Recommendation model training method and display content determination method, device, and medium
By fusing hidden features of multiple advertisement screening models through a shared model, the method enhances consistency and accuracy, improving the selection of advertisements for display.
Patent Information
- Application Number
- CN202210509093.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-10
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-05-10
AI Technical Summary
Existing advertisement screening models lack consistency and accuracy due to the lack of correlation between models trained for different screening steps, leading to poor performance in selecting appropriate advertisements for display.
A method involving a shared model to fuse hidden features of multiple models with task-related relationships, enhancing consistency and accuracy by training models based on predicted and actual interaction information.
Improves the consistency and accuracy of advertisement screening models by correlating features across different screening steps, thereby increasing the conversion rate of displayed content.
Smart Images

Figure CN114819146B_ABST
Abstract
Description
Background Art
[0002] With the rapid development of the Internet, the coverage of Internet advertising is becoming wider and wider. In the prior art, for a certain advertising position, an advertisement matching the user can be screened out from multiple advertisements by an advertisement delivery engine for display. Generally speaking, the steps of screening advertisements may include: targeting, recall, rough sorting, fine sorting, etc. In each step, the advertisements involved in these steps can be screened out by a screening model to screen out the advertisements to be displayed from multiple advertisements.
[0003] However, in the prior art, the screening models are trained for different screening steps, and there is no correlation between the screening models, nor the ability to learn from each other. For example, an advertisement that can be output as a to-be-screened advertisement in the rough screening model cannot be output as a to-be-screened advertisement in the recall screening model. That is, the consistency between the screening models in the prior art is poor, and the accuracy of the advertisements to be displayed obtained by each screening model is also poor.
[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0005] The purpose of the present disclosure is to provide a recommendation model training method, a recommendation model training device, a display content determination method, a display content determination device, a computer-readable storage medium, an electronic device, and a computer program product, which can solve the problem of poor consistency between two models with task-related relationships.
[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.
[0007] According to a first aspect of the present disclosure, a recommendation model training method is provided, comprising: obtaining sample data; wherein the sample data comprises sample content, sample users, and interaction information between the sample content and the sample users; inputting the sample data into a first model to be trained, a second model to be trained, and a shared model, and obtaining hidden features of hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is screening through the first model to be trained, and then screening the screened results through the second model to be trained; the shared model is used to learn the common features of the first model to be trained and the second model to be trained; fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain first predicted interaction information of the first model to be trained, and training the first model to be trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model; fusing the hidden features of the second model to be trained with the hidden features of the shared model to obtain second predicted interaction information of the first model to be trained, and training the second model to be trained according to the second predicted interaction information and actual interaction information in the sample data to obtain a second recommendation model.
[0008] Optionally, based on the aforementioned scheme, the step of fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain the first predicted interaction information of the first model to be trained includes: fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain the first fused features; and determining the first predicted interaction information of the first model to be trained based on the first fused features.
[0009] Optionally, based on the aforementioned scheme, the step of fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain a first fused feature includes: determining the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model according to the sample data; obtaining the first fused feature according to the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model.
[0010] Optionally, based on the aforementioned scheme, the step of fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first fusion features includes: for the first model to be trained, obtaining the current fusion features of the current hidden layer of the first model to be trained based on the previous fusion features of the previous hidden layer of the first model to be trained and the previous fusion features of the shared model of the same layer; obtaining the current fusion features of the shared model based on the previous fusion features of the first model to be trained, the previous fusion features of the shared model of the same layer, and the previous fusion features of the second model to be trained; obtaining the first fusion feature based on the current fusion features of the first model to be trained and the current fusion features of the shared model.
[0011] Optionally, based on the aforementioned scheme, the first model to be trained includes a first sub-training model and a second sub-training model, and the step of fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain a first fusion feature includes: obtaining a first fusion sub-feature corresponding to the first training sub-model based on the hidden features of the first training sub-model and the hidden features of the shared model; obtaining a second fusion sub-feature corresponding to the second training sub-model based on the hidden features of the second training sub-model and the hidden features of the shared model; obtaining the first fusion feature based on the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model.
[0012] Optionally, based on the aforementioned scheme, the step of training the first model to be trained according to the first predicted interaction information and the actual interaction information in the sample data includes: determining the first loss function of the first model to be trained through the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data; updating the neural network parameters in the first model to be trained according to the first loss function of the first model to be trained to obtain a first recommendation model.
[0013] Optionally, based on the aforementioned scheme, the first model to be trained includes a first sub-training model and a second sub-training model, and the step of determining the first loss function of the first model to be trained through the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data includes: inputting the sample data into the first sub-training model to obtain the first sub-prediction interaction information, and determining the first loss sub-function of the first sub-training model through the first sub-prediction interaction information; inputting the sample data into the second sub-training model to obtain the second sub-prediction interaction information, and determining the second loss sub-function of the second sub-training model through the second sub-prediction interaction information; and determining the first loss function of the first model to be trained based on the first sub-loss function of the first training sub-model and the second sub-loss function of the second training sub-model.
[0014] Optionally, based on the aforementioned scheme, the step of updating the neural network parameters in the first model to be trained according to the first loss function of the first model to be trained to obtain the first recommendation model includes: determining the second loss function of the second model to be trained through the second prediction interaction information of the second model to be trained; determining the overall loss function according to the first loss function and the second loss function; and updating the neural network parameters in the first model to be trained according to the overall loss function to train the first model to be trained to obtain the first recommendation model.
[0015] Optionally, the step of determining the overall loss function based on the first loss function and the second loss function includes: obtaining a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function; determining the overall loss function based on the first loss weight, the first loss function, the second loss weight, and the second loss function.
[0016] Optionally, the step of obtaining a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function includes: obtaining the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function by an uncertain loss method.
[0017] According to a second aspect of the present disclosure, a method for determining display content is provided, the method comprising: acquiring a plurality of display contents and relevant information corresponding to the display contents, inputting the relevant information corresponding to the display contents into a first recommendation model to obtain a plurality of first screened display contents; wherein, the relevant information corresponding to the display contents includes the display contents, the display users, and the interaction information between the display contents and the display users, and the first recommendation model is obtained by a recommendation model training method as claimed in any one of claims 1 to 10; inputting the plurality of first screened display contents into a second recommendation model to obtain a plurality of second screened display contents; wherein, the second recommendation model is obtained by a recommendation model training method as claimed in any one of claims 1 to 10; and determining the display content to be displayed according to the plurality of second screened display contents.
[0018] According to a third aspect of the present disclosure, a recommendation model training device is provided, the device comprising: a sample data acquisition unit, configured to execute acquisition of sample data; wherein the sample data comprises sample content, sample users, and interaction information between the sample content and the sample users; a sample data input unit, configured to execute input of the sample data into a first model to be trained, a second model to be trained, and a shared model, to obtain hidden layer features of hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to perform screening through the first model to be trained, and then screen the screened results through the second model to be trained; the shared model is used to The common features of the first model to be trained and the second model to be trained are learned; the first recommendation model training unit is configured to perform a fusion process on the hidden features of the first model to be trained and the hidden features of the shared model to obtain the first predicted interaction information of the first model to be trained, and the first model to be trained is trained according to the first predicted interaction information and the actual interaction information in the sample data to obtain the first recommendation model; the second recommendation model training unit is configured to perform a fusion process on the hidden features of the second model to be trained and the hidden features of the shared model to obtain the second predicted interaction information of the first model to be trained, and the second model to be trained is trained according to the second predicted interaction information and the actual interaction information in the sample data to obtain the second recommendation model.
[0019] According to a fourth aspect of the present disclosure, a display content determination device is provided, the device comprising: a first screening content acquisition unit, configured to execute acquisition of multiple display contents and related information corresponding to the display contents, and input the related information corresponding to the display contents into a first recommendation model to obtain multiple first screened display contents; wherein, the related information corresponding to the display contents includes display contents, display users, and interaction information between the display contents and the display users, and the first recommendation model is obtained by a recommendation model training method such as any one of the above; a second screening content acquisition unit, configured to execute input of the multiple first screened display contents into a second recommendation model to obtain multiple second screened display contents; wherein, the second recommendation model is obtained by a recommendation model training method such as any one of the above; and a display content determination unit, configured to execute determination of the display contents to be displayed according to the multiple second screened display contents.
[0020] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the recommendation model training method of the first aspect and the display content determination method of the second aspect in the above-mentioned embodiments are implemented.
[0021] According to a sixth aspect of the present disclosure, there is provided an electronic device, including:
[0022] Processor; and
[0023] The memory is used to store one or more programs. When the one or more programs are executed by one or more processors, the one or more processors implement the recommendation model training method in the first aspect and the display content determination method in the second aspect in the above-mentioned embodiments.
[0024] According to a seventh aspect of the present disclosure, a computer program product is provided, comprising: when the computer program / instructions are executed by a processor, the method for training a recommendation model as described above or the method for determining display content as described in claim 11 is implemented.
[0025] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:
[0026] In a recommendation model training method provided by an embodiment of the present disclosure, sample data can be obtained, and the sample data can be input into a first model to be trained, a second model to be trained, and a shared model to obtain hidden layer features of the hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model, respectively, and the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain first predicted interaction information of the first model to be trained, the first model to be trained is trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model, the hidden layer features of the second model to be trained and the hidden layer features of the shared model are fused to obtain second predicted interaction information of the first model to be trained, the second model to be trained is trained according to the second predicted interaction information and the actual interaction information in the sample data to obtain a second recommendation model.
[0027] In the embodiments of the present disclosure, sample data can be obtained and input into the first model to be trained, the second model to be trained and the shared model to obtain the hidden features of the hidden layer outputs of the first model to be trained, the second model to be trained and the shared model, and the hidden features of the first model to be trained are fused with the hidden features of the shared model to obtain the first predicted interaction information, and the first model to be trained is trained according to the first predicted interaction information and the actual interaction information to obtain the first recommendation model, and the second recommendation model is similar. On the one hand, the features of the first model to be trained / the second model to be trained and the shared model can be fused so that the first model to be trained and the second model to be trained are correlated, thereby improving the consistency between the first model to be trained and the second model to be trained; on the other hand, since the features of the model with task-related relationships are taken into account when training the model, when screening the display content, the screening accuracy can be made higher, thereby improving the conversion rate of the display content.
