Resource recommendation method and device, electronic device, and storage medium

Abstract

The application provides a resource recommendation method and device, electronic equipment and storage medium. The method comprises the following steps: generating a plurality of candidate sequences based on first feature information related to a user and second feature information of a plurality of candidate resources; for at least one target resource in each candidate sequence, determining an estimated exposure probability and an estimated click rate of the target resource based on corresponding second feature information, corresponding resource display position, second feature information and corresponding resource display position of at least one other target resource in the candidate sequence, and the first feature information; determining a target sequence from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate, and recommending resources according to the corresponding resource display position of each target resource. The plurality of target resources with higher overall matching degree to the user can be obtained, the overall accuracy and effectiveness of the recommended plurality of target resources are improved, and the service quality and user experience of the network resource platform are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information recommendation and artificial intelligence, and particularly relates to a resource recommendation method and device, electronic equipment and a storage medium. BACKGROUND

[0002] With the development of Internet technology, the current e-commerce platform, video playing platform or news platform and the like network resource platform all carry a large amount of resources, such as commodity advertisements, articles or video resources, and how to select appropriate resources from the large amount of resources to recommend to users is the core technical capability of the network resource platform.

[0003] In the related art, for each resource in the large amount of resources, the click probability of the user on the resource is estimated based on complex user features and features of the resource, and all resources are scored based on the click probability of each resource, all resources are sorted from high to low in a greedy manner according to the score, and resource recommendation is performed according to the sorting result. Since multiple resources are recommended, such as multiple resources being displayed in the same area of a page in turn, the multiple resources will affect each other, so that for the same resource, the click probability of the resource is different under different circumstances of other resources that are commonly recommended. The related art only considers the features of each resource itself and ignores the influence of other resources on the resource, which leads to inaccurate scoring of each resource, so that the overall matching degree between the multiple resources recommended by the network resource platform to the user and the user is low, and accordingly, the overall accuracy and effectiveness of the multiple resources recommended are poor, which affects the service quality and user experience of the network resource platform. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the related art to some extent.

[0005] The present application provides a resource recommendation method and device, electronic equipment and a storage medium to solve the technical problem that the overall matching degree between the multiple resources recommended by the network resource platform to the user and the user is low, the overall accuracy and effectiveness of the multiple resources recommended are poor, which affects the service quality and user experience of the network resource platform.

[0006] The first aspect embodiment of the present application provides a resource recommendation method, comprising: obtaining a resource recommendation request, wherein the request comprises first feature information related to a user and a number N of resource display positions, N being an integer greater than 1; obtaining second feature information of a plurality of candidate resources, and generating a plurality of candidate sequences based on the first feature information and each second feature information; each candidate sequence comprises N target resources corresponding to each resource display position selected from the plurality of candidate resources; for at least one target resource in each candidate sequence, determining an estimated exposure probability and an estimated click rate of the target resource based on the corresponding second feature information and the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information; determining a target sequence to be recommended from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and recommending resources according to the corresponding resource display position of each target resource in the target sequence.

[0007] The second aspect embodiment of the present application provides a resource recommendation device, comprising: a first obtaining module configured to obtain a resource recommendation request, wherein the request comprises first feature information related to a user and a number N of resource display positions, N being an integer greater than 1; a generating module configured to obtain second feature information of a plurality of candidate resources, and generate a plurality of candidate sequences based on the first feature information and each second feature information; each candidate sequence comprises N target resources corresponding to each resource display position selected from the plurality of candidate resources; a determining module configured to, for at least one target resource in each candidate sequence, determine an estimated exposure probability and an estimated click rate of the target resource based on the corresponding second feature information and the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information; and a recommendation module configured to determine a target sequence to be recommended from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and recommend resources according to the corresponding resource display position of each target resource in the target sequence.

[0008] The third aspect embodiment of the present application provides an electronic device, comprising: at least one processor; and a memory connected with the at least one processor in communication; 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 at least one processor to perform the resource recommendation method as proposed in the first aspect embodiment of the present application.

[0009] The fourth aspect of the present application provides a non-transitory computer-readable storage medium storing computer instructions, the computer instructions being used to cause the computer to perform the resource recommendation method according to the first aspect of the present application.

[0010] The fifth aspect of the present application provides a computer program product, comprising a computer program, the computer program being used to implement the resource recommendation method according to the first aspect of the present application when executed by a processor.

[0011] The above-mentioned embodiment of the present application has the following advantages or beneficial effects:

[0012] By generating a plurality of candidate sequences based on the first feature information related to the user and the second feature information of each candidate resource, for at least one target resource in the candidate sequence, the estimated exposure probability and the estimated click rate of the target resource are determined based on the second feature information of the target resource and the corresponding resource display position, the second feature information of at least one other target resource in the candidate sequence except the target resource and the corresponding resource display position, and the first feature information. Thus, for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered simultaneously to obtain more accurate estimated exposure probability and estimated click rate. Furthermore, based on the estimated exposure probability and the estimated click rate, the target sequence can be determined from the plurality of candidate sequences, which can obtain a plurality of target resources with higher matching degree to the user, improve the overall accuracy and effectiveness of the recommended plurality of target resources, and improve the service quality and user experience of the network resource platform.

[0013] Additional aspects and advantages of the present application will be made apparent from the following description, which, taken in conjunction with the accompanying drawings, which illustrate by way of example the principles of the application. BRIEF DESCRIPTION OF DRAWINGS

[0014] The above-mentioned and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0015] Figure 1 A flowchart of a resource recommendation method provided by an embodiment of the present application;

[0016] Figure 2 A flowchart of a resource recommendation method provided by another embodiment of the present application;

[0017] Figure 3 An architecture diagram of a sequence generation model provided by an embodiment of the present application;

[0018] Figure 4 A flowchart of a resource recommendation method provided by another embodiment of the present application;

[0019] Figure 5 An architecture diagram of an exposure probability estimation model provided by an embodiment of the present application;

[0020] Figure 6 An architecture diagram of a click rate estimation model provided by an embodiment of the present application;

[0021] Figure 7 A flowchart of a resource recommendation method provided by another embodiment of the present application;

[0022] Figure 8 A structural diagram of a resource recommendation device provided by an embodiment of the present application;

[0023] Figure 9 A structural diagram of a resource recommendation device provided by another embodiment of the present application;

[0024] Figure 10 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present application is shown. DETAILED DESCRIPTION

[0025] Embodiments of the present application are described in detail below with reference to the attached drawings, which show by way of example, embodiments in which the same or similar elements have the same or similar reference numbers throughout the several views. The embodiments described below are examples intended to provide an explanation of the present application and are not intended in any way to restrict the present application.

[0026] It should be noted that in the technical solutions of the present application, the acquisition, storage and application of user personal information involved in the technical solutions comply with relevant laws and regulations and do not violate public order and good customs.

[0027] The matching degree of the overall recommended resources of a network resource platform to a user is low, and the accuracy and effectiveness of the overall recommended resources are poor, which affects the service quality and user experience of the network resource platform. Embodiments of the present application provide a resource recommendation method, device, electronic equipment, storage medium and computer program product. The method comprises: obtaining a resource recommendation request, wherein the request comprises first feature information related to the user and the number N of resource display positions, N is an integer greater than 1; obtaining second feature information of a plurality of candidate resources, and generating a plurality of candidate sequences based on the first feature information and each second feature information; each candidate sequence comprises N target resources corresponding to each resource display position selected from the plurality of candidate resources; for at least one target resource in each candidate sequence, determining the estimated exposure probability and the estimated click rate of the target resource based on the corresponding second feature information and the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information; determining the target sequence to be recommended from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and recommending resources according to the resource display position corresponding to each target resource in the target sequence.

[0028] Therefore, by generating a plurality of candidate sequences based on the first feature information related to the user and the second feature information of each candidate resource, for at least one target resource in the candidate sequence, the estimated exposure probability and the estimated click rate of the target resource are determined based on the second feature information and the corresponding resource display position of the target resource, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information. Therefore, for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered simultaneously to obtain more accurate estimated exposure probability and estimated click rate. Furthermore, based on the estimated exposure probability and the estimated click rate, the target sequence can be determined from the plurality of candidate sequences, and the plurality of target resources with higher matching degree to the user can be obtained, thereby improving the accuracy and effectiveness of the overall recommended target resources, and improving the service quality and user experience of the network resource platform.

[0029] The resource recommendation method, device, electronic equipment, storage medium and computer program product of the embodiments of the present application are described below with reference to the accompanying drawings.

[0030] First, the resource recommendation method provided by the embodiments of the present application is described.

[0031] Figure 1 The flowchart of the resource recommendation method provided by an embodiment of the present application is shown in FIG. 1. Figure 1As shown, the resource recommendation method can include steps 101-104.

[0032] In step 101, a resource recommendation request is obtained, and the request includes first feature information related to a user and a number N of resource display positions, where N is an integer greater than 1.

