Design scheme recommendation method and device, storage medium and computer program product

Abstract

The invention discloses a design scheme recommendation method and device, a storage medium and a computer program product, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a plurality of advertisement pictures clicked at high frequency, inputting each advertisement picture into a pre-trained multi-label extraction model, obtaining a plurality of classification labels of each advertisement picture under different classification dimensions; label co-occurrence analysis is carried out on each classification label to obtain multi-label co-occurrence information, the multi-label co-occurrence information comprises at least one label matching scheme, and each label matching scheme comprises at least two classification labels; and if a design recommendation request input by a user is received, determining a target classification tag corresponding to the design recommendation request, generating a target design scheme according to the target classification tag and the multi-tag co-occurrence information, and taking the target design scheme as a recommendation result. The design efficiency of the advertisement picture is improved, and the design threshold of the advertisement picture is reduced.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a design solution recommendation method, device, storage medium and computer program product. Background Art

[0002] With the rapid development of computer technology and Internet technology, the continuous introduction of new technologies and the diversification of consumers' access to information, advertising images have become the most attractive form of information presentation for potential users and arouse their attention. Therefore, people have higher and higher requirements for the design of advertising images.

[0003] Currently, advertising images are usually designed by designers with certain professional knowledge and then released to the public. However, in this design method, users need to spend a lot of money to hire designers, and designers also need to spend a lot of time looking for existing design solutions through the Internet or other channels for reference, resulting in high thresholds for advertising image design and low design efficiency.

[0004] Therefore, how to lower the design threshold of advertising images and improve the design efficiency of advertising images is a problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of this application is to provide a design scheme recommendation method, device, storage medium and computer program product, aiming to solve the technical problem of how to lower the design threshold of advertising images and improve the design efficiency of advertising images.

[0006] To achieve the above objectives, the present application proposes a design solution recommendation method, which includes:

[0007] Acquire multiple advertisement images with high frequency of clicks, input each of the advertisement images into a pre-trained multi-label extraction model, and obtain multiple classification labels of each of the advertisement images under different classification dimensions, wherein the multi-label extraction model is used to perform feature extraction and feature classification of different classification dimensions on the advertisement images;

[0008] Performing label co-occurrence analysis on each of the classification labels to obtain multi-label co-occurrence information, wherein the multi-label co-occurrence information includes at least one label matching scheme, and each of the label matching schemes includes at least two of the classification labels;

[0009] If a design recommendation request input by a user is received, a target classification label corresponding to the design recommendation request is determined, and a target design solution is generated according to the target classification label and the multi-label co-occurrence information, and the target design solution is used as a recommendation result.

[0010] In one embodiment, the multi-label extraction model includes a deep learning model and a multi-classifier, and the step of inputting each of the advertising images into a pre-trained multi-label extraction model to obtain a plurality of classification labels of different classification dimensions corresponding to each of the advertising images includes:

[0011] For each advertisement picture, extract features of the advertisement picture based on the pre-trained deep learning model to obtain an original feature map;

[0012] The original feature map is input into the pre-trained multi-classifier to obtain multiple classification labels of the advertisement image under different classification dimensions.

[0013] In one embodiment, the step of performing label co-occurrence analysis on each of the classification labels to obtain multi-label co-occurrence information includes:

[0014] Determine the co-occurrence matrix corresponding to all the classification labels, and perform label combination on each of the classification labels to obtain at least one label matching scheme, wherein the co-occurrence matrix is ​​a matrix used to characterize whether each of the classification labels co-appears in the same advertisement image;

[0015] For each of the label collocation schemes, determining the total number of times the classification labels included in the label collocation scheme co-occur in the co-occurrence matrix, and determining the total number as the number of co-occurrences of the label collocation scheme;

[0016] The label collocation schemes and the co-occurrence times corresponding to the label collocation schemes are combined to obtain multi-label co-occurrence information.

