Model training method and device, advertisement putting method and device and electronic equipment

Abstract

An advertisement putting method can be applied to electronic equipment with a display screen. The method comprises the steps of obtaining a first access operation of a user on a first page; in response to the first access operation, a first context feature related to the first page is obtained, and the first context feature is irrelevant to a user portrait and / or a user behavior of the user; based on the first context feature, f advertisements are displayed, and f is larger than or equal to 1. Therefore, the context features are used for replacing the user features, the dependence on the user features is reduced, and accurate advertisement putting can be performed when the features such as the user portrait cannot be obtained.

Description

Model training method, advertising delivery method, device and electronic equipment Technical Field

[0001] The present application relates to the field of artificial intelligence (AI) technology, and in particular to a model training method, an advertising delivery method, a device, and an electronic device. Background Art

[0002] Currently, advertising is often based on user profiles and behavior. However, with the increasing protection of user information, it is becoming increasingly difficult to obtain user identities, making it impossible to obtain user profiles, resulting in poor advertising effectiveness.

[0003] Summary of the Invention

[0004] The present application provides a model training method, an advertising delivery method, an apparatus, an electronic device, a server, a computer storage medium, and a computer product, which can improve the effectiveness of advertising delivery.

[0005] In a first aspect, the present application provides a model training method, which includes: obtaining a first sample set, the first sample set including a first training sample, the first training sample including at least one advertising feature of an advertisement and at least one context feature corresponding to the advertisement; processing the at least one context feature through a first model to obtain a first feature vector, and processing the at least one advertising feature through a second model to obtain a second feature vector; processing the processing results of at least one stage in the first model and the processing results of at least one stage in the second model through a third model to obtain a first correlation score, the first correlation score being used to characterize the degree of matching between at least one context feature and at least one advertising feature; obtaining the loss corresponding to the first training sample based on the first feature vector, the second feature vector, the first correlation score and the sample label of the first training sample; training the first model and the second model based on the loss corresponding to at least one training sample in the first sample set.

[0006] Thus, by using contextual features instead of user features for model training, the reliance on user features in subsequent model use is reduced. Simultaneously, by adding a third model during the training of the first and second models, interactive modeling between the first and second models is achieved, thereby enabling interaction between contextual features and advertising features, enhancing their correlation and improving the representational effectiveness of the resulting first and second models. For example, the trained first model can be used to extract contextual features related to pages visited by users, and the trained second model can be used to extract features from advertisements.

[0007] In one possible implementation, at least one context feature is processed by a first model to obtain a first feature vector, specifically including: encoding at least one context feature by the first model to obtain N third feature vectors, N≥1; wherein, when N=1, the N third feature vectors are processed by the first model to obtain a first feature vector; when N≥2, the N third feature vectors are spliced ​​by the first model to obtain a fourth feature vector, and the fourth feature vector is processed by the first model to obtain the first feature vector.

[0008] In one possible implementation, at least one advertising feature is processed by a second model to obtain a second feature vector, specifically including: encoding at least one advertising feature by a second model to obtain M fifth feature vectors, M≥1; wherein, when M=1, the M fifth feature vectors are processed by the second model to obtain a second feature vector; when M≥2, the M fifth feature vectors are spliced ​​by the second model to obtain a sixth feature vector, and the sixth feature vector is processed by the second model to obtain a second feature vector.

[0009] In one possible implementation, the loss corresponding to the first training sample is obtained according to the first eigenvector, the second eigenvector, the first correlation score, and the sample label of the first training sample, specifically including: obtaining the first loss according to the first eigenvector, the second eigenvector, and the sample label of the first training sample; obtaining the second loss according to the first correlation score and the sample label of the first training sample; and obtaining the loss corresponding to the first training sample according to the first loss and the second loss.

[0010] In one possible implementation, a first loss is obtained based on the first eigenvector, the second eigenvector, and the sample label of the first training sample, specifically including: determining a second correlation score based on the first eigenvector and the second eigenvector, the second correlation score being used to characterize the degree of matching between at least one contextual feature and at least one advertising feature; and obtaining the first loss based on the second correlation score and the sample label of the first training sample.

[0011] In a possible implementation, the method may further include: training a third model based on the loss corresponding to at least one training sample in the first sample set.

[0012] In one possible implementation, the third model is a neural network model, which is used to assist in calculating the loss corresponding to the training samples in the first sample set during the training of the first model and the second model. Exemplarily, the model structure of the third model can be, but is not limited to, deep neural networks (DNN), multi-layer perception (MLP), deep factorization machine (DeepFM), extreme deep factorization machine model (xDeepFM), deep cross network (DCN), or deep interest network (DIN) and other model structures.

[0013] In a second aspect, the present application provides an advertising delivery method, applicable to an electronic device with a display screen. The method may include: obtaining a user's access operation to a first page; in response to the access operation, obtaining a first context feature related to the first page, wherein the first context feature is unrelated to the user's user profile and / or user behavior; and displaying f advertisements based on the first context feature, where f ≥ 1.

[0014] Therefore, when placing advertisements, by using contextual features instead of user features, the dependence on user features is reduced. When user features (such as user portraits, user behaviors, etc.) cannot be obtained, the contextual features related to the pages visited by the user are used to accurately place advertisements, solving the problem of the game between the privacy protection of user information and the accurate placement of advertisements.

[0015] In one possible implementation, after displaying f advertisements, the method may further include: obtaining a second user access operation to a second page; in response to the second access operation, obtaining second contextual features related to the second page, wherein the second contextual features are unrelated to the user's user profile and / or user behavior; and based on the second contextual features, displaying g advertisements, where g ≥ 1, and at least some of the g advertisements are different from the advertisements in the f advertisements. Exemplarily, the content of the first page is different from the content of the second page. Because the content of the two pages is different, the contextual features corresponding to the two pages are also different, resulting in different ads being ultimately displayed.

[0016] In one possible implementation, advertisements are displayed based on contextual features, specifically including: processing the contextual features using a first model to obtain a first feature vector, where the first model is a neural network model, and the contextual features are first contextual features or second contextual features; determining a similarity value between the first feature vector and K second feature vectors to obtain K similarity values, where each second feature vector is used to characterize a feature of an advertisement; selecting Q similarity values ​​from the K similarity values, where each of the Q similarity values ​​is greater than a similarity value of the K similarity values ​​other than the Q similarity values, and 1≤Q≤K; and displaying advertisements corresponding to the Q similarity values, where when the contextual feature is the first contextual feature, the number of advertisements corresponding to the Q similarity values ​​is f, and when the contextual feature is the second contextual feature, the number of advertisements corresponding to the Q similarity values ​​is g.

[0017] In one possible implementation, displaying advertisements based on contextual features specifically includes: sending a first message to a server, the first message including a contextual feature, where the contextual feature is a first contextual feature or a second contextual feature; receiving a second message sent by the server, the second message including Q advertisements or advertisement indication information, where the advertisement indication information is used to generate Q advertisements, wherein when the contextual feature is the first contextual feature, the Q advertisements are f advertisements, and when the contextual feature is the second contextual feature, the Q advertisements are g advertisements; and displaying Q advertisements.

[0018] In one possible implementation, advertisements are displayed based on contextual features, specifically including: processing the contextual features using a first model to obtain a first feature vector, where the first model is a neural network model, and the contextual features are the first contextual features or the second contextual features; sending a first message to a server, where the first message includes the first feature vector; receiving a second message sent by the server, where the second message includes Q advertisements or advertisement indication information, where the advertisement indication information is used to generate Q advertisements, where when the contextual features are the first contextual features, the Q advertisements are f advertisements, and when the contextual features are the second contextual features, the Q advertisements are g advertisements; and displaying Q advertisements.

[0019] In one possible implementation, displaying advertisements based on contextual features specifically includes: sending a first message to a server, the first message including a contextual feature, the first message being used to instruct the server to extract features from the contextual feature, the contextual feature being a first contextual feature or a second contextual feature; receiving a second message sent by the server, the second message including a first feature vector extracted by the server based on the contextual feature; determining a similarity value between the first feature vector and K second feature vectors to obtain K similarity values, each second feature vector being used to characterize features of an advertisement; screening out Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value of the K similarity values ​​other than the Q similarity values, 1≤Q≤K; and displaying advertisements corresponding to each of the Q similarity values, wherein when the contextual feature is the first contextual feature, the number of advertisements corresponding to each of the Q similarity values ​​is f, and when the contextual feature is the second contextual feature, the number of advertisements corresponding to each of the Q similarity values ​​is g.

[0020] In one possible implementation, the first context feature includes one or more of the following: the identifier of the first page, the text in the first page, the identifier of the video in the first page, the identifier of the picture in the first page, the identifier of the audio in the first page, the user's IP regional information, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0021] In one possible implementation, the second context feature includes one or more of the following: the identifier of the second page, the text in the second page, the identifier of the video in the second page, the identifier of the picture in the second page, the identifier of the audio in the second page, the user's IP regional information, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0022] In a third aspect, the present application provides an advertising delivery method, which is applied to a server, and the method includes: receiving a first message sent by an electronic device, the first message including contextual features related to a page visited by a user on the electronic device, and the contextual features are unrelated to the user's user profile and / or user behavior; in response to the first message, based on the contextual features, obtaining R advertisements or information indicating R advertisements, R≥1; sending a second message to the electronic device, the second message being used to instruct the electronic device to display R advertisements.

[0023] In one possible implementation, based on context features, R advertisements or information indicating R advertisements are obtained, specifically including: processing the context features using a first model to obtain a first feature vector, where the first model is a neural network model; determining a similarity value between the first feature vector and K second feature vectors to obtain K similarity values, where each second feature vector is used to characterize the features of an advertisement; screening out Q similarity values ​​from the K similarity values, where each of the Q similarity values ​​is greater than a similarity value of the K similarity values ​​other than the Q similarity values, 1≤Q≤K; and using the advertisements corresponding to each of the Q similarity values ​​as R advertisements, or using the information indicating the advertisements corresponding to each of the Q similarity values ​​as information indicating the R advertisements.