[0033] It should be noted that the resource recommendation method provided in the embodiments of the present application can be executed by a resource recommendation device. The resource recommendation device can be an electronic device or can be configured in an electronic device, so as to obtain a plurality of target resources with higher overall matching degree with the user by executing the method, improve the overall accuracy and effectiveness of the recommended target resources, and improve the service quality and user experience of the network resource platform. The electronic device can be a personal computer (PC), a cloud device, a mobile device, a server, etc., and the present application does not make any limitation in this regard. For example, the resource recommendation device can be a network resource platform such as an e-commerce platform, a video playing platform, or a news platform installed in an electronic device, or a server corresponding to the network resource platform. The embodiments of the present application take the server corresponding to the network resource platform as an example for description.

[0034] In an embodiment of the present application, after the user starts a network resource platform such as an e-commerce platform, the user will first enter the homepage of the e-commerce platform, and resource recommendation can be performed on the homepage. Therefore, the resource recommendation request can be triggered when the network resource platform is started, and the resource recommendation device can obtain the resource recommendation request accordingly. Alternatively, when the user switches from another page of the e-commerce platform to the homepage, the switching operation of the user can also trigger the resource recommendation request.

[0035] The resource can be a product advertisement, an article, a video, etc., and the present application does not make any limitation in this regard.

[0036] The first feature information can include attribute information of the user, such as the age, gender, click behavior in a period of time, whether the user is a high-income group, etc. In addition, the first feature information can also include scene information related to the scene in which the user triggers the resource recommendation request, such as the time, location, and page from which the user enters the homepage of the network resource platform, etc. The present application does not make any limitation on the information included in the first feature information. Moreover, the first feature information is generally provided by the user and is authorized to perform the related resource recommendation business.

[0037] It can be understood that, in order to be able to recommend more resources, the network resource platform usually splits the same area (which can be referred to as a frame) in the home page or other pages for displaying resources into multiple frames, each frame corresponding to display a resource, and each frame displays the corresponding resource in turn according to a certain order. The frame can be understood as a display position in the same area in the page which can display the corresponding resource in turn.

[0038] The resource display position in the embodiment of the present application, i.e. the position for displaying resources, can be a certain area in the home page or other pages of the network resource platform for displaying resources, or a certain frame in a certain area, and the present application does not limit this. Correspondingly, the N resource display positions can be N areas in the home page or other pages of the network resource platform for displaying resources, or N frames in the same area, or other N positions, and the present application does not limit this.

[0039] The N resource display positions can be sequentially arranged according to a preset arrangement mode, such as sequentially arranged according to the display time, or sequentially arranged according to the position relationship, or sequentially arranged according to other arrangement modes, and the present application does not limit this.

[0040] In step 102, the second feature information of the multiple candidate resources is obtained, and based on the first feature information and each second feature information, multiple candidate sequences are generated.

[0041] Each candidate sequence includes N target resources corresponding to each resource display position selected from the multiple candidate resources.

[0042] The second feature information can include attribute information of the candidate resource, such as bid information, a resource type to which the candidate resource belongs, exposure amount in a period of time, click amount in a period of time, and user information for delivery. In the case of a commodity advertisement as the candidate resource, the resource type to which the commodity advertisement belongs can include a charging type such as cpc (Cost Per Click, charging mode according to clicks), cpm (Cost Per Mille, charging mode according to exposure), cpd (cost per day, charging mode according to day-level package), and commodity type such as brand, category, such as home appliance type, clothing type, and food type; and the user for delivery is the advertiser who delivers the commodity advertisement.

[0043] In an embodiment of the present application, N target resources can be selected from the multiple candidate resources based on the first feature information and each second feature information, each target resource corresponding to a resource display position, so as to generate a candidate sequence based on the N target resources. By selecting multiple times in this way, multiple candidate sequences can be generated.

[0044] In step 103, for at least one target resource in each candidate sequence, the estimated exposure probability and the estimated click rate of the target resource are determined based on the corresponding second feature information, the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information.

[0045] Specifically, after the plurality of candidate sequences are obtained, for each candidate sequence, the estimated exposure probability and the estimated click rate of each target resource in the candidate sequence can be determined. The estimated exposure probability is the estimated probability of exposure of the target resource. The estimated click rate is the ratio of the estimated number of clicks to the number of displays.

[0046] In an embodiment of the present application, for each target resource in each candidate sequence, the estimated exposure probability and the estimated click rate of the target resource can be determined based on the second feature information corresponding to the target resource, the resource display position corresponding to the target resource, the second feature information of all other target resources in the candidate sequence except the target resource, the resource display positions corresponding to all other target resources, and the first feature information. That is, for each target resource in each candidate sequence, the estimated exposure probability and the estimated click rate of the target resource can be determined based on the second feature information corresponding to all target resources in the candidate sequence and the resource display positions corresponding to all target resources, and the first feature information.

[0047] For example, it is assumed that the N resource display positions include four resource display positions a, b, c, and d. A candidate sequence includes four target resources A, B, C, and D, wherein A corresponds to a, B corresponds to b, C corresponds to c, and D corresponds to d. In an embodiment of the present application, for any one of A, B, C, and D, the estimated exposure probability and the estimated click rate of the any one target resource can be determined based on the second feature information of A, B, C, and D and the resource display positions corresponding to A, B, C, and D, and the first feature information.

[0048] In an embodiment of the present application, for any one of the partial target resources in each candidate sequence, the estimated exposure probability and the estimated click rate of the target resource can be determined based on the second feature information corresponding to the target resource, the resource display position corresponding to the target resource, the second feature information of at least one other target resource in the candidate sequence except the target resource, the resource display position corresponding to the at least one other target resource, and the first feature information; for any one of another partial target resource in each candidate sequence, the estimated exposure probability and the estimated click rate of the target resource can be determined based only on the second feature information corresponding to the target resource, the resource display position corresponding to the target resource, and the first feature information, or a preset value is determined as the estimated exposure probability and the estimated click rate of the target resource.

[0049] Continuing the above example, assuming that a, b, c, and d are four frame positions arranged in the order of display time, and the corresponding resources are displayed in the order of a, then b, then c, then d, and then a again. In an embodiment of the present application, since a displays the corresponding resource first, for the target resource A corresponding to a, the estimated exposure probability of A can be determined as 1, and the estimated click rate of A can be determined based on the second feature information corresponding to A and the first feature information; since b displays the corresponding resource after a, for the target resource B corresponding to b, the estimated exposure probability and the estimated click rate of B can be determined based on the second feature information of B, the second feature information of the target resource A corresponding to a, and the first feature information; since c displays the corresponding resource after a and b, for the target resource C corresponding to c, the estimated exposure probability and the estimated click rate of C can be determined based on the second feature information of C, the second feature information of the target resource A corresponding to a, the second feature information of the target resource B corresponding to b, and the first feature information; since d displays the corresponding resource after a, b, and c, for the target resource D corresponding to d, the estimated exposure probability and the estimated click rate of D can be determined based on the second feature information of D, the second feature information of the target resource A corresponding to a, the second feature information of the target resource B corresponding to b, the second feature information of the target resource C corresponding to c, and the first feature information.

[0050] In step 104, the target sequence to be recommended is determined from the multiple candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and resource recommendation is performed according to the resource display position corresponding to each target resource in the target sequence.

[0051] In an embodiment of the present application, for each candidate sequence, the candidate sequence can be scored based on the estimated exposure probability and the estimated click rate of each target resource in the candidate sequence, and then the target sequence to be recommended can be determined from the plurality of candidate sequences according to the scores corresponding to the candidate sequences, and then each target resource in the target sequence can be recommended according to the resource display position corresponding to each target resource in the target sequence.

[0052] The resource recommendation method provided by the embodiments of the present application generates a plurality of candidate sequences based on the first feature information related to the user and the second feature information of each candidate resource, determines the estimated exposure probability and the estimated click rate of at least one target resource in the candidate sequence based on the second feature information of the target resource, the corresponding resource display position, the second feature information of at least one other target resource in the candidate sequence except the target resource, the corresponding resource display position, and the first feature information, so that for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered at the same time to obtain more accurate estimated exposure probability and estimated click rate, and then the target sequence can be determined from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate, so that the plurality of target resources with higher matching degree with the user as a whole can be obtained, the overall accuracy and effectiveness of the recommended plurality of target resources are improved, and the service quality and user experience of the network resource platform are improved.

[0053] The resource recommendation method provided by the embodiments of the present application will be further described below. Figure 2 The process of generating a plurality of candidate sequences in the resource recommendation method provided by the embodiments of the present application will be further described.

[0054] Figure 2 The flowchart of the resource recommendation method provided by another embodiment of the present application is shown in FIG. 3. As shown in FIG. 3, the resource recommendation method can include the following steps 301-305. Figure 2

[0055] Step 301, a resource recommendation request is obtained, and the request includes first feature information related to a user and the number N of resource display positions, N is an integer greater than 1.

[0056] The specific implementation process and principles of step 301 can be referred to the description of the above embodiments, which will not be described herein.

