Display information push method, device, electronic device and storage medium

Abstract

The present disclosure relates to a method, device, electronic device, and storage medium for pushing display information. The method comprises: obtaining a display image corresponding to the information to be recommended, a recommended object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended; obtaining image features of the display image, and performing feature screening processing on the image features based on the recommended object category to obtain a category gating feature; obtaining an interaction prediction result of a target user account with respect to the recommended information based on the interaction prediction feature and the category gating feature; and, if the interaction prediction result meets a preset condition, pushing the information to be recommended to the target user account and displaying the display image. The embodiments of the present disclosure can enrich the dimensions of the data used to predict interaction results, improve the accuracy of the interaction prediction results, and recommend information based on the interaction prediction result, effectively improving the accuracy of information recommendation.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device, and storage medium for pushing display information. Background Art

[0002] With the development of computer technology, deep learning technology based on neural networks and feature representation has been widely used in recommendation systems.

[0003] In related technologies, recommendation models are trained based on collected multimedia resource sample data, such as resource identifiers, and corresponding labels, such as resource click-through rates, are used as supervisory signals to constrain recommendation model training, so that the trained recommendation model can output indicator data for recommendation.

[0004] In related technologies, the dimensions of data used for information recommendation are relatively small, and the accuracy of information recommendation is relatively low. Summary of the Invention

[0005] The present disclosure provides a method, device, electronic device, and storage medium for display information push, which at least solves the problem of low accuracy of information recommendation in related technologies. The technical solution of the present disclosure is as follows:

[0006] According to a first aspect of an embodiment of the present disclosure, a method for pushing presentation information is provided, the method comprising:

[0007] Obtaining a display image corresponding to the information to be recommended, a recommendation object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended, wherein the interaction prediction feature represents a predicted interaction status of a target user account with the information to be recommended;

[0008] Obtain the image features of the displayed image, and perform feature screening on the image features based on the recommended object category to obtain category gating features; the category gating features represent the target user account's interaction prediction of the displayed image ；

[0009] Obtaining an interaction prediction result of the target user account for the information to be recommended based on the interaction prediction feature and the category gating feature;

[0010] When the interaction prediction result meets the preset conditions, the information to be recommended is pushed to the target user account, and the display image is displayed.

[0011] In some possible designs, the performing feature screening processing on the image features based on the recommended object category to obtain category-gated features includes:

[0012] Determining category features corresponding to the recommended object category;

[0013] Performing pooling processing on the image features to obtain pooled image features;

[0014] The pooled image feature, the category feature and the image feature are fused to obtain the category gated feature.

[0015] In some possible designs, fusing the pooled image feature, the category feature, and the image feature to obtain the category gated feature includes:

[0016] Splicing the pooled image features with the category features to obtain spliced ​​features;

[0017] performing feature selection processing on the splicing features to obtain selected splicing features;

[0018] The selected splicing features are fused with the image features to obtain the category gating features.

[0019] In some possible designs, the pooled image features include maximum pooling features and average pooling features, and the pooled image features are concatenated with the category features to obtain concatenated features, including:

[0020] Concatenate the category feature with the maximum pooling feature to obtain a first concatenated feature;

[0021] Concatenate the category feature with the average pooling feature to obtain a second concatenated feature;

[0022] The splicing features include the first splicing features and the second splicing features.

[0023] In some possible designs, performing feature selection processing on the splicing features to obtain selected splicing features includes:

[0024] Performing feature selection processing on the first splicing feature and the second splicing feature to obtain selected features corresponding to the first splicing feature and the second splicing feature respectively;

[0025] Averaging the selected features corresponding to the first splicing feature and the second splicing feature to obtain the selected splicing feature.

[0026] In some possible designs, obtaining the image features of the displayed image includes:

[0027] The display image is input into a trained visual model for feature extraction processing to obtain the image features. The visual model is trained using the interaction indicator label data corresponding to the sample display information as supervision information. The interaction indicator label data represents the actual interaction results between the sample user account and the sample display information.

[0028] In some possible designs, obtaining the interaction prediction result of the target user account for the information to be recommended based on the interaction prediction feature and the category gating feature includes:

[0029] Fusing the interaction prediction feature and the category gating feature to obtain a fused interaction prediction feature;

[0030] Inputting the fused interaction prediction feature into a trained interaction prediction model for prediction processing to obtain the interaction prediction result;

[0031] The interaction prediction model is jointly trained with the visual model.

[0032] In some possible designs, the interaction prediction model is a multi-gated hybrid network model, which refers to a neural network model that mixes multiple gating networks and multiple branch networks. The fused interaction prediction feature is input into the trained interaction prediction model for prediction processing to obtain the interaction prediction result, including:

[0033] Inputting the fused features into the multiple branch networks for processing to obtain feature data output by each of the multiple branch networks, wherein the multiple gated networks include a target gated network, and the target gated network corresponds to the interaction indicator data;

[0034] Inputting the fusion features into the target gating network for processing to obtain weight data corresponding to each branch network;

[0035] According to the weight data corresponding to each branch network, the feature data output by each branch network is fused to obtain the interaction index data, and the interaction index data represents the interaction prediction result.

[0036] In some possible designs, the method further includes:

[0037] Obtaining a sample image corresponding to the sample display information, a sample category corresponding to the sample display information, an interaction prediction sample feature corresponding to the sample display information, and the interaction indicator label data, wherein the interaction prediction sample feature represents the interaction situation corresponding to the sample user account;

[0038] Inputting the sample image into the initial visual model for feature extraction processing to obtain sample image features;

[0039] Performing feature screening processing on the sample image features based on the sample category to obtain sample category gating features;

[0040] Inputting the interaction prediction sample features and the sample category gating features into the initial interaction prediction model for prediction processing to obtain interaction index prediction data;

[0041] Determining model loss information based on the interaction indicator prediction data and the interaction indicator label data;

[0042] Parameter adjustment processing is performed on the initial visual model and the initial interaction prediction model based on the model loss information to obtain the trained visual model and the trained interaction prediction model.