[0024] In one possible implementation, the contextual features include one or more of the following: page identification, text in the page, identification of the video in the page, identification of the image in the page, identification of the audio in the page, user's IP region information, access time, device type of the electronic device, operator providing network services for the electronic device, network type of the electronic device, system language of the electronic device, brand of the electronic device, model of the electronic device, scenario in which the electronic device needs to display advertisements, size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0025] In a fourth aspect, the present application provides a model training device, which includes: an acquisition module and a processing module. The acquisition module is used to acquire a first sample set, the first sample set includes a first training sample, and the first training sample includes at least one advertising feature of the advertisement and at least one context feature corresponding to the advertisement. The processing module is used to process at least one context feature through a first model to obtain a first feature vector, and to process at least one advertising feature through a second model to obtain a second feature vector. The processing module is also used to process the processing results of at least one stage in the first model and the processing results of at least one stage in the second model through a third model to obtain a first correlation score, and the first correlation score is used to characterize the degree of matching between at least one context feature and at least one advertising feature. The processing module is also used to obtain the loss corresponding to the first training sample based on the first feature vector, the second feature vector, the first correlation score and the sample label of the first training sample. The processing module is also used to train the first model and the second model based on the loss corresponding to at least one training sample in the first sample set.

[0026] In one possible implementation, when the processing module processes at least one context feature through the first model to obtain a first feature vector, the processing module is specifically used to: encode at least one context feature through the first model to obtain N third feature vectors, N≥1; wherein, when N=1, the N third feature vectors are processed through the first model to obtain the first feature vector; when N≥2, the N third feature vectors are spliced ​​through the first model to obtain a fourth feature vector, and the fourth feature vector is processed through the first model to obtain the first feature vector.

[0027] In one possible implementation, when the processing module processes at least one advertising feature through the second model to obtain a second feature vector, the processing module is specifically used to: encode at least one advertising feature through the second model to obtain M fifth feature vectors, M≥1; wherein, when M=1, the M fifth feature vectors are processed through the second model to obtain the second feature vector; when M≥2, the M fifth feature vectors are spliced ​​through the second model to obtain a sixth feature vector, and the sixth feature vector is processed through the second model to obtain the second feature vector.

[0028] In one possible implementation, when the processing module obtains the loss corresponding to the first training sample based on the first eigenvector, the second eigenvector, the first correlation score and the sample label of the first training sample, it is specifically used to: obtain the first loss based on the first eigenvector, the second eigenvector and the sample label of the first training sample; obtain the second loss based on the first correlation score and the sample label of the first training sample; and obtain the loss corresponding to the first training sample based on the first loss and the second loss.

[0029] In one possible implementation, when the processing module obtains the first loss based on the first eigenvector, the second eigenvector and the sample label of the first training sample, it is specifically used to: determine a second correlation score based on the first eigenvector and the second eigenvector, the second correlation score being used to characterize the degree of matching between at least one contextual feature and at least one advertising feature; and obtain the first loss based on the second correlation score and the sample label of the first training sample.

[0030] In a possible implementation, the processing module is further configured to: train a third model based on a loss corresponding to at least one training sample in the first sample set.

[0031] In one possible implementation, the third model is a neural network model, which is used to assist in calculating the losses corresponding to the training samples in the first sample set during the training of the first model and the second model.

[0032] In one possible implementation, the trained first model is used to extract features of contextual features related to pages visited by users, and the trained second model is used to extract features of advertisements.

[0033] In a fifth aspect, the present application provides an advertising delivery device that can be deployed on an electronic device with a display screen. The device may include: an acquisition module and a display module. The acquisition module is configured to acquire a user's access operation to a first page; and, in response to the access operation, acquire a first contextual feature related to the first page, wherein the contextual feature is unrelated to the user's user profile and / or user behavior. The display module is configured to display f advertisements based on the first contextual feature, where f ≥ 1.

[0034] In one possible implementation, after the display module displays f advertisements, the acquisition module is further configured to acquire a second user access operation to a second page and, in response to the second access operation, acquire a second contextual feature related to the second page, wherein the second contextual feature is unrelated to the user's user profile and / or user behavior. The display module is further configured to display g advertisements based on the second contextual feature, where g ≥ 1, and at least some of the g advertisements are different from the advertisements in the f advertisements.

[0035] In one possible implementation, the device further includes: a processing module for processing the context feature using a first model to obtain a first feature vector, wherein the first model is a neural network model, and the context feature is a first context feature or a second context feature. The processing module is further configured to determine the similarity value between the first feature vector and K second feature vectors to obtain K similarity values, wherein each second feature vector is used to characterize the characteristics of an advertisement. The processing module is further configured to filter out Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than the similarity values ​​of the K similarity values ​​other than the Q similarity values, and 1≤Q≤K. A display module is configured to display advertisements corresponding to each of the Q similarity values, wherein when the context feature is the first context feature, the number of advertisements corresponding to each of the Q similarity values ​​is f, and when the context feature is the second context feature, the number of advertisements corresponding to each of the Q similarity values ​​is g.

[0036] In one possible implementation, the apparatus further includes: a communication module configured to send a first message to a server, the first message including a context feature, where the context feature is either the first context feature or the second context feature. The communication module is further configured to receive a second message from the server, the second message including Q advertisements or advertisement indication information, the advertisement indication information being used to generate Q advertisements, where, when the context feature is the first context feature, the Q advertisements are f advertisements, and when the context feature is the second context feature, the Q advertisements are g advertisements. A display module is configured to display the Q advertisements.

[0037] In one possible implementation, the device further includes: a processing module for processing the context feature using a first model to obtain a first feature vector, wherein the first model is a neural network model, and the context feature is a first context feature or a second context feature. A communication module for sending a first message to a server, wherein the first message includes the first feature vector. The communication module is further configured to receive a second message sent by the server, wherein the second message includes Q advertisements or advertisement indication information, wherein the advertisement indication information is used to generate Q advertisements, wherein when the context feature is the first context feature, the Q advertisements are f advertisements, and when the context feature is the second context feature, the Q advertisements are g advertisements. A display module for displaying the Q advertisements.

[0038] In one possible implementation, the apparatus further includes: a communication module configured to send a first message to a server, the first message including a context feature, the first message instructing the server to extract features from the context feature, the context feature being either a first context feature or a second context feature. The communication module is further configured to receive a second message from the server, the second message including a first feature vector extracted by the server based on the context feature. A processing module is configured to determine similarities between the first feature vector and K second feature vectors to obtain K similarity values, each second feature vector representing a feature of an advertisement. The processing module is further configured to select Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than the similarity values ​​of the K similarity values ​​other than the Q similarity values, with 1≤Q≤K. A display module is configured to display advertisements corresponding to each of the Q similarity values. When the context feature is the first context feature, the number of advertisements corresponding to each of the Q similarity values ​​is f, and when the context feature is the second context feature, the number of advertisements corresponding to each of the Q similarity values ​​is g.

[0039] In one possible implementation, the first context feature includes one or more of the following: the identifier of the first page, the text in the first page, the identifier of the video in the first page, the identifier of the picture in the first page, the identifier of the audio in the first page, the user's IP regional information, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0040] In one possible implementation, the second context feature includes one or more of the following: the identifier of the second page, the text in the second page, the identifier of the video in the second page, the identifier of the picture in the second page, the identifier of the audio in the second page, the user's IP regional information, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0041] In a sixth aspect, the present application provides an advertising delivery device, which is deployed on a server, and the device includes: a communication module and a processing module. Among them, the communication module can be used to receive a first message sent by an electronic device, and the first message includes context features related to the page visited by the user on the electronic device, and the context features are unrelated to the user profile and / or user behavior of the user. The processing module can be used to respond to the first message and, based on the context features, obtain R advertisements or information indicating R advertisements, where R≥1. The communication module can also be used to send a second message to the electronic device, and the second message is used to instruct the electronic device to display R advertisements.

[0042] In one possible implementation, when the processing module obtains R advertisements or information indicating R advertisements based on context features, it is specifically used to: process the context features using a first model to obtain a first feature vector, where the first model is a neural network model; determine the similarity value between the first feature vector and K second feature vectors to obtain K similarity values, where each second feature vector is used to characterize the characteristics of an advertisement; filter out Q similarity values ​​from the K similarity values, where each of the Q similarity values ​​is greater than the similarity values ​​of the K similarity values ​​other than the Q similarity values, 1≤Q≤K; and use the advertisements corresponding to each of the Q similarity values ​​as R advertisements, or use the information indicating the advertisements corresponding to each of the Q similarity values ​​as information indicating the R advertisements.

[0043] In one possible implementation, the contextual features include one or more of the following: page identification, text in the page, identification of the video in the page, identification of the image in the page, identification of the audio in the page, user's IP region information, access time, device type of the electronic device, operator providing network services for the electronic device, network type of the electronic device, system language of the electronic device, brand of the electronic device, model of the electronic device, scenario in which the electronic device needs to display advertisements, size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0044] In the seventh aspect, the present application provides an electronic device comprising: a display screen; at least one memory for storing programs; and at least one processor for executing the programs stored in the memory; wherein, when the program stored in the memory is executed, the processor is used to execute the method described in the second aspect or any possible implementation of the second aspect.

[0045] In an eighth aspect, the present application provides a server comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory; wherein, when the program stored in the memory is executed, the processor is used to execute the method described in the third aspect or any possible implementation of the third aspect.

[0046] In a ninth aspect, the present application provides a computer-readable storage medium storing a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect, or executes the method described in the second aspect or any possible implementation of the second aspect, or executes the method described in the third aspect or any possible implementation of the third aspect.

[0047] In a tenth aspect, the present application provides a computer program product. When the computer program product runs on a processor, it enables the processor to execute the method described in the first aspect or any possible implementation of the first aspect, or execute the method described in the second aspect or any possible implementation of the second aspect, or execute the method described in the third aspect or any possible implementation of the third aspect.

[0048] It can be understood that the beneficial effects of the third to tenth aspects mentioned above can be found in the relevant descriptions of the first to second aspects mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] FIG1 is a schematic diagram of a process of an advertisement recall solution provided in an embodiment of the present application;

[0050] FIG2 is a schematic diagram of an architecture of a user representation model and an advertisement representation model obtained through dual-tower model training according to an embodiment of the present application;

[0051] FIG3 is a schematic diagram of an architecture of a context representation model and an advertisement representation model obtained through training of a dual-tower model according to an embodiment of the present application;

[0052] FIG4 is a schematic diagram of another advertising recall process provided by an embodiment of the present application;

[0053] FIG5 is a flow chart of a model training method provided in an embodiment of the present application;

[0054] FIG6 is a flow chart of an advertisement delivery method according to an embodiment of the present application;

[0055] FIG7 is a schematic diagram of steps for displaying advertisements based on contextual features according to an embodiment of the present application;

[0056] FIG8 is a comparative diagram of an embodiment of the present application providing a method for delivering advertisements to a user browsing the same content using different mobile phones at the same time;

[0057] FIG9 is a comparative diagram of advertising delivery to different users provided by an embodiment of the present application;

[0058] FIG10 is a comparative diagram of an embodiment of the present application providing a method for delivering advertisements to a user browsing different contents at different times using the same mobile phone;

[0059] FIG11 is a schematic structural diagram of a model training device provided in an embodiment of the present application;

[0060] FIG12 is a schematic structural diagram of an advertising delivery device provided in an embodiment of the present application;

[0061] FIG13 is a schematic structural diagram of an advertising delivery device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship, for example, A / B means either A or B.

