Resource recommendation method and device, electronic equipment and storage medium

Abstract

The invention relates to a resource recommendation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a historical delivery effect feature, an image-text feature and a user feature corresponding to each resource creativity in a plurality of resource creativity corresponding to a to-be-recommended resource under the condition of determining to recommend the to-be-recommended resource; for each resource originality, inputting the historical putting effect feature, the image-text feature and the user feature corresponding to the resource originality into a pre-trained originality recommendation model to obtain a predicted putting effect of the resource originality output by the originality recommendation model; and according to the predicted delivery effect corresponding to each resource creativity, determining a target resource creativity from the plurality of resource creativity, and recommending the to-be-recommended resource based on the target resource creativity. Therefore, the accuracy of determining the resource originality can be improved, the accuracy of resource recommendation is improved, and the user experience is improved.

Description

Technical Field

[0001] This application relates to the technical field of resource recommendation, and particularly to a resource recommendation method, apparatus, electronic device, and storage medium. Background Art

[0002] Resource creativity is the first interaction object between users and resource products, which affects users' sensory impressions of resource products. Therefore, it is crucial to screen out appropriate resource creativity during the process of resource placement.

[0003] Currently, for the screening of resource creativity, it mainly relies on users' historical feedback behaviors on different creativities in a historical time period, such as the click-through rate of users on resource creativities. However, this method depends on the historical placement data of resource creativities. When the historical placement data of resource creativities is scarce, the accuracy of the determined resource creativities will be reduced, affecting the user experience. Summary of the Invention

[0004] This application provides a resource recommendation method, apparatus, electronic device, and storage medium to solve the technical problem in the prior art that the screening of resource creativity depends on the historical placement data of resource creativities, and when the historical placement data of resource creativities is scarce, the accuracy of the determined resource creativities will be reduced, affecting the user experience.

[0005] In a first aspect, this application provides a resource recommendation method, and the method includes:

[0006] When it is determined to recommend a resource to be recommended, obtain the historical placement effect characteristics, graphic and text characteristics, and user characteristics corresponding to each resource creativity among the multiple resource creativities corresponding to the resource to be recommended;

[0007] For each resource creativity, input the historical placement effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource creativity into a pre-trained creativity recommendation model to obtain the predicted placement effect of the resource creativity output by the creativity recommendation model;

[0008] According to the predicted placement effect corresponding to each resource creativity, determine a target resource creativity from the multiple resource creativities, and recommend the resource to be recommended based on the target resource creativity.

[0009] As an optional implementation, the resource creativity includes a creative picture, and obtaining the graphic and text characteristics corresponding to each resource creativity among the multiple resource creativities corresponding to the resource to be recommended includes:

[0010] For each resource creativity, preprocess the creative picture corresponding to the resource creativity to obtain a processed picture;

[0011] Process the processed picture through a pre-trained first processing model to obtain an image vector corresponding to the creative picture;

[0012] Identify the text information of the creative picture, and perform word segmentation processing on the text information to obtain a segmented text;

[0013] Process the segmented text through a pre-trained second processing model to obtain a text vector corresponding to the creative picture;

[0014] Perform fusion processing on the image vector and the text vector to obtain the graphic and text features of the resource creativity.

[0015] As an optional implementation manner, the performing fusion processing on the image vector and the text vector to obtain the graphic and text features of the resource creativity includes:

[0016] Concatenate the image vector and the text vector to obtain the graphic and text features of the resource creativity;

[0017] Or,

[0018] Determine the image weight corresponding to the image vector;

[0019] According to the image weight, perform weighted fusion on the image vector and the text vector to obtain the graphic and text features of the resource creativity.

[0020] As an optional implementation manner, the determining the image weight corresponding to the image feature includes:

[0021] Obtain the placement position corresponding to the resource to be recommended;

[0022] According to the placement position, determine the image weight corresponding to the image vector.

[0023] As an optional implementation manner, the creative recommendation model processes the received historical placement effect features, the graphic and text features, and the user features in the following manner:

[0024] According to the historical placement effect features, determine the feature weight corresponding to the historical placement effect features;

[0025] According to the feature weight, perform weighted fusion on the historical placement effect features and the graphic and text features to obtain a fusion feature;

[0026] Process the fusion feature and the user feature to obtain the predicted placement effect corresponding to the resource creativity.

[0027] As an optional implementation manner, the historical placement effect features include the exposure times of the resource creativity;

[0028] Determining the feature weight corresponding to the historical delivery effect feature according to the historical delivery effect feature includes:

[0029] Obtaining the corresponding relationship between the preset feature weight and the number of exposures; wherein, the feature weight and the number of exposures are in a direct proportion relationship;

[0030] Determining the feature weight corresponding to the historical delivery effect feature according to the corresponding relationship and the number of exposures.

[0031] As an optional implementation manner, obtaining the historical delivery effect feature corresponding to each resource creative among the multiple resource creatives corresponding to the to-be-recommended resource includes:

[0032] Obtaining the unique identifier of the resource creative;

[0033] According to the unique identifier, obtaining the user interaction data corresponding to the resource creative from the preset historical log;

[0034] Determining the number of exposures and the number of clicks of the resource creative according to the user interaction data;

[0035] Determining the average click-through rate of the resource creative according to the number of exposures and the number of clicks;

[0036] Determining the number of exposures, the number of clicks, and the average click-through rate as the historical delivery effect feature of the resource creative.

[0037] As an optional implementation manner, the obtaining the user interaction data corresponding to the resource creative from the preset historical log according to the unique identifier includes:

[0038] According to the unique identifier, obtaining the user interaction data corresponding to the resource creative within a preset historical time period from the preset historical log;

[0039] And / or

[0040] Obtaining the preset user screening feature;

[0041] According to the unique identifier and the user screening feature, obtaining the user interaction data corresponding to the resource creative from the preset historical log.

[0042] In a second aspect, the present application provides a resource recommendation device, and the device includes:

[0043] An acquisition module, configured to, when it is determined to recommend a resource to be recommended, acquire the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each of the multiple resource ideas corresponding to the resource to be recommended;

[0044] A model processing module, configured to, for each resource idea, input the historical delivery effect characteristics, the graphic and text characteristics, and the user characteristics corresponding to the resource idea into a pre-trained creative recommendation model, and obtain the predicted delivery effect of the resource idea output by the creative recommendation model;

[0045] A determination module, configured to determine a target resource idea from the multiple resource ideas according to the predicted delivery effect corresponding to each resource idea, and recommend the resource to be recommended based on the target resource idea.