[0043] According to a second aspect of an embodiment of the present disclosure, a display information push device is provided, the device comprising:

[0044] An information acquisition module is configured to acquire a display image corresponding to the information to be recommended, a recommendation object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended, wherein the interaction prediction feature represents a target user account's interaction prediction status with the information to be recommended;

[0045] a feature extraction module configured to obtain image features of the displayed image and perform feature filtering on the image features based on the recommended object category to obtain category-gated features; the category-gated features represent the predicted interaction status of the target user account with the displayed image;

[0046] an interaction prediction module configured to obtain an interaction prediction result of the target user account for the information to be recommended based on the interaction prediction feature and the category gating feature;

[0047] The information recommendation module is configured to push the information to be recommended to the target user account and display the display image when the interaction prediction result meets the preset conditions.

[0048] In some possible designs, the feature extraction module includes:

[0049] A category feature determination submodule is configured to determine the category feature corresponding to the recommended object category;

[0050] An image feature pooling submodule is configured to perform pooling processing on the image features to obtain pooled image features;

[0051] The gated feature determination submodule is configured to perform fusion of the pooled image feature, the category feature and the image feature to obtain the category gated feature.

[0052] In some possible designs, the gating feature determination submodule includes:

[0053] a feature splicing unit, configured to splice the pooled image feature with the category feature to obtain a spliced ​​feature;

[0054] a feature selection unit configured to perform feature selection processing on the splicing features to obtain selected splicing features;

[0055] The feature fusion unit is configured to fuse the selected splicing features with the image features to obtain the category gating features.

[0056] In some possible designs, the pooled image features include maximum pooling features and average pooling features, and the feature splicing unit is specifically configured to perform:

[0057] Concatenate the category feature with the maximum pooling feature to obtain a first concatenated feature;

[0058] Concatenate the category feature with the average pooling feature to obtain a second concatenated feature;

[0059] The splicing features include the first splicing features and the second splicing features.

[0060] In some possible designs, the feature selection unit is specifically configured to perform:

[0061] Performing feature selection processing on the first splicing feature and the second splicing feature to obtain selected features corresponding to the first splicing feature and the second splicing feature respectively;

[0062] Averaging the selected features corresponding to the first splicing feature and the second splicing feature to obtain the selected splicing feature.

[0063] In some possible designs, the feature extraction module further includes:

[0064] The image feature extraction submodule is configured to perform feature extraction processing on the display image input into a trained visual model to obtain the image features. The visual model is trained using the interaction indicator label data corresponding to the sample display information as supervision information. The interaction indicator label data represents the actual interaction results between the sample user account and the sample display information.

[0065] In some possible designs, the interaction prediction module includes:

[0066] a feature fusion submodule, configured to fuse the interaction prediction feature and the category gating feature to obtain a fused interaction prediction feature;

[0067] An interaction prediction submodule is configured to input the fused interaction prediction feature into a trained interaction prediction model for prediction processing to obtain the interaction prediction result;

[0068] The interaction prediction model is jointly trained with the visual model.

[0069] In some possible designs, the interaction prediction model is a multi-gated hybrid network model, which refers to a neural network model that mixes multiple gating networks and multiple branch networks. The interaction prediction submodule includes:

[0070] a feature data extraction unit configured to input the fused feature into the multiple branch networks for processing to obtain feature data output by each of the multiple branch networks, wherein the multiple gated networks include a target gated network, and the target gated network corresponds to the interaction indicator data;

[0071] a weight data determining unit, configured to input the fusion feature into the target gating network for processing, and obtain weight data corresponding to each of the branch networks;

[0072] The indicator data determination unit is configured to perform fusion processing on the feature data output by each branch network according to the weight data corresponding to each branch network, so as to obtain the interaction indicator data, wherein the interaction indicator data represents the interaction prediction result.

[0073] In some possible designs, the apparatus further includes:

[0074] a training data acquisition module configured to acquire a sample image corresponding to the sample display information, a sample category corresponding to the sample display information, an interaction prediction sample feature corresponding to the sample display information, and the interaction indicator label data, wherein the interaction prediction sample feature represents the interaction status corresponding to the sample user account;

[0075] The image feature extraction module is further configured to input the sample image into an initial visual model for feature extraction processing to obtain sample image features;

[0076] The gated feature screening module is further configured to perform feature screening processing on the sample image features based on the sample category to obtain a sample category gated feature;

[0077] The indicator data prediction module is further configured to input the interaction prediction sample features and the sample category gating features into an initial interaction prediction model for prediction processing to obtain interaction indicator prediction data;

[0078] a model loss determination module, configured to determine model loss information based on the interaction indicator prediction data and the interaction indicator label data;

[0079] The model parameter adjustment module is configured to perform parameter adjustment processing on the initial visual model and the initial interaction prediction model based on the model loss information to obtain the trained visual model and the trained interaction prediction model.

[0080] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the display information push method as described in any one of the above-mentioned first aspects.

[0081] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the display information push method described in any one of the first aspects of the embodiment of the present disclosure.

[0082] According to a fifth aspect of the embodiments of the present disclosure, a computer program product comprising instructions is provided, which, when executed on a computer, enables the computer to execute the method for pushing display information according to any one of the first aspects of the embodiments of the present disclosure.

[0083] The technical solutions provided by the embodiments of the present disclosure bring at least the following beneficial effects:

[0084] By filtering the display image features based on the recommended object category corresponding to the information to be recommended, we can obtain visual features that can characterize the target user account's interaction prediction of the display image, namely, category-gated features. Based on the category-gated features and the original interaction prediction features corresponding to the information to be recommended, we can obtain the interaction prediction results of the target object for the recommended information, which enriches the dimensions of the data used to predict the interaction results and improves the accuracy of the interaction prediction results. Recommending information based on the interaction prediction results can effectively improve the accuracy of information recommendation.

[0085] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute an improper limitation of the present disclosure.