[0063] The terms "first" and "second" in this specification and claims are used to distinguish different objects rather than to describe a specific order of objects. For example, "first response message" and "second response message" are used to distinguish different response messages rather than to describe a specific order of response messages.

[0064] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0065] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, for example, multiple processing units means two or more processing units, etc.; multiple elements means two or more elements, etc.

[0066] For example, in order to achieve precise advertising delivery to users, a representation model related to the user portrait can be trained based on the characteristic data of the user portrait and user behavior, and the model can be used to recall the advertisements most relevant to the user to achieve precise advertising delivery. In order to complete the advertising representation and recall technology based on user portraits, it is generally possible to first train the characteristic data of the user side and the advertising side based on a dual-tower structure (user tower, advertising tower) to obtain a representation model on the user side (hereinafter referred to as the "user representation model") and a representation model on the advertising side (hereinafter referred to as the "advertising representation model"). Then, the user representation model and the advertising representation model are used to achieve precise advertising delivery to users.

[0067] Exemplarily, FIG1 shows an advertisement recall process based on user portrait and user behavior. As shown in FIG1 , the process includes two stages, an offline stage and an online stage. In the offline stage, the advertisement representation model is used to process a set of features of each advertisement (such as title, category, brand, label, text content, image content, or video content, etc.) to obtain a set of feature vectors of each advertisement (hereinafter referred to as "advertising side feature vector"). In the online stage, after obtaining the user portrait and user behavior, the user's portrait features (such as gender, age, or province, etc.) and the user's behavior features (such as the category and title of the content clicked) are determined; then, the user representation model is used to process the user's portrait features and behavior features online in real time to obtain the user's feature vector (hereinafter referred to as "user side feature vector").

[0068] After obtaining the user-side feature vector, the similarity between the user-side feature vector and each advertising-side feature vector can be calculated through algorithms such as cosine similarity, and advertisements corresponding to one or more advertising-side feature vectors with higher similarity can be selected for delivery, thereby obtaining advertisements to be delivered. For example, after sorting the similarities from large to small, the advertisements corresponding to the top 5 similarities can be used as advertisements to be delivered, thereby recalling the required advertisements from a pre-set advertisement set.

[0069] The ad recall scheme shown in Figure 1 requires user profiles and / or user behavior. However, with increasing protection of user information, obtaining these profiles and / or user behavior is becoming increasingly difficult, making it difficult to use user representation models for targeted advertising, resulting in poor advertising effectiveness. For example, when user A is browsing the web on a mobile phone, if user A's registration information determines that their user profile is: gender "male," age "18 to 30 years old," and city "S City," and user A's actions determine that user behavior 1 is: browsing content categorized as "Automobile," titled "New Energy Vehicle Price Reduction," and browsing time "8:00 AM," and user behavior 2 is: browsing content categorized as "New Energy," titled "Research on New Energy Vehicles," and browsing time "8:20 AM," then advertisements for new energy vehicles suitable for male drivers aged 20 to 30 in S City can be recommended to user A. However, when user A's user data (such as user information and / or operational behavior, etc.) is protected, user A's user portrait and user behavior cannot be obtained, resulting in an inability to know what kind of advertisements to recommend to user A. In this case, it is very likely that some advertisements that user A is not interested in will be recommended, resulting in poor advertising effectiveness.

[0070] For example, Figure 2 shows a schematic diagram of an architecture for obtaining a user representation model and an advertisement representation model based on dual-tower model training. As shown in Figure 2, dual-tower model training device 200 includes a user tower device 210, an advertisement tower device 220, and a dual-tower loss device 230. The dual-tower model training device 200 can train both the user tower device 210 and the advertisement tower device 220. After training is completed by the dual-tower model training device 200, the user tower device 210 is converted into a user representation model 300 and output by the dual-tower model training device 200, and the advertisement tower device 220 is converted into an advertisement representation model 400 and output by the dual-tower model training device 200.

[0071] The user tower device 210 includes a feature encoding (embedding) device 211, a feature concatenation (concatenation) device 212, and a main network device 213. The feature encoding device 211 is used to encode each user profile feature and user behavior feature in the user feature dataset to represent the sparse original features as dense feature vectors. Exemplarily, the feature encoding device 211 may, but is not limited to, use algorithms such as word2vec or item2vec to encode the features in the user feature dataset. The feature concatenation device 212 is used to concatenate the feature vectors corresponding to the different user features encoded by the feature encoding device 211 into a single feature vector, and its function can also be understood as feature fusion. The main network device 213 is used to control the feature vectors concatenated by the feature concatenation device 212 to perform sufficient feature crosstalk between different dimensions to obtain more effective feature information. The feature vector output by the main network device 213 can be used to characterize a specific user. Exemplarily, the model structure of the main network device 213 can be, but is not limited to, deep neural networks (DNN), multi-layer perception (MLP), deep factorization machine (DeepFM), extreme deep factorization machine model (xDeepFM), deep cross network (DCN), or deep interest network (DIN) and other model structures.

[0072] The advertising tower device 220 includes a feature encoding (embedding) device 221, a feature concatenation (concatenate) device 222 and a main network device 223. The feature encoding device 221 is used to perform feature encoding on each advertising feature in the advertising feature data set to represent the sparse original features as dense feature vectors. The feature concatenation device 222 is used to splice the feature vectors corresponding to the different advertising features encoded by the feature encoding device 221 into one feature vector. The main network device 223 is used to control the feature vectors spliced ​​by the feature concatenation device 222 to perform sufficient feature crossover between different dimensions so as to obtain more effective feature information. Among them, the feature vector output by the main network device 223 can be used to characterize a certain advertisement. Exemplarily, the model structure of the main network device 223 can be, but is not limited to, the same as that of the main network device 213.

[0073] The dual-tower loss device 230 is a loss function in the dual-tower model training device 200 . It evaluates the training degree of the user tower device 210 and the advertising tower device 220 based on similarity comparison methods such as cosine similarity, and back-propagates to the user tower device 210 and the advertising tower device 220 .

[0074] During the process of training the user tower device 210 and the advertising tower device 220 using the dual-tower model training device 200, the model parameters in the user tower device 210 can be continuously adjusted based on the results output by the dual-tower loss device 230 to make the user feature vector representation output by the user tower device 210 more accurate. At the same time, the model parameters in the advertising tower device 220 can be continuously adjusted based on the results output by the dual-tower loss device 230 to make the advertisement feature vector representation output by the advertising tower device 220 more accurate. After the training is completed, the dual-tower model training device 200 can solidify the parameters of the user tower device 210 that achieved the best training effect during the training process to generate a user representation model 300 and output it. At the same time, the parameters of the advertising tower device 220 that achieved the best training effect during the training process can be solidified to generate an advertisement representation model 400 and output it. The user representation model 300 is used to represent online user features as feature vectors, and the advertisement representation model 400 is used to represent offline advertisement features as feature vectors.

[0075] As can be seen in Figure 2, the model training process also relies heavily on user features. Furthermore, during model training, there is no interaction between user tower device 210 and advertising tower device 220. This makes the feature association between user tower device 210 and advertising tower device 220 dependent on user behavior toward advertisements (i.e., user behavior, such as browsing web pages, watching videos, clicking on ads, paying for purchases, etc.). If user behavior cannot be obtained, model training will be impossible, or the trained model will be unusable. Furthermore, during the loss calculation phase, the loss calculation relies solely on the similarity between vectors, which is too simplistic, resulting in a relatively poor representation effect for the resulting model.

[0076] In view of this, an embodiment of the present application provides a model training method that can replace user features with contextual features, thereby reducing dependence on user features. At the same time, the method realizes interactive modeling between the user tower device 210 and the advertising tower device 220 by adding an auxiliary network to the dual-tower model training device 200, thereby realizing the interaction between contextual features and advertising features, enhancing the correlation between the two, solving the information interaction problem of dual-tower features, and improving the representation effect of the final model. Among them, contextual features may include: content context features related to advertising, and / or, environmental context features related to advertising. Exemplarily, content context features may refer to context records of the content contained in an advertisement when a user accesses the advertisement, such as: web page text, web page image identification, video identification, or audio identification, etc.; environmental context features may refer to relevant environmental context records when a user accesses an advertisement, such as: Internet Protocol (IP) regional information (i.e., the location of the IP address), access time, device type of the electronic device, operator providing network services to the electronic device, network type of the electronic device, system language of the electronic device, brand of the electronic device, model of the electronic device, scenario in which the electronic device needs to display advertisements, size of the area used to display advertisements in the display interface of the electronic device, country where the electronic device is located, etc.

[0077] For example, FIG3 shows an architectural diagram of a context representation model and an advertising representation model obtained through dual-tower model training. As shown in FIG3 , the dual-tower model training device 300 includes a user tower device 310, an advertising tower device 320, a dual-tower loss device 330, an auxiliary network model 340, and an auxiliary network loss device 350. The user tower device 310 can refer to the description of the user tower device 210 in FIG2 above, and replace the user feature dataset therein with the context feature dataset. That is, the data required by the user tower device 310 in FIG3 is the context feature, which will not be repeated here. The advertising tower device 320 can refer to the description of the advertising tower device 220 in FIG2 above, which will not be repeated here. The dual-tower loss device 330 can refer to the description of the dual-tower loss device 230 in FIG2 above, which will not be repeated here. Exemplarily, the context feature data set shown in FIG3 may include: content context features related to the advertisement, and / or environmental context features related to the advertisement. The advertisement feature data set may include: one or more of the advertisement's title, category, brand, label, text content, image content, or video content.