[0046] As an optional implementation manner, the resource idea includes a creative picture, and the acquisition module includes:

[0047] A preprocessing sub-module, configured to, for each resource idea, preprocess the creative picture corresponding to the resource idea to obtain a processed picture;

[0048] A first processing sub-module, configured to process the processed picture through a pre-trained first processing model to obtain an image vector corresponding to the creative picture;

[0049] An identification sub-module, configured to identify the text information of the creative picture and perform word segmentation processing on the text information to obtain a segmented text;

[0050] A second processing sub-module, configured to process the segmented text through a pre-trained second processing model to obtain a text vector corresponding to the creative picture;

[0051] A fusion processing sub-module, configured to perform fusion processing on the image vector and the text vector to obtain the graphic and text characteristics of the resource idea.

[0052] As an optional implementation manner, the fusion processing sub-module includes:

[0053] A splicing unit, configured to splice the image vector and the text vector to obtain the graphic and text characteristics of the resource idea;

[0054] Or,

[0055] A determination unit, configured to determine an image weight corresponding to the image vector;

[0056] A fusion unit, configured to perform weighted fusion on the image vector and the text vector according to the image weight to obtain the graphic and text characteristics of the resource idea.

[0057] As an alternative implementation, the determining unit is specifically configured to:

[0058] Obtain the placement position corresponding to the resource to be recommended;

[0059] Determine the image weight corresponding to the image vector according to the placement position.

[0060] As an alternative implementation, the model processing module includes:

[0061] The model processing sub-module is used to process the received historical placement effect features, the text and image features, and the user features by the creative recommendation model in the following manner:

[0062] Determine the feature weight corresponding to the historical placement effect feature according to the historical placement effect feature;

[0063] Perform weighted fusion on the historical placement effect feature and the text and image features according to the feature weight to obtain a fusion feature;

[0064] Process the fusion feature and the user feature to obtain the predicted placement effect corresponding to the resource creativity.

[0065] As an alternative implementation, the historical placement effect feature includes the exposure times of the resource creativity;

[0066] The model processing sub-module is specifically configured to:

[0067] The determining the feature weight corresponding to the historical placement effect feature according to the historical placement effect feature includes:

[0068] Obtain the corresponding relationship between the preset feature weight and the exposure times; wherein, the feature weight and the exposure times are in a direct proportion relationship;

[0069] Determine the feature weight corresponding to the historical placement effect feature according to the corresponding relationship and the exposure times.

[0070] As an alternative implementation, the obtaining module includes:

[0071] The first obtaining sub-module is used to obtain the unique identifier of the resource creativity;

[0072] The second obtaining sub-module is used to obtain the user interaction data corresponding to the resource creativity from the preset historical log according to the unique identifier;

[0073] The first determination sub-module is configured to determine the exposure times and click times of the resource idea according to the user interaction data;

[0074] The second determination sub-module is configured to determine the average click-through rate of the resource idea according to the exposure times and the click times;

[0075] The third determination sub-module is configured to determine the exposure times, the click times, and the average click-through rate as the historical delivery effect characteristics of the resource idea.

[0076] As an optional implementation manner, the second acquisition sub-module is specifically configured to:

[0077] Obtain the user interaction data corresponding to the resource idea within a preset historical time period from a preset historical log according to the unique identifier;

[0078] And / or,

[0079] Obtain a preset user screening feature;

[0080] Obtain the user interaction data corresponding to the resource idea from a preset historical log according to the unique identifier and the user screening feature.

[0081] In a third aspect, the present application provides an electronic device, including: a processor and a memory, where the processor is configured to execute a resource recommendation program stored in the memory to implement the resource recommendation method according to any one of the first aspect.

[0082] In a fourth aspect, the present application provides a storage medium, where the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the resource recommendation method according to any one of the first aspect.

[0083] The technical solution provided by the embodiments of the present application, when it is determined to recommend the resource to be recommended, obtains the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each of the multiple resource ideas corresponding to the resource to be recommended. For each resource idea, the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource idea are input into a pre-trained idea recommendation model to obtain the predicted delivery effect of the resource idea output by the idea recommendation model. According to the predicted delivery effect corresponding to each resource idea, a target resource idea is determined from the multiple resource ideas, and the resource to be recommended is recommended based on the target resource idea. In this technical solution, when determining the target resource idea of the resource to be recommended, the pre-trained idea recommendation model can be used to process the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource idea, so as to determine the target resource idea of the resource to be recommended. Among them, the graphic and text characteristics refer to the characteristics corresponding to the image and text information of the resource idea itself. By obtaining the graphic and text characteristics, the generalization ability of the idea recommendation model to evaluate the resource idea can be improved, while the acquisition of the historical delivery effect and user characteristics can ensure the memory ability of the idea recommendation model to evaluate the resource idea. Therefore, the idea recommendation model can simultaneously have the generalization ability and the memory ability. Then, the target resource idea determined by using the idea recommendation model can be more accurate, realizing the improvement of the accuracy of determining the resource idea, thereby improving the accuracy of resource recommendation and enhancing the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present invention and, together with the specification, are used to explain the principles of the present invention.

[0085] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments or the prior art text. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0086] One or more embodiments are illustrated by way of example in the accompanying drawings, and these exemplary illustrations do not limit the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements, unless otherwise stated, and the drawings in the drawings do not constitute a proportional limitation.

[0087] Figure 1 It is a flowchart of an embodiment of a resource recommendation method provided by an embodiment of the present application;

[0088] Figure 2 It is a flowchart of an embodiment of another resource recommendation method provided by an embodiment of the present application;

[0089] Figure 3 The flowchart of an embodiment of another resource recommendation method provided by an embodiment of this application;

[0090] Figure 4 The flowchart of an embodiment of yet another resource recommendation method provided by an embodiment of this application;

[0091] Figure 5 The block diagram of an embodiment of a resource recommendation device provided by an embodiment of this application;

[0092] Figure 6 The structural schematic diagram of an electronic device provided by an embodiment of this application. Detailed implementation manners