[0057] Step 302, for each candidate resource, a sequence generation model is used to obtain the confidence of the candidate resource displayed at each resource display position based on the first feature information and a plurality of second feature information.

[0058] The confidence of a certain candidate resource displayed at a certain resource display position represents the possibility of the candidate resource displayed at the resource display position. ​

[0059] It can be understood that there are many possibilities for selecting N candidate resources from a plurality of candidate resources to form a sequence, such as selecting 4 candidate resources from 60 candidate resources to form a sequence, and there are A 4 60 (about 100 million) possibilities. If the combination value of each sequence in such a huge full permutation sequence space is estimated, it will seriously exceed the bearing range of the system performance. Therefore, in the embodiments of the present application, a sequence generation model can be used to model the mapping relationship between the plurality of candidate resources and the candidate sequence, so as to reduce the performance consumption of the system.

[0060] The sequence generation model is a neural network model used to generate a plurality of candidate sequences.

[0061] In an embodiment of the present application, with reference to Figure 3 The sequence generation model can include a first embedding layer 301, a first deep neural network layer 302, and a normalization layer 303 connected in sequence. The first deep neural network layer 302 is a deep neural network that considers context information.

[0062] Correspondingly, step 202 can be implemented in the following manner: based on the first embedding layer 301, a first feature vector corresponding to the first feature information and a second feature vector corresponding to each of the second feature information are obtained, and based on each second feature vector, a global feature vector corresponding to all candidate resources is obtained; for each candidate resource, the first feature vector, the global feature vector, and the corresponding second feature vector are spliced to obtain a corresponding first splicing vector; the first splicing vector corresponding to each candidate resource is input into the first deep neural network layer 302 to obtain a first initial confidence of each candidate resource displayed at each resource display position; each first initial confidence is input into the normalization layer 303 to obtain a confidence corresponding to each candidate resource.

[0063] The first feature information can include attribute information of the user and scene information related to the scene in which the user triggers the resource recommendation request.

[0064] In a possible implementation form, the sequence generation model can further include a feature extraction layer connected with the first embedding layer 301. The first feature information including the attribute information of the user and the scene information can be input into the feature extraction layer for feature extraction, and the extracted feature vector is input into the first embedding layer 301, so as to be mapped by the first embedding layer 301 to obtain the first feature vector corresponding to the first feature information. Similarly, the second feature information of each candidate resource can be input into the feature extraction layer for feature extraction, and the extracted feature vector is input into the first embedding layer 301, so as to be mapped by the first embedding layer 301 to obtain the second feature vector corresponding to each of the second feature information of each candidate resource.

[0065] Wherein, the sum of the second feature vectors corresponding to each candidate resource can be determined as the global feature vector corresponding to all candidate resources, or the second feature vectors corresponding to each candidate resource can be spliced to obtain the global feature vector corresponding to all candidate resources, or the global feature vector corresponding to all candidate resources can be obtained by other ways, which are not limited in the present application.

[0066] In step 203, a plurality of target resources corresponding to each resource display position are selected from a plurality of candidate resources by a preset selection manner based on the confidence of each candidate resource displayed at each resource display position, and a plurality of candidate sequences are generated based on the plurality of target resources corresponding to each of the N resource display positions.

[0067] Wherein, each candidate sequence includes N target resources corresponding to each resource display position selected from a plurality of candidate resources.

[0068] Wherein, the preset selection manner can be a Monte Carlo sampling manner, or an importance sampling manner, or other selection manners, which are not limited in the present application.

[0069] In an embodiment of the present application, one target resource corresponding to each of the N resource display positions can be selected from a plurality of candidate resources by a Monte Carlo sampling manner based on the confidence of each candidate resource displayed at each resource display position, and one candidate sequence is generated based on the one target resource corresponding to each of the N resource display positions. A plurality of candidate sequences can be generated by selecting in the same way for multiple times.

[0070] Since the order of each candidate resource is not set, the second feature information of each candidate resource is directly input into the sequence generation model, so that the plurality of optimal combined candidate sequences most matched with the user can be determined from the full permutation sequence space of N candidate resources in the plurality of candidate resources by enumerating the permutations of N candidate resources in the plurality of candidate resources, and by determining the confidence of each candidate resource displayed at each resource display position based on the global feature vector corresponding to all candidate resources in the process of determining the plurality of candidate sequences, the characteristics of other candidate resources can be fully considered for each candidate resource, and the accuracy of the plurality of determined candidate sequences can be improved.

[0071] It should be noted that the way of generating a plurality of candidate sequences by the sequence generation model shown in steps 202 and 203 is only an example, and in actual application, a plurality of candidate sequences can also be generated by other ways, such as some heuristic generation strategies, such as beam search strategy, which are not limited in the present application.

[0072] In step 204, for at least one target resource in each candidate sequence, based on the corresponding second feature information and the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information, the estimated exposure probability and the estimated click rate of the target resource are determined.

[0073] In step 205, based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, the target sequence to be recommended is determined from the plurality of candidate sequences, and resource recommendation is performed according to the corresponding resource display position of each target resource in the target sequence.

[0074] The specific implementation process and principles of steps 204-205 can refer to the description of the above embodiments, which will not be repeated here.

[0075] In summary, the resource recommendation method provided in the embodiments of the present application, for each candidate resource, obtains the confidence of the display of the candidate resource at each resource display position based on the first feature information and the plurality of second feature information through the sequence generation model, selects the plurality of target resources corresponding to each resource display position from the plurality of candidate resources based on the confidence of the display of each candidate resource at each resource display position through the preset selection manner, and generates a plurality of candidate sequences based on the plurality of target resources corresponding to the N resource display positions respectively, which can improve the accuracy of the plurality of candidate sequences determined. For at least one target resource in the candidate sequence, the estimated exposure probability and the estimated click rate of the target resource are determined based on the second feature information of the target resource and the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information, so that for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered at the same time to obtain more accurate estimated exposure probability and estimated click rate, and then the target sequence is determined from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate, so that the plurality of target resources with higher matching degree with the user can be obtained, the overall accuracy and effectiveness of the recommended plurality of target resources are improved, and the service quality and user experience of the network resource platform are improved.

[0076] In a possible implementation form, the N resource display positions can be sequentially arranged, such as a plurality of frame positions can be sequentially arranged according to the display time, and the corresponding resources are displayed in turn. Since in addition to the click rate of the resource, whether the resource is exposed is also a very important information, which will affect the score of each candidate sequence, and thus affect the accuracy of the finally determined target sequence, therefore, in the embodiments of the present application, the exposure probability and the click rate are taken as the basis for scoring the candidate sequence. The following will be described in combination with Figure 4The process of determining the estimated exposure probability and the estimated click rate of each candidate resource in each candidate sequence in the resource recommendation method provided by the embodiment of the application is further described.

[0077] Figure 4 The flowchart of the resource recommendation method provided by another embodiment of the application is shown in FIG. 4. As shown in FIG. 4, the resource recommendation method can include the following steps 401-408. Figure 4

[0078] In step 401, a resource recommendation request is obtained, and the request includes first feature information related to a user and the number N of resource display positions, the N resource display positions are sequentially arranged, and N is an integer greater than 1.

[0079] The N resource display positions can be sequentially arranged according to a preset arrangement mode, for example, sequentially arranged according to the display time, sequentially arranged according to the position relationship, or sequentially arranged according to another arrangement mode, which is not limited in the application.

[0080] In step 402, second feature information of a plurality of candidate resources is obtained, and a plurality of candidate sequences are generated based on the first feature information and the second feature information.

[0081] Each candidate sequence includes N target resources corresponding to one resource display position selected from the plurality of candidate resources.

[0082] The specific implementation process and principles of steps 401-402 can be referred to the description of the above embodiment, which is not described herein.

[0083] In step 403, for the first resource in each candidate sequence, the corresponding active exposure probability is determined as a first preset value.

[0084] It can be understood that for the scenario of sequentially arranging N resource display positions, for example, sequentially arranging a plurality of frames according to the display time, and sequentially displaying the corresponding resources, in the case of no user intervention, the N resource display positions can sequentially display the corresponding resources based on a preset time interval; in the case of user intervention, the corresponding resource can be displayed from a certain resource display position based on the user's sliding behavior, and the corresponding resource can be displayed from another resource display position. Based on this, the estimated exposure probability can be subdivided into active exposure probability and passive exposure probability in the embodiment of the application. The active exposure probability is the exposure probability related to the user's active behavior, and the passive exposure probability is the exposure probability unrelated to the user's active behavior. The user's active behavior can be understood as the user's operation on the resource being displayed, such as sliding or clicking, etc.

[0085] ​The first resource is a target resource in the candidate sequence corresponding to the resource display position arranged in the first place. The first preset value can be set according to needs, for example, can be set to 1 or other values.

[0086] In an embodiment of the present application, the first resource can be set to be necessarily displayed, so that the active exposure probability corresponding to the first resource can be determined as 1.