[0078] Auxiliary network model 340 is used to model the interaction between user tower device 310 and advertising tower device 320, thereby enabling interaction between contextual features and advertising features, enhancing their correlation. The inputs to auxiliary network model 310 can include: the output of user tower device 210 at at least one stage, and the output of advertising tower device 220 at at least one stage. For example, the inputs to auxiliary network model 310 can include the outputs of feature encoding devices 311 and 321, the outputs of feature connection devices 312 and 322, the outputs of primary network devices 313 and 323, the outputs of a network layer within primary network devices 313 and 323, or any combination of these, such as a combination of the outputs of feature encoding device 311 and primary network device 323. The specific method can be determined based on actual circumstances and is not limited here. The output of auxiliary network model 340 can be used to characterize the correlation between contextual features and advertising features. For example, the model structure of auxiliary network model 340 can be, but is not limited to, the same as the model structure of primary network devices 313 or 323. In some embodiments, the auxiliary network model 340 may be pre-trained or to-be-trained (i.e., untrained), depending on the actual situation and not limited herein. For example, when pre-training the auxiliary network model 340, the contextual feature data set and the advertising feature data set may be used as training data to train the auxiliary network model 340.

[0079] The auxiliary network loss device 350 is another loss function in the dual-tower model training device 300. It mainly evaluates the training degree of the user tower device 310, the advertising tower device 320 and the auxiliary network model 340 based on the output results of the auxiliary network model 340 and the sample labels of the pre-set training samples, and back-propagates to the user tower device 310, the advertising tower device 320 and the auxiliary network model 340.

[0080] During the training process of user tower device 310 and advertising tower device 320 using dual-tower model training device 300, model parameters in user tower device 310 can be continuously adjusted based on the results output by dual-tower loss device 330 and the results output by auxiliary network loss device 350 to ensure more accurate representation of the context feature vector output by user tower device 310. Simultaneously, model parameters in advertising tower device 320 can be continuously adjusted based on the results output by dual-tower loss device 330 and the results output by auxiliary network loss device 350 to ensure more accurate representation of the advertisement feature vector output by advertisement tower device 320. After training is complete, dual-tower model training device 300 can solidify the parameters of user tower device 310 that achieved the best training results during the training process to generate and output context representation model 500. Simultaneously, the parameters of advertising tower device 320 that achieved the best training results during the training process can be solidified to generate and output advertisement representation model 600. Context representation model 500 is used to represent online context features as feature vectors, while advertisement representation model 600 is used to represent offline advertisement features as feature vectors.

[0081] After obtaining the context representation model 500 and the advertisement representation model 600, FIG4 shows, for example, a schematic diagram of advertisement delivery using the context representation model 500 and the advertisement representation model 600. As shown in FIG4, the advertisement delivery process also includes two stages: an offline stage and an online stage. In the offline stage, the advertisement representation model 600 is used to process a set of features of each advertisement (e.g., title, category, brand, tag, text content, image content, or video content) to obtain a set of feature vectors for each advertisement (hereinafter referred to as "advertisement-side feature vectors"). In the online stage, when a user browses a webpage, context features related to the page visited by the user can be obtained with the user's authorization. Then, the context features are processed online in real time using the context representation model 500 to obtain a context feature vector. After obtaining the context feature vector, the similarity between the context feature vector and each advertisement-side feature vector can be calculated using algorithms such as cosine similarity, and the advertisement corresponding to the advertisement-side feature vector with the highest similarity is selected for delivery, thereby obtaining the advertisement to be delivered, thereby achieving accurate advertisement delivery without relying on user characteristics.

[0082] Next, based on the contents in Figures 3 and 4, a model training method and an advertising delivery method provided in an embodiment of the present application are introduced.

[0083] For example, FIG5 shows a model training method. It is understood that the method can be performed by any device, equipment, platform, or device cluster with computing and processing capabilities. As shown in FIG5, the model training method includes the following steps:

[0084] S510: Acquire a first sample set, where the first sample set includes a first training sample. The first training sample includes at least one advertisement feature and at least one context feature corresponding to the advertisement.

[0085] In this embodiment, the first sample set may include at least one training sample. This first sample set may be pre-set by the user; alternatively, it may be a collection of training samples collected from the user's operation logs on a terminal device (such as a mobile phone or computer). For example, the operation logs may be related to or unrelated to advertisements. Each training sample in the first sample set may include at least one advertisement feature and at least one contextual feature corresponding to the advertisement. Different training samples may correspond to different advertisements. For example, the first training sample may include the advertisement features of advertisement a and the contextual features corresponding to advertisement a, while the second training sample may include the advertisement features of advertisement b and the contextual features corresponding to advertisement b. For example, each advertisement feature may be any one of the advertisement's title, category, brand, tag, text content, image content, or video content. Each contextual feature may be a content contextual feature corresponding to the advertisement and / or an environmental contextual feature corresponding to the advertisement. Content contextual features may include a contextual record of the content contained in the advertisement when the user accesses the advertisement, such as webpage text, webpage image identifiers, video identifiers, or audio identifiers. Environmental context features may include contextual records of the environment related to when a user accesses an advertisement, such as: IP address information, access time, device type of the electronic device, the network service provider providing the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display the advertisement, the size of the area on the display interface of the electronic device used for displaying advertisements, the country in which the electronic device is located, etc. In some embodiments, the training samples in the first sample set may be positive samples or negative samples, depending on the actual situation.

[0086] Exemplarily, the set of context features contained in the first sample set can be understood as the context feature data set described in the aforementioned FIG. 3 , and the set of advertising features contained in the first sample set can be understood as the advertising feature data set described in the aforementioned FIG. 3 .

[0087] S520: Process the context features included in the first training sample using the first model to obtain a first feature vector, and process the advertisement features included in the first training sample using the second model to obtain a second feature vector.

[0088] In this embodiment, the contextual features included in the first training sample can be processed by a first model to obtain a first feature vector, and the advertising features included in the first training sample can be processed by a second model to obtain a second feature vector. For example, the first model can be understood as the user tower device 310 in FIG. 3 , and the second model can be understood as the advertising tower device 320 in FIG. The first feature vector can be understood as the output of user tower device 310 after inputting the contextual features included in the first training sample into user tower device 310; the second feature vector can be understood as the output of advertising tower device 320 after inputting the advertising features included in the first training sample into advertising tower device 320.

[0089] In some embodiments, the context features contained in the first training sample can be encoded separately by the first model to obtain N feature vectors. Exemplarily, the context features and the feature vectors obtained by encoding can be in a one-to-one correspondence, or of course not in a one-to-one correspondence, which may be determined according to the actual situation. In the case of N=1, the feature vector obtained can be processed by the first model, such as performing sufficient feature crossover in different dimensions, etc., to obtain the first feature vector. In the case of N≥2, the N feature vectors obtained can be spliced ​​by the first model first to splice them into one feature vector, and then the feature vector obtained after splicing can be processed by the first model, such as performing sufficient feature crossover in different dimensions, etc., to obtain the first feature vector.

[0090] In some embodiments, the advertising features contained in the first training sample can be encoded separately by the second model to obtain M feature vectors. Exemplarily, there can be a one-to-one correspondence between the advertising features and the feature vectors obtained by encoding, but it can also be not a one-to-one correspondence, which can be determined according to the actual situation. In the case of M=1, the obtained feature vector can be processed by the second model, such as performing sufficient feature crossover in different dimensions, etc., to obtain a second feature vector. In the case of M≥2, the M feature vectors obtained can be first spliced ​​by the second model to splice them into one feature vector, and then the feature vector obtained after splicing can be processed by the second model, such as performing sufficient feature crossover in different dimensions, etc., to obtain a second feature vector.

[0091] S530. Process the processing results of at least one stage in the first model and the processing results of at least one stage in the second model through a third model to obtain a first correlation score, where the first correlation score is used to characterize the degree of matching between the context features included in the first training sample and the advertising features included in the first training sample.

[0092] In this embodiment, the processing results of at least one stage within the first model and the processing results of at least one stage within the second model can be processed by a third model to obtain a first correlation score. This first correlation score can be used to represent the degree of match between the contextual features included in the first training sample and the advertising features included in the first training sample. Exemplarily, the third model can be understood as the auxiliary network device 340 in Figure 3 . Exemplarily, the third model can be a neural network model that can be used to assist in calculating the loss corresponding to the training samples in the first sample set during the training of the first and second models.

[0093] In some embodiments, the processing result of at least one stage within the first model may refer to the output of any one or more neural network layers within the first model. When the first model is user tower device 310 in FIG. 3 , the processing result of at least one stage within the first model may refer to one or more of the output of feature encoding device 311, the output of feature connection device 312, and the output of master network device 313.

[0094] In some embodiments, the processing result of at least one stage within the second model may refer to the output of any one or more neural network layers within the second model. When the first model is advertising tower device 320 in FIG. 3 , the processing result of at least one stage within the second model may refer to one or more of the output of feature encoding device 321 , the output of feature connection device 322 , and the output of main network device 323 .

[0095] S540: Obtain a loss corresponding to the first training sample according to the first eigenvector, the second eigenvector, the first correlation score, and the sample label of the first training sample.

[0096] In this embodiment, after obtaining the first eigenvector, the second eigenvector and the first correlation score, the loss corresponding to the first training sample can be determined based on the first eigenvector, the second eigenvector, the first correlation score and the sample label of the first training sample.

[0097] In some embodiments, a first loss can be obtained based on the first eigenvector, the second eigenvector, and the sample label of the first training sample. Then, a second loss can be obtained based on the first correlation score and the sample label of the first training sample. Finally, the loss corresponding to the first training sample can be obtained based on the first loss and the second loss. The order of obtaining the first loss and the second loss can be determined based on actual conditions and is not limited here.

[0098] For the first loss, the first feature vector and the second feature vector can be processed based on, but not limited to, an algorithm such as cosine similarity to obtain a correlation score between the two. This correlation score can be used to characterize the degree of match between the contextual features contained in the first training sample and the advertising features contained in the first training sample. Subsequently, this correlation score and the sample label of the first training sample can be processed using a preset loss function to obtain the first loss. Exemplarily, the first loss can be obtained by processing the dual-tower loss device 330 described in FIG. 3 .

[0099] For the second loss, the first correlation score and the sample label of the first training sample can be processed by a preset loss function to obtain the second loss. Exemplarily, the second loss can be obtained by processing the auxiliary network loss device 350 described in FIG3 .

[0100] The loss corresponding to the first training sample can be obtained by adding the first loss and the second loss, or by weighted averaging or multiplying them. The specific method can be determined according to the actual situation and is not limited here.

[0101] The above description is for the processing of one training sample in the first sample set. For other training samples in the first sample set, you can also refer to the processing of the first training sample to obtain the corresponding losses of each training sample, which will not be described here one by one.