[0093] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will, with reference to the accompanying drawings in the embodiments of this application, clearly and completely describe the technical solutions in the embodiments of this application. Apparently, the described embodiments are only a part rather than all of the embodiments of this application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0094] The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, components and settings of specific examples are described below. Of course, they are merely examples and are not intended to limit the present invention. In addition, the present invention may repeat reference numerals and / or letters in different examples. Such repetition is for the purpose of simplification and clarity and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0095] To solve the technical problem that in the prior art, the screening of resource ideas depends on the historical delivery data of resource ideas. When the historical delivery data of resource ideas is scarce, the accuracy of the determined resource ideas will be reduced, affecting the user experience. This application provides a resource recommendation method, device, electronic device, and storage medium. When determining the target resource idea of the resource to be recommended, the pre-trained idea recommendation model can be used to process the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource idea, so as to determine the target resource idea of the resource to be recommended. Among them, the graphic and text characteristics refer to the characteristics corresponding to the image and text information of the resource idea itself. By obtaining the graphic and text characteristics, the generalization ability of the idea recommendation model to evaluate resource ideas can be improved, while the acquisition of historical delivery effects and user characteristics can ensure the memory ability of the idea recommendation model to evaluate resource ideas. Therefore, the idea recommendation model can simultaneously possess generalization ability and memory ability. Then, the target resource idea determined by using this idea recommendation model can be more accurate, achieving the improvement of the accuracy of determining resource ideas, thereby improving the accuracy of resource recommendation and enhancing the user experience.

[0096] The following further explains the resource recommendation method provided by this application with specific examples in conjunction with the accompanying drawings. The examples do not constitute a limitation on the embodiments of the present invention.

[0097] See Figure 1 , which is the flowchart of the embodiment of a resource recommendation method provided by the embodiment of this application. As Figure 1 shown, this process may include the following steps:

[0098] Step 101: In the case of determining to recommend a resource to be recommended, obtain the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each resource idea among the multiple resource ideas corresponding to the resource to be recommended.

[0099] The above-mentioned resource to be recommended refers to the resource determined to be recommended to the user. The resource to be recommended may be an advertisement, a movie or TV drama, a book or a novel, etc. The embodiments of this application do not limit this.

[0100] The above-mentioned resource idea refers to a picture, text, or graphic and text generated by using material basic elements (such as pictures, titles, etc.) to represent the characteristics of the resource.

[0101] The above-mentioned historical delivery effect characteristics refer to the delivery effect characteristics corresponding to the resource when the resource idea is delivered in the historical time period. The historical delivery effect characteristics may include, but are not limited to: the click-through rate of the user on the resource idea, the number of clicks, the number of exposures, and the exposure rate, etc.

[0102] The above-mentioned graphic and text features refer to the features contained in the resource creativity itself. Since resource creativity is generally a picture containing text, the features of the resource creativity itself can be the graphic and text features expressed by the picture corresponding to the resource creativity.

[0103] The above-mentioned user features refer to the corresponding features of the users who view the resource to be delivered during the process of delivering the resource to be delivered in the historical time period, which may include but are not limited to: age features, gender features, and geographical location features, etc.

[0104] In the embodiments of the present application, when it is determined that a resource to be recommended needs to be recommended currently, in order to determine the target resource creativity applied when recommending the resource to be recommended, the execution entity of the embodiments of the present application may determine the historical delivery effect features, graphic and text features, and user features of each resource creativity among the multiple resource creativities corresponding to the resource to be recommended.

[0105] As an optional implementation manner, there may be a preset resource library in the present application. The resource library may include multiple resources, and each resource may correspond to multiple resource creativities. Based on this, when the execution entity of the embodiments of the present application determines that a resource needs to be recommended currently through the operations of the user on the terminal device, it may determine the resource to be delivered from multiple resources according to the current context information, application scenario, or delivery location and other information.

[0106] After that, in order to determine the resource creativity applied to the above-mentioned resource to be delivered, the execution entity of the embodiments of the present application may obtain the historical delivery effect features, graphic and text features, and user features corresponding to each resource creativity among the multiple resource creativities included in the resource to be delivered.

[0107] As an optional implementation manner, when obtaining the user features of each resource creativity, the user information of the users who click on the resource to be delivered during the delivery of the resource to be delivered in the historical time period may be obtained from a preset database. After that, the user feature analysis model trained in advance may be used to analyze the user information to obtain the user features corresponding to the resource to be delivered.

[0108] As for how to obtain the graphic and text features and historical delivery effect features of the resource creativity specifically, it will be described separately in the following through Figure 2 and Figure 4 the processes shown, and will not be elaborated here first.

[0109] Step 102: For each resource creativity, input the historical delivery effect features, graphic and text features, and user features corresponding to the resource creativity into the pre-trained creativity recommendation model to obtain the predicted delivery effect of the resource creativity output by the creativity recommendation model.

[0110] Step 103: Determine a target resource idea from multiple resource ideas according to the predicted delivery effect corresponding to each resource idea, and recommend the above-mentioned resources to be recommended based on the target resource idea.

[0111] The following is a unified description of steps 102 and 103:

[0112] The above-mentioned idea recommendation model refers to a pre-trained model that predicts the delivery effect of a resource idea based on the historical delivery effect characteristics, graphic and text characteristics, and user characteristics of the resource idea. The input of this model can be the historical delivery effect characteristics, graphic and text characteristics, and user characteristics of each resource idea, and the output is the predicted delivery effect corresponding to the resource idea.

[0113] Optionally, the above-mentioned idea recommendation model can be a two-tower structure. In this structure, the idea recommendation model can include a resource tower and a user tower. Among them, the resource tower can be used to process features related to the resource tower, such as the above-mentioned historical delivery effect characteristics and graphic and text characteristics; the above-mentioned user tower can be used to process features related to the user, such as the above-mentioned user characteristics.

[0114] The above-mentioned predicted delivery effect refers to the effect after the resource idea is delivered as predicted by the idea recommendation model, which can be represented by the click-through rate, the number of clicks, or other features. The embodiments of the present application do not limit this.

[0115] In the embodiments of the present application, in order to select an optimal resource idea from multiple resource ideas corresponding to the resources to be delivered as the idea for delivering the resources to be delivered, after the execution subject of the embodiments of the present application obtains the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each resource idea, for each resource idea, the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource idea can be input into the above-mentioned pre-trained idea recommendation model, and the predicted delivery effect of the resource idea output by the idea recommendation model can be obtained.

[0116] As an optional implementation manner, when training the above-mentioned idea recommendation model, a resource feature sample set can be determined from a preset database. The resource feature sample set can include multiple resource feature samples. Each resource feature sample can include a historical delivery effect feature sample, a graphic and text feature sample, and a user feature sample of a resource idea, and each resource feature sample can correspond to a delivery effect sample. The above-mentioned database can be used to store the delivery details information of each resource in the historical time period.