[0017] In one embodiment, the multi-label co-occurrence information further includes the co-occurrence counts of each label collocation scheme, and the step of generating a target design scheme based on the target classification label and the multi-label co-occurrence information includes:

[0018] Querying the candidate tag collocation schemes including the target classification tag in the multi-tag co-occurrence information, and querying the candidate co-occurrence counts of each of the candidate tag collocation schemes;

[0019] A target label collocation scheme is selected from each of the candidate label collocation schemes based on the number of co-occurrences of each of the candidate labels, and the target label collocation scheme is determined as a target design scheme.

[0020] In one embodiment, after the step of querying the candidate co-occurrence counts of each of the candidate tag collocation solutions, the method further includes:

[0021] Acquire the candidate classification labels included in each of the candidate label matching schemes, and group each of the candidate label matching schemes according to the candidate classification labels to obtain at least one candidate label matching scheme group;

[0022] For each of the candidate tag collocation scheme groups, based on the candidate co-occurrence counts of each of the candidate tag collocation schemes included in the candidate tag collocation scheme group, the step of selecting a target tag collocation scheme from each of the candidate tag collocation schemes based on the candidate co-occurrence counts is performed.

[0023] In one embodiment, before the step of obtaining a plurality of advertisement images with a high frequency of clicks, the method further includes:

[0024] Obtaining a preset multi-label extraction model, wherein the multi-label extraction model includes a deep learning model and a multi-classifier, and the multi-classifier includes a plurality of classifiers for different classification dimensions;

[0025] Obtain a training data set, wherein the training data set includes training images and a plurality of image labels corresponding to the training images;

[0026] Inputting the training image into the deep learning model of the multi-label extraction model to obtain a target feature map;

[0027] Inputting the target feature map into each classifier in the multi-classifier of the multi-label extraction model to obtain a classification result output by each classifier;

[0028] The model parameters of the deep learning model and each classifier are adjusted according to each of the classification results and each of the image labels, and the step of inputting the training image into the deep learning model of the multi-label extraction model to obtain the target feature map is returned to the step until the preset iteration termination condition is met, thereby obtaining the trained deep learning model and the trained multi-classifier.

[0029] In one embodiment, the deep learning model includes a regularization module and a squeezing and excitation module, and the step of inputting the training image into the deep learning model of the multi-label extraction model to obtain the target feature map includes:

[0030] Extracting features from the training image using the regularization module, and standardizing the extracted features to obtain an initial feature map;

[0031] The initial feature map is adjusted based on the channel weight of the initial feature map by the squeeze and excitation module to obtain a target feature map.

[0032] In addition, to achieve the above-mentioned purpose, the present application also proposes a design scheme recommendation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the design scheme recommendation method described above.

[0033] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the design scheme recommendation method described above are implemented.

[0034] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the design scheme recommendation method described above are implemented.

[0035] One or more technical solutions proposed in this application have at least the following technical effects:

[0036] By obtaining multiple advertisement images with high frequency of clicks, each advertisement image is input into a pre-trained multi-label extraction model to obtain multiple classification labels for each advertisement image in different classification dimensions, wherein the multi-label extraction model is used to perform feature extraction and feature classification of the advertisement images in different classification dimensions, thereby comprehensively analyzing the high frequency click advertisements from multiple classification dimensions through deep learning and multi-classifiers to extract feature information of the high frequency click advertisement images; performing label co-occurrence analysis on each classification label to obtain multi-label co-occurrence information, wherein the multi-label co-occurrence information includes at least one label matching scheme, and each label matching scheme includes at least two classification labels, thereby statistically analyzing the co-occurrence of the high frequency click advertisement images on multiple labels, so that the multi-label co-occurrence information can be obtained. The advertising image design provides more valuable information for reference. If a design recommendation request input by a user is received, the target classification label corresponding to the design recommendation request is determined, and a target design solution is generated based on the target classification label and multi-label co-occurrence information. The target design solution is used as a recommendation result, so that when the user needs to design an advertising image, a design recommendation request is initiated to request a recommended design solution, without the need to find a professional designer to design the advertising image, thereby lowering the design threshold of the advertising image, and based on the user's design recommendation request and multi-label co-occurrence information, a target design solution is intelligently generated for relevant personnel to refer to the recommended target design solution for advertising image design, without having to spend a lot of time looking for design solutions from websites or other channels as references, thereby improving the design efficiency of advertising images. Therefore, in the embodiment of the present application, it is achieved to recommend relevant design solutions to users based on the user's design recommendation request for user design reference, thereby lowering the design threshold of advertising images and improving the design efficiency of advertising images. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0039] Figure 1 A flowchart of the first embodiment of the method for recommending the design solution of this application is provided;

