Object recommendation method, model training method, device, equipment and medium

Abstract

The present application provides an object recommendation method, a model training method, an apparatus, a device and a medium, the object recommendation method comprising: in the object recommendation model, according to the attribute information of the user and the attribute information of the candidate object, determining the feature vector corresponding to the candidate object, the attribute information of the user includes the target region to which the user belongs, and the attribute information of the candidate object includes the statistical data corresponding to the candidate object in multiple regions; in the object recommendation model, based on the feature vector, the candidate object is scored through the scoring network in the multi-task network and the network branch connected to the scoring network to obtain the target score of the candidate object, and different network branches in the multi-task network correspond to different regions. Therefore, considering the differences in user consumption levels, preferences, etc. in different regions, by paying attention to the user's region and the statistical data corresponding to the object in multiple regions, the accuracy of object recommendation is improved through the multi-task network.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an object recommendation method, a model training method, a device, a equipment and a medium. Background Art

[0002] With the development of Internet technology, Internet merchants provide users with a large number of goods and services, which greatly meets the needs of users. How to accurately recommend objects of interest to users among the massive amount of goods and services is one of the challenges currently faced.

[0003] Usually, the access frequency and transaction frequency of the objects are counted, and based on the access frequency and transaction frequency of the objects, objects that the user may be interested in are selected from a large number of objects, for example, the objects with the highest access frequency are recommended to the user, and the objects with the highest transaction frequency are recommended to the user. It is also possible to select objects that the user may be interested in from a large number of objects based on the user's historical record data, such as the products or services that the user has visited, for example, among a large number of objects, determine the products that are highly similar to the products that the user has visited and recommend them to the user.

[0004] However, the recommendation accuracy of the above object recommendation methods is low. Summary of the invention

[0005] The present application provides an object recommendation method, a model training method, an apparatus, a device and a medium to solve the problem of low accuracy of object recommendation.

[0006] In a first aspect, an embodiment of the present application provides an object recommendation method, comprising: in an object recommendation model, determining a feature vector corresponding to a candidate object based on attribute information of a user and attribute information of a candidate object, wherein the attribute information of the user includes a target region to which the user belongs, and the attribute information of the candidate object includes statistical data corresponding to the candidate object in multiple regions; in the object recommendation model, based on the feature vector, scoring the candidate object through a scoring network in a multi-task network and a network branch connected to the scoring network to obtain a target score for the candidate object, wherein in the multi-task network, there are multiple network branches, and different network branches correspond to different regions; and recommending a target object from the candidate objects to the user based on the target score of the candidate object.

[0007] In a second aspect, an embodiment of the present application provides a model training method, including: determining a training data set, the training data set including multiple training samples and labels of the training samples, the training samples including attribute information of the user and attribute information of the object, the attribute information of the user including the target region to which the user belongs, the attribute information of the object including statistical data corresponding to the object in multiple regions, and the labels of the training samples are used to mark whether the user is interested in the objects in the training samples; according to the training data set, performing supervised training on an object recommendation model, wherein the object recommendation model includes a multi-task network for scoring objects, the multi-task network includes a scoring network and a network branch connected to the scoring network, there are multiple network branches, and different network branches correspond to different regions.

[0008] In a third aspect, an embodiment of the present application provides an object recommendation device, comprising: a feature determination unit, for determining, in an object recommendation model, a feature vector corresponding to a candidate object based on attribute information of a user and attribute information of the candidate object, wherein the attribute information of the user includes a target region to which the user belongs, and the attribute information of the candidate object includes statistical data corresponding to the candidate object in multiple regions; an object scoring unit, for scoring the candidate object in the object recommendation model based on the feature vector, through a scoring network in a multi-task network and a network branch connected to the scoring network, to obtain a target score for the candidate object, wherein in the multi-task network, there are multiple network branches, and different network branches correspond to different regions; an object recommendation unit, for recommending a target object from the candidate objects to the user based on the target score of the candidate object.

[0009] In a fourth aspect, an embodiment of the present application provides a model training device, comprising: a data determination unit, used to determine a training data set, the training data set including multiple training samples and labels of the training samples, the training samples including attribute information of the user and attribute information of the object, the attribute information of the user including the target region to which the user belongs, the attribute information of the object including statistical data corresponding to the object in multiple regions, and the labels of the training samples are used to mark whether the user is interested in the objects in the training samples; a training unit, used to perform supervised training on an object recommendation model according to the training data set, wherein the object recommendation model includes a multi-task network for scoring objects, the multi-task network includes a scoring network and a network branch connected to the scoring network, there are multiple network branches, and different network branches correspond to different regions.

[0010] In a fifth aspect, an embodiment of the present application provides a cloud server, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the cloud server can execute the object recommendation method provided in the first aspect of the embodiment of the present application, and / or so that the cloud server can execute the model training method provided in the second aspect of the embodiment of the present application.

[0011] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the object recommendation method provided in the first aspect of the embodiment of the present application is implemented, and / or, when the computer program is executed by a processor, the model training method provided in the second aspect of the embodiment of the present application is implemented.

[0012] In an embodiment of the present application, the object recommendation model includes a multi-task network, the multi-task network includes a scoring network and a network branch connected to the scoring network, there are multiple network branches in the multi-task network, and different network branches correspond to different regions. The process of using the object recommendation model to recommend objects includes: in the object recommendation model, according to the attribute information of the user and the attribute information of the candidate object, determining the feature vector corresponding to the candidate object, wherein the attribute information of the user includes the target region to which the user belongs, and the attribute information of the candidate object includes the statistical data corresponding to the candidate object in multiple regions; then, based on the feature vector corresponding to the candidate object, in the multi-task network included in the object recommendation model, the candidate object is scored through the scoring network and the network branch connected to the scoring network to obtain the target score of the candidate object; finally, according to the target score of the candidate object, the target object among the candidate objects is recommended to the user.

[0013] It can be seen that in the process of feature design, attention is paid to the region to which the user belongs and the statistical data corresponding to the distribution of objects in multiple regions; in the model structure of the object recommendation model, a multi-task model is used to provide different network branches for different regions. Therefore, starting from the feature design and model structure, the embodiment of the present application fully considers the differences between users in different regions, and adopts the object recommendation model to effectively improve the accuracy of object recommendation and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0015] Figure 1This is an example diagram of an application scenario provided according to an embodiment of the present application;

[0016] Figure 2 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 1 ;

[0017] Figure 3 Schematic diagram of the process of the object recommendation method provided in the embodiment of the present application Figure 1 ;

[0018] Figure 4 Schematic diagram of the process of the object recommendation method provided in the embodiment of the present application Figure 2 ;

[0019] Figure 5 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 2 ;

[0020] Figure 6 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 3 ;

[0021] Figure 7 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 4 ;

[0022] Figure 8 Schematic diagram of the process of the object recommendation method provided in the embodiment of the present application Figure 3 ;

[0023] Fig. 9 A flowchart of a model training method provided in an embodiment of the present application;

[0024] Fig.10 A structural block diagram of an object recommendation device provided in an embodiment of the present application;

[0025] Fig.11 A structural block diagram of a model training device provided in an embodiment of the present application;

[0026] Fig.12 A schematic diagram of the hardware structure of the cloud server provided in an embodiment of the present application.