[0102] After determining the loss for each training sample in the first sample set, S550 may be executed. In some embodiments, S550 may be executed once for each loss obtained, or once for each batch of losses (e.g., two or three). The specific determination may depend on the actual situation and is not limited here. The loss described in S550 at this time is the loss obtained at that time.

[0103] S550: Train the first model and the second model based on the loss corresponding to at least one training sample in the first sample set.

[0104] In this embodiment, after obtaining the losses corresponding to one or more training samples, the first model and the second model can be trained with the goal of minimizing these losses. Of course, the third model can also be trained simultaneously. For example, after obtaining the losses corresponding to one or more training samples, the parameters of the first model and the second model can be updated using, but not limited to, a backpropagation algorithm to complete the training of the first model and the second model.

[0105] In some embodiments, the first model may be used to extract contextual features related to pages visited by the user, and the second model may be used to extract features of advertisements.

[0106] By using contextual features instead of user features for model training, the model's subsequent reliance on user features is reduced. Furthermore, by adding a third model to the training of the first and second models, interactive modeling between the first and second models is achieved, thereby enabling interaction between contextual features and advertising features, strengthening their correlation and improving the representational effectiveness of the resulting first and second models.

[0107] After the first model and the second model are trained, these two models can be used for advertising delivery. The trained first model can be understood as the context representation model 500 described in FIG3 , and the trained second model can be understood as the advertisement representation model 600 described in FIG3 .

[0108] For example, FIG6 shows a method for delivering advertisements. It is understood that the method can be performed by any device, equipment, platform, or device cluster with computing and processing capabilities. For example, the method can be performed by an electronic device with a display screen. As shown in FIG6 , the method for delivering advertisements includes the following steps:

[0109] S610: Obtain the user's access operation on the first page.

[0110] In this embodiment, when a user browses a certain page on an electronic device or issues an instruction to access a certain page, the electronic device can obtain the user's access operation to the page.

[0111] For example, if a user uses a mobile phone as an electronic device, and the user enters a keyword and clicks "search" in the browser configured on the phone, the browser interface presented by the user may be the first page visited by the user. In this case, the user's click "search" can be understood as an access operation. Alternatively, if the user selects content in the browser interface and clicks "view" that content, the browser interface presented by the user may be the first page visited by the user. In this case, the user's click "browse" of the related content can be understood as an access operation.

[0112] S620: In response to an access operation by the user, obtain a first context feature related to the first page, where the first context feature is unrelated to the user profile and / or user behavior of the user.

[0113] In this embodiment, with the user's authorization, information about the page visited by the user can be actively obtained and analyzed to determine contextual features related to the content of the first page, such as the page's identifier, the identifiers of the text, video, image, audio, etc. on the page. And / or, with the user's authorization, contextual information about the environment in which the user accessed the first page can also be obtained to obtain contextual features related to the environment in which the first page was accessed, such as one or more of the user's IP address, access time, device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements on the display interface of the electronic device, the country in which the electronic device is located, etc. Exemplarily, different display interfaces on the electronic device can be used to display different content to the user. In this embodiment, the contextual features related to the content of the first page and / or the environment in which the first page was accessed are unrelated to the user's user profile and / or user behavior. Exemplarily, the first contextual features can include contextual features related to the content of the first page and / or contextual features related to the environment in which the first page was accessed. The first context feature is irrelevant to the user profile and / or user behavior of the user.

[0114] S630: Based on the first context feature, display f advertisements, where f≥1.

[0115] In this embodiment, after the first context feature is acquired, f advertisements may be displayed, where f≥1.

[0116] As a possible implementation, the context representation model 500 (hereinafter referred to as the "first model") trained in FIG3 may be configured in the electronic device. At this time, as shown in FIG7 , displaying f advertisements based on the first context feature may specifically include:

[0117] S6311. Use the first model to process the first context feature to obtain a first feature vector.

[0118] In this embodiment, after the context feature is obtained, the first context feature can be input into the first model obtained through the aforementioned model training, so as to process the context feature using the first model to obtain a first feature vector.

[0119] S6312: Determine similarity values ​​between the first eigenvector and K second eigenvectors to obtain K similarity values, where each second eigenvector is used to characterize a feature of an advertisement.

[0120] In this embodiment, the second model obtained through the aforementioned model training can be used to pre-process the features of the K ads to be delivered to obtain K second feature vectors. Each second feature vector corresponds to one ad. Each second feature vector is used to characterize the features of one ad. After obtaining the first feature vector, the similarity between the first feature vector and the K second feature vectors can be determined using an algorithm such as cosine similarity to obtain K similarity values.

[0121] S6313. Filter out Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than the similarity values ​​other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K.

[0122] In this embodiment, after obtaining K similarity values, these K similarity values ​​can be sorted, such as from large to small or from small to large, etc. Then, Q larger similarity values ​​are selected from them. Each of these Q similarity values ​​is greater than the similarity values ​​of the K similarity values ​​other than the Q similarity values, 1≤Q≤K. Exemplarily, the value of Q can be, but is not limited to, pre-set. For example, if Q is equal to 2, K=4, and the K similarity values ​​are: 2, 4, 6, 7, then the filtered similarity values ​​are 6 and 7. Exemplarily, S6311 to S6313 can be understood as a process of obtaining Q advertisements or obtaining information for indicating Q advertisements (i.e., obtaining Q advertisement indication information) based on contextual features. After filtering out the Q similarity values, the advertisements corresponding to each of the Q similarity values ​​can be used as the advertisements to be delivered, or the information indicating the advertisements corresponding to each of the Q similarity values ​​can be used as the information indicating the advertisements to be delivered. The number of advertisements required to be delivered is the same as the Q value, and the amount of advertisement information required to be delivered is also the same as the Q value. For example, the number of advertisements required to be delivered or the amount of advertisement information required to be delivered can also be expressed by other numbers, such as k, R, g, etc., but these values ​​are the same as Q in the Q similarity values ​​screened out.

[0123] S6314: Display advertisements corresponding to the Q similarity values, where the number of advertisements corresponding to the Q similarity values ​​is f.

[0124] In this embodiment, after the Q similarity values ​​are screened, the advertisements corresponding to the Q similarity values ​​can be used as the advertisements to be delivered this time, and these advertisements can be displayed so that the user can browse the advertisements. The advertisements corresponding to the Q similarity values ​​are the f advertisements described in S630.

[0125] This results in displaying f advertisements based on the first context feature.

[0126] As another possible implementation, steps S6311 to S6313 in FIG. 7 can be executed by another device, such as a server. For example, in this case, after obtaining the first context feature, the electronic device can send a message containing the first context feature to the server. After receiving the message sent by the electronic device, the server can execute steps S6311 to S6313 in FIG. After executing S6313, the server can send a message to the electronic device instructing it to display advertisements corresponding to the Q similarity values. For example, the message sent by the server to the electronic device can include advertisements corresponding to the Q similarity values ​​(i.e., Q advertisements), or advertisement indication information for generating the Q advertisements. After receiving the message sent by the server, the electronic device can display the Q advertisements. For example, the advertisement indication information can include elements, templates, template identifiers, and the like of the Q advertisements. In this implementation, the electronic device may or may not be configured with the first model, which is not limited here. For example, the advertisement indication information can also be referred to as information for indicating advertisements.

[0127] As another possible implementation, S6312 and S6313 in FIG. 7 can be executed by another device, such as a server. Taking server execution as an example, after obtaining the first feature vector, the electronic device can send a message containing the first feature vector to the server. After receiving the message sent by the electronic device, the server can execute S6312 and S6313 in FIG. 7 . After executing S6313, the server can send a message to the electronic device instructing the electronic device to display advertisements corresponding to the Q similarity values. For example, the message sent by the server to the electronic device can include advertisements corresponding to the Q similarity values ​​(i.e., Q advertisements), or advertisement indication information for generating the Q advertisements. When the electronic device receives the message sent by the server, the electronic device can display the Q advertisements. For example, the advertisement indication information can include elements, templates, template identifiers, and the like of the Q advertisements. In this implementation, the electronic device can be configured with a first model. For example, the advertisement indication information can also be referred to as information for indicating advertisements.

[0128] As another possible implementation, S6311 in FIG. 7 can be executed by other devices, such as a server. Taking server execution as an example, at this time, after obtaining the first context feature, the electronic device can send a message containing the first context feature to the server, and the message can be used to instruct the server to perform feature extraction on the first context feature. After the server obtains the message sent by the electronic device, it can execute S6311 in FIG. 7. After executing S6311, the server can send a message containing the first feature vector to the electronic device. Afterwards, the electronic device can execute S6312 to S6314 in FIG. 7.

[0129] Therefore, when placing advertisements, by using contextual features instead of user features, the dependence on user features is reduced. When user features (such as user portraits, user behaviors, etc.) cannot be obtained, the contextual features related to the pages visited by the user are used to accurately place advertisements, solving the problem of the game between the privacy protection of user information and the accurate placement of advertisements.

[0130] In some embodiments, after S630, when the user switches pages on the electronic device, the electronic device may also obtain a second access operation of the user to the second page. Then, the electronic device may obtain a second context feature related to the second page in response to the second access operation, wherein the second context feature is independent of the user's user profile and / or user behavior. Finally, the electronic device may display g advertisements based on the second context feature, g≥1, wherein at least a portion of the g advertisements are different from the advertisements in the f advertisements. Exemplarily, the second context feature may include one or more of the identifier of the second page, the text in the second page, the identifier of the video in the second page, the identifier of the picture in the second page, the identifier of the audio in the second page, the user's IP address, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements (such as a startup scenario or a scenario in which a certain application is used), the size of the area for displaying advertisements in the display interface of the electronic device, the country in which the electronic device is located, etc. In some embodiments, the execution steps after S630 can refer to the description in Figure 6 above, where the first page is replaced by the second page, the first context feature is replaced by the second context feature, and g advertisements are replaced by f advertisements, which will not be repeated here. In addition, for the process of displaying g advertisements based on the second context feature, refer to the description of displaying f advertisements based on the first context feature above, where the first context feature is replaced by the second context feature, the first feature vector is replaced by the second feature vector, and g advertisements are replaced by f advertisements, which will not be repeated here. Among them, the second context feature may include one or more of the following: the identifier of the second page, the text in the second page, the identifier of the video in the second page, the identifier of the picture in the second page, the identifier of the audio in the second page, the user's IP regional information, the access time, the type of electronic device used by the user, the operator and the networking method.