[0028] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:
[0030] Figure 1 A schematic diagram schematically shows an exemplary system architecture of a recommendation model training method in an exemplary embodiment of the present disclosure;
[0031] Figure 2 A flowchart of a recommendation model training method in an exemplary embodiment of the present disclosure is schematically shown;
[0032] Figure 3 Schematically illustrating a flowchart of determining first predicted interaction information of a first to-be-trained model according to a first fusion feature in an exemplary embodiment of the present disclosure;
[0033] Figure 4 A flowchart schematically illustrates a method of obtaining a first fused feature according to a fusion weight corresponding to a hidden layer feature of a first model to be trained, a hidden layer feature of the first model to be trained, a fusion weight corresponding to a hidden layer feature of a shared model, and a hidden layer feature of a shared model in an exemplary embodiment of the present disclosure;
[0034] Figure 5 A schematic diagram schematically illustrates a gating structure in an exemplary embodiment of the present disclosure;
[0035] Figure 6 A flowchart of obtaining a first fused feature according to a current fused feature of a first model to be trained and a current fused feature of a shared model in an exemplary embodiment of the present disclosure is schematically shown;
[0036] Figure 7 A flowchart of obtaining a first fusion feature according to a first fusion sub-feature corresponding to a first training sub-model and a second fusion sub-feature corresponding to a second training sub-model in an exemplary embodiment of the present disclosure is schematically shown;
[0037] Figure 8 A flowchart schematically illustrates updating the neural network parameters in the first to-be-trained model according to the first loss function of the first to-be-trained model in an exemplary embodiment of the present disclosure to obtain a first recommendation model;
[0038] Fig. 9A flowchart schematically illustrates a method of determining a first loss function corresponding to a first to-be-trained model according to a first sub-loss function corresponding to a first training sub-model and a second sub-loss function corresponding to a second training sub-model in an exemplary embodiment of the present disclosure;
[0039] Fig.10 A flowchart schematically illustrates updating the neural network parameters in the first model to be trained according to the overall loss function in an exemplary embodiment of the present disclosure to train the first model to be trained to obtain a first recommendation model;
[0040] Fig.11 Schematically showing a flow chart of determining an overall loss function according to a first loss weight, a first loss function, a second loss weight, and a second loss function in an exemplary embodiment of the present disclosure;
[0041] Fig.12 A schematic diagram schematically illustrates joint training in an exemplary embodiment of the present disclosure;
[0042] Fig.13 A flowchart schematically illustrates determining the presentation content to be displayed according to a plurality of second filtered presentation contents in an exemplary embodiment of the present disclosure;
[0043] Fig.14 A schematic diagram schematically shows the composition of a recommendation model training device in an exemplary embodiment of the present disclosure;
[0044] Fig.15 A schematic diagram schematically shows the composition of a display content determination device in an exemplary embodiment of the present disclosure;
[0045] Fig.16 The structure diagram of a computer system suitable for implementing the electronic device of the exemplary embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION
[0046] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known structures, methods, devices, implementations, materials, or operations are not shown or described in detail to avoid blurring various aspects of the present disclosure.
[0047] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or these functional entities or parts of functional entities may be implemented in one or more software hardened modules, or these functional entities may be implemented in different networks and / or processor devices and / or microcontroller devices.
[0048] Figure 1 A schematic diagram showing an exemplary system architecture of a recommendation model training method to which an embodiment of the present disclosure can be applied.
[0049] like Figure 1 As shown, the system architecture 1000 may include one or more of terminal devices 1001, 1002, 1003, a network 1004, and a server 1005. The network 1004 is used to provide a medium for communication links between the terminal devices 1001, 1002, 1003 and the server 1005. The network 1004 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.
[0050] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 1005 may be a server cluster composed of multiple servers.
[0051] Users can use terminal devices 1001, 1002, 1003 to interact with server 1005 through network 1004 to receive or send messages, etc. Terminal devices 1001, 1002, 1003 can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, portable computers, desktop computers, etc. In addition, server 1005 can be a server that provides various services.
[0052] In one embodiment, the executor of the recommendation model training method disclosed in the present invention may be server 1005, which may obtain sample data sent by terminal devices 1001, 1002, and 1003, and input the sample data into the first model to be trained, the second model to be trained, and the shared model, to obtain hidden features of the hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model, to fuse the hidden features of the first model to be trained with the hidden features of the shared model, to obtain first predicted interaction information of the first model to be trained, to train the first model to be trained according to the first predicted interaction information and actual interaction information in the sample data, to obtain a first recommendation model, to fuse the hidden features of the second model to be trained with the hidden features of the shared model, to obtain second predicted interaction information of the first model to be trained, to train the second model to be trained according to the second predicted interaction information and actual interaction information in the sample data, to obtain a second recommendation model, thereby completing the training process of the recommendation model. In addition, the recommendation model training method of the present disclosure can also be executed through terminal devices 1001, 1002, 1003, etc. to realize the process of training the first model to be trained according to the first predicted interaction information and the actual interaction information in the sample data, and training the second model to be trained according to the second predicted interaction information and the actual interaction information in the sample data.
[0053] In addition, the implementation process of the recommendation model training method disclosed in the present invention can also be jointly implemented by terminal devices 1001, 1002, 1003 and server 1005. For example, the terminal devices 1001, 1002, and 1003 may obtain sample data, wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users, input the sample data into the first model to be trained, the second model to be trained, and the shared model, obtain the hidden layer features of the hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model, and then send the obtained hidden layer features to the server 1005, so that the server 1005 can fuse the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first predicted interaction information of the first model to be trained, train the first model to be trained according to the first predicted interaction information and the actual interaction information in the sample data, obtain the first recommendation model, fuse the hidden layer features of the second model to be trained with the hidden layer features of the shared model, obtain the second predicted interaction information of the first model to be trained, train the second model to be trained according to the second predicted interaction information and the actual interaction information in the sample data, and obtain the second recommendation model.
[0054] With the rapid development of the Internet, the coverage of Internet advertising is becoming wider and wider. In the prior art, for a certain advertising position, an advertisement matching the user can be screened out from multiple advertisements by an advertisement delivery engine for display. Generally speaking, the steps of screening advertisements may include: targeting, recall, rough sorting, fine sorting, etc. In each step, the advertisements involved in these steps can be screened out by a screening model to screen out the advertisements to be displayed from multiple advertisements.
[0055] Specifically, in the prior art, in targeted screening, advertisements can be screened by user preferences; in recall screening, advertisements can be screened by user tags, advertisement tags, user similarity and advertisement similarity, etc.; in rough screening, advertisements can be screened by combining click-through rate (CTR, Click-Through-Rate) and conversion rate (CVR, Conversion Rate) and advertising revenue per thousand impressions (eCPM, effective cost per mille), etc.; in the refined screening stage, advertisements can be screened again by combining click-through rate and conversion rate and advertising revenue per thousand impressions, etc.
[0056] However, in the prior art, the screening models are trained for different screening steps, and there is no correlation between the screening models, nor the ability to learn from each other. For example, an advertisement that can be output as a to-be-screened advertisement in the rough screening model cannot be output as a to-be-screened advertisement in the recall screening model. That is, the consistency between the screening models in the prior art is poor, and the accuracy of the advertisements to be displayed obtained by each screening model is also poor.
[0057] Specifically, during the training process of the screening model in the early step, some fine-grained information may be lost, so in the screening model in the later step, more accurate screening cannot be performed based on the above fine-grained information.
[0058] According to the method for training a recommendation model provided in this exemplary embodiment, sample data can be obtained; wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users, the sample data is input into a first model to be trained, a second model to be trained, and a shared model, hidden features of the hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model are obtained, the hidden features of the first model to be trained and the hidden features of the shared model are fused to obtain first predicted interaction information of the first model to be trained, the first model to be trained is trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model, the hidden features of the second model to be trained and the hidden features of the shared model are fused to obtain second predicted interaction information of the first model to be trained, the second model to be trained is trained according to the second predicted interaction information and the actual interaction information in the sample data to obtain a second recommendation model.
[0059] like Figure 2 As shown, the recommendation model training method may include the following steps S210 to S260:
[0060] Step S210, obtaining sample data; wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users;
[0061] Step S220, inputting sample data into the first model to be trained, the second model to be trained and the shared model, and obtaining hidden layer features of the hidden layer outputs of the first model to be trained, the second model to be trained and the shared model; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to perform screening through the first model to be trained, and then screen the screened results through the second model to be trained; the shared model is used to learn the common features of the first model to be trained and the second model to be trained;
[0062] Step S230, fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain first predicted interaction information of the first model to be trained, and training the first model to be trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model;
[0063] Step S240, fuse the hidden features of the second model to be trained with the hidden features of the shared model to obtain second predicted interaction information of the first model to be trained, train the second model to be trained according to the second predicted interaction information and actual interaction information in the sample data, and obtain a second recommendation model.
[0064] In the embodiments of the present disclosure, sample data can be obtained and input into the first model to be trained, the second model to be trained and the shared model to obtain the hidden features of the hidden layer outputs of the first model to be trained, the second model to be trained and the shared model, and the hidden features of the first model to be trained are fused with the hidden features of the shared model to obtain the first predicted interaction information, and the first model to be trained is trained according to the first predicted interaction information and the actual interaction information to obtain the first recommendation model, and the second recommendation model is similar. On the one hand, the features of the first model to be trained / the second model to be trained and the shared model can be fused so that the first model to be trained and the second model to be trained are correlated, thereby improving the consistency between the first model to be trained and the second model to be trained; on the other hand, since the features of the model with task-related relationships are taken into account when training the model, when screening the display content, the screening accuracy can be made higher, thereby improving the conversion rate of the display content.
[0065] Next, steps S210 to S260 of the joint training in this exemplary embodiment will be described in more detail with reference to the accompanying drawings and embodiments.
[0066] Step S210, obtaining sample data; wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users;
[0067] In an example embodiment of the present disclosure, sample data may be obtained. Specifically, the sample data may include advertisements, homepage recommended content, etc. Among them, the display form of the sample data may include text, pattern, video, or a combination of text, pattern, and video. It should be noted that the present disclosure does not specifically limit the display form of the sample data. Specifically, the sample data includes sample content, sample users, and the interactive information between the sample content and the sample users refers to information that can be used to indicate the characteristics of the sample data. For example, the sample users may include user attribute characteristics, such as age, gender, region, etc.; the interactive information between the sample content and the sample users may include user behavior characteristics, such as historical browsing records, etc.; the sample content may include display content text / voice information, such as text in the display content, or text in a brief description of the display content; the display content related information may also include image information, such as images in the display content.
[0068] It should be noted that the present disclosure does not impose any special limitation on the specific types of sample content, sample users, and interactive information between the sample content and sample users.
[0069] In an exemplary embodiment of the present disclosure, sample data and the sample data corresponding to the sample data, including sample content, sample users, and interactive information between the sample content and the sample users, can be obtained. Specifically, the sample data and the sample data corresponding to the sample data, including sample content, sample users, and interactive information between the sample content and the sample users, can be stored in a sample database. When it is necessary to use the solution of the present disclosure to train a recommendation model, the sample data and the sample data corresponding to the sample data, including sample content, sample users, and interactive information between the sample content and the sample users, can be obtained from the sample database. Alternatively, the sample data and the sample data corresponding to the sample data, including sample content, sample users, and interactive information between the sample content and the sample users, can be obtained in real time.