[0087] In step 404, for the second resource in each candidate sequence, the active exposure probability of the second resource is determined by the exposure probability estimation model based on the corresponding second feature information, the corresponding resource display position, the second feature information of the third resource in the candidate sequence and the corresponding resource display position, and the first feature information.

[0088] The second resource is a target resource in the candidate sequence other than the first resource. The third resource is a target resource in the candidate sequence arranged in the second resource in the corresponding resource display position.

[0089] In an embodiment of the present application, since the active exposure probability is related to the user's active behavior, and the user's active behavior is related to the characteristics of the resource and the characteristics of the user, and for each second resource, the third resource in the candidate sequence corresponding to the resource display position arranged before the second resource will have an impact on the active exposure probability of the second resource, and the target resource in the candidate sequence corresponding to the resource display position arranged after the second resource has less impact on the active exposure probability of the second resource, for each second resource, the active exposure probability of the second resource can be determined by the exposure probability estimation model based on the second feature information corresponding to the second resource, the resource display position corresponding to the second resource, the second feature information of the third resource in the candidate sequence, the resource display position corresponding to the third resource, and the first feature information.

[0090] The exposure probability estimation model is a neural network model for estimating the estimated exposure probability of each candidate resource in the candidate sequence.

[0091] In an embodiment of the present application, referring to Figure 5 The exposure probability estimation model can include a second embedding layer 501, a third embedding layer 502, a splicing layer 503 connected with the second embedding layer 501 and the third embedding layer 502, and a second deep neural network layer 504 connected with the splicing layer 503.

[0092] Correspondingly, for the second resource in each candidate sequence, the active exposure probability of the second resource can be determined in the following manner, that is, step 404 can include: obtaining a third feature vector and a fourth feature vector based on the second embedding layer 501, and obtaining a fifth feature vector corresponding to the first feature information based on the third embedding layer 502; inputting the third feature vector, the fourth feature vector and the fifth feature vector into the splicing layer 503 to obtain a corresponding second splicing vector; inputting the second splicing vector into the second deep neural network layer 504 to obtain the active exposure probability P1 of the second resource.

[0093] Among them, the third feature vector is obtained based on the second feature information of the second resource and the corresponding resource display position; the fourth feature vector is obtained based on the second feature information of the third resource and the corresponding resource display position.

[0094] In an embodiment of the present application, the exposure probability estimation model can also include a feature extraction layer connected with the second embedding layer 501 and the third embedding layer 502. The second feature information of the second resource and the corresponding resource display position, the second feature information of the third resource and the corresponding resource display position can be input into the feature extraction layer for feature extraction, and the extracted feature vectors are input into the second embedding layer 501, so as to obtain the third feature vector corresponding to the second resource and the fourth feature vector corresponding to the third resource through dimension reduction mapping of the second embedding layer 501. Moreover, the first feature information including the attribute information of the user and the scene information can be input into the feature extraction layer for feature extraction, and the extracted feature vectors are input into the third embedding layer 502, so as to obtain the fifth feature vector corresponding to the first feature information through dimension reduction mapping of the third embedding layer 502.

[0095] In an embodiment of the present application, steps 403 and 404 can also be replaced by the following manner: for each candidate sequence, the eighth feature vector corresponding to each target resource is obtained based on the second embedding layer 501, and the fifth feature vector corresponding to the first feature information is obtained based on the third embedding layer 502; for each target resource in each candidate sequence, the eighth feature vectors and the fifth feature vector are input into the splicing layer 503 to splice the eighth feature vector corresponding to the target resource, the eighth feature vectors corresponding to other target resources and the fifth feature vector based on the first weight corresponding to the target resource, the second weight corresponding to other target resources in the candidate sequence and the third weight corresponding to the fifth feature vector, to obtain the third splicing vector corresponding to the target resource, and the third splicing vector corresponding to the target resource is input into the second deep neural network layer 504 to obtain the active exposure probability of the target resource. Among them, the eighth feature vector corresponding to each target resource is obtained based on the second feature information of the corresponding target resource and the corresponding resource display position.

[0096] Thus, for each target resource in each candidate sequence, the characteristics of other target resources in the candidate sequence can be fully considered, and a more accurate active exposure probability can be obtained.

[0097] In an embodiment of the present application, the exposure probability estimation model further comprises a third embedding layer 502 connected to the first embedding layer 501 and the second embedding layer 502. Figure 5 , the exposure probability estimation model further comprises a third deep neural network layer 505 connected to the third embedding layer 502. Accordingly, the passive exposure probability of each target resource in each candidate sequence can be obtained in the following manner: for the first resource in each candidate sequence, the corresponding passive exposure probability is determined as a second preset value; for the second resource in each candidate sequence, the fifth feature vector is input into the third deep neural network layer to obtain the passive exposure probability P2 corresponding to the second resource.

[0098] The second preset value can be set as needed, such as 1 or other values.

[0099] In an embodiment of the present application, the first resource can be set as necessarily displayed, so that the passive exposure probability corresponding to the first resource can be determined as 1.

[0100] In an embodiment of the present application, since the passive exposure probability is irrelevant to the characteristics of the resource, for the second resource in each candidate sequence, the fifth feature vector can be input into the third deep neural network layer, so that the passive exposure probability of the second resource is determined only according to the first feature information of the user.

[0101] Thus, for a scenario in which N resource display positions are arranged in sequence, such as a scenario in which multiple frames are arranged in sequence according to display time and the corresponding resources are displayed in turn, by subdividing the estimated exposure probability into active exposure probability and passive exposure probability, and modeling in a multi-task form that simultaneously determines the active exposure probability and the passive exposure probability, the exposure type information can be introduced to enrich the modeling process, improve the estimation accuracy of the estimated exposure probability of each target resource in the candidate sequence, and further improve the accuracy of subsequent scoring of each candidate sequence and the accuracy of the finally determined target sequence.

[0102] In step 405, for each candidate sequence, a sixth feature vector corresponding to each target resource is obtained based on the fourth embedding layer in the click rate estimation model, and a seventh feature vector corresponding to the first feature information is obtained based on the fifth embedding layer in the click rate estimation model.

[0103] The sixth feature vector corresponding to each target resource is obtained based on the corresponding second feature information and the corresponding resource display position.

[0104] The click rate estimation model is a list-wise model, and is a neural network model for estimating the click rate of each target resource in a candidate sequence. An input space of the click rate estimation model is the features of a candidate sequence and the features of a user.

[0105] In an embodiment of the present application, referring to Figure 6 The click rate estimation model can include a fourth embedding layer 601 and a fifth embedding layer 602.

[0106] In an embodiment of the present application, the click rate estimation model can further include a feature extraction layer connected to the fourth embedding layer 601. The second feature information of each target resource in the candidate sequence and the corresponding resource display position can be input into the feature extraction layer for feature extraction, and the extracted feature vector can be input into the fourth embedding layer 601, so as to obtain the sixth feature vector corresponding to each target resource through dimension reduction mapping of the fourth embedding layer 601. In addition, the click rate estimation model can further include a feature extraction layer connected to the fifth embedding layer 602. The first feature information can be input into the feature extraction layer for feature extraction, and the extracted feature vector can be input into the fifth embedding layer 602, so as to obtain the seventh feature vector corresponding to the first feature information through dimension reduction mapping of the fifth embedding layer 602.

[0107] In step 406, the sixth feature vector is input into the self-attention layer connected to the fourth embedding layer in the click rate estimation model, so as to obtain the fusion feature vector corresponding to each target resource based on the self-attention mechanism.

[0108] Referring to Figure 6 The click rate estimation model can further include a self-attention layer 603 connected to the fourth embedding layer 601. For at least one target resource in each candidate sequence, the self-attention layer 603 can obtain the fusion feature vector corresponding to the target resource based on the self-attention mechanism and in combination with the plurality of sixth feature vectors.

[0109] For example, for each target resource in each candidate sequence, the self-attention layer 603 can determine a fusion feature vector corresponding to the target resource based on the self-attention mechanism and the sixth feature vectors of all target resources in the sequence and the preset weights of the target resources. Alternatively, for a first resource in each candidate sequence, the self-attention layer 603 can determine a fusion feature vector corresponding to the first resource based on the sixth feature vector of the first resource based on the self-attention mechanism, and for a second resource in each candidate sequence, the self-attention layer 603 can determine a fusion feature vector corresponding to the second resource based on the sixth feature vector of the second resource and the sixth feature vector of a third resource corresponding to the resource display position in the candidate sequence that is arranged before the second resource based on the self-attention mechanism. Alternatively, the self-attention layer 603 can determine the fusion feature vector corresponding to each target resource in each candidate sequence in other manners, which are not limited in the present application.

[0110] In step 407, the fusion feature vectors and the seventh feature vector are input into a fourth deep neural network layer connected to the self-attention layer and the fifth embedding layer in the click rate estimation model to obtain the estimated click rates of the target resources in the candidate sequence.