[0131] To facilitate understanding, the following example is used to illustrate.

[0132] As shown in Figure 8, user A carries two mobile phones, namely, mobile phone 1 and mobile phone 2. Mobile phone 1 uses the fourth generation mobile communication technology (4G) for networking, and mobile phone 2 uses the fifth generation mobile communication technology (5G) for networking. Mobile phone 1 and mobile phone 2 have different operators and are manufactured by different manufacturers. When using the advertising delivery method provided in this embodiment, when user A browses the same webpage using mobile phone 1 and mobile phone 2 at the same time, since the contextual features related to the content of the page when browsing the page on mobile phone 1 and the contextual features related to the content of the page when browsing the page on mobile phone 2 are the same, the contextual features related to the page when browsing the page on mobile phone 1 and mobile phone 2 are different. Therefore, when the user browses the same webpage using mobile phone 1 and mobile phone 2, the advertisements recommended to user A may be different. Continuing with Figure 8, the advertisement recommended to user A's mobile phone 1 may be: mid-range pure electric new energy vehicles; the advertisement recommended to user A's mobile phone 2 may be: high-end hybrid electric new energy vehicles.

[0133] In addition, when user A and user B use the same type of mobile phone, for example, the network connection mode, operator and manufacturer of the mobile phones used by both are the same, and visit the same page at the same time and place. Since the advertisement recommendation in this embodiment is based on the contextual features of the pages visited by the users, and the contextual features of the pages visited by both are the same, the same advertisement can be recommended to both. In contrast, when the user characteristics are used for advertisement recommendation, although the contextual features of the pages visited by both are the same, the advertisements recommended to the two are different because the user characteristics of the two are different. In addition, as shown in Figure 9, when the user portrait is turned off, when the content context of the page browsed by two users in different regions is consistent, but the environmental context related to the page is inconsistent, at the same time, two advertisements are stored in the advertisement pool. By analyzing the contextual features of the pages visited by the two users and matching them with the features of the advertisements in the advertisement pool, it can be determined that an advertisement for a mid-range and pure electric new energy vehicle is recommended to user A, and an advertisement for a high-end and hybrid new energy vehicle is recommended to user B.

[0134] In addition, when a user uses the same mobile phone to browse different pages at different times, the content contained in different pages is different. Therefore, when the user switches pages, the advertisements delivered can be changed synchronously. For example, as shown in Figure 10, user A browsed a car-related page at 8:10 am and browsed a game-related page at 8:11 am, that is, the user switched pages while using the mobile phone. There are two advertisements stored in the advertisement pool, Advertisement 1 is a game advertisement, and Advertisement 2 is a car advertisement. Since user A browsed car-related content at 8:10 am, a car-related advertisement, i.e. Advertisement 2, can be delivered at this time. Since user A browsed game-related content at 8:11 am, a game-related advertisement, i.e. Advertisement 1, can be delivered at this time.

[0135] It is understood that the order of execution of the steps in the above embodiments does not necessarily imply a specific order of execution. The order of execution of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. In addition, in some possible implementations, the steps in the above embodiments can be selectively executed according to actual circumstances, and can be executed partially or completely, which is not limited here.

[0136] Based on the method in the above embodiment, the embodiment of the present application also provides a model training device.

[0137] For example, FIG11 illustrates a model training device. As shown in FIG11 , the model training device 1100 includes an acquisition module 1110 and a processing module 1120. Acquisition module 1110 is configured to acquire a first sample set, which includes a first training sample. The first training sample includes at least one advertisement feature and at least one contextual feature corresponding to the advertisement. Processing module 1120 is configured to process the at least one contextual feature using a first model to obtain a first feature vector, and to process the at least one advertisement feature using a second model to obtain a second feature vector. Processing module 1120 is further configured to process the processing results of at least one stage within the first model and at least one stage within the second model using a third model to obtain a first relevance score, which indicates the degree of match between the at least one contextual feature and the at least one advertisement feature. Processing module 1120 is further configured to obtain a loss corresponding to the first training sample based on the first feature vector, the second feature vector, the first relevance score, and the sample label of the first training sample. Processing module 1120 is further configured to train the first model and the second model based on the loss corresponding to the at least one training sample in the first sample set.

[0138] In some embodiments, when the processing module 1120 processes at least one context feature through the first model to obtain a first feature vector, it is specifically used to: encode at least one context feature through the first model to obtain N third feature vectors, N≥1; wherein, when N=1, the N third feature vectors are processed through the first model to obtain the first feature vector; when N≥2, the N third feature vectors are spliced ​​through the first model to obtain a fourth feature vector, and the fourth feature vector is processed through the first model to obtain the first feature vector.

[0139] In some embodiments, when the processing module 1120 processes at least one advertising feature through the second model to obtain a second feature vector, it is specifically used to: encode at least one advertising feature through the second model to obtain M fifth feature vectors, M≥1; wherein, when M=1, the M fifth feature vectors are processed through the second model to obtain the second feature vector; when M≥2, the M fifth feature vectors are spliced ​​through the second model to obtain a sixth feature vector, and the sixth feature vector is processed through the second model to obtain the second feature vector.

[0140] In some embodiments, when the processing module 1120 obtains the loss corresponding to the first training sample based on the first eigenvector, the second eigenvector, the first correlation score and the sample label of the first training sample, it is specifically used to: obtain the first loss based on the first eigenvector, the second eigenvector and the sample label of the first training sample; obtain the second loss based on the first correlation score and the sample label of the first training sample; obtain the loss corresponding to the first training sample based on the first loss and the second loss.

[0141] In some embodiments, when the processing module 1120 obtains the first loss based on the first feature vector, the second feature vector and the sample label of the first training sample, it is specifically used to: determine a second correlation score based on the first feature vector and the second feature vector, the second correlation score being used to characterize the degree of matching between at least one contextual feature and at least one advertising feature; and obtain the first loss based on the second correlation score and the sample label of the first training sample.

[0142] In some embodiments, the processing module 1120 is further configured to: train a third model based on the loss corresponding to at least one training sample in the first sample set.

[0143] In some embodiments, the trained first model is used to extract features of contextual features related to pages visited by users, and the trained second model is used to extract features of advertisements.

[0144] In some embodiments, the third model is a neural network model, which is used to assist in calculating the losses corresponding to the training samples in the first sample set during the training of the first model and the second model.

[0145] It should be understood that the above-mentioned device is used to execute the model training method described in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0146] Based on the method in the above embodiment, the embodiment of the present application further provides an advertisement delivery device.

[0147] For example, FIG12 shows an advertising delivery device. The advertising delivery device can be deployed in an electronic device with a display screen. As shown in FIG12 , the advertising delivery device 1200 includes: an acquisition module 1210 and a display module 1220. The acquisition module is used to obtain a user's access operation to a first page; and, in response to the access operation, obtain a first context feature related to the first page, wherein the context feature is unrelated to the user's user profile and / or user behavior. The display module 1220 is used to display f advertisements based on the first context feature, where f≥1.

[0148] In some embodiments, after display module 1220 displays f advertisements, acquisition module 1210 is further configured to acquire a second user access operation to a second page, and, in response to the second access operation, acquire a second contextual feature related to the second page, wherein the second contextual feature is unrelated to the user's user profile and / or user behavior. Display module 1220 is further configured to display g advertisements based on the second contextual feature, where g ≥ 1, and at least some of the g advertisements are different from the advertisements in the f advertisements.

[0149] In some embodiments, the device further includes: a processing module (not shown in the figure), which is used to process the context feature using a first model to obtain a first feature vector, wherein the first model is a neural network model, and the context feature is a first context feature or a second context feature. The processing module is also used to determine the similarity value between the first feature vector and K second feature vectors to obtain K similarity values, each of which is used to characterize the characteristics of an advertisement. The processing module is also used to filter out Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than the similarity value of the K similarity values ​​other than the Q similarity values, and 1≤Q≤K. The display module 1220 is used to display the advertisements corresponding to each of the Q similarity values, wherein when the context feature is the first context feature, the advertisements corresponding to each of the Q similarity values ​​are f advertisements, and when the context feature is the second context feature, the advertisements corresponding to each of the Q similarity values ​​are g advertisements.

[0150] In some embodiments, the device further includes: a communication module (not shown) configured to send a first message to a server, the first message including a context feature, where the context feature is either a first context feature or a second context feature. The communication module is further configured to receive a second message from the server, the second message including Q advertisements or advertisement indication information, the advertisement indication information being used to generate Q advertisements, where, when the context feature is the first context feature, the Q advertisements are f advertisements, and when the context feature is the second context feature, the Q advertisements are g advertisements. A display module 1220 is configured to display the Q advertisements.

[0151] In some embodiments, the device further includes: a processing module (not shown in the figure), which is used to process the context feature using a first model to obtain a first feature vector, the first model is a neural network model, and the context feature is a first context feature or a second context feature. A communication module is used to send a first message to the server, and the first message includes a first feature vector. The communication module is also used to receive a second message sent by the server, and the second message includes Q advertisements or advertisement indication information, and the advertisement indication information is used to generate Q advertisements, wherein when the context feature is the first context feature, the Q advertisements are f advertisements, and when the context feature is the second context feature, the Q advertisements are g advertisements. Display module 1220 is used to display Q advertisements.

[0152] In some embodiments, the device further includes: a communication module (not shown) configured to send a first message to a server, the first message including a context feature, the first message instructing the server to extract features from the context feature, the context feature being either a first context feature or a second context feature. The communication module is further configured to receive a second message from the server, the second message including a first feature vector extracted by the server based on the context feature. A processing module (not shown) is configured to determine similarities between the first feature vector and K second feature vectors to obtain K similarity values, each of which is used to characterize an advertisement. The processing module is further configured to select Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than the similarity values ​​of the K similarity values ​​other than the Q similarity values, with 1≤Q≤K. A display module 1220 is configured to display advertisements corresponding to each of the Q similarity values. When the context feature is the first context feature, the number of advertisements corresponding to each of the Q similarity values ​​is f, and when the context feature is the second context feature, the number of advertisements corresponding to each of the Q similarity values ​​is g.

