,k (θ n,k ,θ s Let ) represent the loss function for the k-th task under the n-th business. Let be an indicator function, indicating whether sample i is a sample of the k-th task under the n-th business. If so, it is included in the loss of the k-th task under the n-th business. This represents the prediction result of the multi-service, multi-task learning model for sample i. The label represents sample i.

[0040] To achieve the above objectives, a second aspect of this application provides a display device for a target object, the device comprising:

[0041] The collection module is used to collect input features, which include user historical behavior data, user attribute data, item attribute data, and context data.

[0042] The processing module is used to process the input features through a pre-trained multi-service multi-task learning model to predict the evaluation scores of each target object in the application software. The target objects include business categories and sub-business categories under the business categories in the application software.

[0043] The sorting module is used to sort each of the business categories and the sub-business categories under each of the business categories according to the evaluation scores.

[0044] The display module is used to display each of the business categories and each of the sub-business categories on the corresponding display interface of the application software according to the sorting results.

[0045] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0046] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the method described in the first aspect.

[0047] This application proposes a method, apparatus, electronic device, and storage medium for displaying target objects. The method involves collecting input features, including user historical behavior data, user attribute data, item attribute data, and contextual data. A pre-trained multi-service, multi-task learning model processes the input features to predict the evaluation scores of various target objects within an application software. Target objects include business categories and sub-business categories within the application software. The business categories and sub-business categories are then ranked according to their evaluation scores. Based on the ranking results, each business category and sub-business category is displayed on the corresponding interface of the application software. By using a pre-trained multi-service, multi-task learning model to predict input features, the accuracy of predicted behavior values ​​for multiple targets can be maintained even when there is a certain degree of conflict. Simultaneously, it ensures that target objects with higher click-through rates are displayed on the application software's interface, thereby enhancing the accuracy of target object recommendations. Attached Figure Description

[0048] Figure 1 This is a flowchart of the method for displaying the target object provided in the embodiments of this application;

[0049] Figure 2 This is a flowchart illustrating the modeling process of the recommendation algorithm provided in this application embodiment;

[0050] Figure 3 This is a schematic diagram of the structure of the multi-service, multi-task learning model provided in the embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the CGC network structure provided in an embodiment of this application;

[0052] Figure 5 This is a schematic diagram of the structure of the PLE network provided in the embodiments of this application;

[0053] Figure 6 This is a flowchart illustrating the steps of pre-training a multi-service, multi-task learning model provided in the embodiments of this application;

[0054] Figure 7 This is a flowchart illustrating the steps of sorting each business category and the sub-business categories under each business category based on the evaluation score, as provided in the embodiments of this application.

[0055] Figure 8 This application provides a flowchart of steps for displaying each business category and each sub-business category on the corresponding display interface of the application software based on the sorting results.

[0056] Figure 9 This application provides a flowchart of steps that can be executed after training a multi-service, multi-task learning model;

[0057] Figure 10 This is an example diagram of the first sorting result provided in the embodiments of this application;

[0058] Figure 11 This is an example diagram of the second sorting result provided in the embodiments of this application;

[0059] Figure 12 This is a schematic diagram of the structure of the display device for the target object provided in the embodiments of this application;

[0060] Figure 13 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0064] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0065] Recommender systems aim to help users quickly discover the information they are most interested in from a large amount of data, such as music and news. Currently, most recommendation technologies only perform single-objective training, such as click-through rate (CTR) prediction techniques using algorithms like Logistic Regression (LR), Gradient Boosting Decision Tree (GBDT), and deep learning. However, modeling multiple objectives is a common requirement for recommendation systems. For example, because user feedback is of various types, including clicks and full playback, recommendation algorithms that only optimize a single feedback objective may recommend biased information. Models that only optimize CTR are prone to recommending clickbait titles, while models that only optimize full playback rate may easily recommend short music videos. Therefore, the need for multi-objective recommendation has emerged, making it necessary to simultaneously optimize CTR and full playback rate.