[0111] Reference Figure 6 The click rate estimation model can further include a fourth deep neural network layer 604 connected to the self-attention layer 603 and the fifth embedding layer 602. For example, when N is 4, the four fusion feature vectors and the seventh feature vector are input into the fourth deep neural network layer 604 to obtain the estimated click rates pctr1, pctr2, pctr3, and pctr4 of the four target resources in the candidate sequence.

[0112] Since for at least one target resource in the candidate sequence, the self-attention layer in the click rate estimation model can combine multiple sixth feature vectors to obtain a fusion feature vector corresponding to the target resource, and then determine the estimated click rate of each target resource based on the fusion feature vectors corresponding to the target resources, the characteristics of other target resources in the candidate sequence can be fully considered for at least one target resource in the candidate sequence, and a more accurate click rate can be obtained.

[0113] In addition, considering that resources belonging to different resource types have certain differences and connections in actual scenarios, for example, cpc charging type commodity advertisements are usually specific commodities, and cpm charging type commodity advertisements are usually commodity aggregation pages and activity pages, in the present embodiment, a multi-task network structure can be added to the fourth deep neural network layer to predict the estimated click rates of target resources of multiple resource types, so that the estimation accuracy of the estimated click rates of the target resources can be improved in the case of including target resources of multiple resource types in the candidate sequence.

[0114] Step 408, determining the target sequence to be recommended from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and performing resource recommendation according to the resource display position corresponding to each target resource in the target sequence.

[0115] The resource recommendation method provided by the embodiments of the present application generates a plurality of candidate sequences based on the first feature information related to the user and the second feature information of each candidate resource. For at least one target resource in the candidate sequence, the estimated exposure probability and the estimated click rate of the target resource are determined based on the second feature information of the target resource and the corresponding resource display position, the second feature information of at least one other target resource in the candidate sequence except the target resource and the corresponding resource display position, and the first feature information. Thus, for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered simultaneously to obtain more accurate estimated exposure probability and estimated click rate. Then, the target sequence is determined from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate, and a plurality of target resources with higher matching degree to the user as a whole can be obtained, the overall accuracy and effectiveness of the recommended plurality of target resources are improved, and the service quality and user experience of the network resource platform are improved.

[0116] The resource recommendation method provided by the embodiments of the present application will be further described below. Figure 7 The process of determining the target sequence to be recommended from the plurality of candidate sequences in the resource recommendation method provided by the embodiments of the present application will be further described.

[0117] Figure 7 The flowchart of the resource recommendation method provided by another embodiment of the present application is shown. As shown in the figure, the resource recommendation method can include the following steps 701-707. Figure 7

[0118] Step 701, obtaining a resource recommendation request, the request including first feature information related to the user and the number N of resource display positions, N being an integer greater than 1.

[0119] Step 702, obtaining second feature information of a plurality of candidate resources, and generating a plurality of candidate sequences based on the first feature information and each second feature information.

[0120] Each candidate sequence includes N target resources corresponding to one resource display position selected from the plurality of candidate resources.

[0121] ​In step 703, for at least one target resource in each candidate sequence, based on the corresponding second feature information, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information, the estimated exposure probability and the estimated click rate of the target resource are determined.

[0122] The specific implementation process and principles of steps 701-703 can refer to the description of the above embodiments, which will not be repeated here.

[0123] In step 704, for each candidate sequence, the similarity between target resources is determined based on the second feature information of the target resources.

[0124] In step 705, the score corresponding to the candidate sequence is determined by combining the estimated exposure probability, the estimated click rate, the bid information, the resource type to which the target resource belongs, and the similarity between the target resources of each target resource in a preset manner.

[0125] The second feature information of each target resource can include the bid information of the corresponding target resource and the resource type to which the target resource belongs.

[0126] In an embodiment of the present application, for each candidate sequence, the score corresponding to the candidate sequence can be determined in the following formula:

[0127]

[0128] Wherein, pos represents the resource display position; t represents the resource type to which the target resource in the candidate sequence belongs; bid price represents the bid information of the target resource in the candidate sequence; pctr represents the estimated click rate of a certain target resource in the candidate sequence, a t represents the power; boost t represents the weight corresponding to the tth resource type, which can be set as needed, and by setting the weight, the control of resource recommendation of different resource types can be easily realized; imp_prob pos represents the estimated exposure probability of a certain target resource in the candidate sequence, which includes the sum of the active exposure probability and the passive exposure probability; W ecpm and W pctr respectively represent the weights of the consumption and click targets (i.e. the weights corresponding to the first two terms in the above formula), which are the Pareto optimal weights solved by the Pareto optimizer; sim(ad pos_i ,ad pos_j ) represents the similarity between two target resources in the candidate sequence with the corresponding resource display positions pos_i and pos_j.

[0129] At step 706, the target sequence is determined from the plurality of candidate sequences based on the scores corresponding to the plurality of candidate sequences.

[0130] In this embodiment, the candidate sequence with the highest score among the plurality of candidate sequences can be determined as the target sequence.

[0131] It can be understood that when there are multiple target resources that are too similar in a candidate sequence, the multiple resources will seriously affect the browsing experience of the user if recommended to the user. In this embodiment, the score corresponding to the candidate sequence is determined in the manner shown in the above formula, which can penalize the candidate sequence in which multiple target resources that are too similar exist in terms of scores, thereby reducing the possibility of multiple target resources that are too similar existing in the finally determined target sequence and improving the browsing experience of the user. Compared with the manner of deduplication based on a preset rule in the related art, such as when two adjacent product advertisements belong to the same category or the product names corresponding to the two product advertisements are the same, one of the product advertisements is removed, the manner in this embodiment is more flexible and does not have the problem that the deduplication rule is too strict for one user and too loose for another user, which can greatly improve the matching degree of the target sequence and the user and improve the user experience. In addition, by adopting the above manner, the candidate sequence is scored based on the estimated exposure probability, the estimated click-through rate, the bid information, the resource type to which each target resource belongs, and the similarity between the target resources in the candidate sequence, which can also improve the revenue of the network resource platform in recommending resources, consider the browsing experience of the user as well as the revenue of the network resource platform, achieve a balance between the user experience and the revenue, and thereby improve the traffic distribution efficiency of the network resource platform.

[0132] At step 707, the resources are recommended according to the resource display positions corresponding to the target resources in the target sequence.

[0133] In an embodiment of the present application, the sequence generation model, the click-through rate estimation model, and the exposure probability estimation model are trained in the following manner when the plurality of candidate sequences are generated based on the sequence generation model, the estimated click-through rate of each target resource in the candidate sequence is determined based on the click-through rate estimation model, and the estimated exposure probability of each target resource in the candidate sequence is determined based on the exposure probability estimation model in the above embodiment.

[0134] The first sample feature information and the plurality of sample sequences related to each of the plurality of sample users are obtained, each sample sequence includes N sample resources corresponding to each resource display position randomly sampled, and the second sample feature information of each sample resource is obtained; each sample resource is labeled with a corresponding sample exposure probability and a sample sampling rate;

[0135] The initial click rate estimation model and the initial exposure probability estimation model are trained based on the plurality of first sample feature information, the plurality of second sample feature information of the sample resources, the corresponding resource display positions, the corresponding sample exposure probabilities, and the sample sampling rates, to obtain a trained click rate estimation model and a trained exposure probability estimation model.

[0136] The initial sequence generation model is trained based on the plurality of first sample feature information and the plurality of second sample feature information, in combination with the trained click rate estimation model and the trained exposure probability estimation model, to obtain a trained sequence generation model.

[0137] The first sample feature information can include attribute information of the sample user, such as age, gender, click behavior in a period of time, whether the sample user is a high-income group, and any other attribute information, and can include scene information related to a scene in which the sample user triggers a resource recommendation request, such as the time, location, and page from which the sample user enters the homepage of the network resource platform, and any other scene information.

[0138] The second sample feature information can include attribute information of the sample resource, such as bid information, a resource type to which the sample resource belongs, exposure volume in a period of time, click volume in a period of time, and information about a user to which the sample resource is distributed.

[0139] Specifically, the sample exposure probability and the sample sampling rate of each sample resource in each sample sequence can be used as supervision, and the initial click rate estimation model and the initial exposure probability estimation model can be trained based on the plurality of first sample feature information, the plurality of second sample feature information, and the corresponding resource display positions, to obtain a trained click rate estimation model and a trained exposure probability estimation model. The initial sequence generation model can be trained based on the plurality of first sample feature information and the plurality of second sample feature information. In the training of the initial sequence generation model, the trained click rate estimation model can be used to determine the estimated click rate of each resource in a sequence generated by the initial sequence generation model, and the trained exposure probability estimation model can be used to determine the estimated exposure probability of each resource in the sequence generated by the initial sequence generation model. Based on the estimated click rate and the estimated exposure probability, the bid information, the resource type to which each resource in the sequence generated by the initial sequence generation model belongs, and the similarity between two resources, the sequence generated by the initial sequence generation model is scored, and the score is used as supervision to train the initial sequence generation model.