[0027] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] The terms "first", "second", "third", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0030] Under the influence of population base, policies, culture, economic level, logistics and other factors, people in different regions have different consumption levels and product preferences. User traffic in different regions is unevenly distributed (for example, the number of daily visiting users in one region is 10,000, while the number of daily visiting users in another region is 40,000). Different regions also focus on different business goals and the indicator values ​​achieved by the business goals. These differences bring great challenges to object recommendation algorithms.

[0031] Taking commodities as an example, the following solutions can be adopted for commodity recommendations:

[0032] One solution is to build a recommendation model that covers users globally (i.e., all regions), without effectively modeling the differences between different regions. In this solution, regions with more unique visitors (UV) (i.e., regions with a large number of users) dominate the statistical characteristics of products, and the statistical characteristics of products are important dependent characteristics of the recommendation model. This makes regions with more UV dominate the recommendation results, and user needs in regions with fewer UV are difficult to be met and expressed. In the long run, product recommendations will fall into a serious Matthew effect, and the accuracy of product recommendations will be low.

[0033] Another solution is to train and deploy separate models for different regions. Specifically, a recommendation model is designed for each region, and the statistical characteristics of the products in the region are used to train the model and finally deploy the model. However, in actual training, the recommendation models corresponding to some regions face the problems of insufficient training samples and sparse features, and the offline training and online validation of the model in this method are relatively cumbersome.

[0034] Another solution is to use the same recommendation model in different regions, but optimize the recommendation model (i.e., train the model) separately for different regions. In other words, the recommendation model is optimized by region. However, in this solution, the model in regions with more UVs is fully optimized, while the model in regions with fewer UVs is not optimized enough.

[0035] In order to solve the above problems, the embodiments of the present application propose an object recommendation method, a model training method, an apparatus, a device and a medium. Among them, in the object recommendation method, the object recommendation model includes a multi-task network, the multi-task network includes a scoring network and a network branch connected to the scoring network, and there are multiple network branches in the multi-task network, and different network branches correspond to different regions. In the object recommendation model: according to the attribute information of the user and the attribute information of the candidate object, determine the feature vector corresponding to the candidate object, wherein the attribute information of the user includes the target region to which the user belongs, and the attribute information of the candidate object includes the statistical data corresponding to the candidate object in multiple regions; based on the feature vector, the candidate object is scored through the scoring network in the multi-task network and the network branch connected to the scoring network to obtain the target score of the candidate object. Finally, according to the target score of the candidate object, the target object among the candidate objects is recommended to the user.

[0036] Therefore, on the basis of a set of recommendation algorithms (i.e., a set of object recommendation models) covering global users, by designing a multi-task network including a scoring network and network branches corresponding to multiple regions in the object recommendation model, the scoring network is not affected by insufficient samples and sparse features, and the network branches can learn user characteristics and object characteristics of different regions in a targeted manner based on feature vectors, effectively improving the accuracy of object recommendation by the object recommendation model.

[0037] For example, taking the object as a commodity, Figure 1 1 is an example diagram of an application scenario provided according to an embodiment of the present application. Figure 1 As shown, in this application scenario, when users in multiple regions (such as region A and region B in the figure) browse product pages on the user terminal, the object recommendation device ( Figure 1 Taking the object recommendation device as a server as an example, the deployed object recommendation model is used to score the products, and then the products are recommended to the user terminal based on the scores of the products.

[0038] refer to Figure 2 , Figure 2 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 1 .like Figure 2As shown in the structure of the object recommendation model, the object recommendation model includes an input layer and a multi-task network connected to the input layer. The multi-task network includes a scoring network and multiple network branches. The scoring network connects the network branches. Different network branches correspond to different regions, such as Figure 2 The network branch corresponding to region A and the network branch corresponding to region B. The input layer is used to input the attribute information of the user and the attribute information of the object, and output the feature vector corresponding to the object. The multi-task network is used to score the object based on the feature vector corresponding to the object to obtain the score of the object.

[0039] Optionally, in the object recommendation model, the input layer may be a feature embedding layer to perform feature embedding on the model input data.

[0040] based on Figure 2 The object recommendation model shown in Figure 1 In the process of using the object recommendation model to score products, the product can be scored based on the user's attribute information and the product's attribute information in the object recommendation model through the scoring network in the multi-task network and the network branch connected to the scoring network to obtain the product score.

[0041] Among them, the object recommendation device can obtain the user's attribute information from the user terminal, or obtain the user ID from the user terminal, and then obtain the user's attribute information based on the user ID. The user's attribute information includes the region to which the user belongs; the attribute information of the product can be pre-stored in the object recommendation device, or obtained from other devices.

[0042] Next, combined with the above Figure 1 The application scenarios shown are used to explain the technical solutions of the present application in detail through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0043] It should be noted that the execution subject of the embodiments of the present application may be an electronic device, and the electronic device may be a terminal or a server. Among them, the terminal may be a personal digital assistant (PDA) device, a handheld device with wireless communication function (such as a smart phone, a tablet computer), a computing device (such as a personal computer (PC) for short), a vehicle-mounted device, a wearable device (such as a smart watch, a smart bracelet), a smart home device (such as a smart display device), etc. Among them, the server may be a single server, a server cluster, a distributed server, a centralized server, or a cloud server.

[0044] The above is only an exemplary application scenario. The embodiment of the present application can be applied to any object recommendation scenario on the Internet, such as a merchant recommendation scenario, a logistics service recommendation scenario, etc.

[0045] refer to Figure 3 , Figure 3 Schematic diagram of the process of the object recommendation method provided in the embodiment of the present application Figure 1 .like Figure 3 As shown, the object recommendation method provided in the embodiment of the present application specifically includes the following steps:

[0046] S301, in the object recommendation model, a feature vector corresponding to the candidate object is determined based on the user's attribute information and the candidate object's attribute information, wherein the user's attribute information includes the target region to which the user belongs, and the candidate object's attribute information includes statistical data corresponding to the candidate object in multiple regions.

[0047] The object recommendation model is a pre-trained deep learning network for scoring objects. The structure of the object recommendation model can be referred to Figure 2 Specifically, the object recommendation model multi-task network includes a scoring network and multiple network branches. The multiple network branches can be shallow networks, such as multi-layer perceptrons. Different network branches correspond to different regions. In the object recommendation model, the scoring network is used to score objects, and multiple network branches are used to optimize the scores of the scoring network by region to improve the accuracy of the object scores.