[0066] Multi-objective recommendation problems differ from single-objective recommendation problems. Single-objective recommendation problems optimize only a single objective, such as increasing the platform's click-through rate (CTR), but other metrics like the platform's complete playback rate may deteriorate. Multi-objective recommendation problems have multiple objective functions, aiming to find a recommendation method that achieves overall optimization across multiple objectives, such as simultaneously improving the platform's CTR and complete playback rate. Taking a music recommendation system as an example, to analyze user satisfaction with the recommendation results, the system's business metrics include not only the user playback rate but also sharing rate, download rate, and playback duration. A user playing a song does not guarantee a complete playback, sharing, or download. The relationship between playback rate and playback duration, sharing rate, and download rate is not synchronous; sometimes, they even show an inverse trend. Therefore, it is necessary to jointly train the multi-objective recommendation model to find the common optimal solution for multiple objectives, hoping to simultaneously improve the system's music playback rate, complete playback rate, sharing rate, and download rate, better meeting user needs and thus improving user retention and activity.

[0067] However, while multi-task learning (MTL) and multi-scenario learning (MSL) are widely researched and applied in academia and industry, and joint training across multiple tasks can improve application performance (e.g., ESMM, MMoE, PLE), they do not consider the scenario of multiple business scenarios and multiple tasks. This can lead to the rich contextual information of business scenarios being affected by seesaw effects and negative transfer. Developing a separate model for each task in each business scenario would significantly increase development and maintenance costs.

[0068] Based on this, this application proposes a method for displaying target objects. The aim is to predict evaluation scores from input features using a multi-service, multi-task learning model, then rank the target objects according to the evaluation scores, and display each target object on the corresponding display interface of the application software based on the ranking results. This ensures that even when there is a certain degree of conflict between multiple targets, the accuracy of the predicted behavior values ​​for multiple targets can still be guaranteed. Simultaneously, it ensures that target objects with higher click-through rates are displayed on the corresponding interface of the application software, thereby enhancing the accuracy of the recommendation of the corresponding target objects.

[0069] Reference Figure 1 , Figure 1 This is a flowchart of the method for displaying the target object provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0070] Step S101: Collect input features, which include user historical behavior data, user attribute features, item attribute features, and contextual features.

[0071] Recommender systems are a subfield of machine learning, therefore, their problem-solving approach follows the general principles of machine learning. The machine learning process can be viewed as building a machine to produce a certain product. Based on past production experience, a machine is built to produce that product; this past experience is the training set, and the completed machine is the machine learning model. The raw materials are processed and fed into the machine in a certain way; the machine's final output is the desired prediction result. The accuracy of the model is determined by whether the machine is well-constructed and can produce goods with acceptable errors. Based on this simple analogy, the general process of a recommendation algorithm providing personalized recommendations to users is as follows: Figure 2 As shown. Figure 2 This is a flowchart illustrating the modeling process of the recommendation algorithm provided in this application embodiment. Data from different sources is collected and aggregated to form the raw data required by the recommendation algorithm. Feature engineering is used to process the raw data to generate final features. Then, a suitable recommendation algorithm is selected to train the features to obtain the final recommendation model. In the prediction / inference stage, based on a user's characteristics, the features are fed into the model to obtain the recommendation result for that user. Therefore, data and feature engineering are crucial at the starting point of the entire recommendation system's workflow. Data is the raw material; the completeness and quality of the raw materials (recommendation data sources) directly determine whether good products can be produced. The processing of these raw materials (feature engineering) determines whether good products can be produced efficiently, quickly, and with high quality.

[0072] Recommendation systems infer users' interests and preferences based on their actions on products (apps, websites, etc.), ultimately providing personalized recommendations. The recommendation process involves four main data sources: the user themselves, the target item, the user's actions, and the user's context.

[0073] User behavior data encompasses all user actions on a product, such as browsing, clicking, playing, purchasing, searching, saving, liking, sharing, adding to cart, and even swiping, dwell time at a specific location, fast-forwarding, and so on. User actions reflect their true intentions and reveal their current interests. Analyzing user behavior provides profound insights into user preferences. Based on whether a user's behavior directly indicates their interest in a target item, user behavior is generally categorized into explicit and implicit behaviors. Explicit behaviors directly express user interest, such as liking and rating. Implicit behaviors, while not directly indicating user interest, indirectly reflect changes in that interest. Any action other than directly liking or rating is considered implicit feedback, including browsing, clicking, playing, saving, commenting, and sharing.