[0087] Figure 1 is a schematic diagram showing an application environment according to an exemplary embodiment;

[0088] Figure 2 This is a flow chart of a method for displaying information push according to an exemplary embodiment. Figure 1 ;

[0089] Figure 3 This is a flow chart of a method for displaying information push according to an exemplary embodiment. Figure 2 ;

[0090] Figure 4 This is a flow chart of a method for displaying information push according to an exemplary embodiment. Figure 3 ;

[0091] Figure 5 A schematic diagram of a process for determining a category gating feature is shown as an example;

[0092] Figure 6 The following is a schematic diagram of a technical process for predicting interaction index data based on a visual model and an interaction prediction model;

[0093] Figure 7 This is a flowchart of a training method for a display information push model according to an exemplary embodiment;

[0094] Figure 8 This is a block diagram of a device for pushing display information according to an exemplary embodiment;

[0095] Figure 9 The present invention is a block diagram of an electronic device for displaying information push according to an exemplary embodiment. DETAILED DESCRIPTION

[0096] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0097] It should be noted that the terms "first," "second," and the like in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure as detailed in the appended claims.

[0098] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0099] See also Figure 1 , Figure 1 is a schematic diagram showing an application environment according to an exemplary embodiment. Figure 1 As shown, the application environment may include a terminal 100 and a server 200 .

[0100] The terminal 100 can be used to provide information display services to any user. Specifically, the terminal 100 may include, but is not limited to, electronic devices such as smartphones, desktop computers, tablet computers, laptop computers, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices. It may also be software running on these electronic devices, such as applications. Optionally, the operating system running on the electronic device may include, but is not limited to, Android, iOS, Linux, Windows, etc.

[0101] In an optional embodiment, the server 200 can provide background services for the terminal 100 and push relevant display information to the terminal 100. Specifically, the server 200 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0102] In addition, it should be noted that Figure 1What is shown is only one application environment provided by the present disclosure. In actual applications, other application environments may also be included, for example, more terminals may be included.

[0103] In the embodiments of this specification, the terminal 100 and the server 200 may be directly or indirectly connected via wired or wireless communication, which is not limited in this disclosure.

[0104] Figure 2 This is a flow chart of a method for displaying information push according to an exemplary embodiment. Figure 1 Optionally, the display information push method is used in an electronic device. Figure 2 As shown, the method may include the following steps (210-250).

[0105] Step 210 : Obtain a display image corresponding to the information to be recommended, a recommended object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended.

[0106] Optionally, the interaction prediction feature represents the target user account's interaction prediction for the recommended information. Optionally, the interaction prediction feature is determined based on the target user account's interaction information in the presentation environment and can be used alone to determine the target user account's interaction prediction result for the recommended information.

[0107] The information to be recommended may be any information displayed on the display screen of the device, including but not limited to web pages, articles, videos, music, images, advertisements, etc.

[0108] Optionally, the information to be recommended is multimedia information to be presented in an information stream presentation environment.

[0109] Optionally, the above-mentioned display image is an image in the display information, or may be a display image in meta-information corresponding to the display information.

[0110] Optionally, the recommended object category is a specific category corresponding to the information to be recommended, such as the product category corresponding to an advertisement. The recommended object category may or may not overlap with the object category appearing in the displayed image. The recommended object category is independent of the object category appearing in the displayed image.

[0111] Optionally, the interaction prediction feature is a feature extracted based on interaction information corresponding to the information to be recommended. Optionally, the interaction information includes information identification features corresponding to the information to be recommended, account identification features corresponding to the target user account, and feedback data corresponding to the target user account.

[0112] Optionally, the feedback data corresponding to the target user account includes the timestamps corresponding to historical interactive operation instructions sent by the target user account, the types of operation behaviors corresponding to the historical interactive operation instructions, and the sequence of operation behaviors formed by the historical interactive operation instructions. Based on this feedback data, interaction environment features corresponding to the target user account can be extracted. The interaction environment features represent the contextual characteristics of the target user account browsing information in the display environment.

[0113] Optionally, the information identification features corresponding to the information to be recommended, the account identification features corresponding to the target user account, and the interaction environment features are characterized to obtain the interaction prediction features.

[0114] In a possible implementation, the interaction prediction feature is a benchmark feature for predicting interaction index data, and the interaction prediction feature can independently predict the interaction index data corresponding to the information to be recommended.

[0115] Step 220 : Acquire image features of the displayed image, and perform feature screening processing on the image features based on the recommended object category to obtain category gated features.

[0116] The category gated feature represents the target user account's interaction prediction for the displayed image.

[0117] In a possible implementation, feature extraction processing is performed on the displayed image to obtain image features.

[0118] Optionally, the displayed image is input into a trained visual model for feature extraction processing to obtain image features.

[0119] The visual model is trained using the interaction indicator label data corresponding to the sample display information as supervisory information. The interaction indicator label data represents the actual interaction results between the sample user account and the sample display information. For example, if the sample user account has performed an action such as clicking, liking, or purchasing the sample display information, the interaction indicator label data can be 1; if no action has been performed, the interaction indicator label data can be 0.

[0120] Optionally, the visual model is an image classification model, but is trained using the interaction indicator label data as supervisory information. Optionally, the interaction indicator label data is the click-through rate (CTR) corresponding to the sample display information.

[0121] Display images, such as advertising pictures or videos, contain rich information, such as color and the products contained in the image. The pre-trained visual models in related technologies can only capture classification-based information, which has a certain angle with the CTR estimation model target. In the embodiment of the present application, the visual model is trained with the above-mentioned interaction indicator label data as supervisory information, so that the image features extracted by the visual model are visual features that can represent the content of the information to be recommended, and at the same time, they are image features that can be helpful in estimating interaction indicator data, such as CTR, so as to better recommend display information.

[0122] Optionally, the image classification model is a convolutional neural network model, such as a small convolutional neural network such as Mobilenet or Resnet18.

[0123] The technical solution provided in the embodiment of the present application uses interaction indicator label data as supervisory information to train the visual model, which can enable the visual model to learn the interactive relationship between sample user accounts and sample display information, so that the application side can output image features that represent the probability of triggering interactive operations in a focused manner, enriching the dimensions of the data used to predict interaction indicator data and improving the accuracy of interaction indicator data.