[0048] A multi-task network can also be called a multi-task learning network, which can use the commonality of multiple related learning tasks to complete multiple related learning tasks. In the embodiment of the present application, the multiple learning tasks to be completed by the multi-task network include providing targeted object ratings for users in different regions, wherein the rating network is used to learn the common features of objects in multiple regions, and multiple network branches are used to learn the uniqueness of objects in each region.

[0049] The region may be a city, a country, a continent, etc. The target region to which the user belongs may be directly obtained from the user terminal; or, the user ID may be obtained from the user terminal and then queried from a user database based on the user ID; or, the device identification of the user terminal may be obtained and the target region to which the user belongs may be determined based on the device identification, for example, the device identification is an IP address and the target region to which the user belongs may be determined based on the IP address of the user terminal.

[0050] Optionally, in addition to the target region to which the user belongs, the user's attribute information may also include the user's interaction data with the object in a historical time period, wherein the interaction data may include at least one of the following: objects browsed by the user, objects ordered by the user, objects collected by the user, objects commented on by the user, and objects repurchased by the user. In addition, the user's attribute information may also include the user's basic information, such as the user's age range, the user's gender, etc. The historical time period may be, for example, the past week, the past ten days, or the past month, etc.

[0051] Among them, the candidate objects may be all or part of the objects in a massive object set; or, to improve the accuracy and efficiency of object recommendation, the candidate objects may be objects interacted with by users within a historical time period, such as objects browsed and ordered by users within a historical time period; or, to improve the accuracy and efficiency of object recommendation, the candidate objects may be objects interacted with by multiple users in the target area within a historical time period.

[0052] The statistical data corresponding to the candidate objects in multiple regions may include the statistical data of the candidate objects in different regions under the feature dimension, and the feature dimension may include at least one of the following: exposure dimension, click dimension, transaction dimension, favorite dimension, add to shopping list dimension, and profit dimension. As an example, the statistical data of the candidate objects in different regions under the exposure dimension include: the number of times the candidate objects are exposed to users in region A, the number of times the candidate objects are exposed to users in region B, etc.; the statistical data of the candidate objects in different regions under the click dimension include: the number of times the candidate objects are clicked by users in region A, the number of times the candidate objects are clicked by users in region B; the remaining dimensions are not given examples one by one here.

[0053] Optionally, the attribute information of the candidate object also includes identification information of the candidate object to distinguish different candidate objects.

[0054] Optionally, the attribute information of the candidate object also includes total statistical data of the candidate object in all regions, such as the sum of exposure times of the candidate object in all regions, the sum of click times of the candidate object in all regions, etc. Thus, by providing statistical data of the candidate object in all regions and statistical data by region, the accuracy of object recommendation is improved.

[0055] In this embodiment, the user's attribute information can be determined in response to an object recommendation request from a user terminal (for example, when a user browses a product recommendation page, the user terminal sends an object recommendation request to an object recommendation device). In response to the object recommendation request, the attribute information of the candidate object is obtained from a pre-collected object database. Afterwards, the user's attribute information and the attribute information of the candidate object can be input into the object recommendation model. In the object recommendation model, the user's attribute information and the attribute information of the candidate object are encoded to obtain a feature vector corresponding to the candidate object. If there are multiple candidate objects, for each candidate object, the user's attribute information and the attribute information of the candidate object can be input into the object recommendation model, and the user's attribute information and the attribute information of the candidate object are encoded in the object recommendation model to obtain a feature vector corresponding to the candidate object. In this way, a feature vector corresponding to each candidate object is obtained.

[0056] Optionally, when the object recommendation model includes a feature embedding layer, the user's attribute information and the candidate object's attribute information can be input into the feature embedding layer, and feature embedding processing is performed on the user's attribute information and the candidate object's attribute information in the feature embedding layer to obtain a feature vector with the candidate object. Thus, the user's attribute information and the candidate object's attribute information are reduced to a feature vector with a fixed dimension through feature embedding, thereby improving the feature learning effect of the subsequent multi-task network.

[0057] S302, in the object recommendation model, based on the feature vector corresponding to the candidate object, the candidate object is scored through the scoring network in the multi-task network and the network branch connected to the scoring network to obtain a target score for the candidate object.

[0058] Among them, the network branches connected to the scoring object include network branches corresponding to the target area.

[0059] In this embodiment, the feature vector of the candidate object is input into the multi-task network, and the feature vector is processed through the scoring network and the network branch connected to the scoring network to obtain the target score of the candidate object.

[0060] In the process of performing feature processing on the feature vector through the scoring network and the network branches connected to the scoring network, the feature vector can be feature processed through the scoring network and the network branches corresponding to the target area, and then the score obtained through the network branch corresponding to the target area is determined as the target score of the candidate object; or, the feature vector can be feature processed through the scoring network and multiple network branches, and then the scores obtained through the multiple network branches are weighted, such as weighted summation, to obtain the target score of the candidate object. In the weighted processing, a relatively high weight can be assigned to the network branch corresponding to the target area so that the target score is mainly affected by the learning effect of the network branch.

[0061] S303: recommending target objects among the candidate objects to the user according to the target scores of the candidate objects.

[0062] In this embodiment, after obtaining the target score of the candidate object, the target object among the candidate objects can be recommended to the user in a descending order of the target score. Specifically, in one method, the target object can be determined to be a candidate object with a target score greater than a score threshold, and the target object can be recommended to the user; in another method, the candidate objects can be sorted in a descending order of the target score, and a preset number of target objects can be selected from the sorted candidate objects. After determining the target object, the relevant information of the target object (such as pictures, prices, etc.) can be sent to the user's terminal in a descending order of the target score, and the user's terminal can display the relevant information of the target object on the display screen to achieve object recommendation.

[0063] In the embodiment of the present application, attention is paid to the target region to which the user belongs, and data statistics of the candidate objects are performed by region. In the object recommendation model, a multi-task network including a scoring network and multiple network branches is designed. Through the scoring network and multiple network branches in the multi-task network, the commonalities and differences in the characteristic performance of the candidate objects in multiple regions are learned. In this way, the object recommendation model can be applied to all regions, and users in various regions can provide ratings of objects in a targeted manner, that is, recommend objects in a targeted manner. Moreover, fewer samples are required for network branch training, which solves the problem that the model effect is poor due to fewer training samples in regions with fewer UVs. Therefore, the accuracy of object recommendations in various regions is effectively improved.

[0064] Reference Figure 4 , Figure 4 Schematic diagram of the process of the object recommendation method provided in the embodiment of the present application Figure 2 .like Figure 4 As shown, the object recommendation method provided in the embodiment of the present application specifically includes the following steps:

[0065] S401, in the object recommendation model, a feature vector corresponding to the candidate object is determined based on the user's attribute information and the candidate object's attribute information, wherein the user's attribute information includes the target region to which the user belongs, and the candidate object's attribute information includes statistical data corresponding to the candidate object in multiple regions.