[0074] User attribute data, also known as user demographic data, refers to the attributes a user possesses, such as age, gender, region, education level, family composition, and occupation. This data is generally stable (e.g., gender) or changes slowly (e.g., age). Humans are social beings; different user attributes determine their social class or social circle. Different social classes or social circles have different behavioral characteristics, lifestyles, and preferences. Within the same social circle, there is a certain degree of similarity, which provides a certain guarantee for personalized recommendations.

[0075] The most important "participant" in a recommendation system is the target object (such as an item) to be recommended. Items themselves contain many features and attributes. For videos, metadata includes the production company, director, cast, main actors, country of origin, year, language, awards received, plot, and poster image. For e-commerce products, category, purpose, price, place of origin, and brand are also crucial attributes. By analyzing user interactions with items, the features of the items can be assigned to users with certain weights. These features construct the user's interests and preferences, essentially tagging the user (e.g., someone who likes "horror movies"). Based on these interests and preferences, personalized recommendations can then be made to the user.

[0076] Contextual data refers to the overall environmental characteristics and state of a user when interacting with items, such as the user's geographical location, the time, the weather, the user's mood, and the path the user is on. This contextual data is crucial, even decisive, for user decision-making. For example, location-based services like Meituan and Ele.me recommend restaurants that are near the user's current or specified location. Appropriately using contextual data and integrating it into recommendation algorithms can provide more accurate and context-specific personalized recommendations.

[0077] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0078] Step S102: The input features are processed by a pre-trained multi-business multi-task learning model to predict the evaluation scores of each target object in the application software. The target objects include business categories and sub-business categories under the business categories in the application software.

[0079] In this embodiment, multi-business scenario learning aims to solve business scenarios with multiple identical task labels and similar spatial structures. For example, it captures business scenario relevance through a star topology, including shared central parameters and business scenario-specific parameters. Multi-task learning focuses on modeling the correlations and differences between tasks. Examples include click-through rate (CTR) prediction tasks and conversion rate (CVR) prediction tasks, credit risk prediction tasks and credit limit prediction tasks. For videos, it also includes VTR (whether the entire video was watched) tasks and VCR (the percentage of viewers who watched the entire video). This embodiment is based on the multi-task learning model PLE, constructing a multi-business multi-task learning model to automatically learn the correlations between business scenario tasks and ensure positive benefits for each business scenario.

[0080] In this embodiment, taking click-through rate (CTR) prediction and conversion rate (CVR) prediction tasks as examples, the click-through rate is the probability that a user clicks on a target object, and the conversion information includes the conversion rate or conversion duration. The conversion rate is the probability that a user performs a conversion operation on the target object after clicking on it, and the conversion duration includes the duration the user stays on the target object after performing the conversion operation. Therefore, in this embodiment, multi-task learning can be performed simultaneously on multiple business objectives and multiple sub-objectives under multiple businesses, thereby predicting the evaluation scores of each business scenario and the evaluation scores of each sub-objective under each business scenario.

[0081] It should be noted that, in the embodiments of this application, the evaluation score is calculated by a scoring function, which represents a weighted sum of various indicators.

[0082] Reference Figure 3 , Figure 3 This is a schematic diagram of the structure of the multi-service, multi-task learning model provided in an embodiment of this application. For example... Figure 3 As shown, the collected input features are first fed into the Embedding layer, transforming them into vectors of equal dimension. Then, they are fed into the concatenation layer, where all vectors of equal dimension are concatenated using convolutional kernels. Next, they are fed into the fully connected (FC) layer for nonlinear transformation, and finally into the multi-service PLE network. In this embodiment, the number of services is first determined to be N. Then, a multi-task learning PLE network is constructed based on the CGC network for each service scenario. The multi-task learning PLE networks for each service scenario are then superimposed to construct the multi-service PLE network. Finally, the outputs of the multi-service PLE network layer and the FC fully connected layer are fed into the merge layer to fuse the output features of the multi-service PLE network layer and the FC fully connected layer.