[0070] It should be noted that this public card does not specifically limit the specific method of obtaining sample data and the sample data corresponding to the sample data, including sample content, sample users, and interactive information between sample content and sample users.
[0071] In an exemplary embodiment of the present disclosure, after obtaining the sample data and the sample data corresponding to the sample data, including the sample content, the sample user, and the interactive information between the sample content and the sample user, through the above steps, the sample data and the sample data corresponding to the sample data, including the sample content, the sample user, and the interactive information between the sample content and the sample user, can be converted into a sample data feature vector. Specifically, the existing vector conversion method can be used to convert the display content related information into the sample data feature vector.
[0072] Furthermore, for different types of sample data and the sample data's corresponding information including sample content, sample users, and interaction information between sample content and sample users, different vector conversion methods can be used to convert the sample data and the sample data's corresponding information including sample content, sample users, and interaction information between sample content and sample users into sample data feature vectors.
[0073] For example, for image information, a vector conversion model based on convolutional neural networks (CNN) can be used to convert sample data and the sample data's corresponding information including sample content, sample users, and the interaction information between sample content and sample users into sample data feature vectors; for another example, for text / voice information, a vector conversion model based on a pre-trained language representation model (BERT, Bidirectional Encoder Representation from Transformers) can be used to convert sample data and the sample data's corresponding information including sample content, sample users, and the interaction information between sample content and sample users into sample data feature vectors; for another example, for interaction information, a vector conversion model based on graph neural networks (GNN, Graph Neural Networks) can be used to convert sample data and the sample data's corresponding information including sample content, sample users, and the interaction information between sample content and sample users into sample data feature vectors.
[0074] It should be noted that the present disclosure does not specifically limit the specific method of converting sample data and the sample data corresponding to the sample data, including sample content, sample users, and interaction information between the sample content and the sample users, into a sample data feature vector.
[0075] Step S220, inputting sample data into the first model to be trained, the second model to be trained and the shared model, and obtaining hidden layer features of the hidden layer outputs of the first model to be trained, the second model to be trained and the shared model; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to perform screening through the first model to be trained, and then screen the screened results through the second model to be trained; the shared model is used to learn the common features of the first model to be trained and the second model to be trained;
[0076] In an exemplary embodiment of the present disclosure, after the sample data is obtained through the above steps, the sample data can be input into the first model to be trained, the second model to be trained and the shared model. Specifically, when the first model to be trained and the second model to be trained are trained in the scheme of the present disclosure, the first model to be trained and the second model to be trained need to be jointly trained. Specifically, the display content feature vector input into the shared model is the sample data, and the sample data input into the first model to be trained or the second model to be trained is all or part of the sample data.
[0077] It should be noted that the present disclosure does not impose any special limitation on the specific sample data input into the first model to be trained and the second model to be trained.
[0078] Among them, the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to filter through the first model to be trained, and then filter the filtered results through the second model to be trained; the shared model is used to learn the common features of the first model to be trained and the second model to be trained. Specifically, the task-related relationship means that there is a correlation between the tasks to be performed by the first model to be trained and the second model to be trained. For example, the first model to be trained recommends display content based on click-through rate, and the second model to be trained recommends display content based on the duration of the display content, and the recommendation is first made through the first model to be trained, and then through the second model to be trained, that is, at this time, the first model to be trained and the second model to be trained have a task-related relationship; for another example, the first model to be trained is a recall model, and the display content is filtered through user tags, advertising tags, user similarity and advertising similarity, etc., and the second model to be trained is a rough ranking model, and the display content is filtered in combination with the click-through rate and conversion rate and the advertising revenue per thousand impressions, and the recall model is first used to filter, and then the rough ranking model is used to filter, that is, at this time, the first model to be trained and the second model to be trained have a task-related relationship.
[0079] It should be noted that the present disclosure does not specifically limit the specific form of the task correlation relationship between the first model to be trained and the second model to be trained.
[0080] In addition, the first model to be trained and the second model to be trained in the present disclosure do not consider the order of use after the training is completed. For example, through the scheme of the present disclosure, the first model to be trained can be trained as the first recommendation model, and the second model to be trained can be trained as the second recommendation model. When using the first recommendation model and the second recommendation model, the first recommendation model can be used first, and then the second recommendation model; or, the second recommendation model can be used first, and then the first recommendation model; or, the first recommendation model and the second recommendation model can be used at the same time.
[0081] It should be noted that the present disclosure does not impose any special limitation on the order of using the first recommendation model and the second recommendation model.
[0082] Step S230, fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain first predicted interaction information of the first model to be trained, and training the first model to be trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model;
[0083] In an exemplary embodiment of the present disclosure, after the sample data is input into the first model to be trained, the second model to be trained and the shared model through the above steps, the hidden layer features of the first model to be trained and the hidden layer features of the shared model can be fused to obtain the first prediction interaction information of the first model to be trained. Specifically, the sample data input into the first model to be trained can be calculated by the hidden layer in the first model to be trained to obtain the hidden layer calculation result of the first model to be trained, and then the sample data input into the shared model can be calculated by the hidden layer in the shared model to obtain the hidden layer calculation result of the shared model. Among them, the hidden layer in the first model to be trained or the hidden layer of the shared model is a hidden layer in the neural network structure, which can include a convolutional layer, a pooling layer, an excitation layer, a fully connected layer, etc.
[0084] It should be noted that the present disclosure does not specifically limit the specific structures of the hidden layer of the first model to be trained and the hidden layer of the shared model.
[0085] In an exemplary embodiment of the present disclosure, after the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused through the above steps, the first prediction interaction information of the first model to be trained can be obtained. Specifically, the first prediction interaction information can be used to indicate the prediction value corresponding to the sample data.
[0086] In an exemplary embodiment of the present disclosure, after obtaining the first predicted interaction information of the first model to be trained through the above steps, the first model to be trained can be trained according to the first predicted interaction information and the actual interaction information in the sample data. Specifically, the predicted value obtained through the above steps can be compared with the true value corresponding to the sample data, and the first model to be trained can be trained according to the comparison result.
[0087] For example, the true value corresponding to the sample data is whether a certain display content wins, that is, whether the display content will be exhibited in the end. The first model to be trained can be used to predict whether the display content wins to obtain first predicted interaction information, and the first predicted interaction information is compared with the true value to train the first model to be trained.
[0088] Step S240, fuse the hidden features of the second model to be trained with the hidden features of the shared model to obtain second predicted interaction information of the first model to be trained, train the second model to be trained according to the second predicted interaction information and actual interaction information in the sample data, and obtain a second recommendation model.
[0089] In an exemplary embodiment of the present disclosure, after the sample data is input into the first model to be trained, the second model to be trained and the shared model through the above steps, the hidden layer features of the second model to be trained and the hidden layer features of the shared model can be fused to obtain the second prediction interaction information of the second model to be trained. Specifically, the sample data input into the second model to be trained can be calculated by the hidden layer in the second model to be trained to obtain the hidden layer calculation result of the second model to be trained, and then the sample data input into the shared model can be calculated by the hidden layer in the shared model to obtain the hidden layer calculation result of the shared model. Among them, the hidden layer in the first model to be trained or the hidden layer of the shared model is a hidden layer in the neural network structure, which can include a convolution layer, a pooling layer, an excitation layer, a fully connected layer, etc.
[0090] It should be noted that the present disclosure does not specifically limit the specific structures of the hidden layer of the second model to be trained and the hidden layer of the shared model.
[0091] In an exemplary embodiment of the present disclosure, after the hidden layer features of the second model to be trained are fused with the hidden layer features of the shared model through the above steps, the second prediction interaction information of the second model to be trained can be obtained. Specifically, the second prediction interaction information can be used to indicate the prediction value corresponding to the sample data.
[0092] In an exemplary embodiment of the present disclosure, after obtaining the second predicted interaction information of the second model to be trained through the above steps, the second model to be trained can be trained according to the second predicted interaction information and the actual interaction information in the sample data. Specifically, the predicted value obtained through the above steps can be compared with the true value corresponding to the sample data, and the second model to be trained can be trained according to the comparison result.
[0093] For example, the true value corresponding to the sample data is whether a certain display content wins, that is, whether the display content will be exhibited in the end. The second model to be trained can be used to predict whether the display content wins to obtain second predicted interaction information, and the second predicted interaction information is compared with the true value to train the second model to be trained.
[0094] In an exemplary embodiment of the present disclosure, the hidden layer features of the first model to be trained and the hidden layer features of the shared model may be fused to obtain a first fused feature, and the first predicted interaction information of the first model to be trained may be determined according to the first fused feature. Figure 3 As shown, referring to determining the first predicted interaction information of the first to-be-trained model according to the first fusion feature, the following steps S310 to S320 may be included:
[0095] Step S310, fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain a first fused feature;
[0096] Step S320: determining first predicted interaction information of the first model to be trained according to the first fusion feature.
[0097] In an exemplary embodiment of the present disclosure, the hidden layer features of the first model to be trained and the hidden layer features of the shared model may be fused to obtain the first fused features. Specifically, the hidden layer calculation results of the first model to be trained and the hidden layer calculation results of the shared model may be fused to obtain the first fused features.
[0098] For example, during fusion, the hidden layer features of the first model to be trained and the hidden layer features of the shared model may be fused by dot multiplication to obtain the first fused features.
[0099] It should be noted that the present disclosure does not specifically limit the specific method of fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first fused features.
[0100] In an exemplary embodiment of the present disclosure, after obtaining the first fused feature through the above steps, the first predicted interaction information of the first model to be trained can be determined according to the first fused feature. Specifically, the first predicted interaction information of the first model to be trained can be obtained by predicting the first fused feature for the first model to be trained.
[0101] For example, the sample data is display content, and the actual interaction information in the sample data is that the display content needs to be displayed. At this time, the first fusion feature can be predicted by the first model to be trained to obtain the first predicted interaction information of the first model to be trained (predicting whether the display content needs to be displayed).
[0102] It should be noted that the present disclosure does not specifically limit the specific method of determining the first predicted interaction information of the first model to be trained according to the first fusion feature.
[0103] Through the above steps S310 to S320, the hidden layer features of the first model to be trained and the hidden layer features of the shared model can be fused to obtain the first fused features, and the first predicted interaction information of the first model to be trained can be determined based on the first fused features.