[0066] The implementation principle and technical effects of S401 may refer to the aforementioned embodiments and will not be described in detail.

[0067] S402, inputting the feature vector corresponding to the candidate object into the scoring network in the multi-task network, scoring the candidate object based on the feature vector in the scoring network, and obtaining a score output by the scoring network.

[0068] The scoring network can also be called an expert network. The scoring network can include multiple network layers. Furthermore, the network layer in the scoring network can be a fully connected network layer. The input data of the scoring network is the feature vector corresponding to the candidate object, and the output data of the scoring network is the score corresponding to the feature vector.

[0069] In this embodiment, the feature vector corresponding to the candidate object is input into the scoring network, where the feature vector undergoes feature processing of multiple network layers, such as convolution processing, and finally, a score output by the scoring network is obtained, which can be regarded as the initial score of the candidate object.

[0070] S403: input the score output by the scoring network into the network branch, and optimize the score output by the scoring network in the network branch to obtain a target score.

[0071] The scoring network can be connected to one or more network branches to provide scores for one or more network branches. In the multi-task network, different network branches correspond to different regions, and the network branch corresponding to a region is trained based on the difference between the preference of the object by the users in the region and the score of the object by the object recommendation model.

[0072] In this embodiment, the score output by the scoring network is input into the network branch connected to the scoring network, and the score output by the scoring network is optimized through multiple network layers in the network branch. Since the network branch has regional differences, the score after the network branch optimization is the score of the candidate in the region corresponding to the network branch. For example, the score after the network branch optimization corresponding to region A is the score of the candidate in region A, and the score after the network branch optimization corresponding to region B is the score of the candidate in region B. Finally, based on the score after the network branch optimization, the target score of the candidate is obtained.

[0073] In one possible model structure, reference Figure 5 , Figure 5 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 2 , the scoring network is multiple, and different network branches share all scoring networks. In other words, each network branch is connected to all scoring networks. In the object recommendation process, the scoring network is equivalent to the judges. The accuracy of scoring by multiple judges is obviously higher than the accuracy of scoring by one judge. Therefore, setting multiple scoring networks connected to all network branches in the object recommendation model is equivalent to providing multiple judges for each region. Moreover, different network branches share all scoring networks, which means that the scoring network can be fully trained based on samples from all regions. There is no problem of insufficient samples and sparse features. Therefore, this model structure can effectively improve the recommendation accuracy of the object recommendation model.

[0074] based on Figure 5 The object recommendation model shown in FIG. 1 may be implemented in a manner such as Figure 4 As shown, S402 includes: S4021, inputting the feature vectors corresponding to the candidate objects into each scoring network in the multi-task network, scoring the candidate objects based on the feature vectors in each scoring network, and obtaining the scores output by each scoring network. S403 includes: S4031, for each network branch, inputting the scores output by all scoring networks into the network branch, fusing the scores of all scoring networks in the network branch, and obtaining the scores output by the network branch; S4032, obtaining the target scores according to the scores output by the network branches. Thus, multiple scoring networks are used to learn the global feature performance of the candidate objects, and the feature performance of the candidate objects in each region is learned through the network branches corresponding to the regions, thereby improving the accuracy of object recommendation.

[0075] The scores of all scoring networks are integrated in the network branch. For example, all scoring networks are weighted and adjusted in the network branch according to the network parameters in the network branch.

[0076] Optionally, any model result based on the object recommendation model, such as Figure 5 The object recommendation model shown in Figure 5 As shown, the network branch includes a weighted network and an optimization network. The weighted network is used to fuse the scores from multiple scoring networks, and the optimization network is used to further optimize the scores fused by the weighted network, so as to improve the scoring accuracy of the object recommendation model by weighting and further optimization.

[0077] In another possible model structure, reference Figure 6 , Figure 6 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 3The multi-task network also includes a mask network, which is used to filter the output data of multiple network branches to obtain the output data of the network branch corresponding to the target area from the output data of multiple network branches, that is, to mask the output data of other network branches. The network parameters of the mask network are determined by the target area to which the user belongs.

[0078] based on Figure 6 In the object recommendation model shown, in one possible implementation, S4032 includes: initializing a mask network according to a target region; in a multi-task network, inputting the scores output by all network branches into the initialized mask network, and filtering out the scores output by the target network branches corresponding to the target region from the initialized mask network; and determining the target score according to the scores output by the target network branches.

[0079] For the sake of simplicity, the network branch corresponding to the target area is referred to as the target network branch.

[0080] In this implementation, the network parameters of the mask network include network parameters corresponding to each network branch. In the process of initializing the mask network, the network parameters corresponding to the target network branch can be initialized to 1 in the network parameters of the mask network, and the network parameters corresponding to the remaining network branches can be initialized to 0. After that, the network parameters in the mask network are weighted with the scores output by the corresponding network branches to screen out the scores output by the target network branches. After that, the target score of the candidate object can be determined as the score output by the target network branch. Therefore, in this implementation, when the user belongs to the target area and users in different areas have different interests in the object, the score of the candidate object in the target area is obtained through the above network structure, which improves the accuracy of the score and thus improves the accuracy of the object recommendation.

[0081] S404: Recommend target objects among the candidate objects to the user according to the target scores of the candidate objects.

[0082] The implementation principle and technical effects of S404 may refer to the aforementioned embodiments and will not be described in detail.

[0083] In the embodiment of the present application, the target region to which the user belongs is paid attention, and the data statistics of the candidate objects are divided into regions. In the object recommendation model, a multi-task network including a scoring network and multiple network branches is designed, and the candidate objects are scored by the scoring network in the multi-task network, and the scores output by the scoring network are optimized by regions through multiple network branches. In this way, a set of recommendation algorithms is implemented to be applicable to all regions, and the scores of objects are provided to users in various regions in a targeted manner, that is, the object recommendation model is compatible with regional differences, and the scoring network can be fully trained under global samples. As a shallow network, the network branches require fewer samples for training, which solves the problem that the model effect is poor due to fewer training samples in regions with fewer UVs. Therefore, the accuracy of object recommendation in various regions is effectively improved.

[0084] In another possible model structure, reference Figure 7 , Figure 7 Schematic diagram of the structure of the object recommendation model provided in the embodiment of the present application Figure 4 There are multiple multi-task networks, where different multi-task networks correspond to different business goals, for example, business goals related to visits, transactions, and profits. Each multi-task network includes a scoring network and multiple network branches. In each multi-task network, different network branches correspond to different regions. Through the multi-task network, the target score of the object under the business goal corresponding to the multi-task network can be obtained.

[0085] Among them, the structures of different multi-task networks can be the same.