[0083] Specifically, refer to Figure 4 , Figure 4 This is a schematic diagram of the CGC network structure provided in this application embodiment. The underlying network of CGC mainly includes a shared expert network and task-specific expert networks. For task A, m... A A dedicated expert network for each task; for task B, there is m B A dedicated network of experts for each task; and m SA shared expert network is shared by tasks A and B. The input to each subtask tower is a weighted sum of several expert vectors from both the task-specific and shared expert networks (the attention weights are also determined by the gate). This explicit partitioning of the expert networks ensures that the task-specific expert networks are only affected by the gradients of their own tasks and are not interfered with by other tasks; while only the shared expert networks are affected by the mixed gradients of multiple tasks. Here, the judging network uses the input as a selector, obtaining the attention weights assigned to different expert networks through a single fully connected layer and softmax.

[0084] PLE is essentially a multi-layered stack of the aforementioned CGC network, designed to achieve richer representation capabilities. The specific network structure is as follows: Figure 5 As shown, in the underlying Extraction network, in addition to the gating networks of each task-specific expert network, the shared expert network also has a gating network. Thus, the output of the task-specific expert network only incorporates the task's own expert network and the shared expert network; while the output of the shared expert network incorporates all expert networks. When the outputs of Task A, Task B, and the shared expert network are input to the next layer, the next layer's gating network uses these three outputs as the selector, rather than using the input itself. This allows for better utilization of more abstract, higher-level information.

[0085] In this embodiment of the application, a system is constructed based on each business scenario, such as... Figure 5 The multi-task learning PLE network shown is used as an example. Then, by superimposing the multi-task learning PLE networks of various business scenarios, a multi-business multi-task learning model can be constructed.

[0086] Reference Figure 6 , Figure 6 This is a flowchart illustrating the steps of pre-training a multi-service, multi-task learning model provided in this application embodiment. In this application embodiment, after constructing the multi-service, multi-task learning model, it is necessary to train the model, including but not limited to steps S601 to S605.

[0087] Step S601: Construct a training sample set based on user historical behavior data;

[0088] Step S602: Obtain the training sample set to train the multi-service multi-task learning model;

[0089] Step S603: Determine the loss function;

[0090] Step S604: Update the parameters of the multi-service, multi-task learning model through backpropagation according to the loss function;

[0091] Step S605: Determine whether the training of the multi-service multi-task learning model meets the termination condition. If the termination condition is met, output the multi-service multi-task learning model. If the termination condition is not met, continue training.

[0092] In this embodiment, several training sample sets can be constructed based on users' historical behavior data. These training sample sets are then input into a multi-service, multi-task learning model for training. The most crucial aspect of the training process is parameter updating. First, a loss function needs to be determined. Then, based on the loss function, the parameters of the multi-service, multi-task learning model are updated through backpropagation. Specifically, the weights corresponding to each service and each task are first determined. Then, the sample space for each task under each service is obtained. The loss function for each task under each service is calculated. Based on the weights corresponding to each service, the weights corresponding to each task, the sample space for each task under each service, and the loss function for each task under each service, the loss function is calculated. Finally, it is determined whether the training of the multi-service, multi-task learning model meets the termination condition. If the termination condition is met, the multi-service, multi-task learning model is output; otherwise, training continues.

[0093] In this embodiment, the weights between different tasks under different services adopt an adaptive scheme, continuously adjusting the weights between different tasks under different services during training. Wherein, the weight corresponding to service n is... In the formula, ω n,0 The initial weights of samples for business n are represented by γ, t is the number of training iterations, and γ is the initial weights of samples for business n. n ω represents the sample weight update factor for business n. n,0 and γ n These are all network hyperparameters. Similarly, the weights corresponding to task k are... In the formula, ω k,0 The initial weights of the samples for task k are represented by γ, t is the number of training iterations, and γ is the initial weights of the samples for task k. k ω represents the sample weight update factor for task k. k,0 and γ k All of these are network hyperparameters. Furthermore, considering the different sample spaces between different tasks under different services, this embodiment unifies all samples into the same sample space, namely, the sample space jointly constituted by all services and all tasks. The expression for the loss function is then determined as follows:

[0094]

[0095] in,

[0096] In the formula, L N,K (θ N,K ,θ s ) represents the loss function of the multi-service, multi-task learning model, θ N,K Let θ represent the characteristics of N business processes and K tasks, where N represents the total number of business processes and K represents the total number of tasks. n,k Let θ represent the characteristics of the k-th task under the n-th business. s Represents shared features, ω n ω represents the weight of business n. k L represents the weight of task k. n,k (θ n,k ,θ s Let ) represent the loss function for the k-th task under the n-th business. Let be an indicator function, indicating whether sample i is a sample of the k-th task under the n-th business. If so, it is included in the loss of the k-th task under the n-th business. This represents the prediction result of the multi-service, multi-task learning model for sample i. The label represents sample i.