[0117] As an optional implementation manner, when determining the resource feature sample set from the database, the delivery details information corresponding to each resource sample can be obtained from the database, and the delivery details information can be processed to obtain the above-mentioned resource feature sample set.

[0118] After that, the above resource feature sample set can be used to train a preset initial creative recommendation model, and the initial predicted placement effect output by the initial creative recommendation model obtained from each training can be obtained.

[0119] After that, the current loss value can be calculated according to a preset loss value function for the above initial predicted placement effect and the corresponding placement effect sample.

[0120] Optionally, when the loss value is less than a preset threshold or the number of training times reaches a preset number threshold, it can be determined that the current training is completed, and a creative recommendation model is obtained.

[0121] Optionally, when the loss value is greater than or equal to the preset threshold, it means that the training of the initial creative recommendation model is not completed at this time. Therefore, the parameters in the initial creative recommendation model can be adjusted according to a preset step size, and after the adjustment, the preset initial creative recommendation model is continuously trained using the resource feature sample set until the obtained loss value is less than the preset threshold or the number of training times reaches the preset number threshold.

[0122] In the embodiment of the present application, after obtaining the predicted placement effect of each resource creative corresponding to the resource to be placed, the execution subject of the embodiment of the present application can determine the target resource creative from multiple resource creatives according to the predicted placement effect corresponding to each resource creative, and recommend the above-mentioned resource to be recommended based on the target resource creative.

[0123] As an optional implementation manner, the resource creative corresponding to the best predicted placement effect among multiple resource creatives can be determined as the target resource creative.

[0124] As an exemplary implementation manner, the above predicted placement effect can be the predicted click-through rate when the resource to be placed is placed with this resource creative. Based on this, the execution subject of the embodiment of the present application can determine the resource creative with the largest click-through rate as the target resource creative.

[0125] After that, the resource to be recommended can be recommended with this target resource creative, so that the target resource creative can be applied when the resource to be placed is placed.

[0126] The technical solution provided by the embodiments of the present application, when it is determined to recommend a resource to be recommended, obtains, for each of the multiple resource ideas corresponding to the resource to be recommended, the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each resource idea. For each resource idea, the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource idea are input into a pre-trained idea recommendation model, and the predicted delivery effect of the resource idea output by the idea recommendation model is obtained. According to the predicted delivery effect corresponding to each resource idea, a target resource idea is determined from the multiple resource ideas, and the above-mentioned resource to be recommended is recommended based on the target resource idea. In this technical solution, when determining the target resource idea of the resource to be recommended, the pre-trained idea recommendation model can be used to process the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to the resource idea, so as to determine the target resource idea of the resource to be recommended. Among them, the graphic and text characteristics refer to the characteristics corresponding to the image and text information of the resource idea itself. By obtaining the graphic and text characteristics, the generalization ability of the idea recommendation model to evaluate the resource idea can be improved, while the acquisition of the historical delivery effect and user characteristics can ensure the memory ability of the idea recommendation model to evaluate the resource idea. Therefore, the idea recommendation model can simultaneously have generalization ability and memory ability, and then the target resource idea determined by using the idea recommendation model can be more accurate, realizing the improvement of the accuracy of determining the resource idea, thereby improving the accuracy of resource recommendation and enhancing the user experience.

[0127] See Figure 2 , which is a flowchart of an embodiment of another resource recommendation method provided by the embodiments of the present application. Figure 2 The process shown Figure 1 On the basis of the process shown, it describes how to obtain the graphic and text characteristics corresponding to each resource idea specifically. As Figure 2 shown, the process may include the following steps:

[0128] Step 201, when the resource idea includes a creative picture, for each resource idea, preprocess the creative picture corresponding to the resource idea to obtain a processed picture.

[0129] The above-mentioned creative picture refers to the picture corresponding to the resource idea, which may include text or may not include text. The embodiments of the present application do not limit this.

[0130] The above-mentioned preprocessing refers to preprocessing the creative picture before extracting the characteristics of the creative picture. The preprocessing may include, but is not limited to: scaling the creative picture, updating the pixels of the creative picture so that the pixels of the creative picture are within a preset pixel range, etc.

[0131] The above-mentioned processed picture is the picture obtained after preprocessing the creative picture.

[0132] In the embodiments of the present application, in order to more accurately extract the image features of the creative picture, the execution subject of the embodiments of the present application may first preprocess the creative picture so that the creative picture meets the preset feature extraction conditions.

[0133] As an alternative implementation, the above feature extraction conditions may include that the size of the creative picture is smaller than a preset size threshold. Based on this, the execution subject of the embodiments of the present application may determine for each creative picture whether the size of the creative picture is smaller than the above size threshold.

[0134] Optionally, when it is determined that the size of the creative picture is smaller than the above size threshold, it may be determined that the creative picture meets the feature extraction conditions, so the creative picture may not be preprocessed.

[0135] Optionally, when it is determined that the size of the creative picture is greater than or equal to the above size threshold, it may be determined that the creative picture does not meet the feature extraction conditions, so the creative picture may be scaled so that the size of the creative picture is smaller than the above size threshold.

[0136] As another alternative implementation, the above feature extraction conditions may include that the pixel value of the creative picture is within a preset pixel range. Based on this, the execution subject of the embodiments of the present application may determine for each creative picture whether the pixel value of the creative picture is within the above pixel range.

[0137] Optionally, when it is determined that the pixel value of the creative picture is within the above pixel range, it may be determined that the creative picture meets the feature extraction conditions, so the creative picture may not be preprocessed.

[0138] Optionally, when it is determined that the pixel value of the creative picture is not within the above pixel range, it may be determined that the creative picture does not meet the feature extraction conditions, so the resolution of the creative picture may be updated so that the pixel value of the creative picture is within the above pixel range.

[0139] As yet another alternative implementation, the above feature extraction conditions may include that the size of the creative picture is smaller than a preset size threshold and the pixel value of the creative picture is within a preset pixel range. Based on this, the execution subject of the embodiments of the present application may determine for each creative picture whether the size of the creative picture is smaller than the above size threshold and whether the pixel value of the creative picture is within the above pixel range.

[0140] Optionally, when it is determined that the size of the creative picture is smaller than the above size threshold and the pixel value of the creative picture is within the above pixel range, it may be determined that the creative picture meets the feature extraction conditions, so the creative picture may not be preprocessed.

[0141] Optionally, when it is determined that the size of the creative picture is greater than or equal to the above size threshold and the pixel value of the creative picture is within the above pixel range, it can be determined that the creative picture does not meet the feature extraction condition. Therefore, the creative picture can be scaled so that the size of the creative picture is less than the above size threshold.