[0104] In an exemplary embodiment of the present disclosure, the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model can be determined according to the sample data, and the first fusion feature can be obtained according to the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model. Figure 4 As shown, obtaining the first fused feature according to the fusion weight corresponding to the hidden layer feature of the first model to be trained, the hidden layer feature of the first model to be trained, the fusion weight corresponding to the hidden layer feature of the shared model, and the hidden layer feature of the shared model may include the following steps S410 to S420:
[0105] Step S410, determining the fusion weights corresponding to the hidden layer features of the first model to be trained and the fusion weights corresponding to the hidden layer features of the shared model according to the sample data;
[0106] Step S420, obtaining a first fused feature according to the fusion weight corresponding to the hidden layer feature of the first model to be trained, the hidden layer feature of the first model to be trained, the fusion weight corresponding to the hidden layer feature of the shared model, and the hidden layer feature of the shared model.
[0107] In an exemplary embodiment of the present disclosure, after obtaining the hidden layer features of the hidden layer outputs of the first model to be trained, the second model to be trained and the shared model through the above steps, the fusion weights corresponding to the hidden layer features of the first model to be trained and the fusion weights corresponding to the hidden layer features of the shared model can be determined according to the sample data.
[0108] The sample data input into the first model to be trained, the second model to be trained and the shared model includes the hidden features of the first model to be trained and the hidden features of the shared model. At this time, the hidden features of the first model to be trained and the hidden features of the shared model can be fused to obtain the first fused features.
[0109] For example, the hidden layer feature of the first model to be trained is vector1, and the hidden layer feature of the shared model is vector2. At this time, the hidden layer feature of the first model to be trained and the hidden layer feature of the shared model can be fused to obtain the first fused feature, and vector1 and vector2 can be fused to obtain the first fused feature.
[0110] Specifically, when the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain the first fused features, the fusion can be performed in a dot product manner.
[0111] It should be noted that the present disclosure does not specifically limit the specific method of fusing the hidden features of the first model to be trained and the hidden features of the shared model to obtain the first fused features.
[0112] Specifically, the hidden features corresponding to the first model to be trained and the shared model can be obtained respectively through the above steps, and the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model can be obtained. Specifically, the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model can be obtained through sample data training, wherein the fusion weights corresponding to the hidden features of the first model to be trained can be used to indicate the fusion weights of the hidden features of the first model to be trained when the hidden features of the first model to be trained are fused with the shared model, and the fusion weights corresponding to the hidden features of the shared model can be used to indicate the fusion weights of the hidden features of the shared model when the hidden features of the shared model are fused with the shared model.
[0113] It should be noted that the present disclosure does not specifically limit the specific method of obtaining the fusion weights corresponding to the hidden layer features of the first model to be trained and the fusion weights corresponding to the hidden layer features of the shared model through sample data training.
[0114] In an exemplary embodiment of the present disclosure, after obtaining the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model through sample data training through the above steps, the first fusion feature can be obtained according to the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model. Specifically, the first fusion feature can be calculated by configuring the corresponding fusion weights for the hidden features of the first model to be trained and the hidden features of the shared model.
[0115] In an exemplary embodiment of the present disclosure, the hidden layer features of the first model to be trained and the hidden layer features of the shared model can be fused into the first fused features through a gate structure. Figure 5 As shown, the sample data can be input into the gated structure, and after processing, it can be output to the fully connected layer, and then the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights (w1, w2) corresponding to the hidden features of the shared model are determined by the SOFTMAX function (normalized exponential function). At this time, after obtaining the hidden features of the first model to be trained and the hidden features of the shared model (vector 1, vector 2), the first fusion feature can be obtained according to the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model.
[0116] It should be noted that the present disclosure does not specifically limit the specific method of obtaining the first fusion feature based on the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model.
[0117] Through the above steps S410 to S420, the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model can be determined according to the sample data, and the first fusion features can be obtained according to the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model.
[0118] In an exemplary embodiment of the present disclosure, for the first model to be trained, the current fusion feature of the current hidden layer of the first model to be trained can be obtained according to the previous fusion feature of the previous hidden layer of the first model to be trained and the previous fusion feature of the shared model at the same layer, the current fusion feature of the shared model can be obtained according to the previous fusion feature of the first model to be trained, the previous fusion feature of the shared model at the same layer and the previous fusion feature of the second model to be trained, and the first fusion feature can be obtained according to the current fusion feature of the first model to be trained and the current fusion feature of the shared model. Figure 6 As shown, obtaining the first fusion feature according to the current fusion feature of the first model to be trained and the current fusion feature of the shared model may include the following steps S610-S620:
[0119] Step S610, for the first model to be trained, obtaining the current fusion feature of the current hidden layer of the first model to be trained according to the previous fusion feature of the previous hidden layer of the first model to be trained and the previous fusion feature of the shared model at the same layer;
[0120] In an exemplary embodiment of the present disclosure, the current fusion features of the current hidden layer of the first model to be trained can be obtained for the first model to be trained based on the previous fusion features of the previous hidden layer of the first model to be trained and the previous fusion features of the shared model of the same layer. Specifically, the first model to be trained may include multiple hidden layers, and the previous fusion features can be calculated by the previous hidden layer of the first model to be trained. The shared model may include multiple hidden layers, and the previous fusion features can be calculated by the previous hidden layer of the shared model. The current fusion features of the current hidden layer of the first model to be trained are obtained by the previous fusion features of the previous hidden layer of the first model to be trained and the previous fusion features of the shared model of the same layer. For example, the current fusion features of the current hidden layer of the first model to be trained can be obtained by the dot product of the previous fusion features of the previous hidden layer of the first model to be trained and the previous fusion features of the shared model of the same layer.
[0121] It should be noted that the present disclosure does not specifically limit the specific method of obtaining the current fused features of the current hidden layer of the first model to be trained by using the previous fused features of the previous hidden layer of the first model to be trained and the previous fused features of the shared model at the same layer.
[0122] Step S620, obtaining current fusion features of the shared model according to the previous fusion features of the first model to be trained, the previous fusion features of the shared model at the same layer, and the previous fusion features of the second model to be trained;
[0123] In an exemplary embodiment of the present disclosure, after obtaining the current fusion feature of the current hidden layer of the first model to be trained through the above steps, the current fusion feature of the shared model can be obtained according to the previous fusion feature of the first model to be trained, the previous fusion feature of the shared model of the same layer, and the previous fusion feature of the second model to be trained. Specifically, the current fusion feature of the shared model can be obtained by determining the previous fusion feature of the first model to be trained, the previous fusion feature of the shared model of the same layer, and the previous fusion feature of the second model to be trained through the current hidden layer of the first model to be trained. Specifically, the shared model may include multiple hidden layers, and the previous fusion feature can be calculated by the previous hidden layer of the shared model, the previous fusion feature of the shared model of the same layer, and the previous fusion feature of the second model to be trained to obtain the current fusion feature of the shared model. For example, the previous fusion feature of the first model to be trained, the previous fusion feature of the shared model of the same layer, and the previous fusion feature of the second model to be trained can be fused by dot multiplication to obtain the current fusion feature of the shared model.
[0124] It should be noted that the present disclosure does not specifically limit the specific method of obtaining the current fusion features of the shared model based on the previous fusion features of the first model to be trained, the previous fusion features of the shared model at the same layer, and the previous fusion features of the second model to be trained.
[0125] Step S630, obtaining a first fusion feature according to the current fusion feature of the first model to be trained and the current fusion feature of the shared model.
[0126] In an exemplary embodiment of the present disclosure, after obtaining the current fusion features of the current hidden layer of the first model to be trained and the current fusion features of the shared model through the above steps, the first fusion feature can be obtained based on the current fusion features of the first model to be trained and the current fusion features of the shared model. Specifically, the first model to be trained includes multiple hidden layers, and the current fusion features of the current hidden layer of the first model to be trained and the current fusion features of the shared model can be fused through the current hidden layer to obtain the first fusion feature. For example, the current fusion features of the first model to be trained and the current fusion features of the shared model can be fused through dot multiplication to obtain the first fusion feature.
[0127] It should be noted that the present disclosure does not specifically limit the specific method of obtaining the first fusion feature based on the current fusion feature of the first model to be trained and the current fusion feature of the shared model.
[0128] Through the above steps S610 to S630, the current fusion features of the current hidden layer of the first model to be trained can be obtained according to the previous fusion features of the previous hidden layer of the first model to be trained and the previous fusion features of the shared model of the same layer for the first model to be trained; the current fusion features of the shared model can be obtained according to the previous fusion features of the first model to be trained, the previous fusion features of the shared model of the same layer and the previous fusion features of the second model to be trained; and the first fusion features can be obtained according to the current fusion features of the first model to be trained and the current fusion features of the shared model.
[0129] In an exemplary embodiment of the present disclosure, the first fusion sub-feature corresponding to the first training sub-model can be obtained according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model, the second fusion sub-feature corresponding to the second training sub-model can be obtained according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model, and the first fusion feature can be obtained according to the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model. Figure 7 As shown, obtaining the first fusion feature according to the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model may include the following steps S710 to S730:
[0130] Step S710, obtaining a first fusion sub-feature corresponding to the first training sub-model according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model;
[0131] Step S720, obtaining a second fusion sub-feature corresponding to the second training sub-model according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model;
[0132] In an exemplary embodiment of the present disclosure, after the hidden layer features of the first model to be trained and the hidden layer features of the shared model are obtained through the above steps, the first fused sub-features corresponding to the first training sub-model can be obtained according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model, and the second fused sub-features corresponding to the second training sub-model can be obtained according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model. The first model to be trained may include a first sub-training model and a second sub-training model.
[0133] During training, the hidden layer features of the first sub-training model can be calculated through the hidden layer of the first sub-training model, the hidden layer features of the shared model can be calculated through the hidden layer of the shared model, and the hidden layer features of the first sub-training model are fused with the hidden layer features of the shared model to obtain the first fused sub-features corresponding to the first training sub-model.
[0134] Specifically, when the hidden layer features of the first sub-training model are fused with the hidden layer features of the shared model to obtain the first fused sub-features corresponding to the first training sub-model, the hidden layer features of the first training sub-model and the hidden layer features of the shared model can be fused by point multiplication to obtain the first fused sub-features corresponding to the first training sub-model.
[0135] In an exemplary embodiment of the present disclosure, during training, the hidden layer features of the second sub-training model can be calculated through the hidden layer of the second sub-training model, and the hidden layer features of the second training sub-model are fused with the hidden layer features of the shared model to obtain the second fused sub-features corresponding to the second training sub-model.
[0136] Specifically, when the hidden layer features of the second training sub-model are fused with the hidden layer features of the shared model to obtain the second fused sub-features corresponding to the second training sub-model, the hidden layer features of the second training sub-model and the hidden layer features of the shared model can be fused by point multiplication to obtain the second fused sub-features corresponding to the second training sub-model.
[0137] The first sub-training model and the second sub-training model may include models based on network structures in the prior art. For example, the first sub-training model and the second sub-training model may be models based on a DNN (Deep Neural Networks) network structure, or the first sub-training model and the second sub-training model may be models based on an R-drop (Regularized Dropout for Neural Networks) network structure.