[0097] In this embodiment, the trained multi-business, multi-task learning model still needs to be applied to specific downstream tasks for parameter tuning. For example, when applied to an e-commerce sub-network, multiple tasks are set for the e-commerce sub-network, including click-through rate (CTR) prediction and conversion rate (CVR) prediction. Shared features of the e-commerce sub-network need to be extracted through a shared expert network, and task-specific expert networks need to be used to extract the specific features of the e-commerce sub-network for learning. Then, the model parameters are tuned by continuously minimizing the loss function through backpropagation to obtain the final multi-business, multi-task learning model applicable to the e-commerce sub-network. Similarly, the multi-business, multi-task learning model can also be applied to an insurance sub-network. Multiple tasks can be set for the insurance sub-network, including credit risk prediction and credit limit prediction. Shared features of the insurance sub-network need to be extracted through a shared expert network, and task-specific expert networks need to be used to extract the specific features of the insurance sub-network for learning. Then, the model parameters are tuned by continuously minimizing the loss function through backpropagation to obtain the final multi-business, multi-task learning model applicable to the insurance sub-network.

[0098] Step S103: Sort each business category and its sub-business categories according to the evaluation scores.

[0099] In this embodiment, after processing the input features using a pre-trained multi-service, multi-task model, the evaluation scores of each target object in the application software can be predicted. The target objects include business categories and sub-business categories within the application software. Then, based on the evaluation scores, each business category and its sub-business categories are ranked.

[0100] Reference Figure 7 , Figure 7 This is a flowchart of the steps for sorting each business category and the sub-business categories under each business category based on the evaluation score, provided in the embodiments of this application, including but not limited to steps S701 to S704.

[0101] Step S701: Sort the evaluation scores corresponding to each business category in descending order of value to obtain the first sorted list;

[0102] Step S702, or, sort the evaluation scores corresponding to each business category in ascending order of numerical value to obtain a second sorted list;

[0103] Step S703: Sort the evaluation scores corresponding to each sub-business category in descending order of numerical value to obtain the third sorted list;

[0104] Step S704, or, sort the evaluation scores corresponding to each sub-business category in ascending order of numerical value to obtain the fourth sorting list.

[0105] In this embodiment, the evaluation is calculated using a scoring function, which is a weighted sum of various task indicators. The scoring function provides the evaluation score for each business category and the evaluation scores for the sub-business categories within each business category.

[0106] For example, taking the Ping An Digital Bank APP as an example, based on users' historical behavior data, a multi-business, multi-task learning model can predict the evaluation scores of various application services within the Ping An Digital Bank APP. For instance, it can predict that user A will have higher click rates on financial services, legal services, procurement services, loan services, and account services while using the Ping An Digital Bank APP. Since each of these application services has its own evaluation score, the evaluation scores can be sorted from highest to lowest value to obtain a first sorted list, or they can be sorted from lowest to highest value to obtain a second sorted list. Using either the first or second sorted list, the order of the application services with higher click rates for user A can be clearly obtained. Similarly, based on users' historical behavioral data, a multi-business, multi-task learning model can predict the evaluation scores of sub-business categories under various application services in the Ping An Digital Bank APP. For example, it can predict the evaluation scores of sub-businesses such as legal services (law retrieval, case retrieval, intelligent Q&A, and legal calculator). Then, the evaluation scores corresponding to each sub-business category can be sorted in descending order of value to obtain a third sorting list, or the evaluation scores corresponding to each sub-business category can be sorted in ascending order of value to obtain a fourth sorting list. Likewise, through the third or fourth sorting list, the ranking of each sub-business under the legal services category with high click-through rates for user A can be clearly obtained.

[0107] Step S104: Based on the sorting results, display each business category and each sub-business category on the corresponding display interface of the application software.

[0108] In this embodiment of the application, after sorting each business category and the sub-business categories under each business category, each business category and each sub-business category can be displayed on the corresponding display interface of the application software according to the sorting results.