[0124] Optionally, after determining the above-mentioned image features, feature screening processing may be performed on the image features based on the recommended object category to obtain category-gated features.

[0125] Optionally, the category gating feature is a visual feature representing the probability of triggering an interactive operation. The interactive operation trigger probability refers to the probability that the target user account will trigger an interactive operation instruction based on the recommended information. The interactive operation trigger probability can reflect the target user account's predicted interaction with the recommended information. For example, after the recommended information is displayed, the target user account may trigger an interactive operation based on the information.

[0126] Optionally, the interactive operation instruction may be an instruction triggered by the terminal in response to an interactive operation. Optionally, the interactive operation includes but is not limited to sliding, clicking, liking, commenting, and the like. Accordingly, the interactive operation instruction includes but is not limited to sliding operation instructions, clicking operation instructions, liking operation instructions, commenting operation instructions, and the like.

[0127] When the above-mentioned visual model is trained, there may be more than one interaction indicator label data corresponding to the same sample image. Different interaction indicator label data respectively represent the interaction relationship between different label sample accounts and the sample display information, which can indirectly represent the degree of interest of different sample user accounts in the sample display information.

[0128] For example, a sample image is a sample image from a product A advertisement, which includes both product A and object B. Sample user account 1 is interested in product A and clicks on the product A advertisement. Therefore, the interaction indicator label data for sample user account 1 for product A advertisement can be 1. However, sample user account 2 is not interested in product A, or even interested in object B, but because it is a product A advertisement, does not click on it. In this case, the interaction indicator label data for sample user account 2 for product A advertisement can be 0.

[0129] In a practical application, the above-mentioned visual model is trained end-to-end. Under this training method, the visual model is trained directly based on sample images corresponding to multiple label data, which is prone to collapse. The phenomenon is that for any image, the features output by the visual model are relatively similar, which is also called model over-smoothing. Therefore, in order to solve the problem of over-smoothing of the visual model, the embodiment of the present application introduces the above-mentioned recommended object category to perform feature screening on the features output by the visual model, strengthens the importance of the model output features, and makes the category gating features used to predict the interaction index data tend to be the visual features of the above-mentioned recommended object category. At the same time, it is also beneficial to estimate the visual features of the interaction index data, such as CTR, so as to better perform display information recommendation.

[0130] In an exemplary embodiment, as Figure 3 As shown, the above step 220 includes the following steps (221-223): Figure 3 This is a flow chart of a method for displaying information push according to an exemplary embodiment. Figure 2 .

[0131] Step 221: Determine the category features corresponding to the recommended object category.

[0132] Optionally, the category feature identifier corresponding to the recommended object category is determined as the above-mentioned category feature. Optionally, the category feature identifier corresponding to the recommended object category is subjected to characterization processing to obtain the above-mentioned category feature.

[0133] The above-mentioned category features can be category feature vectors or category feature data sequences, which is not limited in the embodiments of the present disclosure.

[0134] Step 222: Perform pooling processing on the image features to obtain pooled image features.

[0135] Optionally, the image features are subjected to maximum pooling processing to obtain maximum pooling features; and the image features are subjected to average pooling processing to obtain average pooling features. The pooled image features include the maximum pooling features and the average pooling features.

[0136] Optionally, the above-mentioned image features are feature maps corresponding to the displayed images; accordingly, the above-mentioned feature maps are subjected to maximum pooling processing to obtain maximum pooling features; and the above-mentioned feature maps are subjected to average pooling processing to obtain average pooling features.

[0137] In step 223 , the pooled image features, the category features, and the image features are fused to obtain category gated features.

[0138] The technical solution provided by the embodiments of the present disclosure can achieve category screening of image features and improve the accuracy of extracting category gating features by pooling image features and fusing the pooled image features with category features and original image features.

[0139] The embodiment of the present disclosure does not limit the above-mentioned feature fusion method, and can be configured according to actual application scenarios.

[0140] In an exemplary embodiment, as Figure 4 As shown, the above step 223 includes the following steps (2231-2233): Figure 4 This is a flow chart of a method for displaying information push according to an exemplary embodiment. Figure 3 .

[0141] In step 2231, the pooled image features and the category features are concatenated to obtain concatenated features.

[0142] Optionally, the category feature is concatenated with the maximum pooling feature to obtain a first concatenated feature; the category feature is concatenated with the average pooling feature to obtain a second concatenated feature.

[0143] The splicing features include a first splicing feature and a second splicing feature.

[0144] The technical solution provided by the embodiment of the present disclosure can extract richer feature information by performing different pooling processes on image features and splicing different pooling features with category features respectively, which helps to improve the accuracy of category gated features.

[0145] Step 2232: Perform feature selection processing on the splicing features to obtain selected splicing features.

[0146] Feature selection, also known as feature subset selection or attribute selection, involves selecting N features from a set of M existing features to optimize a specific system metric. It involves selecting the most effective features from the original set to reduce the dimensionality of the dataset. This helps improve the performance of learning algorithms and is a key data preprocessing step in pattern recognition (the classification model mentioned above is a pattern recognition model).

[0147] Optionally, feature selection processing is performed on the first splicing feature and the second splicing feature to obtain selected features corresponding to each of the first splicing feature and the second splicing feature; and average processing is performed on the selected features corresponding to each of the first splicing feature and the second splicing feature to obtain the selected splicing feature.

[0148] The technical solution provided by the embodiment of the present disclosure can obtain selected features of different dimensions by performing feature selection processing on different splicing features respectively, and then averagely fuse the selected features of different dimensions, so as to extract richer and more important feature information with a focus, which helps to improve the accuracy of category gating features.

[0149] In step 2233, the selected splicing features are fused with the image features to obtain the category gated features.

[0150] The technical solution provided by the embodiments of the present disclosure can extract effective features from the spliced ​​features by performing feature selection on the spliced ​​features between the pooled image features and the category features, reduce the data dimension, improve the accuracy of the category gated features, and reduce the model data processing volume.