[0086] based on Figure 7 The object recommendation model shown in Figure 8 , Figure 8 Schematic diagram of the process of the object recommendation method provided in the embodiment of the present application Figure 3 The object recommendation method provided in the embodiment of the present application specifically includes the following steps:

[0087] S801. In the object recommendation model, a feature vector corresponding to the candidate object is determined based on the user's attribute information and the candidate object's attribute information, wherein the user's attribute information includes the target region to which the user belongs, and the candidate object's attribute information includes statistical data corresponding to the candidate object in multiple regions.

[0088] The implementation principle and technical effects of S801 may refer to the aforementioned embodiments and will not be described in detail.

[0089] S802. In the object recommendation model, the feature vector is input into each multi-task network. Based on the feature vector, the candidate object is scored by a scoring network in the multi-task network and a network branch in the multi-task network connected to the scoring network to obtain a target score of the candidate object under the business goal corresponding to the multi-task network.

[0090] In this embodiment, the feature vectors corresponding to the candidate objects are respectively input into each multi-task network of the object recommendation model. In each multi-task network, the feature vectors are processed through the scoring network and multiple network branches to achieve scoring of the candidate objects, and the scores output by each multi-task network are obtained, that is, the target scores of the candidate objects under the business objectives corresponding to each multi-task network. The process of scoring the candidate objects through the scoring network and multiple network branches in the multi-task network can refer to the above-mentioned embodiment and will not be repeated here.

[0091] S803: Determine a final target score of the candidate object according to the target score of the candidate object under the business goals corresponding to each multi-task network.

[0092] In this embodiment, the target score of the candidate object can be determined to include the target score of the candidate object under the business goals corresponding to each multi-task network, or the target scores of the candidate object under the business goals corresponding to each multi-task network can be fused, such as by performing calculations through some preset calculation formulas to obtain the final target score of the candidate object.

[0093] Optionally, the business objectives include click-through rate (CTR) and conversion rate (ConversionRate), and the statistical data of the candidate objects in multiple regions include the number of users who click on the candidate objects in multiple regions (such as the number of users who click on the candidate objects in region A and the number of users who click on the candidate objects in region B) and the number of users who have completed transactions with the candidate objects in multiple regions, and, if Figure 7 As shown, the multi-task network includes a first multi-task network corresponding to the click-through rate and a second multi-task network corresponding to the conversion rate. At this time, based on the number of users who clicked and the number of users who completed transactions for the candidate objects in multiple regions, the first multi-task network and the second multi-task network can be used to score the candidate objects in terms of click-through rate and conversion rate respectively.

[0094] Among them, click-through rate and conversion rate are two basic business goals, based on which other business goals can be converted. Therefore, in the object recommendation model, a first multi-task network corresponding to the click-through rate and a second multi-task network corresponding to the conversion rate are designed. In addition to obtaining the target scores of candidate objects under the click-through rate and conversion rate, the target scores of candidate objects under other business goals can also be further calculated based on the target scores of candidate objects under the click-through rate and conversion rate. In turn, the object recommendation model can be applied to different business scenarios and different regions, improving the scope of application of the object recommendation model and the accuracy of object recommendation based on the object recommendation model.

[0095] Based on this optional solution, in a possible implementation, S802 includes: inputting the feature vector into the first multi-task network, scoring the candidate object through the scoring network in the first multi-task network and the network branch in the first multi-task network connected to the scoring network, and obtaining the target score of the candidate object under the click-through rate; inputting the feature vector into the second multi-task network, scoring the candidate object through the scoring network in the second multi-task network and the network branch in the second multi-task network connected to the scoring network, and obtaining the target score of the candidate object under the conversion rate. At this time, S803 includes: obtaining the final target score of the candidate object based on the target score of the candidate object under the click-through rate and the target score of the candidate object under the conversion rate.

[0096] In one example, Figure 7 As shown, the target score of the candidate object under the click-through rate is multiplied by the target score of the candidate object under the conversion rate to obtain the final target score of the candidate object.

[0097] S804: Recommend target objects among the candidate objects to the user according to the target scores of the candidate objects.

[0098] The implementation principle and technical effects of S804 may refer to the aforementioned embodiments and will not be described in detail.

[0099] In an embodiment of the present application, attention is paid to the target region to which the user belongs, and data statistics of the candidate objects are performed by region. In the object recommendation model, multiple multi-task networks are designed, and each multi-task network includes a scoring network and multiple network branches. The candidate objects are scored under multiple business objectives in the multi-task network, which takes into account the regional differences and the diversity of business objectives, so that the object recommendation method can be applied to a variety of recommendation scenarios and improves the accuracy of object recommendation in various regions.

[0100] Based on any of the foregoing embodiments, optionally, the model architecture of the object recommendation model can adopt a multi-task learning task relationship model based on a multi-gate mixture of experts (Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts, MMOE), and the multi-task network can adopt an Entire Space Multi-task Model (ESMM). In other words, within the framework of ESMM, the scoring tasks under multiple business objectives can be fitted respectively with a regionally customized MMOE structure. Further, in the case where the business objectives include CTR (click-through rate) and CVR (conversion rate), within the framework of ESMM, the scoring tasks under CTR and CVR are fitted respectively with a regionally customized MMOE structure. Thus, combined with the advantages of MMOE being compatible with multiple multi-task networks that are related and different and the advantages of ESMM being better in solving the problems of data sparsity and sample selection bias, the recommendation relevance and accuracy are improved, and the area of ​​the user benefit group is expanded.

[0101] refer to Fig. 9 , Fig. 9 The model training method provided in the embodiment of the present application specifically comprises the following steps:

[0102] S901, determine a training data set, the training data set includes multiple training samples and labels of the training samples, the training samples include attribute information of the user and attribute information of the object, the attribute information of the user includes the target area to which the user belongs, the attribute information of the object includes statistical data corresponding to the object in multiple areas, and the label of the training sample is used to mark whether the user is interested in the object in the training sample.

[0103] Among them, the user's attribute information, the object's attribute information, and the statistical data corresponding to the object in multiple regions can refer to the description of the user's attribute information, the candidate object's attribute information, and the candidate object's statistical data corresponding to multiple regions in the aforementioned embodiment, and will not be repeated here. Whether the user is interested in the object in the training sample can be reflected in different user behaviors in different recommendation scenarios. For example, in the recommendation scenario of improving the click-through rate of the object, whether the user is interested in the object is reflected in whether the user clicks on the object (i.e., whether the object is visited). In the recommendation scenario of improving the transaction rate of the object, whether the user is interested in the object is reflected in whether the user purchases the object.

[0104] In this embodiment, a pre-collected training data set may be obtained from a database.

[0105] S902, performing supervised training on the object recommendation model according to the training data set, wherein the recommendation model includes a multi-task network for scoring the objects, the multi-task network includes a scoring network and a network branch connected to the scoring network, there are multiple network branches, and different network branches correspond to different regions.