[0040] Figure 2 A schematic diagram of a method for initiating a design recommendation request involved in an embodiment of a design solution recommendation method of the present application;

[0041] Figure 3 A schematic diagram of the structure of a multi-label extraction model involved in an embodiment of the recommended method for designing a solution of this application;

[0042] Figure 4 Another structural diagram of a multi-label extraction model involved in an embodiment of the recommended method of the design solution of this application;

[0043] Figure 5 A schematic diagram of the device structure of the hardware operating environment involved in the design scheme recommendation method in the embodiment of the present application.

[0044] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0046] It should be noted that the executor of each embodiment of the design scheme recommendation method of the present application can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or a design scheme recommendation device that can realize the above functions. The embodiments of the design scheme recommendation method of the present application do not impose specific restrictions on this.

[0047] In advertising image design, designing attractive advertising images is the key to attracting users to click and improving advertising effectiveness. Highly clicked advertising images often have certain common features, such as color matching, style design, background selection, and font type. However, it is difficult for advertising designers to systematically analyze and refer to the characteristics of these high-click advertising images when designing new advertising images.

[0048] Currently, it mainly relies on manual experience or simple statistical analysis, which is unable to comprehensively and systematically analyze the characteristics of advertising images. In addition, advertising images are usually only classified in a single classification dimension, making it difficult to capture the characteristic information of advertising images in multiple classification dimensions. It is also impossible to effectively integrate multi-label classification results into a systematic design reference, which makes it difficult for designers to fully utilize the potential information fed back by user clicks.

[0049] Based on this, the first embodiment of the recommended method for the design scheme of this application is proposed below. Figure 1 , Figure 1 This is a flow chart of the first embodiment of the recommended method for the design solution of this application.

[0050] In this embodiment, the design scheme recommendation method includes steps S10 to S30:

[0051] Step S10, obtaining a plurality of advertisement images with high frequency of clicks, inputting each of the advertisement images into a pre-trained multi-label extraction model, and obtaining a plurality of classification labels of each of the advertisement images under different classification dimensions, wherein the multi-label extraction model is used to perform feature extraction and feature classification of different classification dimensions on the advertisement images;

[0052] It should be noted that the frequently clicked advertisement images are advertisement images with click times greater than or equal to a preset number threshold.

[0053] Furthermore, the collected advertising pictures can be preprocessed, and the preprocessed advertising pictures can be input into a pre-trained multi-label extraction model to improve the quality of the advertising pictures, thereby improving the classification effect of the model. The picture preprocessing includes but is not limited to unifying the picture resolution, normalization, image denoising, etc., and this embodiment does not specifically limit this.

[0054] It should be noted that the classification dimensions for classifying the advertising images can be preset in advance. For example, the advertising images can be classified according to four classification dimensions: color, background, style, and font, so as to classify the advertising images in multiple ways based on different classification dimensions.

[0055] The multi-label extraction model is a pre-trained model that can realize multiple classification of advertising images. It should be noted that one classification dimension corresponds to dividing the advertising image into one classification label.

[0056] Step S20, performing label co-occurrence analysis on each of the classification labels to obtain multi-label co-occurrence information, wherein the multi-label co-occurrence information includes at least one label matching scheme, and each of the label matching schemes includes at least two of the classification labels;

[0057] The multi-label co-occurrence information includes at least one label matching scheme, and each label matching scheme includes at least two classification labels. Preferably, each classification label included in each label matching scheme belongs to a different classification dimension. For example, "big character poster + building + orange + Founder bold" represents a label matching scheme, wherein "big character poster" is a classification label of the style classification dimension, "building" is a label of the background classification dimension, "orange" is a classification label of the color classification dimension, and "Founder bold" is a label of the font classification dimension.