[0151] In one example, if Figure 5 As shown in the figure, it shows an exemplary process diagram for determining category gating features. The whole process can be divided into three steps:

[0152] Step 1: For the output feature map 51 of the visual model, the size of the feature map 51 is CxHxW, and average pooling and maximum pooling are performed respectively to obtain average pooling features and maximum pooling features, and then the two are respectively spliced ​​with the product category features.

[0153] Step 2: The two concatenated vectors in step 1 are passed through a feature selection module 52 to output their respective Cx1x1 features, where C is the number of channels of the convolutional network feature output size.

[0154] Step 3: Average the two selected features in Step 2 to obtain the final weight vector 53, i.e., the selected feature, with a size of Cx1x1. Then multiply the weight vector 53 by the original feature map 51 to obtain the final product category gated feature 54.

[0155] Step 230 : Obtain the interaction prediction result of the target user account for the recommended information based on the interaction prediction feature and the category gating feature.

[0156] Optionally, the interaction prediction result includes interaction indicator data. The interaction indicator data may represent the probability of triggering an interaction operation. Optionally, the interaction operation triggering probability refers to the probability that the target user account triggers an interaction operation based on the information to be recommended.

[0157] In an exemplary embodiment, the interaction prediction feature and the category gating feature are fused to obtain a fused interaction prediction feature, which is then input into a trained interaction prediction model for prediction processing to obtain an interaction prediction result, such as the aforementioned interaction indicator data.

[0158] Among them, the interaction prediction model is jointly trained with the visual model.

[0159] By inputting the fused interaction prediction features into the prediction model jointly trained with the visual model, the interaction prediction model can predict interaction indicator data from more dimensions based on more diverse feature data, thereby improving the accuracy of predicting the above interaction indicator data.

[0160] The following describes interaction index data in specific scenarios. For example, consider a display image from a Q product advertisement, which includes both the Q product and an E object. User account 11 is interested in the Q product, while user account 12 is not interested in the Q product but is interested in the E object. Therefore, the interaction index data corresponding to the Q product advertisement for user account 11 is higher than the interaction index data corresponding to the Q product advertisement for user account 12. Category gating reduces the interference of E objects in the image that are unrelated to the Q product on the predicted interaction index data.

[0161] In one possible embodiment, the interactive prediction model is a multi-gated hybrid network model, which refers to a neural network model that is a mixture of multiple gating networks and multiple branch networks. Different gating networks correspond to different indicator data prediction tasks, and the multiple gating networks include a target gating network corresponding to the interactive indicator data. The multiple branch networks can be multiple feature extraction networks with shared parameters. Optionally, the multi-gated hybrid network model is an MMoE (Multi-gate Mixture-of-Experts) + MLP (Multilayer Perceptron) model.

[0162] Accordingly, the process of predicting interaction index data based on the multi-gated hybrid network model is as follows:

[0163] The fused features are input into multiple branch networks for processing to obtain feature data output by each of the multiple branch networks. The multiple gated networks include a target gated network, which corresponds to the interaction indicator data.

[0164] The fused features are input into the target gating network for processing, yielding weight data corresponding to each branch network. Different indicator data prediction tasks focus on different feature data, so different gating networks are used to determine the weight information of each branch network for the corresponding task. The target gating network is the gating network corresponding to the interaction indicator data, so the weight data output by the target gating network is the weight data suitable for predicting the interaction indicator data.

[0165] Based on the weights corresponding to each branch network, the feature data output by each branch network is fused to generate interaction index data. The interaction index data is obtained by weighted fusion of the feature data output by a branch network according to the weights corresponding to each branch network. This interaction index data can represent the interaction prediction results.

[0166] The technical solution provided by the embodiments of the present disclosure deploys a target gating network in a multi-task model, i.e., the above-mentioned multi-gated hybrid network model, so that the multi-task model can, while predicting other data indicators, fuse the feature data output by each branch network according to the weight output by the target gating network, thereby predicting the interaction indicator data, thereby improving the efficiency of predicting the interaction indicator data, and is also easy to expand, meeting the requirements of the recommendation system for multi-dimensional recommendations.

[0167] In one example, if Figure 6As shown, a schematic diagram of a technical process for predicting interaction indicator data based on a visual model and an interaction prediction model is shown. A visual model 61 (convolutional neural network model) and an interaction prediction model 62 (MMoE+MLP) jointly form a display information push model that utilizes a heterogeneous graph-based feature learning scheme. Specifically, an end-to-end visual model 61 is embedded within a baseline CTR prediction model, namely the interaction prediction model 62. The input to the visual model 61 is the original advertisement image 63. The visual model 61 can be a predefined convolutional network model, such as a small network like MobileNet or ResNet18. The visual model 61 extracts image features corresponding to the original advertisement image 63 and uses the advertised product category to gate and filter these image features, thereby obtaining visual features favored by the interaction prediction model 62, namely category-gated features, which also correspond to the estimated product category. The display information push model also includes an interaction feature representation layer 64. This interaction feature representation layer 64 and the interaction prediction model 62 form a baseline display information push model, i.e., a model that predicts the aforementioned interaction indicator data based solely on interaction prediction features. By embedding the above-mentioned visual model 61 in the baseline display information push model, a display information push model that can achieve end-to-end visual perception can be obtained. The interactive feature representation layer 64 corresponds to three types of inputs, one is the advertising identification feature, the second is the target user account feature, which can be an account identification feature or an account attribute feature, and the third is the interactive environment feature, also known as contextual feature, such as timestamp, user behavior type, behavior sequence and other features. After the above three features are input into the above-mentioned interactive feature representation layer 64, the interactive prediction feature used by the baseline model can be output. After splicing the category gated feature with the interactive prediction feature, it is input into the interactive prediction model 62 for prediction processing, and the click-through rate corresponding to the original advertisement image 63 relative to the target user account can be predicted. On the basis of the original baseline model, the CTR prediction accuracy is further improved by introducing the visual information increment.

[0168] Step 240 : If the interactive prediction result meets the preset conditions, the recommended information is pushed to the target user account and a display image is displayed.

[0169] Optionally, the information to be recommended is used for presentation in a presentation environment.