[0109] Reference Figure 8 , Figure 8 This application provides a flowchart of steps for displaying each business category and each sub-business category on the corresponding display interface of the application software based on the sorting results, including but not limited to steps S801 to S804.

[0110] Step S801: Select the top N business categories from the first sorting list and display them on the corresponding display interface;

[0111] Step S802, or, select the business category that is ranked last N in the second sorting list and display it on the corresponding display interface, where N is a positive integer;

[0112] Step S803: Select the top M sub-business categories from the third sorting list and display them on the corresponding display interface.

[0113] Step S804, or, select the last M sub-business categories in the fourth sorting list and display them on the corresponding display interface, where M is a positive integer.

[0114] In this embodiment, based on the ranking results, multiple application services with the highest evaluation scores can be displayed on the corresponding interface of the application software. Similarly, multiple application sub-services with the highest evaluation scores under each application service can be displayed on the corresponding interface of the application software.

[0115] For example, taking the Ping An Digital Bank APP as an example, since step S103 has already sorted the multiple application services in the Ping An Digital Bank APP, only the top 5 application services in the first sorted list need to be selected and displayed on the first display interface of the Ping An Digital Bank APP, such as financial services, legal services, procurement services, loan services, and account services. Other application services can be hidden in the "More" icon on the first display interface; that is, the other application services will only be displayed when the user clicks the "More" icon. Similarly, if the various application sub-services under legal services have also been sorted accordingly, only the top 5 application sub-services in the sorted list need to be selected and displayed on the second display interface of the Ping An Digital Bank APP, such as legal search, case search, intelligent Q&A, legal calculator, and lawyer services. This display method allows for the display of the application services and application sub-services that user A is most likely to click on within the limited display space.

[0116] Reference Figure 9 , Figure 9 The flowchart of steps that can be executed after training a multi-service, multi-task learning model is provided in the embodiments of this application, including but not limited to steps S901 to S904.

[0117] Step S901: Obtain the first ranking results of each target object that matches the search keywords;

[0118] Step S902: Predict each target object using a pre-trained multi-service multi-task learning model to obtain the score results for each target object;

[0119] Step S903: Based on the scoring results, the first ranking result is corrected to obtain the second ranking result;

[0120] Step S904: Based on the second sorting result, display each target object on the corresponding display interface of the application software.

[0121] In this embodiment, based on users' historical behavior data, a pre-trained multi-service, multi-task learning model can not only predict the multiple application services or sub-services that users are most likely to click on, but also be applied to the fine-grained ranking of user search processes. Specifically, based on the ranking model, a first ranking result is obtained for each target object matching the search keywords. Then, the pre-trained multi-service, multi-task learning model is used to predict the rating results for each target object. Based on the rating results, the first ranking result is corrected to obtain a second ranking result. Finally, based on the second ranking result, each target object is displayed on the corresponding display interface of the application software.

[0122] For example, again using the Ping An Digital Bank APP as an example, refer to Figure 10 , Figure 10 This is an example diagram of the first sorting result provided in this application embodiment. When a user enters the search term "insurance," after processing by the sorting model, the redirected display interface will show multiple target objects matching "insurance." These target objects include application businesses and their sub-businesses. For example, the display interface shows corporate insurance products and their sub-products: Employer's Peace of Mind Insurance and Micro-Enterprise Insurance; insurance services: policies and claims; insurance zone: Ping An Student Accident Insurance; and personal insurance products: Ping An-Driver and Passenger Accident Insurance, Ping An-Family Fortune Insurance, and Ping An-Transportation Accident Insurance. Among these, corporate insurance products, insurance services, insurance zone, and personal insurance products are application businesses; Employer's Peace of Mind Insurance and Micro-Enterprise Insurance are sub-businesses under the corporate insurance product application business; policies and claims are sub-businesses under the insurance services application business; Ping An Student Accident Insurance is a sub-business under the insurance zone application business; and Ping An-Driver and Passenger Accident Insurance, Ping An-Family Fortune Insurance, and Ping An-Transportation Accident Insurance are sub-businesses under the personal insurance product application business. Then, the graph is processed through a pre-trained multi-service, multi-task learning model. Figure 10 The various application services and their sub-services shown are predicted to obtain scores for each application service and its sub-services. Based on these scores, the first ranking result is corrected to obtain a second ranking result. Finally, based on the second ranking result, the target objects are displayed on the corresponding interface of the application software as follows: Figure 11 As shown.