[0142] Optionally, when it is determined that the size of the creative picture is less than the above size threshold and the pixel value of the creative picture is not within the above pixel range, it can be determined that the creative picture does not meet the feature extraction condition. Therefore, the resolution of the creative picture can be updated so that the pixel value of the creative picture is within the above pixel range.

[0143] Optionally, when it is determined that the size of the creative picture is greater than or equal to the above size threshold and the pixel value of the creative picture is not within the above pixel range, it can be determined that the creative picture does not meet the feature extraction condition. Therefore, the creative picture can be scaled so that the size of the creative picture is less than the above size threshold, and the resolution of the creative picture can be updated so that the pixel value of the creative picture is within the above pixel range.

[0144] Step 202: Process the above processed picture through a pre-trained first processing model to obtain an image vector corresponding to the creative picture.

[0145] The above first processing model refers to a pre-trained model for extracting picture features, and this model can be trained based on the CN-CLIP (Chinese Contrastive Language Image Pretraining) large model.

[0146] The above image vector refers to the picture features extracted from the processed picture.

[0147] In the embodiments of the present application, the execution entity of the embodiments of the present application can pre-train a picture feature extraction model for extracting picture features (for ease of description, hereinafter referred to as the "first processing model").

[0148] Based on this, when the execution entity of the embodiments of the present application extracts the picture features in the processed picture corresponding to the resource creativity, the above processed picture can be processed through the first processing model to obtain an image vector corresponding to the creative picture.

[0149] As an implementation manner, the above processed picture can be input into the first processing model, and an image vector corresponding to the processed picture output by the first processing model can be obtained.

[0150] Step 203: Identify the text information of the creative picture and perform word segmentation on the above text information to obtain segmented text.

[0151] Step 204: Process the above tokenized text through a pre-trained second processing model to obtain the text vector corresponding to the creative picture.

[0152] The following provides a unified description of steps 203 and 204:

[0153] The above text information refers to the text content contained in the creative picture. In order to more accurately describe the resource to be recommended, the creative picture corresponding to the resource to be recommended generally may include corresponding text description information, such as a title, a brief introduction, etc.

[0154] The above second processing model refers to a pre-trained model for extracting text features of text information. This model can be trained based on the CN-CLIP (Chinese Contrastive Language Image Pretraining) large model.

[0155] The above text vector refers to the text features corresponding to the text information contained in the creative picture.

[0156] In the embodiments of the present application, in order to more accurately extract the own content of the resource creativity, the execution subject of the embodiments of the present application, in addition to extracting the picture features of the resource creativity, can also extract the text features of the resource creativity.

[0157] As an optional implementation manner, the execution subject of the embodiments of the present application can pre-train a text feature extraction model for extracting text features (hereinafter referred to as the "second processing model" for convenience of description). Based on this, when extracting the text features of the resource creativity, the text information in the above creative picture can be recognized and tokenized to obtain the corresponding tokenized text.

[0158] After that, the above tokenized text can be input into the pre-trained second processing model, and the text vector of the creative picture output by the second processing model can be obtained.

[0159] Step 205: Perform a fusion process on the above image vector and text vector to obtain the graphic and text features of the resource creativity.

[0160] The above graphic and text features refer to the comprehensive features of the creative picture that simultaneously include the image features and text features of the creative picture.

[0161] In the embodiments of the present application, after obtaining the image vector and text vector of the creative picture, in order to simplify the processing flow, the execution subject of the embodiments of the present application can fuse the above image vector and text vector to obtain the graphic and text features of the creative picture.

[0162] As an alternative implementation, the above-mentioned image vector and text vector can be directly concatenated and fused to obtain a multi-modal vector, and this multi-modal vector can be determined as the graphic and text feature of the resource creativity.

[0163] As another alternative implementation, the image weight corresponding to the image vector can be determined, and based on the above-mentioned image weight, the above-mentioned image vector and text vector can be weighted and fused to obtain the graphic and text feature of the resource creativity.

[0164] As an exemplary implementation, each creative picture can correspond to a preset weight value. Therefore, the execution entity of the embodiment of the present application can directly obtain this image weight.

[0165] As another exemplary implementation, the execution entity of the embodiment of the present application can obtain the placement position corresponding to the resource to be recommended, and based on this placement position, determine the image weight corresponding to the image vector.

[0166] As an implementation, a weight analysis model can be pre-trained, and this placement position can be input into the above-mentioned weight analysis model to obtain the image weight output by the weight analysis model.

[0167] As yet another exemplary implementation, the execution entity of the embodiment of the present application can analyze the text information in the above-mentioned creative picture to determine the text content type corresponding to the text information.

[0168] After that, the target text weight corresponding to this text content type can be determined from the pre-set correspondence between the text content type and the text weight.

[0169] Finally, the preset value (such as 1) can be subtracted from the above-mentioned target text weight to obtain the image weight corresponding to the image vector.

[0170] In an embodiment, when performing weighted fusion on the image vector and the text vector according to the image weight, the preset value can be subtracted from the image weight to obtain the text weight. After that, the above-mentioned image vector and the image weight can be multiplied to obtain a first vector, and the above-mentioned text vector and the text weight can be multiplied to obtain a second vector. Finally, the above-mentioned first vector and the second vector can be added to obtain the graphic and text feature of the resource creativity.

[0171] The technical solution provided by the embodiments of the present application, when the resource idea includes an idea picture, for each resource idea, preprocess the idea picture corresponding to the resource idea to obtain a processed picture, process the above-mentioned processed picture through a pre-trained first processing model to obtain an image vector corresponding to the idea picture, identify the text information of the idea picture, and perform word segmentation processing on the above-mentioned text information to obtain a segmented text. Process the above-mentioned segmented text through a pre-trained second processing model to obtain a text vector corresponding to the idea picture, and perform fusion processing on the above-mentioned image vector and text vector to obtain the graphic and text features of the resource idea. This technical solution, by pre-training a first processing model for extracting image vectors and a second processing model for extracting text vectors, and using the first processing model and the second processing model to extract the image vectors and text vectors of the idea picture respectively, so as to fuse the image vectors and text vectors to obtain the graphic and text features of the idea picture. By extracting the features of the idea picture through the model, the accuracy and efficiency of feature extraction can be improved, and the graphic and text features of the idea picture can be accurately and efficiently extracted.