[0170] Optionally, the preset condition refers to a condition where the interaction indicator data is greater than or equal to a threshold. That is, if the interaction indicator data is greater than or equal to the threshold, the recommended information is pushed to the target user account. Optionally, the interaction prediction feature is determined in real time based on real-time feedback data. If the current interaction indicator data is greater than or equal to the threshold, the recommended information can be pushed to the target user account.

[0171] Optionally, the display environment may be an environment such as an information flow environment displayed by a device, which is not limited in the embodiment of the present application.

[0172] To sum up, the technical solution provided by the embodiments of the present disclosure filters the display image features through the recommendation object category corresponding to the information to be recommended, so as to obtain visual features that can characterize the target user account's interaction prediction situation with the display image, namely, category gating features. Based on the category gating features and the original interaction prediction features corresponding to the information to be recommended, the interaction prediction results of the target object with respect to the recommended information can be obtained, which enriches the dimension of the data used to predict the interaction results and improves the accuracy of the interaction prediction results. Recommending information based on the interaction prediction results can effectively improve the accuracy of information recommendation.

[0173] Figure 7 1 is a flow chart of a method for training a display information push model according to an exemplary embodiment. Optionally, the display information push model includes a visual model and an interaction prediction model. Optionally, the method for training the display information push model is used in an electronic device. Figure 7 As shown, the method may include the following steps (710-750).

[0174] Step 710 : Obtain the sample image corresponding to the sample display information, the sample category corresponding to the sample display information, the interaction prediction sample feature corresponding to the sample display information, and the interaction indicator label data.

[0175] Optionally, the interaction prediction sample feature represents the interaction situation corresponding to the sample user account.

[0176] The sample display information may be historical display information collected, including but not limited to web pages, articles, videos, music, images, advertisements, etc.

[0177] Optionally, the sample image is an image in the sample display information, or may be a display image in the meta information corresponding to the sample display information.

[0178] Optionally, the sample category is a specific category corresponding to the sample display information, such as a product category corresponding to an advertisement. The sample category may or may not overlap with the object category appearing in the sample image. The sample category is independent of the object category appearing in the sample image.

[0179] Optionally, the interaction prediction sample features are features extracted based on historical interaction information corresponding to the sample display information. Optionally, the historical interaction information includes sample information identification features corresponding to the sample information, account identification features corresponding to the sample user account, and sample feedback data corresponding to the sample user account.

[0180] Optionally, the sample feedback data corresponding to the sample user account includes the timestamp corresponding to the historical interactive operation instructions sent by the sample user account, the operation behavior type corresponding to the historical interactive operation instructions, and the operation behavior sequence formed by the historical interactive operation instructions. Based on this sample feedback data, the interaction environment sample features corresponding to the sample user account can be extracted. The interaction environment sample features represent the contextual characteristics of the sample user account browsing information in the display environment.

[0181] Optionally, the information identification features corresponding to the sample display information, the account identification features corresponding to the sample user account, and the interaction environment sample features are characterized to obtain the interaction prediction sample features.

[0182] In a possible implementation, the interaction prediction sample feature is a benchmark feature for predicting interaction index data corresponding to the sample display information. The interaction prediction sample feature can independently predict the interaction index data corresponding to the sample display information.

[0183] Step 720: Input the sample image into the initial visual model for feature extraction to obtain sample image features.

[0184] Optionally, the initial visual model refers to a visual model that has not been trained.

[0185] Step 730 : Perform feature screening processing on the sample image features based on the sample category to obtain sample category gated features.

[0186] In an exemplary embodiment, a sample category feature corresponding to a sample category is determined; the sample image feature is pooled to obtain a pooled sample image feature; and the pooled sample image feature, the sample category feature, and the sample image feature are fused to obtain a sample category gated feature.

[0187] Optionally, the pooled sample image features include maximum pooled sample features and average pooled sample features.

[0188] Optionally, the pooled sample image features are concatenated with the sample category features to obtain a concatenated sample feature. For example, the sample category features are concatenated with the maximum pooled sample features to obtain a first concatenated sample feature; the sample category features are concatenated with the average pooled sample features to obtain a second concatenated sample feature; wherein the concatenated sample features include the first concatenated sample feature and the second concatenated sample feature.

[0189] Perform feature selection processing on the spliced ​​sample features to obtain selected spliced ​​sample features. Optionally, perform feature selection processing on the first spliced ​​sample features and the second spliced ​​sample features to obtain selected features corresponding to the first spliced ​​sample features and the second spliced ​​sample features; and perform averaging processing on the selected features corresponding to the first spliced ​​sample features and the second spliced ​​sample features to obtain the selected spliced ​​sample features.

[0190] The selected spliced ​​sample features are fused with the sample image features to obtain the sample category gating features.

[0191] Step 740: Input the interaction prediction sample features and the sample category gating features into the initial interaction prediction model for prediction processing to obtain interaction index prediction data.

[0192] Optionally, the initial interaction prediction model is a trained baseline interaction prediction model, which predicts interaction index data using a single interaction prediction feature as input.

[0193] Step 750: Determine model loss information based on the interaction indicator prediction data and the interaction indicator label data.

[0194] Optionally, the sample display information and the sample user account are paired. Optionally, each interaction indicator label data item corresponds to a sample user account, and each interaction indicator label data item is used to represent the interaction relationship between the sample display information and the corresponding sample user account. The interaction relationship includes the existence of an interaction relationship or the absence of an interaction relationship.

[0195] For example, sample pair 1: sample user account a, picture c, interaction index label data (CTR) = 1; sample pair 2: sample user account b, picture c, interaction index label data (CTR) = 0. 1 indicates that there is an interaction relationship, and 0 indicates that there is no interaction relationship.

[0196] Step 760 : Perform parameter adjustment processing on the initial visual model and the initial interaction prediction model based on the model loss information to obtain a trained visual model and a trained interaction prediction model.