[0123] In this embodiment, the trained multi-service, multi-task learning model can also be applied to the ranking of search results. It can correct the coarse-grained ranking results processed by the ranking model, and can obtain user interests and preferences through historical user behavior data, thereby displaying application services and application sub-services that users are more likely to be interested in, matching the search terms, within a limited display interface.

[0124] Please see Figure 12 This application embodiment also provides a target object display device 120, which can implement the above-described target object display method. The device includes:

[0125] The collection module 121 is used to collect input features, which include user historical behavior data, user attribute data, item attribute data, and context data.

[0126] The processing module 122 is used to process the input features through a pre-trained multi-business multi-task learning model and predict the evaluation scores of each target object in the application software. The target objects include business categories and sub-business categories under the business categories in the application software.

[0127] The sorting module 123 is used to sort each business category and the sub-business categories under each business category according to the evaluation score.

[0128] The display module 124 is used to display each business category and each sub-business category on the corresponding display interface of the application software according to the sorting results.

[0129] The specific implementation of the display device for the target object is basically the same as the specific implementation of the display method for the target object described above, and will not be repeated here.

[0130] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method for displaying the target object. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0131] Please see Figure 13 , Figure 13 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0132] The processor 1301 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0133] The memory 1302 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1302 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1302 and is called and executed by the processor 1301 to execute the method for displaying the target object in the embodiments of this application.

[0134] The input / output interface 1303 is used to implement information input and output;

[0135] The communication interface 1304 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0136] Bus 1305 transmits information between various components of the device (e.g., processor 1301, memory 1302, input / output interface 1303, and communication interface 1304);

[0137] The processor 1301, memory 1302, input / output interface 1303 and communication interface 1304 are connected to each other within the device via bus 1305.

[0138] This application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described method for displaying the target object.

[0139] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0140] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0141] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0142] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0143] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0144] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0145] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0146] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0147] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0148] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0149] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0150] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for displaying a target object, characterized in that, The method includes: The input features are collected, including user historical behavior data, user attribute data, item attribute data, and contextual data; The input features are processed by a pre-trained multi-service multi-task learning model to predict the evaluation scores of each target object in the application software. The target objects include business categories and sub-business categories within the application software. The multi-service multi-task learning model includes an Embedding layer, a concatenation layer, a fully connected (FC) layer, a multi-service PLE network layer, and a merge layer. The multi-service PLE network layer is constructed by building a multi-task learning PLE network based on a CGC network for each business scenario, and then stacking the multi-task learning PLE networks of each business scenario. The processing of the input features by the pre-trained multi-service multi-task learning model includes: first inputting the input features into the Embedding layer... The input features are transformed into vectors of equal dimensions by a ding layer, then input into the concaten layer to concatenate all vectors of equal dimensions through convolution operations of convolution kernels, then input into the fully connected (FC) layer for nonlinear transformation, and finally input into the multi-service PLE network layer. Finally, the outputs of the multi-service PLE network layer and the FC fully connected layer are input into the merge layer to fuse the output features of the multi-service PLE network layer and the FC fully connected layer. The evaluation score is calculated by a scoring function, which is a weighted sum of various task indicators. Through this scoring function, the evaluation scores corresponding to each service category and the evaluation scores of sub-service categories under each service category can be obtained. The business categories are sorted according to the evaluation scores, and the sub-business categories under each business category are sorted. Based on the sorting results, each of the business categories and each of the sub-business categories are displayed on the corresponding display interface of the application software.

2. The method according to claim 1, characterized in that, After training the multi-service, multi-task learning model, the method further includes: Retrieve the first ranking results for each target object that matches the search keywords; The target objects are predicted by a pre-trained multi-service multi-task learning model to obtain the scoring results of each target object; Based on the scoring results, the first ranking result is corrected to obtain the second ranking result; Based on the second sorting result, each of the target objects is displayed on the corresponding display interface of the application software.