[0172] See Figure 3 , which is a flowchart of an embodiment of another resource recommendation method provided by the embodiments of the present application. Figure 3 The process shown Figure 1 On the basis of the process shown, it describes how the creative recommendation model specifically processes the received historical delivery effect features, graphic and text features, and user features. As Figure 3 shown, the process may include the following steps:

[0173] Step 301: Determine the feature weight corresponding to the historical delivery effect feature according to the historical delivery effect feature.

[0174] Step 302: Perform weighted fusion on the historical delivery effect feature and the graphic and text feature according to the above feature weight to obtain a fusion feature.

[0175] The following is a unified description of Step 301 and Step 302:

[0176] The above-mentioned fusion feature refers to the feature obtained by fusing the historical delivery effect feature and the graphic and text feature, that is, the fusion feature contains both the historical delivery effect feature and the graphic and text feature of the resource idea.

[0177] In the embodiments of the present application, since both the historical delivery effect feature and the graphic and text feature belong to the relevant features of the resource idea delivery, the above-mentioned historical delivery effect feature and the graphic and text feature can be fused to obtain the corresponding fusion feature.

[0178] As an alternative implementation, in order to balance the historical delivery effect features and the graphic features, the creative recommendation model may determine the feature weights corresponding to the historical delivery effect features, and fuse the historical delivery effect features and the graphic features according to the above feature weights to obtain fused features.

[0179] As an exemplary implementation, the above historical delivery effect features may include the number of exposures of the creative resource during historical delivery, and the number of exposures refers to the number of times the resource creative is shown to the audience.

[0180] Based on this, when determining the feature weights corresponding to the historical delivery effect features according to the historical delivery effect features, the corresponding relationship between the preset feature weights and the number of exposures can be obtained. Among them, the above feature weights may be in a proportional relationship with the number of exposures, that is, the greater the number of exposures of the resource creative, the greater the corresponding feature weight, so as to strengthen the memory ability of the model; the smaller the number of exposures of the resource creative model, the more the model can rely on the features of the creative itself, that is, the graphic features. Therefore, the above feature weights are reduced to enhance the generalization ability of the model.

[0181] After that, the feature weights corresponding to the historical delivery effect features can be determined according to the above corresponding relationship and the number of exposures.

[0182] As an implementation, the above number of exposures can be used as a keyword to search for the above corresponding relationship to obtain the target corresponding relationship including the above keyword. After that, the weight value included in the target corresponding relationship can be determined as the feature weight corresponding to the historical delivery effect feature.

[0183] Step 303: Process the above fused features and user features to obtain the predicted delivery effect corresponding to the resource creative.

[0184] In the embodiments of the present application, the creative recommendation model may be a two-tower model, that is, it may include two sub-models: a creative sub-model and a user sub-model. Among them, the creative sub-model can be used to process the relevant features of the resource creative, and the user sub-model can be used to process the features related to the user features.

[0185] Based on this, the creative recommendation model can use the above creative sub-model and user sub-model to process the above fused features and user features respectively. After that, the predicted delivery effect of the resource creative model can be determined according to the processing results of the creative sub-model and the user sub-model.

[0186] The technical solution provided by the embodiments of the present application determines the feature weight corresponding to the historical placement effect feature according to the historical placement effect feature, and performs weighted fusion on the historical placement effect feature and the graphic and text feature according to the above feature weight to obtain a fusion feature, and processes the above fusion feature and the user feature to obtain the predicted placement effect corresponding to the resource creative. This technical solution dynamically determines the feature weight of the historical placement effect feature according to the historical placement effect feature, so as to balance the historical placement effect feature and the graphic and text feature according to the feature weight, thereby achieving a dynamic balance between the memory ability and the generalization ability of the creative recommendation model, and realizing the improvement of the accuracy of the creative recommendation model in predicting the placement effect of the resource creative.

[0187] See Figure 4 , which is a flowchart of an embodiment of another resource recommendation method provided by the embodiments of the present application. Figure 4 The process shown Figure 1 On the basis of the process shown, it describes how to obtain the historical placement effect feature of each resource creative specifically. As Figure 4 shown, the process may include the following steps:

[0188] Step 401: Obtain the unique identifier of the resource creative.

[0189] Step 402: According to the above unique identifier, obtain the user interaction data corresponding to the resource creative from the preset historical log.

[0190] The following is a unified description of Step 401 and Step 402:

[0191] The above unique identifier refers to the identifier used to identify the resource creative. In order to identify each resource creative, each resource creative may correspond to a unique identifier.

[0192] The above historical log is used to record the interaction information between the terminal device and the server during the process of placing the resource to be placed in the historical time period, and the above terminal device is the terminal device where the resource to be placed is placed.

[0193] In the embodiments of the present application, since each resource creative may have a unique identifier, and the resource creative is recorded in the historical log with the above unique identifier, therefore, when the execution subject of the embodiments of the present application determines the historical placement effect feature of the resource to be placed, it may obtain the unique identifier of the resource creative.

[0194] After that, according to the unique identifier, the user interaction data corresponding to the resource creative can be obtained from the above preset historical log.

[0195] As an optional implementation, the data in the historical log can be searched using the above unique identifier as a keyword to determine the user interaction data corresponding to the unique identifier.

[0196] As an alternative implementation, in order to obtain the historical delivery effect characteristics of a resource idea in a recent period of time, the execution entity of the embodiment of the present application can obtain, according to the above unique identifier, the user interaction data corresponding to the resource idea within a preset historical time period from a preset historical log.

[0197] As another alternative implementation, in order to obtain the historical delivery effect characteristics of the resource idea for a preset user, the execution entity of the embodiment of the present application can obtain preset user screening characteristics. Subsequently, according to the above unique identifier and the user screening characteristics, the user interaction data corresponding to the resource idea can be obtained from the preset historical log. The above user screening characteristics refer to preset user characteristics for screening specific users, which may include, but are not limited to: age characteristics, gender characteristics, and location characteristics, etc.

[0198] As yet another alternative implementation, in order to obtain the historical delivery effect characteristics of a resource idea for a specific user in a recent period of time, the execution entity of the embodiment of the present application can obtain preset user screening characteristics. Subsequently, according to the above unique identifier and the user screening characteristics, the user interaction data corresponding to the resource idea within a preset historical time period can be obtained from the preset historical log.

[0199] Step 403: Determine the exposure times and click times of the resource idea according to the above user interaction data.

[0200] Step 404: Determine the average click-through rate of the resource idea according to the above exposure times and click times.

[0201] Step 405: Determine the exposure times, click times, and the above average click-through rate as the historical delivery effect characteristics of the resource idea.