[0153] In some embodiments, the first context feature includes one or more of the following: the identification of the first page, the text in the first page, the identification of the video in the first page, the identification of the picture in the first page, the identification of the audio in the first page, the user's IP regional information, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0154] In some embodiments, the second contextual feature includes one or more of the following: identification of the second page, text in the second page, identification of the video in the second page, identification of the picture in the second page, identification of the audio in the second page, user's IP regional information, access time, device type of the electronic device, operator providing network services for the electronic device, network type of the electronic device, system language of the electronic device, brand of the electronic device, model of the electronic device, scenario in which the electronic device needs to display advertisements, size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

[0155] It should be understood that the above-mentioned device is used to execute the advertising delivery method described in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0156] Based on the method in the above embodiment, the embodiment of the present application further provides an advertisement delivery device.

[0157] Exemplarily, FIG13 shows an advertisement delivery device. The advertisement delivery device can be deployed in a server. As shown in FIG13 , the advertisement delivery device 1300 includes: a communication module 1310 and a processing module 1320. Among them, the communication module 1310 can be used to receive a first message sent by an electronic device, and the first message includes context features related to the page visited by the user on the electronic device, and the context features are unrelated to the user's user profile and / or user behavior. The processing module 1320 can be used to respond to the first message and, based on the context features, obtain R advertisements or information indicating R advertisements, where R≥1. The communication module 1310 can also be used to send a second message to the electronic device, and the second message is used to instruct the electronic device to display R advertisements.

[0158] In some embodiments, when the processing module 1320 obtains R advertisements or information indicating R advertisements based on context features, it is specifically used to: process the context features using a first model to obtain a first feature vector, where the first model is a neural network model; determine the similarity value between the first feature vector and K second feature vectors to obtain K similarity values, where each second feature vector is used to characterize the characteristics of an advertisement; filter out Q similarity values ​​from the K similarity values, where each of the Q similarity values ​​is greater than the similarity values ​​of the K similarity values ​​other than the Q similarity values, 1≤Q≤K; use the advertisements corresponding to each of the Q similarity values ​​as R advertisements, or use the information indicating the advertisements corresponding to each of the Q similarity values ​​as information indicating the R advertisements.

[0159] In some embodiments, the contextual features include one or more of the following: page identification, text in the page, identification of the video in the page, identification of the picture in the page, identification of the audio in the page, user's IP region information, access time, type of electronic device used by the user, operator and networking method.

[0160] It should be understood that the above-mentioned device is used to execute the advertising delivery method described in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0161] Based on the method in the above embodiment, an embodiment of the present application provides an electronic device. The electronic device may include: a display screen; at least one memory for storing a program; and at least one processor for executing the program stored in the memory. Wherein, when the program stored in the memory is executed, the processor is used to execute the method described in the above embodiment. Exemplarily, the electronic device may be a mobile phone, a tablet computer, a desktop computer, a laptop computer, a handheld computer, a notebook computer, a server, an ultra-mobile personal computer (UMPC), a netbook, a cellular phone, a personal digital assistant (PDA), an augmented reality (AR) device, a virtual reality (VR) device, an artificial intelligence (AI) device, a wearable device, a vehicle-mounted device, a smart home device and / or a smart city device. The embodiment of the present application does not impose any special restrictions on the specific type of the electronic device.

[0162] Based on the methods in the above embodiments, embodiments of the present application provide a server. The server may include: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the programs stored in the memory are executed, the processor is configured to execute the methods described in the above embodiments. Exemplarily, the server may be, but is not limited to, a cloud server, a virtual machine, a hardware server, and the like. Embodiments of the present application do not impose any particular restrictions on the specific type of the server.

[0163] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0164] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0165] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0166] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0167] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)).

[0168] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

Claims

1. A model training method, characterized in that: The method comprises: Acquire a first sample set, where the first sample set includes a first training sample, and the first training sample includes at least one advertisement feature and at least one context feature corresponding to the advertisement; Processing the at least one context feature through a first model to obtain a first feature vector, and processing the at least one advertisement feature through a second model to obtain a second feature vector; Processing the processing results of at least one stage in the first model and the processing results of at least one stage in the second model using a third model to obtain a first relevance score, where the first relevance score is used to indicate a degree of match between the at least one contextual feature and the at least one advertisement feature; Obtaining a loss corresponding to the first training sample according to the first eigenvector, the second eigenvector, the first correlation score, and the sample label of the first training sample; The first model and the second model are trained based on the loss corresponding to at least one training sample in the first sample set.

2. The method according to claim 1, characterized in that The processing of the at least one context feature by the first model to obtain a first feature vector specifically includes: Encoding the at least one context feature using the first model to obtain N third feature vectors, where N≥1; Wherein, when N=1, the N third eigenvectors are processed by the first model to obtain the first eigenvector; In the case of N≥2, the N third eigenvectors are concatenated using the first model to obtain a fourth eigenvector, and the fourth eigenvector is processed using the first model to obtain the first eigenvector.

3. The method according to claim 1 or 2, characterized in that The processing of the at least one advertisement feature by the second model to obtain a second feature vector specifically includes: Encoding the at least one advertisement feature using the second model to obtain M fifth feature vectors, where M≥1; Wherein, in the case of M=1, the M fifth eigenvectors are processed by the second model to obtain the second eigenvector; In the case of M≥2, the M fifth eigenvectors are concatenated using the second model to obtain a sixth eigenvector, and the sixth eigenvector is processed using the second model to obtain the second eigenvector.

4. The method according to any one of claims 1 to 3, characterized in that: Obtaining the loss corresponding to the first training sample according to the first eigenvector, the second eigenvector, the first correlation score, and the sample label of the first training sample specifically includes: Obtaining a first loss according to the first feature vector, the second feature vector, and the sample label of the first training sample; Obtaining a second loss according to the first relevance score and the sample label of the first training sample; A loss corresponding to the first training sample is obtained according to the first loss and the second loss.

5. The method according to claim 4, characterized in that Obtaining a first loss according to the first eigenvector, the second eigenvector, and the sample label of the first training sample specifically includes: determining a second relevance score based on the first feature vector and the second feature vector, wherein the second relevance score is used to represent a degree of matching between the at least one context feature and the at least one advertisement feature; The first loss is obtained according to the second correlation score and the sample label of the first training sample.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: The third model is trained based on the loss corresponding to at least one training sample in the first sample set.

7. The method according to any one of claims 1 to 6, characterized in that: The third model is a neural network model, which is used to assist in calculating the losses corresponding to the training samples in the first sample set during the training of the first model and the second model.

8. The method according to any one of claims 1 to 7, characterized in that: The trained first model is used to extract features of contextual features related to pages visited by users, and the trained second model is used to extract features of advertisements.

9. An advertisement delivery method, characterized in that: Applied to an electronic device having a display screen, the method includes: Obtaining the user's first access operation on the first page; In response to the first access operation, obtaining a first context feature related to the first page, wherein the first context feature is unrelated to the user profile and / or user behavior of the user; Based on the first context feature, f advertisements are displayed, where f≥1.

10. The method according to claim 9, characterized in that After displaying f advertisements, the method further includes: Obtaining a second access operation of the user to the second page; In response to the second access operation, obtaining a second context feature related to the second page, wherein the second context feature is unrelated to the user profile and / or user behavior of the user; Based on the second context feature, g advertisements are displayed, where g≥1, wherein at least some of the g advertisements are different from advertisements in the f advertisements.

11. The method according to claim 9 or 10, characterized in that Display ads based on contextual characteristics, including: Processing the context feature using a first model to obtain a first feature vector, where the first model is a neural network model, and the context feature is the first context feature or the second context feature; Determining similarity values ​​between the first feature vector and K second feature vectors to obtain K similarity values, each second feature vector being used to characterize a feature of an advertisement; Filtering Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K; The advertisements corresponding to the Q similarity values ​​are displayed, wherein when the context feature is the first context feature, the advertisements corresponding to the Q similarity values ​​are the f advertisements, and when the context feature is the second context feature, the advertisements corresponding to the Q similarity values ​​are the g advertisements.

12. The method according to claim 9 or 10, characterized in that Display ads based on contextual characteristics, including: Sending a first message to a server, where the first message includes the context feature, and the context feature is the first context feature or the second context feature; receiving a second message sent by the server, the second message including Q advertisements or advertisement indication information, the advertisement indication information being used to generate the Q advertisements, wherein when the context feature is the first context feature, the Q advertisements are the f advertisements, and when the context feature is the second context feature, the Q advertisements are the g advertisements; The Q advertisements are displayed.

13. The method according to claim 9 or 10, characterized in that Display ads based on contextual characteristics, including: Processing the context feature using a first model to obtain a first feature vector, where the first model is a neural network model, and the context feature is the first context feature or the second context feature; Sending a first message to a server, where the first message includes the first feature vector; receiving a second message sent by the server, the second message including Q advertisements or advertisement indication information, the advertisement indication information being used to generate the Q advertisements, wherein when the context feature is the first context feature, the Q advertisements are the f advertisements, and when the context feature is the second context feature, the Q advertisements are the g advertisements; The Q advertisements are displayed.

14. The method according to claim 9 or 10, characterized in that Display ads based on contextual characteristics, including: Sending a first message to a server, where the first message includes the context feature, and the first message is used to instruct the server to extract the context feature, where the context feature is the first context feature or the second context feature; receiving a second message sent by the server, where the second message includes a first feature vector extracted by the server based on the context feature; Determining similarity values ​​between the first feature vector and K second feature vectors to obtain K similarity values, each second feature vector being used to characterize a feature of an advertisement; Filtering Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K; The advertisements corresponding to the Q similarity values ​​are displayed, wherein when the context feature is the first context feature, the advertisements corresponding to the Q similarity values ​​are the f advertisements, and when the context feature is the second context feature, the advertisements corresponding to the Q similarity values ​​are the g advertisements.

15. The method according to any one of claims 9 to 14, characterized in that: The first context feature includes one or more of the following: the identifier of the first page, the text in the first page, the identifier of the video in the first page, the identifier of the picture in the first page, the identifier of the audio in the first page, the IP regional information of the user, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

16. The method according to any one of claims 10 to 15, characterized in that: The second context feature includes one or more of the following: the identifier of the second page, the text in the second page, the identifier of the video in the second page, the identifier of the picture in the second page, the identifier of the audio in the second page, the IP regional information of the user, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

17. An advertisement delivery method, characterized in that: Applied to a server, the method includes: Receiving a first message sent by an electronic device, the first message including contextual features related to a page visited by a user on the electronic device, the contextual features being unrelated to a user profile and / or user behavior of the user; In response to the first message, obtaining R advertisements or information indicating R advertisements based on the context feature, where R≥1; A second message is sent to the electronic device, where the second message is used to instruct the electronic device to display the R advertisements.