[0138] It should be noted that the present disclosure does not specifically limit the network structures of the first sub-training model and the second sub-training model.
[0139] Step S730, obtaining a first fused feature according to the first fused sub-feature corresponding to the first training sub-model and the second fused sub-feature corresponding to the second training sub-model.
[0140] In an exemplary embodiment of the present disclosure, after obtaining the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model through the above steps, the first fusion feature can be obtained according to the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model. Specifically, the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model can be fused by a dot multiplication method to obtain the first fusion feature.
[0141] It should be noted that the present disclosure does not specifically limit the specific method of obtaining the first fusion feature according to the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model.
[0142] Through the above steps S710 to S730, the first fused sub-feature corresponding to the first training sub-model is obtained according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model, the second fused sub-feature corresponding to the second training sub-model is obtained according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model, and the first fused feature is obtained according to the first fused sub-feature corresponding to the first training sub-model and the second fused sub-feature corresponding to the second training sub-model.
[0143] In an exemplary embodiment of the present disclosure, the first model to be trained includes a first sub-model to be trained and a second sub-model to be trained, the second model to be trained includes a first sub-model to be trained and a second sub-model to be trained, and the two sub-training models in the first model to be trained and the two sub-training models in the second model to be trained present a double-tower structure.
[0144] In an exemplary embodiment of the present disclosure, the first loss function of the first model to be trained can be determined by the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data, and the neural network parameters in the first model to be trained can be updated according to the first loss function of the first model to be trained to obtain the first recommendation model. Figure 8 As shown, updating the neural network parameters in the first to-be-trained model according to the first loss function of the first to-be-trained model to obtain the first recommendation model may include the following steps S810-S820:
[0145] Step S810, determining a first loss function of the first model to be trained by using first predicted interaction information of the first model to be trained and actual interaction information of the sample data;
[0146] Step S820: updating the neural network parameters in the first model to be trained according to the first loss function of the first model to be trained to obtain a first recommendation model.
[0147] In an exemplary embodiment of the present disclosure, after obtaining the first predicted interaction information of the first model to be trained through the above steps, the first loss function of the first model to be trained can be determined by the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data. Specifically, the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data can be compared, the predicted difference between the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data can be calculated, and the first loss function of the first model to be trained can be determined based on the predicted difference.
[0148] For example, the sample data is display content, and the actual interaction information corresponding to the sample data is the display content that needs to be displayed. At this time, the actual interaction information of the sample data can be compared with the first predicted interaction information to obtain the first loss function corresponding to the first model to be trained.
[0149] It should be noted that the present disclosure does not specifically limit the specific method of determining the first loss function of the first model to be trained by using the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data.
[0150] In an exemplary embodiment of the present disclosure, after obtaining the first loss function of the first model to be trained through the above steps, the first model to be trained can be trained through the first loss function of the first model to be trained. For example, the neural network parameters in the first model to be trained can be trained through a back propagation algorithm, and at the end of the training, a first recommendation model (obtained by training the first model to be trained) is obtained.
[0151] It should be noted that the present disclosure does not specifically limit the method of updating the neural network parameters in the first model to be trained according to the first loss function of the first model to be trained to obtain the first recommendation model.
[0152] Through the above steps S810 to S820, the first loss function of the first model to be trained can be determined by the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data, and the neural network parameters in the first model to be trained are updated according to the first loss function of the first model to be trained to obtain a first recommendation model.
[0153] In an exemplary embodiment of the present disclosure, sample data may be input into the first sub-training model to obtain first sub-prediction interaction information, and the first loss sub-function of the first sub-training model may be determined by the first sub-prediction interaction information. Sample data may be input into the second sub-training model to obtain second sub-prediction interaction information, and the second loss sub-function of the second sub-training model may be determined by the second sub-prediction interaction information. The first loss function corresponding to the first training sub-model and the second sub-loss function corresponding to the second training sub-model may be determined. Fig. 9 As shown, determining the first loss function corresponding to the first to-be-trained model according to the first sub-loss function corresponding to the first training sub-model and the second sub-loss function corresponding to the second training sub-model may include the following steps S910 to S930:
[0154] Step S910, inputting sample data into the first sub-training model to obtain first sub-prediction interaction information, and determining a first loss sub-function of the first sub-training model according to the first sub-prediction interaction information;
[0155] In an exemplary embodiment of the present disclosure, sample data may be input into the first sub-training model to obtain first sub-prediction interaction information, and the first loss sub-function of the first sub-training model may be determined by the first sub-prediction interaction information. Specifically, sample data may be input into the first sub-training model for prediction to obtain first sub-prediction interaction information, and the first sub-prediction interaction information may be compared with actual interaction information in the sample data to obtain a prediction difference between the first sub-prediction interaction information and the actual interaction information in the sample data, and the first sub-loss function corresponding to the first training sub-model may be determined according to the prediction difference.
[0156] It should be noted that the present disclosure does not specifically limit the specific method of inputting sample data into the first sub-training model to obtain the first sub-prediction interaction information and determining the first loss sub-function of the first sub-training model through the first sub-prediction interaction information.
[0157] Step S920, inputting the sample data into the second sub-training model to obtain second sub-prediction interaction information, and determining a second loss sub-function of the second sub-training model according to the second sub-prediction interaction information;
[0158] In an exemplary embodiment of the present disclosure, sample data may be input into the second sub-training model to obtain second sub-prediction interaction information, and the second loss sub-function of the second sub-training model may be determined by the second sub-prediction interaction information. Specifically, sample data may be input into the second sub-training model for prediction to obtain second sub-prediction interaction information, and the second sub-prediction interaction information may be compared with the actual interaction information in the sample data to obtain the predicted difference between the second sub-prediction interaction information and the actual interaction information in the sample data, and the second sub-loss function corresponding to the second training sub-model may be determined according to the predicted difference.
[0159] It should be noted that the present disclosure does not specifically limit the specific method of inputting sample data into the second sub-training model to obtain the second sub-prediction interaction information and determining the second loss sub-function of the second sub-training model through the second sub-prediction interaction information.
[0160] Step S930, determining the first loss function of the first model to be trained according to the first sub-loss function corresponding to the first training sub-model and the second sub-loss function of the second training sub-model.
[0161] In an exemplary embodiment of the present disclosure, after obtaining the first loss sub-function of the first sub-training model and the second loss sub-function of the second sub-training model through the above steps, the first loss function of the first model to be trained can be determined according to the first sub-loss function of the first training sub-model and the second sub-loss function of the second training sub-model. Specifically, the first loss function of the first training sub-model and the second sub-loss function of the second training sub-model can be added to obtain the first loss function of the first model to be trained.
[0162] Furthermore, loss weights can be assigned to the first sub-loss function of the first training sub-model and the second sub-loss function of the second training sub-model, respectively, and the first loss function corresponding to the first model to be trained can be obtained according to the first sub-loss function of the first training sub-model, the loss weight corresponding to the first sub-loss function, the second sub-loss function of the second training sub-model, and the loss weight corresponding to the second sub-loss function.
[0163] It should be noted that the present disclosure does not specifically limit the specific method of determining the first loss function of the first model to be trained based on the first sub-loss function corresponding to the first training sub-model and the second sub-loss function of the second training sub-model.
[0164] Through the above steps S910 to S930, the sample data can be input into the first sub-training model to obtain the first sub-prediction interaction information, the first loss sub-function of the first sub-training model can be determined through the first sub-prediction interaction information, the sample data can be input into the second sub-training model to obtain the second sub-prediction interaction information, the second loss sub-function of the second sub-training model can be determined through the second sub-prediction interaction information, and the first loss function corresponding to the first training sub-model and the second sub-model can be determined according to the first loss function corresponding to the first training sub-model and the second loss function corresponding to the second training sub-model.
[0165] In an exemplary embodiment of the present disclosure, the second loss function of the second model to be trained can be determined by the second prediction interaction information of the second model to be trained, the overall loss function can be determined according to the first loss function and the second loss function, and the neural network parameters in the first model to be trained can be updated according to the overall loss function to train the first model to be trained to obtain the first recommendation model. Fig.10 As shown, updating the neural network parameters in the first model to be trained according to the overall loss function to train the first model to be trained to obtain the first recommendation model may include the following steps S1010 to S1020:
[0166] Step S1010, determining a second loss function of the second model to be trained through second predicted interaction information of the second model to be trained;
[0167] In an exemplary embodiment of the present disclosure, the second loss function of the second model to be trained can be determined by the second predicted interaction information of the second model to be trained, the overall loss function can be determined according to the first loss function and the second loss function, and the neural network parameters in the first model to be trained can be updated according to the overall loss function to train the first model to be trained to obtain the first recommendation model. Specifically, the sample data can be input into the second model to be trained, the second predicted interaction information can be obtained by the second model to be trained, and the second predicted interaction information can be compared with the actual interaction information in the sample data to obtain the predicted difference between the second predicted interaction information and the actual interaction information in the sample data, and the second loss function of the second model to be trained can be determined according to the predicted difference.
[0168] Step S1020, determining an overall loss function according to the first loss function and the second loss function;
[0169] In an exemplary embodiment of the present disclosure, after the second loss function of the second model to be trained is obtained through the above steps, the overall loss function can be determined according to the first loss function and the second loss function. Specifically, the first loss function of the first model to be trained and the second loss function of the second model to be trained can be added to obtain the overall loss function, or the first loss function of the first model to be trained and the second loss function of the second model to be trained can be respectively assigned weights to obtain the overall loss function.
[0170] It should be noted that the present disclosure determines the overall loss function based on the first loss function and the second loss function.
[0171] Step S1030, updating the neural network parameters in the first model to be trained according to the overall loss function, so as to train the first model to be trained to obtain a first recommendation model.
[0172] In an exemplary embodiment of the present disclosure, after the overall loss function is obtained through the above steps, the overall loss function can be determined according to the first loss function and the second loss function. Specifically, the first model to be trained can be trained by the overall loss function. For example, the neural network parameters in the first model to be trained can be trained by the back propagation algorithm, and at the end of the training, the first recommendation model (obtained by training the first model to be trained) is obtained.
[0173] It should be noted that the present disclosure does not specifically limit the method of updating the neural network parameters in the first model to be trained according to the overall loss function to train the first model to be trained to obtain the first recommended model.
[0174] Through the above steps S1010 to S1030, the second loss function of the second model to be trained can be determined by the second prediction interaction information of the second model to be trained, the overall loss function can be determined according to the first loss function and the second loss function, and the neural network parameters in the first model to be trained can be updated according to the overall loss function to train the first model to be trained to obtain the first recommendation model.