[0202] The following provides a unified description of steps 403 to 405:

[0203] The above exposure times refer to the total number of times the resource idea is displayed to users.

[0204] The above click times refer to the total number of times users click on the resource to be delivered corresponding to the displayed resource idea.

[0205] The above average click-through rate refers to the probability that the resource to be delivered using the resource idea is clicked after being exposed, and it can be the ratio of the above click times to the exposure times.

[0206] In the embodiments of the present application, after obtaining the user interaction data corresponding to the resource idea, the user interaction data can be analyzed to determine the exposure times and click times corresponding to the resource idea.

[0207] After that, the average click-through rate corresponding to the resource idea can be determined according to the above-mentioned exposure times and click times.

[0208] As an optional implementation manner, the above-mentioned click times can be divided by the above-mentioned exposure times to obtain the average click-through rate corresponding to the resource idea.

[0209] Finally, the above-mentioned exposure times, click times, and average click-through rate can be determined as the historical delivery effect characteristics of the resource idea.

[0210] In addition, through Figure 3 the process shown, the resource recommendation model needs to fuse the historical delivery effect characteristics and the text and image characteristics. Since there are many characteristics included in the historical delivery effect characteristics, the vector corresponding to the historical delivery effect characteristics is a vector with a dimension higher than the dimension of the text and image characteristics. Therefore, an embedding layer can be included in the resource recommendation model, and the embedding layer can convert the vector of the historical delivery effect characteristics to obtain a low-dimensional vector. After that, the low-dimensional vector can be input into a deep neural network for processing, so as to fuse the low-dimensional historical delivery effect characteristics and the text and image characteristics through the deep neural network to obtain the fused characteristics.

[0211] The technical solution provided by the embodiments of the present application obtains the unique identifier of the resource idea, obtains the user interaction data corresponding to the resource idea from the preset historical log according to the above-mentioned unique identifier, determines the exposure times and click times of the resource idea according to the above-mentioned user interaction data, determines the average click-through rate of the resource idea according to the above-mentioned exposure times and click times, and determines the above-mentioned exposure times, click times, and the above-mentioned average click-through rate as the historical delivery effect characteristics of the resource idea. This technical solution realizes the fast and accurate determination of the historical delivery effect characteristics of the resource idea by obtaining the user interaction data related to the resource idea within the historical time period according to the unique identifier of the resource idea, and then determining the historical delivery effect characteristics of the resource idea according to the historical interaction data.

[0212] See Figure 5 for the block diagram of an embodiment of a resource recommendation device provided by the embodiments of the present application. As Figure 5 shown, the device may include:

[0213] An obtaining module 51, configured to obtain, when it is determined to recommend a to-be-recommended resource, the historical delivery effect characteristics, text and image characteristics, and user characteristics corresponding to each resource idea among the multiple resource ideas corresponding to the to-be-recommended resource;

[0214] A model processing module 52, configured to input, for each resource idea, the historical delivery effect features, the graphic features, and the user features corresponding to the resource idea into a pre-trained creative recommendation model, and obtain the predicted delivery effect of the resource idea output by the creative recommendation model;

[0215] A determination module 53, configured to determine a target resource idea from multiple resource ideas according to the predicted delivery effect corresponding to each resource idea, and recommend the resource to be recommended based on the target resource idea.

[0216] As an optional implementation manner, the resource idea includes a creative picture, and the acquisition module 51 includes:

[0217] A preprocessing sub-module, configured to preprocess the creative picture corresponding to each resource idea to obtain a processed picture;

[0218] A first processing sub-module, configured to process the processed picture through a pre-trained first processing model to obtain an image vector corresponding to the creative picture;

[0219] An identification sub-module, configured to identify the text information of the creative picture and perform word segmentation processing on the text information to obtain a segmented text;

[0220] A second processing sub-module, configured to process the segmented text through a pre-trained second processing model to obtain a text vector corresponding to the creative picture;

[0221] A fusion processing sub-module, configured to perform fusion processing on the image vector and the text vector to obtain the graphic features of the resource idea.

[0222] As an optional implementation manner, the fusion processing sub-module includes:

[0223] A splicing unit, configured to splice the image vector and the text vector to obtain the graphic features of the resource idea;

[0224] Or,

[0225] A determination unit, configured to determine an image weight corresponding to the image vector;

[0226] A fusion unit, configured to perform weighted fusion on the image vector and the text vector according to the image weight to obtain the graphic features of the resource idea.

[0227] As an optional implementation manner, the determination unit is specifically configured to:

[0228] Obtain the delivery position corresponding to the resource to be recommended;

[0229] Determine the image weight corresponding to the image vector according to the placement position.

[0230] As an optional implementation, the model processing module 52 includes:

[0231] A model processing sub-module for the creative recommendation model to process the received historical placement effect features, the graphic and text features, and the user features in the following manner:

[0232] Determine the feature weight corresponding to the historical placement effect feature according to the historical placement effect feature;

[0233] Perform weighted fusion on the historical placement effect feature and the graphic and text feature according to the feature weight to obtain a fusion feature;

[0234] Process the fusion feature and the user feature to obtain the predicted placement effect corresponding to the resource creativity.

[0235] As an optional implementation, the historical placement effect feature includes the exposure times of the resource creativity;

[0236] The model processing sub-module is specifically used for:

[0237] The determining the feature weight corresponding to the historical placement effect feature according to the historical placement effect feature includes:

[0238] Obtain the correspondence between the preset feature weight and the exposure times; wherein, the feature weight is in a direct proportion relationship with the exposure times;

[0239] Determine the feature weight corresponding to the historical placement effect feature according to the correspondence and the exposure times.

[0240] As an optional implementation, the obtaining module 51 includes:

[0241] A first obtaining sub-module for obtaining the unique identifier of the resource creativity;

[0242] A second obtaining sub-module for obtaining the user interaction data corresponding to the resource creativity from the preset historical log according to the unique identifier;

[0243] A first determining sub-module for determining the exposure times and click times of the resource creativity according to the user interaction data;

[0244] A second determining sub-module for determining the average click-through rate of the resource creativity according to the exposure times and the click times;

[0245] A third determination sub-module, configured to determine the number of exposures, the number of clicks, and the average click-through rate as the historical delivery effect characteristics of the resource creative.

[0246] As an optional implementation manner, the second acquisition sub-module is specifically configured to:

[0247] Obtain, from a preset historical log according to the unique identifier, user interaction data corresponding to the resource creative within a preset historical time period;

[0248] And / or

[0249] Obtain a preset user screening feature;

[0250] Obtain, from a preset historical log according to the unique identifier and the user screening feature, user interaction data corresponding to the resource creative.