[0197] The technical solution provided by the embodiment of the present disclosure can prevent the collapse of visual model training by screening the sample image features output by the visual model through the sample category corresponding to the sample display information, avoid the visual model outputting relatively similar image features for any sample image, and strengthen the focus of the visual model output features; the sample gating features obtained by screening are input into the prediction model together with the interaction prediction sample features, namely the category gating features, and the prediction model is supervisedly trained with the interaction indicator label data as the supervision signal. On the one hand, the visual model can learn visual features that can characterize the probability of triggering the interactive operation. On the other hand, due to the introduction of the category gating features, the prediction model can also learn more diverse features under the constraints of the supervision signal, thereby outputting more accurate interaction indicator prediction data. The visual model and the prediction model obtained by the above joint training can predict more accurate interaction indicator data based on the multivariate feature data on the application side, thereby improving the accuracy of information recommendation.

[0198] Figure 8 FIG. 1 is a block diagram of a device for pushing information according to an exemplary embodiment. Figure 8 , the apparatus 800 comprises:

[0199] The information acquisition module 810 is configured to acquire a display image corresponding to the information to be recommended, a recommendation object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended, wherein the interaction prediction feature represents a target user account's interaction prediction with the information to be recommended;

[0200] The feature extraction module 820 is configured to obtain image features of the displayed image and perform feature filtering on the image features based on the recommended object category to obtain category-gated features; the category-gated features represent the target user account's predicted interaction with the displayed image;

[0201] The interaction prediction module 830 is configured to obtain an interaction prediction result of the target user account for the information to be recommended based on the interaction prediction feature and the category gating feature;

[0202] The information recommendation module 840 is configured to push the information to be recommended to the target user account and display the display image when the interaction prediction result meets the preset conditions.

[0203] In some possible designs, the feature extraction module includes:

[0204] A category feature determination submodule is configured to determine the category feature corresponding to the recommended object category;

[0205] An image feature pooling submodule is configured to perform pooling processing on the image features to obtain pooled image features;

[0206] The gated feature determination submodule is configured to perform fusion of the pooled image feature, the category feature and the image feature to obtain the category gated feature.

[0207] In some possible designs, the gating feature determination submodule includes:

[0208] a feature splicing unit, configured to splice the pooled image feature with the category feature to obtain a spliced ​​feature;

[0209] a feature selection unit configured to perform feature selection processing on the splicing features to obtain selected splicing features;

[0210] The feature fusion unit is configured to fuse the selected splicing features with the image features to obtain the category gating features.

[0211] In some possible designs, the pooled image features include maximum pooling features and average pooling features, and the feature splicing unit is specifically configured to perform:

[0212] Concatenate the category feature with the maximum pooling feature to obtain a first concatenated feature;

[0213] Concatenate the category feature with the average pooling feature to obtain a second concatenated feature;

[0214] The splicing features include the first splicing features and the second splicing features.

[0215] In some possible designs, the feature selection unit is specifically configured to perform:

[0216] Performing feature selection processing on the first splicing feature and the second splicing feature to obtain selected features corresponding to the first splicing feature and the second splicing feature respectively;

[0217] Averaging the selected features corresponding to the first splicing feature and the second splicing feature to obtain the selected splicing feature.

[0218] In some possible designs, the feature extraction module further includes:

[0219] The image feature extraction submodule is configured to perform feature extraction processing on the display image input into a trained visual model to obtain the image features. The visual model is trained using the interaction indicator label data corresponding to the sample display information as supervision information. The interaction indicator label data represents the actual interaction results between the sample user account and the sample display information.

[0220] In some possible designs, the interaction prediction module includes:

[0221] a feature fusion submodule, configured to fuse the interaction prediction feature and the category gating feature to obtain a fused interaction prediction feature;

[0222] An interaction prediction submodule is configured to input the fused interaction prediction feature into a trained interaction prediction model for prediction processing to obtain the interaction prediction result;

[0223] The interaction prediction model is jointly trained with the visual model.

[0224] In some possible designs, the interaction prediction model is a multi-gated hybrid network model, which refers to a neural network model that mixes multiple gating networks and multiple branch networks. The interaction prediction submodule includes:

[0225] a feature data extraction unit configured to input the fused feature into the multiple branch networks for processing to obtain feature data output by each of the multiple branch networks, wherein the multiple gated networks include a target gated network, and the target gated network corresponds to the interaction indicator data;

[0226] a weight data determining unit, configured to input the fusion feature into the target gating network for processing, and obtain weight data corresponding to each of the branch networks;

[0227] The indicator data determination unit is configured to perform fusion processing on the feature data output by each branch network according to the weight data corresponding to each branch network, so as to obtain the interaction indicator data, wherein the interaction indicator data represents the interaction prediction result.

[0228] In some possible designs, the apparatus further includes:

[0229] a training data acquisition module configured to acquire a sample image corresponding to the sample display information, a sample category corresponding to the sample display information, an interaction prediction sample feature corresponding to the sample display information, and the interaction indicator label data, wherein the interaction prediction sample feature represents the interaction status corresponding to the sample user account;

[0230] The image feature extraction module is further configured to input the sample image into an initial visual model for feature extraction processing to obtain sample image features;

[0231] The gated feature screening module is further configured to perform feature screening processing on the sample image features based on the sample category to obtain a sample category gated feature;

[0232] The indicator data prediction module is further configured to input the interaction prediction sample features and the sample category gating features into an initial interaction prediction model for prediction processing to obtain interaction indicator prediction data;

[0233] a model loss determination module, configured to determine model loss information based on the interaction indicator prediction data and the interaction indicator label data;

[0234] The model parameter adjustment module is configured to perform parameter adjustment processing on the initial visual model and the initial interaction prediction model based on the model loss information to obtain the trained visual model and the trained interaction prediction model.

[0235] To sum up, the technical solution provided by the embodiments of the present disclosure filters the display image features through the recommendation object category corresponding to the information to be recommended, so as to obtain visual features that can characterize the target user account's interaction prediction situation with the display image, namely, category gating features. Based on the category gating features and the original interaction prediction features corresponding to the information to be recommended, the interaction prediction results of the target object with respect to the recommended information can be obtained, which enriches the dimension of the data used to predict the interaction results and improves the accuracy of the interaction prediction results. Recommending information based on the interaction prediction results can effectively improve the accuracy of information recommendation.