[0058] The multi-tag co-occurrence information includes but is not limited to at least one tag matching scheme, and may also include the co-occurrence times of each tag matching scheme.

[0059] The co-occurrence count of a label pairing scheme represents the historical frequency of the label pairing scheme, which can be specifically the number of advertisement images that contain all the classification labels in the label pairing scheme among the advertisement images with high frequency of clicks. For example, assuming that a total of 1,000 advertisement images with high frequency of clicks are obtained, a label pairing scheme is "big character poster + architecture", and there are 100 advertisement images that contain both the classification labels "big character poster" and "architecture" among the 1,000 advertisement images, then the co-occurrence count of the label pairing scheme "big character poster + architecture" is 100.

[0060] Exemplarily, the multi-label co-occurrence information includes at least one label matching scheme and the number of co-occurrences of each label matching scheme to illustrate and describe the embodiments of the present application. Further, to assist in understanding the multi-label co-occurrence information, a specific example of the multi-label co-occurrence information is given. In this specific example, the multi-label co-occurrence information is the associated information shown in Table 1 below.

[0061]

[0062]

[0063] Table 1

[0064] Step S30: If a design recommendation request input by a user is received, a target classification label corresponding to the design recommendation request is determined, and a target design solution is generated according to the target classification label and the multi-label co-occurrence information, and the target design solution is used as a recommendation result.

[0065] When a user needs a design recommendation, he or she can initiate a design recommendation request. In a specific implementation, the initiation method of the design recommendation request can be set according to actual needs, for example, Figure 2 As shown, in the retrieval scenario, the user can initiate a design recommendation request by clicking the search icon 100 .

[0066] It should be noted that the design requirement data can be obtained according to the design recommendation request, and the label of the design requirement data can be extracted to obtain the target classification label. The design requirement data can be the requirement data input by the user, for example, Figure 2 In the search scenario shown, the user enters "big character poster style" in the input box, and the "big character poster style" is the design requirement data.

[0067] Furthermore, if the user does not input the requirement data, such as in the search scenario, the user directly clicks the search icon without inputting any data to initiate a design recommendation request, the design requirement data may specifically be preset default design requirement data. The preset default design requirement data is any design requirement data set in advance, such as "orange font", "big poster style + architectural background", etc., and this embodiment does not specifically limit this.

[0068] Furthermore, corresponding preset default design requirement data can be set for different users. Specifically, the target historical requirement data with the most cumulative input times among the historical requirement data input by the user can be determined, and the target historical requirement data can be used as the preset default design requirement data corresponding to the user, so that when the user does not input the requirement data, the corresponding design scheme can also be recommended according to the user's design habits, realizing intelligent and personalized recommendation of design schemes.

[0069] The target classification label represents the label extracted from the design requirement data. In a specific implementation, when the advertising image is multi-classified based on four classification dimensions such as color, background, style, and font, the target classification label is extracted from the design requirement data based on the four classification dimensions. For example, if the design requirement data is "big-character poster style", the extracted target classification label is "big-character poster"; if the design requirement data is "big-character poster style plus architectural background", the extracted target classification labels are "big-character poster" and "architecture"; if the design requirement data is "what kind of background is more suitable for advertising images in the big-character poster style?", the extracted target classification labels are "big-character poster" and "background".

[0070] In a possible implementation, the step of extracting target classification labels from the design requirement data includes steps A10 to A30:

[0071] Step A10: if the design requirement data is text data, extract labels from the design requirement data to obtain target classification labels;

[0072] Specifically, the user may set the tag extraction method in advance, and this embodiment does not impose any specific limitation on this.

[0073] As one of the implementation methods, the text data of the design requirement data may be segmented, such as by using the jieba segmentation tool, and the segmentation result matching the preset label type in the query segmentation result is determined as the target classification label.