[0175] In an exemplary embodiment of the present disclosure, for the first loss function and the second loss function, a cross entropy loss may be used, or a KL divergence loss function may be used, or a focal loss may be used. The focal loss function may weight samples with low confidence, thereby eliminating the negative impact of these samples on training, thereby improving the training effect. Specifically, the expression of the focal loss is as follows, where L ft is the focus loss, y′ is the output of the activation function, α is the balance factor, and γ is the weight factor:
[0176]
[0177] The training method of the second model to be trained mentioned in the present disclosure is the same as that of the first model to be trained, which will not be elaborated here.
[0178] In an exemplary embodiment of the present disclosure, a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function may be obtained, and an overall loss function may be determined according to the first loss weight, the first loss function, the second loss weight, and the second loss function. Fig.11 As shown, determining the overall loss function according to the first loss weight, the first loss function, the second loss weight, and the second loss function may include the following steps S1110 to S1120:
[0179] Step S1110, obtaining a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function;
[0180] In an exemplary embodiment of the present disclosure, after determining the first loss function corresponding to the first training model and the second loss function corresponding to the second training model through the above steps, the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function can be obtained. Specifically, the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function can be configured manually, or the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function can be determined by machine learning.
[0181] It should be noted that the present disclosure does not specifically limit the method for obtaining the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function.
[0182] Step S1120, determining an overall loss function according to the first loss weight, the first loss function, the second loss weight, and the second loss function.
[0183] In an exemplary embodiment of the present disclosure, after obtaining the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function through the above steps, the first loss weight can be configured for the first loss function, and the second loss weight can be configured for the second loss function, and the two can be added together to obtain the overall loss function.
[0184] For example, the first loss function is loss1, the second loss function is loss2, the first loss weight corresponding to the first loss function is w1, and the second loss weight corresponding to the second loss function is w2. At this time, the overall loss function is w=w1loss1·w2loss2.
[0185] It should be noted that the present disclosure does not specifically limit the specific method of determining the overall loss function based on the first loss weight, the first loss function, the second loss weight, and the second loss function.
[0186] In an exemplary embodiment of the present disclosure, a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function may also be obtained by an uncertain loss method.
[0187] For example, the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function can be automatically learned by uncertainty weighted loss. The expression for determining the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function by uncertainty weight loss is as follows, where σ 1 is the variance of the task corresponding to the first model to be trained, σ2 is the variance of the task corresponding to the second model to be trained, L 1 (W) is the first loss function corresponding to the first model to be trained, L 2 (W) is the second loss function corresponding to the second model to be trained, L is the overall loss function, when the loss of a task increases, its weight decreases, when the loss of a task decreases, its weight increases:
[0188]
[0189] Through the above steps S1110 to S1120, the first loss weight corresponding to the first loss function and the second loss weight corresponding to the second loss function can be obtained, and the overall loss function is determined according to the first loss weight, the first loss function, the second loss weight, and the second loss function.
[0190] In an exemplary embodiment of the present disclosure, the first model to be trained includes a first sub-model to be trained and a second sub-model to be trained, the second model to be trained includes a first sub-model to be trained and a second sub-model to be trained, and the two sub-training models in the first model to be trained and the two sub-training models in the second model to be trained present a double-tower structure.
[0191] like Fig.12As shown, sample data can be obtained, and the display content feature vector can be input into the first model to be trained, the second model to be trained and the shared model. The sample data can be processed in the hidden layers of the first model to be trained, the second model to be trained and the shared model. The numbers in the figure can be used to indicate the dimensions of the processed sample data. After the calculation of the hidden layers of the first model to be trained, the second model to be trained and the shared model, the hidden layer features of the first model to be trained (including the hidden layer features corresponding to the sub-training model in the first model to be trained, i.e., 256 in the first model to be trained), the hidden layer features of the shared model (i.e., 256 in the shared model to be trained), and the hidden layer features of the second model to be trained (including the hidden layer features corresponding to the sub-training model in the second model to be trained, i.e., 256 in the second model to be trained) are obtained. At this time, the hidden layer features of the first model to be trained and the hidden layer features of the shared model can be fused through a gate structure (gate method) to obtain the hidden layer features of the first model to be trained (i.e., 1 in the first model to be trained). 28), and obtain the hidden features of the shared model (i.e., 128 in the shared model) by combining the hidden features of the first model to be trained and the hidden features of the shared model through a gate structure (gate method), and determine the hidden features of the second model to be trained (i.e., 128 in the first model to be trained) by combining the hidden features of the second model to be trained and the hidden features of the shared model through a gate structure (gate method). By analogy, multiple groups of hidden features can be calculated to determine the first fusion feature corresponding to the first model to be trained, wherein the first sub-loss function of the first sub-model to be trained of the first model to be trained is delivery_loss1, and the second sub-loss function of the second sub-model to be trained of the first model to be trained is delivery_loss2, and the loss function weight w1 of the first sub-loss function and the loss function weight w2 of the second sub-loss function are obtained. The first loss function corresponding to the first model to be trained can be determined by the loss function weight of the first sub-loss function and the loss function weight of the second sub-loss function:
[0192] L1(W)=w1*delivery_loss1+w2*delivery_loss2
[0193] The second fusion feature corresponding to the second model to be trained is determined by the dot product method, and the second loss function of the second model to be trained is determined to be delivery_loss3 by the second predicted interaction information of the second model to be trained and the actual interaction information of the sample data. On this basis, the overall loss function is determined by the uncertain weight loss method:
[0194]
[0195] It should be noted that in the embodiments of the present disclosure, the dimensions of the sample data processed in the hidden layers of the first model to be trained, the second model to be trained, and the shared model may be different. When the sample data needs to be fused, the sample data processed in the hidden layers of the first model to be trained, the second model to be trained, and the shared model may be processed into the same dimension.
[0196] It should be noted that the present disclosure does not impose any special restrictions on the number of sub-training models in the model to be trained and the number of models to be trained for joint training.
[0197] In an exemplary embodiment of the present disclosure, multiple display contents and related information corresponding to the display contents can be obtained, and the related information corresponding to the display contents can be input into a first recommendation model to obtain multiple first screened display contents, and the multiple first screened display contents can be input into a second recommendation model to obtain multiple second screened display contents, wherein the second recommendation model is obtained by any of the recommendation model training methods described above, and the display contents to be displayed are determined according to the multiple second screened display contents. Fig.13 As shown, determining the display content to be displayed according to the plurality of second filtered display contents may include the following steps S1310 to S1330:
[0198] Step S1310, obtaining a plurality of display contents and related information corresponding to the display contents, inputting the related information corresponding to the display contents into a first recommendation model to obtain a plurality of first screened display contents; wherein the related information corresponding to the display contents includes the display contents, the display users, and the interaction information between the display contents and the display users, and the first recommendation model is obtained by any of the above-mentioned recommendation model training methods;
[0199] Step S1320, inputting the plurality of first screened display contents into a second recommendation model to obtain a plurality of second screened display contents; wherein the second recommendation model is obtained by any of the above-mentioned recommendation model training methods;
[0200] Step S1330: Determine the display content to be displayed according to the plurality of second filtered display contents.
[0201] In an example embodiment of the present disclosure, it can be applied to a scenario where it is necessary to recommend display content to a user when the user browses a web page. Specifically, multiple display contents and display content related information can be obtained from a display content pool, and the display content related information can be input into a first recommendation model to obtain multiple first screened display contents, and the first recommendation model is obtained by the recommendation model training method described in the above embodiment. For example, the first recommendation model can include a network model for realizing a recall task.
[0202] The obtained first screened display content is then input into a second recommendation model, and the second recommendation model is obtained by the recommendation model training method described in the above embodiment. For example, the first recommendation model may include a network model for implementing a rough sorting task.
[0203] In an exemplary embodiment of the present disclosure, after obtaining multiple second filtered display contents through the above steps, display contents to be displayed to the user can be determined based on the multiple second filtered display contents. For example, the multiple second filtered display contents can be determined as the display contents to be displayed, or, on this basis, further filtering can be performed to obtain the display contents to be displayed.
[0204] It should be noted that the present disclosure does not specifically limit the specific method of determining the display content to be displayed based on the plurality of second filtered display contents.
[0205] Through the above steps S1310 to S1330, multiple display contents and related information corresponding to the display contents can be obtained, and the related information corresponding to the display contents is input into the first recommendation model to obtain multiple first screened display contents, and the multiple first screened display contents are input into the second recommendation model to obtain multiple second screened display contents, wherein the second recommendation model is obtained by a recommendation model training method such as any one of the above items, and the display content to be displayed is determined according to the multiple second screened display contents.
[0206] In the embodiments of the present disclosure, sample data can be obtained and input into the first model to be trained, the second model to be trained and the shared model to obtain the hidden features of the hidden layer outputs of the first model to be trained, the second model to be trained and the shared model, and the hidden features of the first model to be trained are fused with the hidden features of the shared model to obtain the first predicted interaction information, and the first model to be trained is trained according to the first predicted interaction information and the actual interaction information to obtain the first recommendation model, and the second recommendation model is similar. On the one hand, the features of the first model to be trained / the second model to be trained and the shared model can be fused so that the first model to be trained and the second model to be trained are correlated, thereby improving the consistency between the first model to be trained and the second model to be trained; on the other hand, since the features of the model with task-related relationships are taken into account when training the model, when screening the display content, the screening accuracy can be made higher, thereby improving the conversion rate of the display content.
[0207] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.
[0208] In addition, in an exemplary embodiment of the present disclosure, a recommendation model training device is also provided. Fig.14 As shown, a recommendation model training device 1400 includes: a sample data acquisition unit 1410, a sample data input unit 1420, a first recommendation model training unit 1430, and a second recommendation model training unit 1440.
[0209] The sample data acquisition unit is configured to acquire sample data; the sample data includes sample content, sample users, and interaction information between sample content and sample users; the sample data input unit is configured to input sample data into the first model to be trained, the second model to be trained, and the shared model to obtain hidden layer features of hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model; the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to filter by the first model to be trained, and then filter the filtered results by the second model to be trained; the shared model is used to learn the first model to be trained and the second model to be trained. The common features of the training model; the first recommendation model training unit is configured to perform a fusion processing on the hidden layer features of the first model to be trained and the hidden layer features of the shared model to obtain the first predicted interaction information of the first model to be trained, and the first model to be trained is trained according to the first predicted interaction information and the actual interaction information in the sample data to obtain the first recommendation model; the second recommendation model training unit is configured to perform a fusion processing on the hidden layer features of the second model to be trained and the hidden layer features of the shared model to obtain the second predicted interaction information of the first model to be trained, and the second model to be trained is trained according to the second predicted interaction information and the actual interaction information in the sample data to obtain the second recommendation model.
[0210] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain the first predicted interaction information of the first model to be trained, and the recommendation model training device also includes: a first fused feature acquisition unit, which fuses the hidden layer features of the first model to be trained and the hidden layer features of the shared model to obtain the first fused features; and a predicted interaction information acquisition unit, which is configured to determine the first predicted interaction information of the first model to be trained based on the first fused features.