[0106] Among them, the object recommendation model can be applied to the object recommendation method provided by any of the aforementioned embodiments, and its model structure can refer to the description of any of the aforementioned embodiments and will not be repeated here.

[0107] In this embodiment, the training samples in the training data set are used as input data, and the objects in the training samples are scored by the object recommendation model to obtain the target score of the objects in the training samples. The target score of the object reflects the probability that the user is interested in the object. Therefore, based on the target score of the object and the label of the training sample, the model parameters of the object recommendation model can be adjusted to achieve supervised training of the object recommendation model. The object recommendation model can be trained multiple times in a supervised manner to improve the scoring accuracy of the object recommendation model, that is, to improve the accuracy of the object recommendation model for object recommendation.

[0108] In a possible implementation, S902 includes: in the object recommendation model, determining a feature vector corresponding to the object in the training sample based on the attribute information of the user and the attribute information of the object in the training sample; in the object recommendation model, scoring the objects in the training sample based on the feature vector corresponding to the object in the training sample through a scoring network in a multi-task network and a network branch connected to the scoring network to obtain a target score for the object in the training sample; determining a loss value of the object recommendation model based on the label of the training sample and the target score of the object in the training sample; and adjusting the model parameters of the object recommendation model based on the loss value of the object recommendation model.

[0109] Among them, the process of scoring the object by using the object recommendation model can refer to the process of scoring the candidate objects by using the object recommendation model in the aforementioned embodiment, which will not be repeated here.

[0110] In this implementation, based on the attribute information of the user and the attribute information of the object in the training sample, the object in the training sample is scored by the object recommendation model. After obtaining the target score of the object in the training sample, the loss value of the object recommendation model can be determined based on the difference between the label of the training sample and the target score of the object in the training sample. Then, based on the loss value, the model parameters of the object recommendation model are adjusted using an optimization algorithm.

[0111] The optimization algorithm is, for example, a gradient descent method, and no limitation is imposed on the optimization algorithm.

[0112] In another possible implementation, when the object recommendation model includes multiple multi-task networks, S902 includes: in the object recommendation model, determining a feature vector corresponding to the object in the training sample based on the attribute information of the user and the attribute information of the object in the training sample; in the object recommendation model, scoring the objects in the training sample based on the feature vector corresponding to the object in the training sample through the scoring network in each multi-task network and the network branch connected to the scoring network to obtain the target score of the object in the training sample under the business objectives corresponding to each multi-task network; determining multiple loss values ​​of the object recommendation model based on the label of the training sample and the target score of the object in the training sample under the business objectives corresponding to each multi-task network; and adjusting the model parameters of the object recommendation model based on the multiple loss values.

[0113] Among them, the process of scoring the object by using multiple multi-task networks in the object recommendation model can refer to the process of scoring the candidate object by using multiple multi-task networks in the object recommendation model in the aforementioned embodiment, which will not be repeated here. The label of the training sample includes the actual indicator value of the object in the training sample under the business objectives corresponding to the multiple multi-task networks respectively.

[0114] In this implementation, for each multi-task network, the network parameters of the multi-task network can be adjusted based on the difference between the actual indicator values ​​of the objects in the training samples under the business objectives corresponding to the multi-task network and the target scores of the objects in the training samples under the business objectives, so as to improve the accuracy of parameter adjustment of the multi-task network.

[0115] Furthermore, in the case where the business objectives include click-through rate and conversion rate, the labels of the training samples may include the actual click-through rate and actual conversion rate of the objects in the training samples, the target score of the object under the click-through rate is equivalent to the predicted click-through rate of the object, and the target score of the object under the conversion rate is equivalent to the predicted conversion rate of the object. Therefore, based on the difference between the target score of the object under the click-through rate and the actual click-through rate of the object, the network parameters of the first multi-task network corresponding to the click-through rate can be adjusted, and based on the difference between the target score of the object under the conversion rate and the actual conversion rate of the object, the network parameters of the second multi-task network corresponding to the conversion rate can be adjusted.

[0116] Based on any of the aforementioned parameter adjustment schemes, optionally, since different network branches in the multi-task network correspond to different regions, in the process of adjusting the model parameters of the object recommendation model, the scoring network in the multi-task network and the target network branch corresponding to the target region can be adjusted according to the loss value.

[0117] In this optional method, based on the attribute information of users in the target area and the attribute information of the object, the object recommendation model is used to score the object. After obtaining the target score of the object, the target score of the object reflects the target score of the object in the target area, that is, it reflects the interest of users in the target area in the object. Considering the differences among users in different regions, the loss value can be determined based on the difference between the target score of the object in the training sample and the label corresponding to the training sample. Based on the loss value, the scoring network in the multi-task network and the target network branch corresponding to the target area are adjusted so that the target network branch learns the features related to the preference of users in the target area for the object.

[0118] The estimation optimization process of object C in different areas (area A and area B) is taken as an example to illustrate as follows:

[0119] Based on the attribute information of object C (such as the total statistical data of object C in all regions and the statistical data of object C in region A) and the attribute information of users in region A, the target scores (under the two business goals of click-through rate and conversion rate) of object C in region A can be obtained through the object recommendation subnetwork of region A (for example, when the business goals include click-through rate and conversion rate, the object recommendation subnetwork of region A includes a multi-task network corresponding to the click-through rate, a multi-task network corresponding to the conversion rate, and a feature embedding layer); the difference between the actual indicators of object C in region A (the actual click-through rate and the actual transaction rate) and the above target scores can be used to calculate the loss value; based on the loss value, in the multi-task network corresponding to the click-through rate and the multi-task network corresponding to the conversion rate (which can also be referred to as the click-through rate estimation network and the conversion rate estimation network, respectively), the parameters of the scoring network and the network branches corresponding to region A are adjusted.

[0120] Similarly, the target score of object C in area B can also be obtained by inputting the attribute information of object C (such as the total statistical data of object C in all areas and the statistical data of object C in area B) and the attribute information of users in area B into the object recommendation subnetwork of area B, and obtaining the difference between the target score of object C in area B and the actual index of object C in area B, and the corresponding loss value of object C in area B can be calculated; based on the loss value, in the multi-task network, the parameters of the scoring network and the network branch corresponding to area B are adjusted.

[0121] Among them, since the scoring network is shared by multiple network branches, and the network branches are unique to a region, when the parameters are updated, the scoring network shared by all network branches can be updated, and the network branches unique to a region can only be updated by the region's own losses. In other words, the loss of region A cannot update the network branch of region B, but can only update the network branch of region A; the loss of region B cannot update the network branch of region A, but can only update the network branch of region B.

[0122] Based on any of the aforementioned parameter adjustment schemes, optionally, when the object recommendation model includes a feature embedding layer, the network parameters of the feature embedding layer can also be adjusted based on the loss value to improve the accuracy of the embedded representation of the data input into the object recommendation model, thereby improving the accuracy of object recommendation.