[0074] As another implementation, the query text may be first processed to remove stop words and then to perform word segmentation, and then the word segmentation results that match the preset label type in the query word segmentation results are determined as the target classification labels.

[0075] Step A20: if the design requirement data is image data, the design requirement data is input into a pre-trained multi-label extraction model to obtain a target classification label;

[0076] Step A30: If the design requirement data is audio data, speech recognition is performed on the design requirement data to obtain a recognition result, and a label is extracted from the recognition result to obtain a target classification label.

[0077] If the design requirement data is audio data, speech recognition is performed on the design requirement data to obtain a recognition result. It can be understood that the recognition result of speech recognition is text data. At this time, the recognition result can be labeled according to the label extraction method of text data to obtain the target classification label.

[0078] For data in modalities such as text, images, and audio, the target classification labels can be obtained through the corresponding label extraction methods. In this way, the data modality of the design requirement data that can be input is not limited to a single modality. Users can input design requirement data in different ways according to actual conditions, such as voice input, picture input, text input, etc., so that the design requirement data that can be processed is not limited to a single modality, which improves the universality and flexibility of the design solution recommendation method.

[0079] After the target classification label is extracted, a target design solution is generated based on the target classification label and the multi-label co-occurrence information. Specifically, as one implementation method, the target design solution can be a label matching solution that includes the target classification label in the multi-label co-occurrence information. The label matching solution that includes the target classification label is queried from the multi-label co-occurrence information so that the recommended design solution meets the design requirements.

[0080] As another implementation, the multi-label co-occurrence information also includes the co-occurrence count of each label collocation scheme, and the step of generating a target design scheme based on the target classification label and the multi-label co-occurrence information includes steps S301 to S302:

[0081] Step S301, querying the candidate tag collocation schemes including the target classification tag in the multi-tag co-occurrence information, and querying the candidate co-occurrence times of each of the candidate tag collocation schemes;

[0082] The candidate label collocation scheme is a label collocation scheme that includes the target classification label in all label collocation schemes of the multi-label co-occurrence information. Specifically, all label collocation schemes that include the target classification label can be selected from the multi-label co-occurrence information as candidate label collocation schemes, or, in order to reduce the search scope, improve the search efficiency, and thus improve the timeliness of the design solution recommendation response, a preset number of label collocation schemes that include the target classification label can be selected from the multi-label co-occurrence information as candidate label collocation schemes.

[0083] Exemplarily, taking the database described in Table 1 with multi-label co-occurrence information as an example, selecting all label matching schemes containing the target classification label from the multi-label co-occurrence information as the selection method of the candidate label matching scheme is an example. If the target classification label is "big-character poster", the candidate label matching schemes include "big-character poster + building", "big-character poster + street", "big-character poster + orange", "big-character poster + black", "big-character poster + Founder bold", "big-character poster + regular script", "big-character poster + building + orange", "big-character poster + street + orange", "big-character poster + building + orange + Founder bold", "big-character poster + street + orange + regular script"; if the target classification labels are "big-character poster" and "background", the candidate label matching schemes include "big-character poster + building", "big-character poster + street", "big-character poster + building + orange", "big-character poster + street + orange", "big-character poster + building + orange + Founder bold", "big-character poster + street + orange + regular script".

[0084] Step S302 : selecting a target label collocation scheme from each of the candidate label collocation schemes based on the number of co-occurrences of each of the candidate labels, and determining the target label collocation scheme as a target design scheme.

[0085] A target label collocation scheme is selected from the candidate label collocation schemes based on the co-occurrence counts of the candidate label collocation schemes. The target label collocation scheme may be one or more, and correspondingly, the target design scheme may also be one or more, which is not specifically limited in this embodiment.

[0086] In a possible implementation, the step of selecting a target tag collocation scheme from each of the candidate tag collocation schemes based on each of the candidate co-occurrence times includes steps S3021 to S3022:

[0087] Step S3021, sorting the co-occurrence times in a non-decreasing order to obtain a co-occurrence time sequence;

[0088] Step S3022: Select the first preset number of co-occurrence times in the co-occurrence times sequence as the target co-occurrence times, and determine the candidate tag matching scheme corresponding to the target co-occurrence times as the target tag matching scheme.