[0211] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the hidden features of the first model to be trained and the hidden features of the shared model are fused to obtain a first fused feature, and the recommendation model training device also includes: a fusion weight acquisition unit, which is configured to execute the fusion weights corresponding to the hidden features of the first model to be trained and the fusion weights corresponding to the hidden features of the shared model according to the sample data; a second fusion feature acquisition unit, which is configured to execute the fusion weights corresponding to the hidden features of the first model to be trained, the hidden features of the first model to be trained, the fusion weights corresponding to the hidden features of the shared model, and the hidden features of the shared model to obtain the first fused feature.
[0212] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain a first fused feature, and the recommendation model training device also includes: a first current fused feature acquisition unit, configured to execute, for the first model to be trained, the current fused feature of the current hidden layer of the first model to be trained based on the previous fused features of the previous hidden layer of the first model to be trained and the previous fused features of the shared model of the same layer to obtain the current fused feature of the shared model; a second current fused feature acquisition unit, configured to execute, for the first model to be trained, the current fused feature of the current hidden layer of the first model to be trained based on the previous fused features of the first model to be trained, the previous fused features of the shared model of the same layer, and the previous fused features of the second model to be trained to obtain the current fused feature of the shared model; a third fused feature acquisition unit, configured to execute, for the first model to be trained, the current fused feature of the shared model based on the current fused features of the first model to be trained and the current fused features of the shared model to obtain the first fused feature.
[0213] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the first model to be trained includes a first sub-training model and a second sub-training model, and the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain a first fused feature. The recommendation model training device also includes: a first fused sub-feature acquisition unit, configured to execute the first fused sub-feature corresponding to the first training sub-model according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model; a second fused sub-feature acquisition unit, configured to execute the second fused sub-feature corresponding to the second training sub-model according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model; and a fourth fused feature acquisition unit, configured to execute the first fused sub-feature corresponding to the first training sub-model and the second fused sub-feature corresponding to the second training sub-model to obtain the first fused feature.
[0214] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the first model to be trained is trained according to the first predicted interaction information and the actual interaction information in the sample data, and the recommendation model training device also includes: a first recommendation model training unit, configured to execute a first loss function of the first model to be trained by determining the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data; a first training unit, configured to execute an update of the neural network parameters in the first model to be trained according to the first loss function of the first model to be trained to obtain a first recommendation model.
[0215] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the first model to be trained includes a first sub-training model and a second sub-training model, and the first loss function of the first model to be trained is determined by the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data, and the recommendation model training device also includes: a first loss sub-function determination unit, configured to execute input of sample data into the first sub-training model to obtain first sub-prediction interaction information, and determine the first loss sub-function of the first sub-training model through the first sub-prediction interaction information; a second loss sub-function determination unit, configured to execute input of sample data into the second sub-training model to obtain second sub-prediction interaction information, and determine the second loss sub-function of the second sub-training model through the second sub-prediction interaction information; and a second recommendation model training unit, configured to execute determination of the first loss function corresponding to the first model to be trained according to the first sub-loss function corresponding to the first training sub-model and the second sub-loss function corresponding to the second training sub-model.
[0216] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the neural network parameters in the first model to be trained are updated according to the first loss function of the first model to be trained to obtain a first recommendation model, and the recommendation model training device also includes: a third loss function determination unit, configured to execute the second loss function of the second model to be trained through the second predicted interaction information of the second model to be trained; a first overall loss function determination unit, configured to execute the overall loss function according to the first loss function and the second loss function; a second training unit, configured to execute the update of the neural network parameters in the first model to be trained according to the overall loss function to train the first model to be trained to obtain the first recommendation model.
[0217] In an exemplary embodiment of the present disclosure, based on the aforementioned scheme, the overall loss function is determined according to the first loss function and the second loss function, and the recommendation model training device also includes: a loss weight acquisition unit, configured to execute acquisition of a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function; a second overall loss function determination unit, configured to execute determination of the overall loss function according to the first loss weight, the first loss function, the second loss weight, and the second loss function.
[0218] Since the various functional modules of the recommendation model training device of the example embodiment of the present disclosure correspond to the steps of the example embodiment of the recommendation model training method described above, for details not disclosed in the embodiment of the device of the present disclosure, please refer to the embodiment of the recommendation model training method described above in the present disclosure.
[0219] In addition, in an exemplary embodiment of the present disclosure, a display content determination device is also provided. Fig.15 As shown, a display content determination device 1500 includes: a first screening content acquisition unit 1510 , a second screening content acquisition unit 1520 , and a display content determination unit 1530 .
[0220] Among them, the first screening content acquisition unit is configured to execute the acquisition of multiple display contents and related information corresponding to the display contents, and input the related information corresponding to the display contents into the first recommendation model to obtain multiple first screening display contents; wherein, the related information corresponding to the display contents includes the display contents, the display users, and the interaction information between the display contents and the display users, and the first recommendation model is obtained by any of the recommendation model training methods mentioned above; the second screening content acquisition unit is configured to execute the input of multiple first screening display contents into the second recommendation model to obtain multiple second screening display contents; wherein, the second recommendation model is obtained by any of the recommendation model training methods mentioned above; the display content determination unit is configured to execute the determination of the display content to be displayed according to the multiple second screening display contents.
[0221] Since the functional modules of the display content determination device of the exemplary embodiment of the present disclosure correspond to the steps of the exemplary embodiment of the display content determination method described above, for details not disclosed in the device embodiment of the present disclosure, please refer to the embodiment of the display content determination method described above.
[0222] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.
[0223] In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above-mentioned recommendation model training method or display content determination method is also provided.
[0224] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware embodiments, complete software embodiments (including firmware, microcode, etc.), or embodiments combining hardware and software aspects, which may be collectively referred to herein as "circuits", "modules" or "systems".
[0225] Refer to the following Fig.16 16 is a diagram to describe an electronic device 1600 according to such an embodiment of the present disclosure. Fig.16 The electronic device 1600 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.
[0226] like Fig.16 As shown, the electronic device 1600 is in the form of a general computing device. The components of the electronic device 1600 may include, but are not limited to: the at least one processing unit 1610, the at least one storage unit 1620, a bus 1630 connecting different system components (including the storage unit 1620 and the processing unit 1610), and a display unit 1640.
[0227] The storage unit stores program codes, which can be executed by the processing unit 1610, so that the processing unit 1610 performs the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 1610 can perform the following steps: Figure 2 In step S210, sample data is obtained; wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users; in step S220, the sample data is input into the first model to be trained, the second model to be trained, and the shared model to obtain the hidden layer features of the hidden layer outputs of the first model to be trained, the second model to be trained, and the shared model; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to filter by the first model to be trained, and then filter the filtered results by the second model to be trained; the shared model is used to learn the first model to be trained and the second model to be trained. The common features of the two models to be trained; step S230, the hidden features of the first model to be trained are fused with the hidden features of the shared model to obtain the first predicted interaction information of the first model to be trained, and the first model to be trained is trained according to the first predicted interaction information and the actual interaction information in the sample data to obtain the first recommendation model; step S240, the hidden features of the second model to be trained are fused with the hidden features of the shared model to obtain the second predicted interaction information of the first model to be trained, and the second model to be trained is trained according to the second predicted interaction information and the actual interaction information in the sample data to obtain the second recommendation model. Alternatively, the following can also be performed: Fig.12In step S1210 shown in the figure, a plurality of display contents and relevant information corresponding to the display contents are obtained, and the relevant information corresponding to the display contents are input into a first recommendation model to obtain a plurality of first screened display contents; wherein, the relevant information corresponding to the display contents includes display contents, display users, and interaction information between display contents and display users, and the first recommendation model is obtained by a recommendation model training method such as any one of the above; in step S1220, the plurality of first screened display contents are input into a second recommendation model to obtain a plurality of second screened display contents; wherein, the second recommendation model is obtained by a recommendation model training method such as any one of the above; in step S1230, the display contents to be displayed are determined according to the plurality of second screened display contents.
[0228] For example, electronic devices can achieve Figure 2 and Fig.12 The steps shown.
[0229] The storage unit 1620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 1621 and / or a cache memory unit 1622 , and may further include a read-only memory unit (ROM) 1623 .
[0230] The storage unit 1620 may also include a program / utility 1624 having a set (at least one) of program modules 1625, such program modules 1625 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.
[0231] Bus 1630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.
[0232] The electronic device 1600 may also communicate with one or more external devices 1670 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1600, and / or communicate with any device that enables the electronic device 1600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1650. Furthermore, the electronic device 1600 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1660. As shown, the network adapter 1660 communicates with other modules of the electronic device 1600 via a bus 1630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0233] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.
[0234] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory including instructions, and the above instructions can be executed by a processor of the device to complete the above method. Optionally, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0235] In an exemplary embodiment, a computer program product is also provided, including a computer program / instruction, which implements the recommendation model training method or the display content determination method in the above-mentioned embodiment when the computer program / instruction is executed by a processor.
[0236] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0237] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A recommendation model training method, characterized in that: The method comprises: Acquire sample data; wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users; The sample data is input into a first model to be trained, a second model to be trained and a shared model, and hidden layer features of hidden layer outputs of the first model to be trained, the second model to be trained and the shared model are obtained; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is screened by the first model to be trained, and then the screened result is screened by the second model to be trained; the shared model is used to learn the common features of the first model to be trained and the second model to be trained; Fusing the hidden features of the first model to be trained with the hidden features of the shared model to obtain first predicted interaction information of the first model to be trained, and training the first model to be trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model; Fusing the hidden features of the second model to be trained with the hidden features of the shared model to obtain second predicted interaction information of the first model to be trained, and training the second model to be trained according to the second predicted interaction information and actual interaction information in the sample data to obtain a second recommendation model; The fusing the hidden layer features of the first model to be trained and the hidden layer features of the shared model includes: The hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain a first fused feature, wherein the first fused feature is obtained by fusing the calculation results of the hidden layer features of the first model to be trained and the calculation results of the hidden layer features of the shared model.
2. The method according to claim 1, characterized in that The step of fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first predicted interaction information of the first model to be trained also includes: Determine first predicted interaction information of the first model to be trained according to the first fusion feature.
3. The method according to claim 2, characterized in that The step of fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first fused features comprises: Determine, according to the sample data, a fusion weight corresponding to the hidden layer features of the first model to be trained and a fusion weight corresponding to the hidden layer features of the shared model; A first fused feature is obtained according to the fusion weight corresponding to the hidden layer feature of the first model to be trained, the hidden layer feature of the first model to be trained, the fusion weight corresponding to the hidden layer feature of the shared model, and the hidden layer feature of the shared model.