[0123] The object recommendation device provided in the embodiment of the present application is used to execute the technical solution in any method embodiment of the above-mentioned object recommendation method. Its implementation principle and technical effect are similar and will not be repeated here.

[0124] refer to Fig.10 , Fig.10 FIG. 1 is a structural block diagram of an object recommendation device 100 provided in an embodiment of the present application. Fig.10 As shown, the object recommendation device 100 provided in the embodiment of the present application includes: a feature determination unit 101, an object scoring unit 102 and an object recommendation unit 103, wherein:

[0125] The feature determination unit 101 is used to determine a feature vector corresponding to a candidate object according to the attribute information of the user and the attribute information of the candidate object in the object recommendation model, wherein the attribute information of the user includes the target region to which the user belongs, and the attribute information of the candidate object includes statistical data corresponding to the candidate object in multiple regions;

[0126] The object scoring unit 102 is used to score the candidate object based on the feature vector in the object recommendation model through the scoring network in the multi-task network and the network branch connected to the scoring network to obtain a target score of the candidate object, wherein the multi-task network has multiple network branches, and different network branches correspond to different regions;

[0127] The object recommendation unit 103 is used to recommend a target object among the candidate objects to the user according to the target scores of the candidate objects.

[0128] In one embodiment of the present application, the object scoring unit 102 is specifically used to: input the feature vector into the scoring network, score the candidate object based on the feature vector in the scoring network, and obtain the score output by the scoring network; input the score output by the scoring network into the network branch, optimize the score output by the scoring network in the network branch, and obtain the target score.

[0129] In one embodiment of the present application, there are multiple scoring networks, and different network branches share all the scoring networks. When the scores output by the scoring networks are input into the network branches, the object scoring unit 102 is specifically used to: for each network branch, input the scores output by all the scoring networks into the network branch, fuse the scores of all the scoring networks in the network branch, and obtain the scores output by the network branch; and obtain the target scores according to the scores output by the network branches.

[0130] In one embodiment of the present application, the multi-task network also includes a mask network, and the object scoring unit 102 is specifically used to: initialize the mask network according to the target area; in the multi-task network, input the scores output by all network branches into the initialized mask network, and filter out the scores output by the target network branches corresponding to the target area from the initialized mask network; determine the target score according to the scores output by the target network branches.

[0131] In one embodiment of the present application, there are multiple multi-task networks, and different multi-task networks correspond to different business objectives. The object scoring unit 102 is specifically used to: in the object recommendation model, input the feature vector into each multi-task network, and score the candidate objects based on the feature vector through the scoring network in the multi-task network and the network branch in the multi-task network connected to the scoring network to obtain the target score of the candidate object under the business objective corresponding to the multi-task network; determine the final target score of the candidate object according to the target score of the candidate object under the business objective corresponding to each multi-task network.

[0132] In one embodiment of the present application, business objectives include click-through rate and conversion rate, statistical data include the number of users who click and the number of users who complete transactions, the multi-task network includes a first multi-task network corresponding to the click-through rate and a second multi-task network corresponding to the conversion rate, and the object scoring unit 102 is specifically used to: input the feature vector into the first multi-task network, score the candidate objects through the scoring network in the first multi-task network and the network branch in the first multi-task network connected to the scoring network, and obtain the target score of the candidate objects under the click-through rate; input the feature vector into the second multi-task network, score the candidate objects through the scoring network in the second multi-task network and the network branch in the second multi-task network connected to the scoring network, and obtain the target score of the candidate objects under the conversion rate.

[0133] refer to Fig.11 , Fig.11 This is a structural block diagram of the model training device 110 provided in the embodiment of the present application. Fig.11 As shown, the model training device 110 provided in the embodiment of the present application includes: a determination unit 111 and a training unit 112, wherein:

[0134] The data determination unit 111 is used to determine a training data set, wherein the training data set includes a plurality of training samples and labels of the training samples, wherein the training samples include attribute information of the user and attribute information of the object, wherein the attribute information of the user includes the target region to which the user belongs, and the attribute information of the object includes statistical data corresponding to the object in a plurality of regions, and the label of the training sample is used to mark whether the user is interested in the object in the training sample;

[0135] The training unit 112 is used to perform supervised training on the object recommendation model according to the training data set, wherein the object recommendation model includes a multi-task network for scoring objects, the multi-task network includes a scoring network and a network branch connected to the scoring network, there are multiple network branches, and different network branches correspond to different regions.

[0136] In one embodiment of the present application, the training unit 112 is specifically used to: in the object recommendation model, determine the feature vector corresponding to the object in the training sample based on the attribute information of the user and the attribute information of the object in the training sample; in the object recommendation model, based on the feature vector, score the objects in the training sample through the scoring network in the multi-task network and the network branch connected to the scoring network to obtain the target score of the object in the training sample; determine the loss value of the object recommendation model based on the label and the target score of the training sample; and adjust the model parameters of the object recommendation model based on the loss value.

[0137] In one embodiment of the present application, the training unit 112 is specifically used to: adjust parameters of a scoring network in the multi-task network and a target network branch in the multi-task network corresponding to the target region according to the loss value.

[0138] The model training device provided in the embodiment of the present application is used to execute the technical solution in any method embodiment of the above-mentioned model training method. Its implementation principle and technical effects are similar and will not be repeated here.

[0139] Among them, the technical solution provided in the embodiment of the present application can be implemented on a cloud server.

[0140] Fig.12 A schematic diagram of the structure of a cloud server provided for an exemplary embodiment of the present application. The cloud server is used to run an object recommendation method, and is used to score candidate objects based on user attribute information and candidate object attribute information through an object recommendation model to obtain a target score for the candidate object, and recommend a target object from the candidate objects to the user based on the target score for the candidate object; and / or, the cloud server is used to run a model training method to train an object recommendation model. Fig.12 As shown, the cloud server includes: a memory 123 and a processor 124 .

[0141] The memory 123 is used to store computer programs and can be configured to store various other data to support operations on the cloud server. The memory 123 can be an object storage service (OSS).

[0142] Memory 123 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0143] The processor 124 is coupled to the memory 123 and is used to execute the computer program in the memory 123 to execute the object recommendation method and / or model training method provided in any of the above embodiments.

[0144] Further, if Fig.12 As shown, the cloud server also includes: a firewall 121, a load balancer 122, a communication component 125, a power supply component 126 and other components. Fig.12 The following only shows some components, which does not mean that the cloud server only includes Fig.12 Components shown.

[0145] Accordingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, which, when the computer program / instructions are executed by a processor, causes the processor to implement the steps in the above method embodiment.

[0146] Accordingly, an embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is caused to implement the steps in the above method embodiment.

[0147] Above Fig.12 The communication component 125 is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component 125 is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component 125 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 125 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0148] Above Fig.12 The power supply component 126 in the device provides power to various components of the device where the power supply component 126 is located. The power supply component 126 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device where the power supply component 126 is located.