[0140] By training the sequence generation model, the exposure probability estimation model and the click rate estimation model in the foregoing manner, the prediction results of the trained exposure probability estimation model and the click rate estimation model can be fed back to the sequence generation model, thereby reducing the possibility that the sequence generated by the sequence generation model contains too similar resources, making the sequence generated by the sequence generation model more matched with the user, and improving the prediction accuracy of the sequence generation model.

[0141] Figure 8 A structural diagram of a resource recommendation device provided by an embodiment of the present application is shown in the figure.

[0142] As shown in the figure, the resource recommendation device 800 can include a first acquisition module 801, a generation module 802, a determination module 803 and a recommendation module 804. Figure 8

[0143] It should be noted that the resource recommendation device provided by the embodiments of the present application can execute the resource recommendation method of the foregoing embodiments. The resource recommendation device can be an electronic device or can be configured in an electronic device, so as to obtain a plurality of target resources with higher overall matching degree with the user by executing the method, improve the overall accuracy and effectiveness of the recommended target resources, and improve the service quality and user experience of the network resource platform. The electronic device can be a PC, a cloud device, a mobile device, a server, etc., which is not limited in the present application.

[0144] The first acquisition module 801 is configured to acquire a resource recommendation request, wherein the request includes first feature information related to a user and a number N of resource display positions, and N is an integer greater than 1.

[0145] The generation module 802 is configured to acquire second feature information of a plurality of candidate resources, and generate a plurality of candidate sequences based on the first feature information and the second feature information.

[0146] The determination module 803 is configured to, for at least one target resource in each candidate sequence, determine an estimated exposure probability and an estimated click rate of the target resource based on the corresponding second feature information, the corresponding resource display position, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information.

[0147] The recommendation module 804 is configured to determine a target sequence to be recommended from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and recommend resources according to the corresponding resource display positions of the target resources in the target sequence.

[0148] ​It should be noted that the explanation of the foregoing embodiment of the resource recommendation method is also applicable to the resource recommendation device of this embodiment, which will not be described here.

[0149] The resource recommendation device of the embodiment of the present application generates a plurality of candidate sequences based on the first feature information related to the user and the second feature information of each candidate resource. For at least one target resource in the candidate sequence, the second feature information of the target resource and the corresponding resource display position, the second feature information of at least one other target resource in the candidate sequence except the target resource and the corresponding resource display position, and the first feature information are used to determine the estimated exposure probability and the estimated click rate of the target resource. Thus, for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered to obtain more accurate estimated exposure probability and estimated click rate. Then, based on the estimated exposure probability and the estimated click rate, the target sequence is determined from the plurality of candidate sequences, and the plurality of target resources with higher overall matching degree with the user can be obtained, the overall accuracy and effectiveness of the recommended plurality of target resources are improved, and the service quality and user experience of the network resource platform are improved.

[0150] Figure 9 The structure diagram of the resource recommendation device provided by another embodiment of the present application.

[0151] As Figure 9 shown, the resource recommendation device 900 can include a first acquisition module 901, a generation module 902, a determination module 903, and a recommendation module 904. Among them, Figure 9 The first acquisition module 901, the generation module 902, the determination module 903, and the recommendation module 904 in the Figure 8 have the same functions and structures as the first acquisition module 801, the generation module 802, the determination module 803, and the recommendation module 804 in the

[0152] It should be noted that the resource recommendation device 900 provided by the embodiment of the present application can execute the resource recommendation method of the foregoing embodiment. The resource recommendation device can be an electronic device or can be configured in an electronic device. By executing the method, the plurality of target resources with higher overall matching degree with the user can be obtained, the overall accuracy and effectiveness of the recommended plurality of target resources are improved, and the service quality and user experience of the network resource platform are improved. The electronic device can be a PC, a cloud device, a mobile device, a server, etc., which is not limited in the present application.

[0153] In a possible implementation manner of the embodiment of the present application, the generation module 902 includes:

[0154] The first obtaining unit 9021 is configured to, for each candidate resource, obtain a confidence degree of display of the candidate resource at each resource display position based on the first feature information and the plurality of second feature information by using a sequence generation model.

[0155] The generating unit 9022 is configured to select, by using a preset selection manner, a plurality of target resources corresponding to each resource display position from the plurality of candidate resources based on the confidence degrees of display of the candidate resources at the resource display positions, and generate a plurality of candidate sequences based on the plurality of target resources corresponding to the N resource display positions.

[0156] In another possible implementation manner of the embodiment of the application, the sequence generation model comprises a first embedding layer, a first deep neural network layer and a normalization layer connected in sequence; the first obtaining unit 9021 is configured to:

[0157] obtain a first feature vector corresponding to the first feature information and a second feature vector corresponding to each second feature information based on the first embedding layer, and obtain a global feature vector corresponding to all candidate resources based on the second feature vectors;

[0158] for each candidate resource, splice the first feature vector, the global feature vector and the corresponding second feature vector to obtain a corresponding first spliced vector;

[0159] input the first spliced vector corresponding to each candidate resource into the first deep neural network layer to obtain a first initial confidence degree of display of each candidate resource at each resource display position;

[0160] input the first initial confidence degrees into the normalization layer to obtain the confidence degrees corresponding to the candidate resources.

[0161] In another possible implementation manner of the embodiment of the application, the estimated exposure probability comprises an active exposure probability related to a user's active behavior, and the N resource display positions are arranged in sequence; the determining module 903 comprises:

[0162] The first determining unit 9031 is configured to, for a first resource in each candidate sequence, determine the corresponding active exposure probability as a first preset value; the first resource is a target resource corresponding to a resource display position arranged in the first position in the candidate sequence.

[0163] The second determining unit 9032 is configured to determine, for the second resource in each candidate sequence, an active exposure probability of the second resource by using the exposure probability estimation model based on the corresponding second feature information, the corresponding resource display position, the second feature information of the third resource in the candidate sequence and the corresponding resource display position, and the first feature information; the second resource is a target resource other than the first resource in the candidate sequence; and the third resource is a target resource arranged before the second resource in the corresponding resource display position in the candidate sequence.

[0164] In another possible implementation of the embodiments of the present application, the exposure probability estimation model comprises a second embedding layer, a third embedding layer, a concatenation layer connected with the second embedding layer and the third embedding layer, and a second deep neural network layer connected with the concatenation layer.

[0165] The second determining unit 9032 is configured to:

[0166] The third feature vector and the fourth feature vector are obtained based on the second embedding layer, and the fifth feature vector corresponding to the first feature information is obtained based on the third embedding layer; the third feature vector is obtained based on the second feature information of the second resource and the corresponding resource display position; and the fourth feature vector is obtained based on the second feature information of the third resource and the corresponding resource display position.

[0167] The third feature vector, the fourth feature vector and the fifth feature vector are input into the concatenation layer to obtain a corresponding second concatenation vector.

[0168] The second concatenation vector is input into the second deep neural network layer to obtain the active exposure probability of the second resource.

[0169] In another possible implementation of the embodiments of the present application, the estimated exposure probability further comprises a passive exposure probability irrelevant to the active behavior of the user; and the exposure probability estimation model further comprises a third deep neural network layer connected with the third embedding layer.

[0170] The determining module 903 further comprises:

[0171] The third determining unit 9033 is configured to determine, for the first resource in each candidate sequence, the corresponding passive exposure probability as a second preset value.

[0172] The second obtaining unit 9034 is configured to input, for the second resource in each candidate sequence, the fifth feature vector into the third deep neural network layer to obtain the corresponding passive exposure probability.

[0173] In another possible implementation of the embodiments of the present application, the determining module 903 comprises:

[0174] The third obtaining unit 9035 is configured to, for each candidate sequence, obtain a sixth feature vector corresponding to each target resource based on a fourth embedding layer in the click rate estimation model, and obtain a seventh feature vector corresponding to the first feature information based on a fifth embedding layer in the click rate estimation model; the sixth feature vector corresponding to each target resource is obtained based on corresponding second feature information and a corresponding resource display position;

[0175] The fourth obtaining unit 9036 is configured to input each sixth feature vector into a self-attention layer connected with the fourth embedding layer in the click rate estimation model, to obtain a fusion feature vector corresponding to each target resource based on a self-attention mechanism;

[0176] The fifth obtaining unit 9037 is configured to input each fusion feature vector and the seventh feature vector into a fourth deep neural network layer connected with the self-attention layer and the fifth embedding layer in the click rate estimation model, to obtain an estimated click rate of each target resource in the candidate sequence.

[0177] In another possible implementation manner of the embodiment of the application, the second feature information includes bid information of the corresponding target resource and a resource type to which the corresponding target resource belongs; the recommendation module 904 includes:

[0178] The fourth determining unit is configured to, for each candidate sequence, determine a similarity between target resources based on second feature information of the target resources;

[0179] The fifth determining unit is configured to determine a score corresponding to a candidate sequence by combining, in a preset manner, an estimated exposure probability, an estimated click rate, bid information, a resource type to which each target resource belongs, and a similarity between the target resources of each target resource;

[0180] The sixth determining unit is configured to determine a target sequence from the multiple candidate sequences based on scores corresponding to the multiple candidate sequences.