4. The method according to claim 2, characterized in that: The step of fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first fused features comprises: For the first model to be trained, obtaining a current fused feature of a current hidden layer of the first model to be trained according to a previous fused feature of a previous hidden layer of the first model to be trained and a previous fused feature of the shared model at the same layer; Obtaining current fusion features of the shared model according to the previous fusion features of the first model to be trained, the previous fusion features of the shared model at the same layer, and the previous fusion features of the second model to be trained; The first fused feature is obtained according to the current fused feature of the first model to be trained and the current fused feature of the shared model.
5. The method according to claim 2, characterized in that: The first model to be trained includes a first sub-training model and a second sub-training model, and the step of fusing the hidden layer features of the first model to be trained with the hidden layer features of the shared model to obtain the first fused features includes: Obtaining a first fusion sub-feature corresponding to the first training sub-model according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model; Obtaining a second fusion sub-feature corresponding to the second training sub-model according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model; A first fusion feature is obtained according to a first fusion sub-feature corresponding to the first training sub-model and a second fusion sub-feature corresponding to the second training sub-model.
6. The method according to claim 2, characterized in that The step of training the first to-be-trained model according to the first predicted interaction information and the actual interaction information in the sample data comprises: Determine a first loss function of the first model to be trained by using the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data; The neural network parameters in the first model to be trained are updated according to the first loss function of the first model to be trained to obtain a first recommendation model.
7. The method according to claim 6, characterized in that The first model to be trained includes a first sub-training model and a second sub-training model, and the step of determining a first loss function of the first model to be trained by using the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data includes: Inputting the sample data into the first sub-training model to obtain first sub-prediction interaction information, and determining a first loss sub-function of the first sub-training model according to the first sub-prediction interaction information; Inputting the sample data into the second sub-training model to obtain second sub-prediction interaction information, and determining a second loss sub-function of the second sub-training model according to the second sub-prediction interaction information; Determine the first loss function of the first model to be trained according to the first sub-loss function of the first training sub-model and the second sub-loss function of the second training sub-model.
8. The method according to claim 6, characterized in that The step of updating the neural network parameters in the first to-be-trained model according to the first loss function of the first to-be-trained model to obtain a first recommendation model includes: Determine a second loss function of the second model to be trained by using the second predicted interaction information of the second model to be trained; Determine an overall loss function according to the first loss function and the second loss function; The neural network parameters in the first model to be trained are updated according to the overall loss function to train the first model to be trained to obtain a first recommendation model.
9. The method according to claim 8, characterized in that The step of determining the overall loss function according to the first loss function and the second loss function comprises: Obtaining a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function; The overall loss function is determined according to the first loss weight, the first loss function, the second loss weight, and the second loss function.
10. The method according to claim 9, characterized in that The step of obtaining a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function comprises: A first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function are obtained through an uncertain loss method.
11. A method for determining display content, characterized in that: The method comprises: Acquire multiple display contents and related information corresponding to the display contents, input the related information corresponding to the display contents into a first recommendation model to obtain multiple first screened display contents; wherein the related information corresponding to the display contents includes display contents, display users, and interaction information between the display contents and the display users, and the first recommendation model is obtained by the recommendation model training method according to any one of claims 1 to 10; Inputting the plurality of first screened display contents into a second recommendation model to obtain a plurality of second screened display contents; wherein the second recommendation model is obtained by the recommendation model training method according to any one of claims 1 to 10; The display content to be displayed is determined according to the plurality of second filtered display contents.
12. A recommendation model training device, characterized in that: include: A sample data acquisition unit, configured to acquire sample data; wherein the sample data includes sample content, sample users, and interaction information between the sample content and the sample users; A sample data input unit is configured to input the sample data into a first model to be trained, a second model to be trained and a shared model, and obtain hidden layer features of hidden layer outputs of the first model to be trained, the second model to be trained and the shared model; wherein the first model to be trained and the second model to be trained have a task-related relationship; the task-related relationship is to filter by the first model to be trained, and then filter the filtered results by the second model to be trained; the shared model is used to learn the common features of the first model to be trained and the second model to be trained; A first recommendation model training unit is configured to perform fusion processing on the hidden layer features of the first model to be trained and the hidden layer features of the shared model to obtain first predicted interaction information of the first model to be trained, and train the first model to be trained according to the first predicted interaction information and actual interaction information in the sample data to obtain a first recommendation model; A second recommendation model training unit is configured to perform fusion processing on the hidden layer features of the second model to be trained and the hidden layer features of the shared model to obtain second predicted interaction information of the first model to be trained, and train the second model to be trained according to the second predicted interaction information and actual interaction information in the sample data to obtain a second recommendation model; The first fusion feature acquisition unit is configured to perform a fusion process on the hidden layer features of the first model to be trained and the hidden layer features of the shared model to obtain a first fusion feature, wherein the first fusion feature is obtained by fusing the calculation results of the hidden layer features of the first model to be trained and the calculation results of the hidden layer features of the shared model.
13. The recommendation model training device according to claim 12, characterized in that: The hidden layer features of the first model to be trained are fused with the hidden layer features of the shared model to obtain first predicted interaction information of the first model to be trained, and the recommendation model training device further includes: The predicted interaction information acquisition unit is configured to determine the first predicted interaction information of the first model to be trained according to the first fusion feature.
14. The recommendation model training device according to claim 13, characterized in that: The first fusion feature is obtained by fusing the hidden layer feature of the first model to be trained with the hidden layer feature of the shared model, and the recommendation model training device further includes: A fusion weight acquisition unit is configured to determine, according to the sample data, a fusion weight corresponding to the hidden layer features of the first model to be trained and a fusion weight corresponding to the hidden layer features of the shared model; The second fusion feature acquisition unit is configured to execute the first fusion feature according to the fusion weight corresponding to the hidden layer feature of the first model to be trained, the hidden layer feature of the first model to be trained, the fusion weight corresponding to the hidden layer feature of the shared model, and the hidden layer feature of the shared model.
15. The recommendation model training device according to claim 13, characterized in that: The hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain the first fused features, and the recommendation model training device further includes: The first current fusion feature acquisition unit is configured to execute, for the first model to be trained, obtaining the current fusion feature of the current hidden layer of the first model to be trained according to the previous fusion feature of the previous hidden layer of the first model to be trained and the previous fusion feature of the shared model at the same layer; A second current fusion feature acquisition unit is configured to obtain the current fusion feature of the shared model according to the previous fusion feature of the first model to be trained, the previous fusion feature of the shared model at the same layer, and the previous fusion feature of the second model to be trained; The third fusion feature acquisition unit is configured to obtain the first fusion feature according to the current fusion feature of the first model to be trained and the current fusion feature of the shared model.
16. The recommendation model training device according to claim 14, characterized in that: The first model to be trained includes a first sub-training model and a second sub-training model, the hidden layer features of the first model to be trained and the hidden layer features of the shared model are fused to obtain a first fused feature, and the recommendation model training device also includes: A first fusion sub-feature acquisition unit is configured to obtain a first fusion sub-feature corresponding to the first training sub-model according to the hidden layer features of the first training sub-model and the hidden layer features of the shared model; A second fusion sub-feature acquisition unit is configured to obtain a second fusion sub-feature corresponding to the second training sub-model according to the hidden layer features of the second training sub-model and the hidden layer features of the shared model; The fourth fusion feature acquisition unit is configured to obtain a first fusion feature according to the first fusion sub-feature corresponding to the first training sub-model and the second fusion sub-feature corresponding to the second training sub-model.
17. The recommendation model training device according to claim 14, characterized in that: The first to-be-trained model is trained according to the first predicted interaction information and the actual interaction information in the sample data, and the recommendation model training device further comprises: A first recommendation model training unit is configured to determine a first loss function of the first model to be trained by using first predicted interaction information of the first model to be trained and actual interaction information of the sample data; The first training unit is configured to update the neural network parameters in the first model to be trained according to the first loss function of the first model to be trained to obtain a first recommendation model.
18. The recommendation model training device according to claim 17, characterized in that: The first model to be trained includes a first sub-training model and a second sub-training model, and the first loss function of the first model to be trained is determined by the first predicted interaction information of the first model to be trained and the actual interaction information of the sample data. The recommendation model training device also includes: A first loss sub-function determining unit is configured to input the sample data into the first sub-training model to obtain first sub-prediction interaction information, and determine a first loss sub-function of the first sub-training model through the first sub-prediction interaction information; A second loss sub-function determining unit is configured to input the sample data into the second sub-training model to obtain second sub-prediction interaction information, and determine a second loss sub-function of the second sub-training model through the second sub-prediction interaction information; The second recommendation model training unit is configured to determine the first loss function corresponding to the first model to be trained according to the first sub-loss function corresponding to the first training sub-model and the second sub-loss function corresponding to the second training sub-model.
19. The recommendation model training device according to claim 17, characterized in that: The neural network parameters in the first model to be trained are updated according to the first loss function of the first model to be trained to obtain a first recommendation model, and the recommendation model training device further includes: A third loss function determining unit is configured to determine a second loss function of the second to-be-trained model through second predicted interaction information of the second to-be-trained model; A first overall loss function determining unit, configured to determine an overall loss function according to the first loss function and the second loss function; The second training unit is configured to update the neural network parameters in the first model to be trained according to the overall loss function to train the first model to be trained to obtain a first recommendation model.
20. The recommendation model training device according to claim 19, characterized in that: The overall loss function is determined according to the first loss function and the second loss function, and the recommendation model training device further comprises: A loss weight acquisition unit, configured to acquire a first loss weight corresponding to the first loss function and a second loss weight corresponding to the second loss function; The second overall loss function determination unit is configured to determine the overall loss function according to the first loss weight, the first loss function, the second loss weight, and the second loss function.
21. A display content determination device, characterized in that: include: A first screening content acquisition unit is configured to acquire a plurality of display contents and related information corresponding to the display contents, and input the related information corresponding to the display contents into a first recommendation model to obtain a plurality of first screening display contents; wherein the related information corresponding to the display contents includes display contents, display users, and interaction information between the display contents and the display users, and the first recommendation model is obtained by the recommendation model training method according to any one of claims 1 to 10; A second screening content acquisition unit is configured to input the plurality of first screening display contents into a second recommendation model to obtain a plurality of second screening display contents; wherein the second recommendation model is obtained by the recommendation model training method according to any one of claims 1 to 10; The display content determination unit is configured to determine the display content to be displayed according to the plurality of second filtered display contents.
22. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the executable instructions to implement the recommendation model training method described in any one of claims 1 to 10 or the display content determination method described in claim 11.
23. A computer-readable storage medium, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to execute the recommendation model training method described in any one of claims 1 to 10 or the display content determination method described in claim 11.
24. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the recommendation model training method described in any one of claims 1 to 10 or the display content determination method described in claim 11 is implemented.