[0251] As Figure 6 shown, a schematic structural diagram of an electronic device provided by an embodiment of the present application includes a processor 61, a communication interface 62, a memory 63, and a communication bus 64. Among them, the processor 61, the communication interface 62, and the memory 63 complete mutual communication through the communication bus 64.

[0252] The memory 63 is used to store a computer program;

[0253] In an embodiment of the present application, when the processor 61 is used to execute the program stored on the memory 63, it implements the resource recommendation method provided by any one of the foregoing method embodiments, including:

[0254] When it is determined to recommend a resource to be recommended, obtain the historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each resource creative among the multiple resource creatives corresponding to the resource to be recommended;

[0255] For each resource creative, input the historical delivery effect characteristics, the graphic and text characteristics, and the user characteristics corresponding to the resource creative into a pre-trained creative recommendation model, and obtain the predicted delivery effect of the resource creative output by the creative recommendation model;

[0256] According to the predicted delivery effect corresponding to each resource creative, determine a target resource creative from the multiple resource creatives, and recommend the resource to be recommended based on the target resource creative.

[0257] An embodiment of the present application further provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the resource recommendation method provided by any one of the foregoing method embodiments.

[0258] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple grid units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0259] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the related technology, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a grid device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0260] It should be understood that the terms used in this text are only for the purpose of specific example embodiments of the text and are not intended to be restrictive. Unless otherwise clearly indicated in the context, the singular forms "a", "an", and "the" used in this text may also represent the plural form. The terms "comprising", "including", "containing", and "having" are inclusive and thus specify the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations in this text are not to be construed as necessarily requiring them to be executed in the specific order stated or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps can be used.

[0261] The above description is only the specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in this text, but will conform to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A resource recommendation method, characterized in that: The method comprises: In the case of determining to recommend the resource to be recommended, obtaining historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each of the resource creatives among multiple resource creatives corresponding to the resource to be recommended; For each resource creative idea, the historical delivery effect features, the graphic features, and the user features corresponding to the resource creative idea are input into a pre-trained creative recommendation model to obtain a predicted delivery effect of the resource creative idea output by the creative recommendation model; According to the predicted delivery effect corresponding to each resource idea, a target resource idea is determined from the plurality of resource ideas, and the resource to be recommended is recommended based on the target resource idea.

2. The method according to claim 1, characterized in that The resource creativity includes a creative picture, and obtaining the graphic features corresponding to each of the resource creativity among the multiple resource creativity corresponding to the resource to be recommended includes: For each resource creative idea, pre-processing the creative image corresponding to the resource creative idea to obtain a processed image; Processing the processing picture by using a pre-trained first processing model to obtain an image vector corresponding to the creative picture; Identify the text information of the creative image, and perform word segmentation processing on the text information to obtain a word segmentation text; Processing the segmented text by a pre-trained second processing model to obtain a text vector corresponding to the creative picture; The image vector and the text vector are fused to obtain the graphic and text features of the resource creativity.

3. The method according to claim 2, characterized in that The fusing process of the image vector and the text vector to obtain the graphic and text features of the resource creativity includes: splicing the image vector and the text vector to obtain the graphic and text features of the resource creativity; or, Determining an image weight corresponding to the image vector; According to the image weight, the image vector and the text vector are weightedly fused to obtain the graphic and text features of the resource creativity.

4. The method according to claim 3, characterized in that The determining the image weight corresponding to the image feature includes: Obtaining the placement location corresponding to the resource to be recommended; An image weight corresponding to the image vector is determined according to the placement position.

5. The method according to claim 1, characterized in that The creative recommendation model processes the received historical delivery effect features, the image and text features, and the user features in the following manner: Determine, according to the historical delivery effect characteristics, the feature weight corresponding to the historical delivery effect characteristics; According to the feature weights, weighted fusion is performed on the historical delivery effect features and the image and text features to obtain fusion features; The fusion features and the user features are processed to obtain a predicted delivery effect corresponding to the resource creativity.

6. The method according to claim 5, characterized in that The historical delivery effect characteristics include the number of exposures of the resource creativity; The determining, according to the historical delivery effect feature, a feature weight corresponding to the historical delivery effect feature comprises: Obtaining a correspondence between a preset feature weight and the number of exposures; wherein the feature weight and the number of exposures are in direct proportion; According to the corresponding relationship and the number of exposures, a feature weight corresponding to the historical delivery effect feature is determined.

7. The method according to claim 1, characterized in that The historical delivery effect characteristics corresponding to each of the resource creatives among the multiple resource creatives corresponding to the resource to be recommended are obtained, including: Obtain a unique identifier for the resource idea; According to the unique identifier, obtaining user interaction data corresponding to the resource creativity from a preset historical log; Determine the number of exposures and clicks of the resource creative according to the user interaction data; Determine an average click-through rate of the resource creative according to the number of exposures and the number of clicks; The number of exposures, the number of clicks, and the average click rate are determined as historical delivery effect characteristics of the resource creative.

8. The method according to claim 7, characterized in that The acquiring, according to the unique identifier, user interaction data corresponding to the resource creativity from a preset historical log includes: According to the unique identifier, obtaining user interaction data corresponding to the resource creativity within a preset historical time period from a preset historical log; and / or, Get the preset user filtering features; According to the unique identifier and the user screening feature, user interaction data corresponding to the resource creativity is obtained from a preset historical log.

9. A resource recommendation device, characterized in that: The device comprises: An acquisition module, for acquiring, when determining to recommend a resource to be recommended, historical delivery effect characteristics, graphic and text characteristics, and user characteristics corresponding to each of the plurality of resource creatives corresponding to the resource to be recommended; A model processing module is used to input the historical delivery effect characteristics, the graphic and text characteristics, and the user characteristics corresponding to each resource creative into a pre-trained creative recommendation model to obtain a predicted delivery effect of the resource creative output by the creative recommendation model; The determination module is used to determine a target resource idea from a plurality of resource ideas according to the predicted delivery effect corresponding to each resource idea, and recommend the resource to be recommended based on the target resource idea.

10. An electronic device, characterized in that: include: A processor and a memory, wherein the processor is used to execute a resource recommendation program stored in the memory to implement the resource recommendation method according to any one of claims 1 to 8.

11. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the resource recommendation method according to any one of claims 1 to 8.

Images