3. The method according to claim 1, characterized in that, The ranking of each business category and the ranking of the sub-business categories under each business category based on the evaluation scores include: The evaluation scores corresponding to each of the aforementioned business categories are sorted in descending order of numerical value to obtain a first sorted list; Alternatively, the evaluation scores corresponding to each of the aforementioned business categories can be sorted in ascending order of numerical value to obtain a second sorting list; The evaluation scores corresponding to each of the aforementioned sub-business categories are sorted in descending order of numerical value to obtain a third sorting list; Alternatively, the evaluation scores corresponding to each of the sub-business categories can be sorted in ascending order of numerical value to obtain a fourth sorting list.

4. The method according to claim 3, characterized in that, The step of displaying each business category and each sub-business category on the corresponding display interface of the application software according to the sorting result includes: Select the top N business categories from the first sorting list and display them on the corresponding display interface; Alternatively, select the last N business categories from the second sorting list and display them on the corresponding display interface, where N is a positive integer; Select the top M sub-business categories from the third sorting list and display them on the corresponding display interface; Alternatively, the last M sub-business categories in the fourth sorting list can be selected and displayed on the corresponding display interface, where M is a positive integer.

5. The method according to claim 1, characterized in that, Pre-training the multi-service, multi-task learning model includes: A training sample set is constructed based on users' historical behavior data; The training sample set is used to train the multi-service multi-task learning model. Determine the loss function; Based on the loss function, the parameters of the multi-service multi-task learning model are updated through backpropagation; Determine whether the training of the multi-service multi-task learning model meets the termination condition. If the termination condition is met, output the multi-service multi-task learning model. If the termination condition is not met, continue training.

6. The method according to claim 5, characterized in that, The determined loss function includes: Determine the weights for each business function and each task. Obtain the sample space for each task under each business; Calculate the loss function for each task under each business; The loss function is calculated based on the weights corresponding to each business, the weights corresponding to each task, the sample space of each task under each business, and the loss function of each task under each business.

7. The method according to claim 6, characterized in that, The loss function is calculated based on the weights corresponding to each business, the weights corresponding to each task, the sample space of each task under each business, and the loss function of each task under each business. The calculation is performed using the following formula: ； in, ; In the formula, This represents the loss function of a multi-service, multi-task learning model. express Business The characteristics of each task Indicates the total number of transactions. Indicates the total number of tasks. Indicates the first The first business The characteristics of each task Indicates shared features, Indicates business The weight, Indicates task The weight, Indicates the first The first business The loss function for each task. The indicator function represents the sample. Is it the first The first business If the sample of the task is yes, then it is counted in the first task. The first business The loss of each task This indicates that the multi-service, multi-task learning model is applied to samples. The prediction results Indicates sample The tag.

8. A display device for a target object, characterized in that, The device includes: The collection module is used to collect input features, which include user historical behavior data, user attribute data, item attribute data, and context data. The processing module is used to process the input features using a pre-trained multi-service multi-task learning model to predict the evaluation scores of each target object in the application software. The target objects include business categories and sub-business categories within the business categories in the application software. The multi-service multi-task learning model includes an embedding layer, a concatenation layer, a fully connected (FC) layer, a multi-service PLE network layer, and a merge layer. The multi-service PLE network layer is constructed by building a multi-task learning PLE network based on a CGC network for each business scenario, and then superimposing the multi-task learning PLE networks of each business scenario. The processing of the input features using the pre-trained multi-service multi-task learning model includes: first inputting the input features into the E... The input features are transformed into vectors of equal dimensions by an mbedding layer, then input into the concatenating layer to concatenate all vectors of equal dimensions through convolution operations of convolution kernels, then input into the fully connected (FC) layer for nonlinear transformation, and finally input into the multi-service PLE network layer. The outputs of the multi-service PLE network layer and the FC fully connected layer are then input into the merge layer to fuse the output features of the multi-service PLE network layer and the FC fully connected layer. The evaluation score is calculated by a scoring function, which is a weighted sum of various task indicators. Through this scoring function, the evaluation scores corresponding to each service category and the evaluation scores of sub-service categories under each service category can be obtained. The sorting module is used to sort each of the business categories and the sub-business categories under each of the business categories according to the evaluation scores. The display module is used to display each of the business categories and each of the sub-business categories on the corresponding display interface of the application software according to the sorting results.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images