18. The method according to claim 17, characterized in that The method further comprises: Processing the context feature using a first model to obtain a first feature vector, where the first model is a neural network model; Determining similarity values ​​between the first feature vector and K second feature vectors to obtain K similarity values, each second feature vector being used to characterize a feature of an advertisement; Filtering Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K; The advertisements corresponding to the Q similarity values ​​are used as the R advertisements, or the information indicating the advertisements corresponding to the Q similarity values ​​are used as the information for indicating the R advertisements.

19. The method according to claim 17 or 18, characterized in that The context features include one or more of the following: the identification of the page, the text in the page, the identification of the video in the page, the identification of the picture in the page, the identification of the audio in the page, the IP regional information of the user, the access time, the type of electronic device used by the user, the operator and the networking method.

20. A model training device, characterized in that: The device comprises: An acquisition module, configured to acquire a first sample set, wherein the first sample set includes a first training sample, and the first training sample includes at least one advertisement feature and at least one context feature corresponding to the advertisement; a processing module, configured to process the at least one context feature using a first model to obtain a first feature vector, and to process the at least one advertisement feature using a second model to obtain a second feature vector; The processing module is further configured to process the processing results of at least one stage in the first model and the processing results of at least one stage in the second model using a third model to obtain a first relevance score, wherein the first relevance score is used to represent a degree of matching between the at least one context feature and the at least one advertisement feature; The processing module is further configured to obtain a loss corresponding to the first training sample based on the first eigenvector, the second eigenvector, the first correlation score, and the sample label of the first training sample; The processing module is further used to train the first model and the second model based on the loss corresponding to at least one training sample in the first sample set.

21. The device according to claim 20, characterized in that When the processing module processes the at least one context feature through the first model to obtain the first feature vector, the processing module is specifically configured to: Encoding the at least one context feature using the first model to obtain N third feature vectors, where N≥1; Wherein, when N=1, the N third eigenvectors are processed by the first model to obtain the first eigenvector; In the case of N≥2, the N third eigenvectors are concatenated using the first model to obtain a fourth eigenvector, and the fourth eigenvector is processed using the first model to obtain the first eigenvector.

22. The device according to claim 20 or 21, characterized in that When the processing module processes the at least one advertisement feature using the second model to obtain a second feature vector, the processing module is specifically configured to: Encoding the at least one advertisement feature using the second model to obtain M fifth feature vectors, where M≥1; Wherein, in the case of M=1, the M fifth eigenvectors are processed by the second model to obtain the second eigenvector; In the case of M≥2, the M fifth eigenvectors are concatenated using the second model to obtain a sixth eigenvector, and the sixth eigenvector is processed using the second model to obtain the second eigenvector.

23. The device according to any one of claims 20 to 22, characterized in that: When the processing module obtains the loss corresponding to the first training sample according to the first eigenvector, the second eigenvector, the first correlation score, and the sample label of the first training sample, the processing module is specifically configured to: Obtaining a first loss according to the first feature vector, the second feature vector, and the sample label of the first training sample; Obtaining a second loss according to the first relevance score and the sample label of the first training sample; A loss corresponding to the first training sample is obtained according to the first loss and the second loss.

24. The device according to claim 23, characterized in that When the processing module obtains the first loss according to the first feature vector, the second feature vector, and the sample label of the first training sample, the processing module is specifically configured to: determining a second relevance score based on the first feature vector and the second feature vector, wherein the second relevance score is used to represent a degree of matching between the at least one context feature and the at least one advertisement feature; The first loss is obtained according to the second correlation score and the sample label of the first training sample.

25. The device according to any one of claims 20 to 24, characterized in that The processing module is further configured to: The third model is trained based on the loss corresponding to at least one training sample in the first sample set.

26. The device according to any one of claims 20 to 25, characterized in that The third model is a neural network model, which is used to assist in calculating the losses corresponding to the training samples in the first sample set during the training of the first model and the second model.

27. The device according to any one of claims 20 to 26, characterized in that The trained first model is used to extract features of contextual features related to pages visited by users, and the trained second model is used to extract features of advertisements.

28. An advertisement delivery device, characterized in that: Deployed in an electronic device with a display screen, the device includes: An acquisition module, configured to acquire a first access operation of a user to a first page; The acquisition module is further configured to acquire, in response to the first access operation, a first context feature related to the first page, wherein the first context feature is unrelated to the user profile and / or user behavior of the user; The display module is configured to display f advertisements based on the first context feature, where f≥1.

29. The device according to claim 28, characterized in that After the display module displays f advertisements, the acquisition module is further configured to acquire a second access operation of the user to a second page, and, in response to the second access operation, acquire a second context feature related to the second page, wherein the second context feature is unrelated to the user profile and / or user behavior of the user; The display module is further configured to display g advertisements based on the second context feature, where g≥1, and at least a portion of the g advertisements are different from advertisements in the f advertisements.

30. The device according to claim 28 or 29, characterized in that The device further comprises: a processing module, configured to process the context feature using a first model to obtain a first feature vector, wherein the first model is a neural network model, and the context feature is the first context feature or the second context feature; The processing module is further configured to determine a similarity value between the first feature vector and K second feature vectors to obtain K similarity values, wherein each second feature vector is configured to represent a feature of an advertisement; The processing module is further configured to select Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K; The display module is used to display the advertisements corresponding to the Q similarity values, wherein when the context feature is the first context feature, the advertisements corresponding to the Q similarity values ​​are the f advertisements, and when the context feature is the second context feature, the advertisements corresponding to the Q similarity values ​​are the g advertisements.

31. The device according to claim 28 or 29, characterized in that The device further comprises: a communication module, configured to send a first message to a server, where the first message includes the context feature, and the context feature is the first context feature or the second context feature; The communication module is further configured to receive a second message sent by the server, the second message including Q advertisements or advertisement indication information, the advertisement indication information being used to generate the Q advertisements, wherein when the context feature is the first context feature, the Q advertisements are the f advertisements, and when the context feature is the second context feature, the Q advertisements are the g advertisements; The display module is used to display the Q advertisements.

32. The device according to claim 28 or 29, characterized in that The device further comprises: a processing module, configured to process the context feature using a first model to obtain a first feature vector, wherein the first model is a neural network model, and the context feature is the first context feature or the second context feature; A communication module, configured to send a first message to a server, where the first message includes the first feature vector; The communication module is further configured to receive a second message sent by the server, the second message including Q advertisements or advertisement indication information, the advertisement indication information being used to generate the Q advertisements, wherein when the context feature is the first context feature, the Q advertisements are the f advertisements, and when the context feature is the second context feature, the Q advertisements are the g advertisements; The display module is used to display the Q advertisements.

33. The device according to claim 28 or 29, characterized in that The device further comprises: a communication module, configured to send a first message to a server, wherein the first message includes the context feature, and the first message is used to instruct the server to extract the context feature, where the context feature is the first context feature or the second context feature; The communication module is further configured to receive a second message sent by the server, where the second message includes a first feature vector extracted by the server based on the context feature; a processing module, configured to determine a similarity value between the first feature vector and K second feature vectors to obtain K similarity values, wherein each second feature vector is used to characterize a feature of an advertisement; The processing module is further configured to select Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K; The display module is used to display the advertisements corresponding to the Q similarity values, wherein when the context feature is the first context feature, the advertisements corresponding to the Q similarity values ​​are the f advertisements, and when the context feature is the second context feature, the advertisements corresponding to the Q similarity values ​​are the g advertisements.

34. The device according to any one of claims 28 to 33, characterized in that The first context feature includes one or more of the following: the identifier of the first page, the text in the first page, the identifier of the video in the first page, the identifier of the picture in the first page, the identifier of the audio in the first page, the IP regional information of the user, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

35. The device according to any one of claims 29 to 34, characterized in that The second context feature includes one or more of the following: the identifier of the second page, the text in the second page, the identifier of the video in the second page, the identifier of the picture in the second page, the identifier of the audio in the second page, the IP regional information of the user, the access time, the device type of the electronic device, the operator providing network services for the electronic device, the network type of the electronic device, the system language of the electronic device, the brand of the electronic device, the model of the electronic device, the scenario in which the electronic device needs to display advertisements, the size of the area for displaying advertisements in the display interface of the electronic device, and the country where the electronic device is located.

36. An advertising delivery device, characterized in that: Deployed on a server, the device includes: a communication module, configured to receive a first message sent by an electronic device, wherein the first message includes contextual features related to a page visited by a user on the electronic device, the contextual features being unrelated to a user profile and / or user behavior of the user; a processing module, configured to obtain, in response to the first message and based on the context feature, R advertisements or information indicating R advertisements, where R≥1; The communication module is further configured to send a second message to the electronic device, where the second message is configured to instruct the electronic device to display the R advertisements.

37. The device according to claim 36, characterized in that When the processing module obtains R advertisements or information indicating R advertisements based on the context feature, it is specifically configured to: Processing the context feature using a first model to obtain a first feature vector, where the first model is a neural network model; Determining similarity values ​​between the first feature vector and K second feature vectors to obtain K similarity values, each second feature vector being used to characterize a feature of an advertisement; Filtering Q similarity values ​​from the K similarity values, wherein each of the Q similarity values ​​is greater than a similarity value other than the Q similarity values ​​in the K similarity values, and 1≤Q≤K; The advertisements corresponding to the Q similarity values ​​are used as the R advertisements, or the information indicating the advertisements corresponding to the Q similarity values ​​are used as the information indicating the R advertisements.

38. The device according to claim 36 or 37, characterized in that The context features include one or more of the following: the identification of the page, the text in the page, the identification of the video in the page, the identification of the picture in the page, the identification of the audio in the page, the IP regional information of the user, the access time, the type of electronic device used by the user, the operator and the networking method.

39. An electronic device, characterized in that: include: Display screen; at least one memory for storing a program; at least one processor, configured to execute the program stored in the memory; When the program stored in the memory is executed, the processor is configured to execute the method according to any one of claims 9 to 16.

40. A server, characterized in that: include: at least one memory for storing a program; at least one processor, configured to execute the program stored in the memory; When the program stored in the memory is executed, the processor is configured to execute the method according to any one of claims 17 to 19.

41. A computer-readable storage medium storing a computer program, wherein when the computer program is executed on a processor, the processor is caused to execute the method according to any one of claims 1 to 19.

42. A computer program product, characterized in that When the computer program product is run on a processor, the processor is caused to perform the method according to any one of claims 1 to 19.

Images