[0089] The preset number may be any integer greater than zero set in advance, such as 1, 2, 3, 10, etc., and this embodiment does not impose any specific limitation on this.

[0090] The candidate tag matching schemes corresponding to the first preset number of co-occurrence times are selected from the co-occurrence frequency sequence as the target tag matching scheme. The higher the historical matching frequency of the target tag matching scheme, the higher the frequency of the target tag matching scheme. Then, the target tag matching schemes with higher historical matching frequency are further selected from the candidate tag matching schemes that meet the design requirements and recommended to the user, which simplifies the number of recommended design schemes and makes the recommended design schemes conform to the mainstream design trend, improves the quality of the recommended design schemes, and thus achieves high-quality recommendation of design schemes.

[0091] In another implementation, the candidate tag matching scheme corresponding to the highest co-occurrence number among the co-occurrence numbers may also be selected as the target tag matching scheme.

[0092] In another implementation, if the user wants to design a relatively niche advertising image, the user may also select the candidate label matching scheme corresponding to the lowest co-occurrence number among the co-occurrence numbers as the target label matching scheme.

[0093] It should be noted that the above are only several feasible implementation methods of selecting the target tag matching scheme provided in this embodiment, and this embodiment does not specifically limit the specific implementation method of selecting the target tag matching scheme.

[0094] Furthermore, after obtaining the recommendation result, the recommendation result can also be output so that the user can refer to the recommended design scheme to design the advertisement picture.

[0095] This embodiment obtains multiple advertisement images with high frequency of clicks, inputs each advertisement image into a pre-trained multi-label extraction model, and obtains multiple classification labels for each advertisement image under different classification dimensions, wherein the multi-label extraction model is used to perform feature extraction and feature classification of the advertisement images in different classification dimensions, thereby comprehensively analyzing the advertisements with high frequency of clicks from multiple classification dimensions through deep learning and multiple classifiers to extract feature information of the advertisement images with high frequency of clicks; performs label co-occurrence analysis on each classification label to obtain multi-label co-occurrence information, wherein the multi-label co-occurrence information includes at least one label matching scheme, and each label matching scheme includes at least two classification labels, thereby statistically analyzing the co-occurrence of advertisement images with high frequency of clicks on multiple labels, so that the multi-label co-occurrence information It can provide more valuable information for advertising image design. If a design recommendation request input by a user is received, the target classification label corresponding to the design recommendation request is determined, and a target design solution is generated based on the target classification label and multi-label co-occurrence information. The target design solution is used as a recommendation result, so that when the user needs to design an advertising image, a design recommendation request is initiated to request a recommended design solution, without the need to find a professional designer to design the advertising image, thereby lowering the design threshold of the advertising image, and based on the user's design recommendation request and multi-label co-occurrence information, a target design solution is intelligently generated for relevant personnel to refer to the recommended target design solution for advertising image design, without having to spend a lot of time looking for design solutions from websites or other channels as references, thereby improving the design efficiency of advertising images. Therefore, in the embodiment of the present application, it is achieved to recommend relevant design solutions to users based on the user's design recommendation request for user design reference, thereby lowering the design threshold of advertising images and improving the design efficiency of advertising images.

[0096] Based on the above-mentioned first embodiment, a second embodiment of the design scheme recommendation method of this application is proposed. In this embodiment, the same or similar contents as those of the above-mentioned first embodiment can be referred to the above introduction and will not be repeated later. In this embodiment, the multi-label extraction model includes a deep learning model and a multi-classifier, and the step of inputting each of the advertising images into a pre-trained multi-label extraction model to obtain multiple classification labels of different classification dimensions corresponding to each of the advertising images includes:

[0097] Step B10, for each advertisement picture, extracting features of the advertisement picture based on the pre-trained deep learning model to obtain an original feature map;

[0098] The deep learning model is used to extract features from advertising images, and the multi-classifier is used to perform multi-classification based on the features extracted by deep learning.