[0236] It should be noted that the apparatus provided in the above embodiments, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0237] Figure 9 This is a block diagram of an electronic device for displaying information push according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 9As shown. The electronic device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for displaying information push is implemented. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0238] Those skilled in the art will understand that Figure 9 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present disclosure, and does not constitute a limitation on the electronic device to which the scheme of the present disclosure is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0239] In an exemplary embodiment, an electronic device is further provided, including: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the display information push method in the embodiment of the present disclosure.

[0240] In an exemplary embodiment, a computer-readable storage medium is further provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the method for pushing display information in the embodiment of the present disclosure.

[0241] In an exemplary embodiment, a computer program product including instructions is further provided. When the computer program product is run on a computer, the computer is enabled to execute the method for pushing presentation information in the embodiment of the present disclosure.

[0242] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, which can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0243] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0244] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An information recommendation method, characterized in that: The method comprises: Obtaining a display image corresponding to the information to be recommended, a recommendation object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended, wherein the interaction prediction feature represents a predicted interaction status of a target user account with the information to be recommended; Inputting the displayed image into a trained visual model for feature extraction to obtain image features, and performing feature screening on the image features based on the recommended object category to obtain category-gated features; the category-gated features represent the target user account's predicted interaction with the displayed image; The interaction prediction feature and the category gating feature are fused and input into a trained interaction prediction model for prediction processing to obtain the interaction prediction result of the target user account for the information to be recommended; the visual model is trained using the interaction indicator label data corresponding to the sample display information as supervision information; the interaction prediction model is jointly trained with the visual model; the interaction indicator label data represents the actual interaction result between the sample user account and the sample display information; When the interaction prediction result meets the preset conditions, the information to be recommended is pushed to the target user account, and the display image is displayed.

2. The method according to claim 1, characterized in that The performing feature screening processing on the image features based on the recommended object category to obtain category gating features includes: Determining category features corresponding to the recommended object category; Performing pooling processing on the image features to obtain pooled image features; The pooled image feature, the category feature and the image feature are fused to obtain the category gated feature.

3. The method according to claim 2, characterized in that The fusing the pooled image feature, the category feature, and the image feature to obtain the category gated feature includes: Splicing the pooled image features with the category features to obtain spliced ​​features; performing feature selection processing on the splicing features to obtain selected splicing features; The selected splicing features are fused with the image features to obtain the category gating features.

4. The method according to claim 3, characterized in that The pooled image features include maximum pooling features and average pooling features, and the pooled image features are spliced ​​with the category features to obtain spliced ​​features, including: Concatenate the category feature with the maximum pooling feature to obtain a first concatenated feature; Concatenate the category feature with the average pooling feature to obtain a second concatenated feature; The splicing features include the first splicing features and the second splicing features.

5. The method according to claim 4, characterized in that The performing feature selection processing on the splicing features to obtain selected splicing features includes: Performing feature selection processing on the first splicing feature and the second splicing feature to obtain selected features corresponding to the first splicing feature and the second splicing feature respectively; Averaging the selected features corresponding to the first splicing feature and the second splicing feature to obtain the selected splicing feature.

6. The method according to claim 1, characterized in that The method further comprises: Fusing the interaction prediction feature and the category gating feature to obtain a fused interaction prediction feature; The fused interaction prediction feature is input into a trained interaction prediction model for prediction processing to obtain the interaction prediction result.

7. The method according to claim 6, characterized in that The interaction prediction model is a multi-gated hybrid network model, which refers to a neural network model that mixes multiple gating networks and multiple branch networks. The fused interaction prediction feature is input into the trained interaction prediction model for prediction processing to obtain the interaction prediction result, including: Inputting the fused features into the multiple branch networks for processing to obtain feature data output by each of the multiple branch networks, wherein the multiple gated networks include a target gated network, and the target gated network corresponds to the interaction indicator data; Inputting the fusion features into the target gating network for processing to obtain weight data corresponding to each branch network; According to the weight data corresponding to each branch network, the feature data output by each branch network is fused to obtain the interaction index data, and the interaction index data represents the interaction prediction result.

8. The method according to claim 1, characterized in that The method further comprises: Obtaining a sample image corresponding to the sample display information, a sample category corresponding to the sample display information, an interaction prediction sample feature corresponding to the sample display information, and the interaction indicator label data, wherein the interaction prediction sample feature represents the interaction situation corresponding to the sample user account; Inputting the sample image into the initial visual model for feature extraction processing to obtain sample image features; Performing feature screening processing on the sample image features based on the sample category to obtain sample category gating features; Inputting the interaction prediction sample features and the sample category gating features into the initial interaction prediction model for prediction processing to obtain interaction index prediction data; Determining model loss information based on the interaction indicator prediction data and the interaction indicator label data; Parameter adjustment processing is performed on the initial visual model and the initial interaction prediction model based on the model loss information to obtain the trained visual model and the trained interaction prediction model.

9. A display information push device, characterized in that: The device comprises: An information acquisition module is configured to acquire a display image corresponding to the information to be recommended, a recommendation object category corresponding to the information to be recommended, and an interaction prediction feature corresponding to the information to be recommended, wherein the interaction prediction feature represents a target user account's interaction prediction status with the information to be recommended; a feature extraction module configured to perform feature extraction processing on the displayed image by inputting it into a trained visual model to obtain image features, and to perform feature screening processing on the image features based on the recommended object category to obtain category-gated features; the category-gated features represent the target user account's predicted interaction with the displayed image; The interaction prediction module is configured to perform prediction processing by fusing the interaction prediction feature with the category gating feature and inputting the result into a trained interaction prediction model to obtain a prediction result of the target user account's interaction with the information to be recommended; the visual model is trained using the interaction indicator label data corresponding to the sample display information as supervision information; the interaction prediction model is jointly trained with the visual model; the interaction indicator label data represents the actual interaction result between the sample user account and the sample display information; The information recommendation module is configured to push the information to be recommended to the target user account and display the display image when the interaction prediction result meets the preset conditions.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instruction to implement the display information pushing method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the display information pushing method according to any one of claims 1 to 8.

Images