[0149] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0150] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0151] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0152] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0153] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0154] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0155] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0156] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0157] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. An object recommendation method, characterized in that: include: In the object recommendation model, a feature vector corresponding to the candidate object is determined according to the attribute information of the user and the attribute information of the candidate object, wherein the attribute information of the user includes the target region to which the user belongs, and the attribute information of the candidate object includes the statistical data corresponding to the candidate object in multiple regions; In the object recommendation model, based on the feature vector, the candidate object is scored through a scoring network in a multi-task network and a network branch connected to the scoring network to obtain a target score for the candidate object, wherein in the multi-task network, there are multiple network branches, and different network branches correspond to different regions; According to the target scores of the candidate objects, target objects among the candidate objects are recommended to the user.

2. The object recommendation method according to claim 1, characterized in that: In the object recommendation model, based on the feature vector, the candidate object is scored by a scoring network in a multi-task network and a network branch connected to the scoring network to obtain a target score for the candidate object, including: Inputting the feature vector into the scoring network, scoring the candidate object based on the feature vector in the scoring network, and obtaining a score output by the scoring network; The score output by the scoring network is input into the network branch, and the score output by the scoring network is optimized in the network branch to obtain the target score.

3. The object recommendation method according to claim 2, characterized in that: The scoring network is multiple, and different network branches share all the scoring networks. The scores output by the scoring network are input into the network branches, and the scores output by the scoring network are optimized in the network branches to obtain the target scores, including: For each network branch, the scores output by all scoring networks are input into the network branch, and the scores of all scoring networks are integrated in the network branch to obtain the score output by the network branch; The target score is obtained according to the score output by the network branch.

4. The object recommendation method according to claim 3, characterized in that: The multi-task network further includes a mask network, and obtaining the target score according to the score output by the network branch includes: Initializing the mask network according to the target area; In the multi-task network, the scores output by all network branches are input into the initialized mask network, and the scores output by the target network branches corresponding to the target area are screened out through the initialized mask network; The target score is determined according to the score output by the target network branch.

5. The object recommendation method according to any one of claims 1 to 4, characterized in that: There are multiple multi-task networks, and different multi-task networks correspond to different business objectives. In the object recommendation model, based on the feature vector, the candidate object is scored through a scoring network in the multi-task network and a network branch connected to the scoring network to obtain a target score for the candidate object, including: In the object recommendation model, the feature vector is input into each multi-task network, and based on the feature vector, the candidate object is scored by a scoring network in the multi-task network and a network branch in the multi-task network connected to the scoring network, so as to obtain a target score of the candidate object under the business target corresponding to the multi-task network; According to the target scores of the candidate objects under the business goals corresponding to each multi-task network, a final target score of the candidate objects is determined.

6. The object recommendation method according to claim 5, characterized in that: The business objectives include click-through rate and conversion rate, the statistical data include the number of users who click and the number of users who complete transactions, the multi-task network includes a first multi-task network corresponding to the click-through rate and a second multi-task network corresponding to the conversion rate, and in the object recommendation model, the feature vector is input into each multi-task network, and based on the feature vector, the candidate object is scored by a scoring network in the multi-task network and a network branch in the multi-task network connected to the scoring network to obtain a target score of the candidate object under the business objectives corresponding to the multi-task network, including: Inputting the feature vector into the first multi-task network, scoring the candidate object through a scoring network in the first multi-task network and a network branch in the first multi-task network connected to the scoring network, to obtain a target score of the candidate object under a click-through rate; The feature vector is input into the second multi-task network, and the candidate object is scored through a scoring network in the second multi-task network and a network branch in the second multi-task network connected to the scoring network to obtain a target score of the candidate object at a conversion rate.

7. A model training method, characterized in that: include: Determine a training data set, wherein the training data set includes a plurality of training samples and labels of the training samples, wherein the training samples include attribute information of a user and attribute information of an object, wherein the attribute information of the user includes a target region to which the user belongs, and the attribute information of the object includes statistical data corresponding to the object in a plurality of regions, and the labels of the training samples are used to mark whether the user is interested in the object in the training sample; According to the training data set, supervised training is performed on the object recommendation model, wherein the object recommendation model includes a multi-task network for scoring objects, the multi-task network includes a scoring network and a network branch connected to the scoring network, the network branches are multiple, and different network branches correspond to different regions.

8. The model training method according to claim 7, characterized in that: The step of performing supervised training on the object recommendation model according to the training data set includes: In the object recommendation model, determining a feature vector corresponding to the object in the training sample according to the attribute information of the user and the attribute information of the object in the training sample; In the object recommendation model, based on the feature vector, the objects in the training sample are scored through a scoring network in the multi-task network and a network branch connected to the scoring network to obtain a target score for the objects in the training sample; Determining a loss value of the object recommendation model according to the label of the training sample and the target score; According to the loss value, the model parameters of the object recommendation model are adjusted.

9. The model training method according to claim 8, characterized in that: The adjusting the model parameters of the object recommendation model according to the loss value includes: According to the loss value, parameters of a scoring network in the multi-task network and a target network branch in the multi-task network corresponding to the target region are adjusted.

10. An object recommendation device, characterized in that: include: a feature determination unit, configured to determine, in an object recommendation model, a feature vector corresponding to the candidate object according to attribute information of the user and attribute information of the candidate object, wherein the attribute information of the user includes a target region to which the user belongs, and the attribute information of the candidate object includes statistical data corresponding to the candidate object in multiple regions; an object scoring unit, configured to score the candidate object in the object recommendation model based on the feature vector through a scoring network in a multi-task network and a network branch connected to the scoring network to obtain a target score for the candidate object, wherein in the multi-task network, there are multiple network branches, and different network branches correspond to different regions; The object recommendation unit is used to recommend a target object among the candidate objects to the user according to the target scores of the candidate objects.

11. A model training device, characterized in that: include: a data determination unit, configured to determine a training data set, wherein the training data set includes a plurality of training samples and labels of the training samples, wherein the training samples include attribute information of a user and attribute information of an object, wherein the attribute information of the user includes a target region to which the user belongs, and the attribute information of the object includes statistical data corresponding to the object in a plurality of regions, and wherein the labels of the training samples are used to mark whether the user is interested in the object in the training sample; A training unit is used to perform supervised training on an object recommendation model based on the training data set, wherein the object recommendation model includes a multi-task network for scoring objects, the multi-task network includes a scoring network and a network branch connected to the scoring network, the network branches are multiple, and different network branches correspond to different regions.

12. A cloud server, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the cloud server to execute the object recommendation method described in any one of claims 1 to 6, and / or to enable the cloud server to execute the model training method described in any one of claims 7 to 9.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the object recommendation method described in any one of claims 1 to 6, and / or, when the computer program is executed by a processor, it implements the model training method described in any one of claims 7 to 9.

Images