[0181] In another possible implementation manner of the embodiment of the application, the resource recommendation apparatus 900 further includes:

[0182] The second obtaining module is configured to obtain first sample feature information and multiple sample sequences related to each sample user, each sample sequence including N sample resources corresponding to each resource display position obtained by random sampling, and second sample feature information of each sample resource; each sample resource is labeled with a corresponding sample exposure probability and a sample sampling rate;

[0183] The first training module is configured to train the initial click rate estimation model and the initial exposure probability estimation model based on the plurality of first sample feature information, the plurality of second sample feature information of the plurality of sample resources, the corresponding resource display positions, the corresponding sample exposure probabilities, and the sample sampling rate, to obtain the trained click rate estimation model and the trained exposure probability estimation model.

[0184] The second training module is configured to train the initial sequence generation model based on the plurality of first sample feature information and the plurality of second sample feature information, in combination with the trained click rate estimation model and the trained exposure probability estimation model, to obtain the trained sequence generation model.

[0185] The resource recommendation apparatus of the embodiments of the present application generates a plurality of candidate sequences based on the first feature information related to the user and the second feature information of each candidate resource. For at least one target resource in the candidate sequence, the estimated exposure probability and the estimated click rate of the target resource are determined based on the second feature information of the target resource and the corresponding resource display position, the second feature information of at least one other target resource in the candidate sequence except the target resource and the corresponding resource display position, and the first feature information. Therefore, for at least one target resource in the candidate sequence, the characteristics of the target resource and the characteristics of other target resources in the candidate sequence can be considered simultaneously to obtain more accurate estimated exposure probability and estimated click rate. Furthermore, based on the estimated exposure probability and the estimated click rate, the target sequence is determined from the plurality of candidate sequences, which can obtain a plurality of target resources with higher matching degree to the user as a whole, improve the overall accuracy and effectiveness of the recommended plurality of target resources, and improve the service quality and user experience of the network resource platform.

[0186] To implement the above-mentioned embodiments, the present application further provides an electronic device, comprising: at least one processor; and a memory communicatively connected with 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 at least one processor to perform the resource recommendation method according to any one of the preceding embodiments of the present application.

[0187] The electronic device can be a personal computer (PC), a cloud device, a mobile device, a server, etc. The mobile device can be any hardware device such as a mobile phone, a tablet computer, a personal digital assistant, a wearable device, a vehicle-mounted device, etc. The present application does not limit the mobile device.

[0188] To implement the above-mentioned embodiments, the present application further provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the resource recommendation method according to any one of the preceding embodiments of the present application.

[0189] To achieve the above-mentioned embodiments, the present application further provides a computer program product comprising a computer program which, when executed by a processor, implements the resource recommendation method as proposed by any of the preceding embodiments of the present application.

[0190] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application described and / or claimed in this document.

[0191] As shown in Figure 10 The electronic device 1000 can include a computing unit 1001 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. Various programs and data required for the operation of the device 1000 can also be stored in the RAM 1003. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0192] Various components in the device 1000 are connected to the I / O interface 1005, including an input unit 1006, such as a keyboard, a mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a magnetic disk, an optical disk, etc.; and a communication unit 1009, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0193] The computing unit 1001 can be various general purpose and / or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs various methods and processes described above, such as the resource recommendation method. For example, in some embodiments, the resource recommendation method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded onto the RAM 1003 and executed by the computing unit 1001, one or more steps of the resource recommendation method described above can be performed. Alternatively, in other embodiments, the computing unit 1001 can be configured to perform the resource recommendation method by any other suitable means, such as by means of firmware.

[0194] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0195] Program code for carrying out methods of the present application can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0196] In the context of this application, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include a tangible, non-transitory memory such as a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0197] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0198] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0199] The computer system can include clients and servers. This relationship can be. The servers are generally remote from the users and can be accessed via the Internet using a communication network. The relationship can be a client-server relationship over a communications network, and as such, the aforementioned devices can be considered clients and servers. The servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are mainframe products in the cloud computing service system, and solve the defects of large management difficulty and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS").

[0200] It should be understood that the various forms of flow shown above can be reordered, added to, or have steps deleted. For example, the steps described in the present disclosure can be performed in parallel, in series, or in a different order, as long as the desired results of the technical solutions disclosed in the present application can be achieved, and the present disclosure is not limited herein.

[0201] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method of recommending a resource, characterized by, The method comprises: obtaining a resource recommendation request, the request comprising first feature information related to a user and a number N of resource display positions, N being an integer greater than 1; obtaining second feature information of a plurality of candidate resources, and generating a plurality of candidate sequences based on the first feature information and the second feature information; each candidate sequence comprising N target resources selected from the plurality of candidate resources, each target resource corresponding to a resource display position; for at least one target resource in each candidate sequence, determining an estimated exposure probability and an estimated click rate of the target resource based on the corresponding second feature information, the second feature information and the corresponding resource display position of at least one other target resource in the candidate sequence other than the target resource, and the first feature information; determining a target sequence to be recommended from the plurality of candidate sequences based on the estimated exposure probability and the estimated click rate of each target resource in each candidate sequence, and recommending resources according to the corresponding resource display position of each target resource in the target sequence.

2. The method of claim 1, wherein, The method comprises: for each candidate resource, obtaining a confidence degree of display of the candidate resource at each resource display position based on the first feature information and the plurality of second feature information through a sequence generation model; selecting a plurality of target resources corresponding to each resource display position from the plurality of candidate resources based on the confidence degree of display of each candidate resource at each resource display position through a preset selection manner, and generating a plurality of candidate sequences based on the plurality of target resources corresponding to the N resource display positions.

3. The method of claim 2, wherein, The sequence generation model comprises a first embedding layer, a first deep neural network layer and a normalization layer connected in sequence; and the method comprises: obtaining a first feature vector corresponding to the first feature information and a second feature vector corresponding to each second feature information based on the first embedding layer, and obtaining a global feature vector corresponding to all candidate resources based on each second feature vector; for each candidate resource, splicing the first feature vector, the global feature vector and the corresponding second feature vector to obtain a corresponding first spliced vector; inputting the first spliced vector corresponding to each candidate resource into the first deep neural network layer to obtain a first initial confidence degree of display of each candidate resource at each resource display position; inputting each first initial confidence degree into the normalization layer to obtain each confidence degree corresponding to each candidate resource.

4. The method according to claim 2 or 3, characterized in that, The estimated exposure probability comprises an active exposure probability related to a user's active behavior, and the N resource display positions are arranged in sequence; The method comprises: For a first resource in each of the candidate sequences, the corresponding active exposure probability is determined as a first preset value; the first resource is a target resource corresponding to a resource display position arranged in a first place in the candidate sequence; For a second resource in each of the candidate sequences, the active exposure probability of the second resource is determined by an exposure probability estimation model based on corresponding second feature information and a corresponding resource display position, second feature information of a third resource in the candidate sequence and a corresponding resource display position of the third resource, and the first feature information; the second resource is a target resource other than the first resource in the candidate sequence; the third resource is a target resource arranged before the second resource in the corresponding resource display position in the candidate sequence.

5. The method of claim 4, wherein, The exposure probability estimation model comprises a second embedding layer, a third embedding layer, a concatenation layer connected with the second embedding layer and the third embedding layer, and a second deep neural network layer connected with the concatenation layer; The determination of the active exposure probability of the second resource comprises: Based on the second embedding layer, a third feature vector and a fourth feature vector are obtained, and based on the third embedding layer, a fifth feature vector corresponding to the first feature information is obtained; the third feature vector is obtained based on the second feature information of the second resource and the corresponding resource display position; the fourth feature vector is obtained based on the second feature information of the third resource and the corresponding resource display position; The third feature vector, the fourth feature vector, and the fifth feature vector are input into the concatenation layer to obtain a corresponding second concatenation vector; The second concatenation vector is input into the second deep neural network layer to obtain the active exposure probability of the second resource.

6. The method of claim 5, wherein, The estimated exposure probability also includes a passive exposure probability unrelated to the user's active behavior; the exposure probability estimation model further comprises a third deep neural network layer connected with the third embedding layer; The method further comprises: For a first resource in each of the candidate sequences, the corresponding passive exposure probability is determined as a second preset value; For a second resource in each of the candidate sequences, the fifth feature vector is input into the third deep neural network layer to obtain the corresponding passive exposure probability.

7. The method of claim 4, wherein, The determination of the estimated exposure probability and the estimated click rate of the target resource comprises: For each of the candidate sequences, based on a fourth embedding layer in a click rate estimation model, a sixth feature vector corresponding to each of the target resources is obtained, and based on a fifth embedding layer in the click rate estimation model, a seventh feature vector corresponding to the first feature information is obtained; the sixth feature vector corresponding to each of the target resources is obtained based on the corresponding second feature information and the corresponding resource display position; Each of the sixth feature vectors is input into a self-attention layer connected with the fourth embedding layer in the click rate estimation model to obtain a fusion feature vector corresponding to each of the target resources based on a self-attention mechanism; The fusion feature vectors and the seventh feature vector are input into a fourth deep neural network layer connected with the self-attention layer and the fifth embedding layer of the click rate estimation model, so as to obtain the estimated click rates of the target resources in the candidate sequence.

8. The method of claim 1, wherein, The second feature information includes bid information and a resource type of a corresponding target resource; and the target sequence to be recommended is determined from the candidate sequences based on the estimated exposure probabilities and the estimated click rates of the target resources in the candidate sequences, including: For each candidate sequence, the similarity between the target resources is determined based on the second feature information of the target resources in the candidate sequence; The score corresponding to the candidate sequence is determined by combining the estimated exposure probability, the estimated click rate, the bid information, the resource type, and the similarity between the target resources of the target resources in a predetermined manner; The target sequence is determined from the candidate sequences based on the scores corresponding to the candidate sequences.

9. The method of claim 7, wherein, The sequence generation model, the click rate estimation model, and the exposure probability estimation model are trained in the following manner: First sample feature information and sample sequences related to a plurality of sample users are obtained, each sample sequence including N sample resources corresponding to each resource display position randomly sampled, and second sample feature information of each sample resource is obtained; each sample resource is labeled with a corresponding sample exposure probability and a sample sampling rate; Based on the first sample feature information, the second sample feature information of the sample resources, the corresponding resource display position, the corresponding sample exposure probability, and the sample sampling rate, the initial click rate estimation model and the initial exposure probability estimation model are trained respectively to obtain the trained click rate estimation model and the trained exposure probability estimation model; Based on the first sample feature information and the second sample feature information, the trained click rate estimation model and the trained exposure probability estimation model are combined to train the initial sequence generation model to obtain the trained sequence generation model.

10. A resource recommendation apparatus characterized by comprising: The device includes: A first obtaining module is configured to obtain a resource recommendation request, wherein the request includes first feature information related to a user and a number N of resource display positions, and N is an integer greater than 1; A generating module is configured to obtain second feature information of a plurality of candidate resources, and generate a plurality of candidate sequences based on the first feature information and the second feature information; each candidate sequence includes N target resources corresponding to each resource display position selected from the plurality of candidate resources. determining, for at least one target resource in each of the candidate sequences, a predicted exposure probability and a predicted click rate of the target resource based on corresponding second feature information and corresponding resource display position, second feature information and corresponding resource display position of at least one other target resource in the candidate sequence except the target resource, and the first feature information; recommending, based on the predicted exposure probability and the predicted click rate of each target resource in each candidate sequence, a target sequence to be recommended from the plurality of candidate sequences, and recommending resources in accordance with the corresponding resource display position of each target resource in the target sequence.

11. The apparatus of claim 10, wherein, The generation module comprises: A first acquisition unit is configured to, for each candidate resource, acquire, by a sequence generation model, a confidence degree of display of the candidate resource at each resource display position based on the first feature information and the plurality of second feature information. A generation unit is configured to, by a preset selection manner, select, based on the confidence degree of display of each candidate resource at each resource display position, a plurality of target resources corresponding to each resource display position from the plurality of candidate resources, and generate a plurality of candidate sequences based on the plurality of target resources corresponding to each of the N resource display positions.

12. The apparatus of claim 11, wherein, The sequence generation model comprises a first embedding layer, a first deep neural network layer and a normalization layer connected in sequence. The first acquisition unit is configured to: acquire, based on the first embedding layer, a first feature vector corresponding to the first feature information and a second feature vector corresponding to each second feature information, and acquire, based on each second feature vector, a global feature vector corresponding to all candidate resources; for each candidate resource, concatenate the first feature vector, the global feature vector and the corresponding second feature vector to obtain a corresponding first concatenated vector; input the first concatenated vector corresponding to each candidate resource into the first deep neural network layer to obtain a first initial confidence degree of display of each candidate resource at each resource display position; 13. The apparatus of claim 11 or 12, wherein, input each first initial confidence degree into the normalization layer to obtain each confidence degree corresponding to each candidate resource. The predicted exposure probability comprises an active exposure probability related to a user's active behavior, and the N resource display positions are arranged in sequence. The determining module comprises: A first determination unit is configured to, for a first resource in each candidate sequence, determine a corresponding active exposure probability as a first preset value; the first resource is a target resource corresponding to a resource display position arranged in the first position in the candidate sequence. The second determining unit is configured to determine, for a second resource in each of the candidate sequences, the active exposure probability of the second resource by using an exposure probability estimation model based on corresponding second feature information and corresponding resource display position, second feature information and corresponding resource display position of a third resource in the candidate sequence, and the first feature information; the second resource is a target resource other than the first resource in the candidate sequence; and the third resource is a target resource arranged before the second resource in the corresponding resource display position in the candidate sequence.

14. The apparatus of claim 13, wherein, The exposure probability estimation model comprises a second embedding layer, a third embedding layer, a concatenation layer connected with the second embedding layer and the third embedding layer, and a second deep neural network layer connected with the concatenation layer. The second determining unit is configured to: obtain a third feature vector and a fourth feature vector based on the second embedding layer, and obtain a fifth feature vector corresponding to the first feature information based on the third embedding layer; the third feature vector is obtained based on second feature information and corresponding resource display position of the second resource; and the fourth feature vector is obtained based on second feature information and corresponding resource display position of the third resource; input the third feature vector, the fourth feature vector, and the fifth feature vector into the concatenation layer to obtain a corresponding second concatenation vector; input the second concatenation vector into the second deep neural network layer to obtain the active exposure probability of the second resource.

15. The apparatus of claim 14, wherein, The estimated exposure probability further comprises a passive exposure probability irrelevant to the active behavior of the user; and the exposure probability estimation model further comprises a third deep neural network layer connected with the third embedding layer. The determining module further comprises: a third determining unit configured to determine, for the first resource in each of the candidate sequences, the passive exposure probability corresponding to the first resource as a second preset value; a second obtaining unit configured to input the fifth feature vector into the third deep neural network layer to obtain the passive exposure probability corresponding to the second resource in each of the candidate sequences.

16. The apparatus of claim 13, wherein, The determining module comprises: a third obtaining unit configured to obtain, for each of the candidate sequences, a sixth feature vector corresponding to each of the target resources based on a fourth embedding layer in a click rate estimation model, and obtain a seventh feature vector corresponding to the first feature information based on a fifth embedding layer in the click rate estimation model; the sixth feature vector corresponding to each of the target resources is obtained based on corresponding second feature information and corresponding resource display position; a fourth obtaining unit configured to input each of the sixth feature vectors into a self-attention layer connected with the fourth embedding layer in the click rate estimation model to obtain a fusion feature vector corresponding to each of the target resources based on a self-attention mechanism; a fifth obtaining unit configured to input each of the fusion feature vectors and the seventh feature vector into a fourth deep neural network layer connected with the self-attention layer and the fifth embedding layer in the click rate estimation model to obtain the estimated click rate of each of the target resources in the candidate sequence.

17. The apparatus of claim 10, wherein, The second feature information includes bid information of the corresponding target resource and a resource type to which the target resource belongs; and the recommendation module includes: A fourth determination unit configured to determine, for each of the candidate sequences, similarity between the target resources based on second feature information of the target resources; A fifth determination unit configured to determine, by a preset manner, a score corresponding to the candidate sequence by combining the estimated exposure probability, the estimated click rate, the bid information, the resource type to which the target resource belongs, and the similarity between the target resources of each of the target resources; A sixth determination unit configured to determine the target sequence from the candidate sequences based on the scores corresponding to the candidate sequences.

18. The apparatus of claim 16, wherein, The apparatus further includes: A second acquisition module configured to acquire first sample feature information and sample sequences related to each of a plurality of sample users, each of the sample sequences including N sample resources corresponding to each of the resource display positions and obtained by random sampling, and second sample feature information of each of the sample resources; and each of the sample resources is labeled with a corresponding sample exposure probability and a sample sampling rate; A first training module configured to train an initial click rate estimation model and an initial exposure probability estimation model based on the first sample feature information, the second sample feature information of the sample resources, the corresponding resource display positions, the corresponding sample exposure probabilities, and the sample sampling rates, to obtain a trained click rate estimation model and a trained exposure probability estimation model; A second training module configured to train an initial sequence generation model based on the first sample feature information and the second sample feature information, in combination with the trained click rate estimation model and the trained exposure probability estimation model, to obtain a trained sequence generation model.

19. An electronic device, comprising: comprise: at least one processor; and a memory connected with the at least one processor in communication; 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 at least one processor to perform the method of any one of claims 1-9.

20. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method of any one of claims 1-9. The computer instructions are used to enable the computer to perform the method of any one of claims 1-9.

Images