[0099] For the specific structure of the multi-label extraction model, as one of the implementation methods, the multi-label extraction model includes a deep learning model and multiple classifiers. Figure 3 As shown, the multiple classifiers include classifier 1, classifier 2, .... classifier n, the input of each classifier is connected to the output of the deep learning model, and each classifier classifies the advertising image from a classification dimension to obtain classification label 1, classification label 2, .... classification label n. In a specific embodiment, when the advertising image is classified from the four classification dimensions of color, font, background, and style, the multiple classifiers may include classifier 1, classifier 2, classifier 3, and classifier 4. These four classifiers classify the advertising images from the four classification dimensions of color, font, background, and style, such as classifier 1 classifies the advertising images from the color classification dimension, classifier 2 classifies the advertising images from the font classification dimension, classifier 3 classifies the advertising images from the background classification dimension, and classifier 4 classifies the advertising images from the style classification dimension.

[0100] As another implementation, the multi-label extraction model includes multiple deep learning models and multiple classifiers, and the number of deep learning models is consistent with the number of classifiers, and the deep learning models and classifiers are connected one-to-one. Figure 4 As shown, the deep learning model includes deep learning model 1, deep learning model 2, ... deep learning model n, the multiple classifiers include classifier 1, classifier 2 ... classifier n, the deep learning model is connected to the multiple classifiers in a one-to-one correspondence, and each classifier classifies the advertising image from a classification dimension to obtain classification label 1, classification label 2 ... classification label n.

[0101] The deep learning model is used to extract image features. In a specific implementation, the deep learning model may be a Resnet model (Residual Network).

[0102] In another specific embodiment, the deep learning model may be a SE-Resnet model (Squeeze-and-Excitation Resnet), which is a Resnet model that introduces a SE (Squeeze-and-Excitation) module. Specifically, the SE module may be inserted after the last convolutional layer of the residual block of the Resnet network and before the skip connection to obtain a model.

[0103] Furthermore, regularization methods such as dropout can be used in deep learning models to prevent network overfitting and improve the generalization ability of the network.

[0104] Step B20, inputting the original feature map into the pre-trained multi-classifier to obtain multiple classification labels of the advertisement image under different classification dimensions.

[0105] This embodiment realizes multi-classification of advertising images based on a pre-trained deep learning model and a pre-trained multi-classifier, that is, it can comprehensively and systematically analyze the characteristics of advertising images from multiple classification dimensions, and there is no need to rely on manual experience to classify advertising images, thereby realizing intelligent and automatic classification of advertising images and improving classification efficiency.

[0106] Based on the above-mentioned first and / or second embodiments, a third embodiment of the design scheme recommendation method of this application is proposed. In this embodiment, the same or similar contents as those of the above-mentioned first and second embodiments can be referred to the above introduction, and will not be repeated later. In this embodiment, the step of performing label co-occurrence analysis on each of the classification labels to obtain multi-label co-occurrence information includes:

[0107] Step C10, determining the co-occurrence matrix corresponding to all the classification labels, and combining the classification labels to obtain at least one label matching scheme, wherein the co-occurrence matrix is ​​a matrix used to characterize whether the classification labels co-appear in the same advertisement image;

[0108] It should be noted that the co-occurrence of multiple labels of the classification label can be analyzed to obtain a co-occurrence matrix, so as to perform subsequent co-occurrence analysis based on the co-occurrence matrix. For example, as shown in Table 2 below, the horizontal direction of the co-occurrence matrix is ​​the label name, and the vertical direction is different advertising pictures. The content is filled with 0 or 1 to obtain the co-occurrence matrix, where 0 indicates that it is not the label and 1 indicates that it is the label. If the values ​​of the classification labels in the same row are all 1, it is determined that the classification labels co-appear in the same advertising picture.

[0109]

[0110] Table 2

[0111] Combine each classification label to obtain multiple label matching solutions. Specifically, as one implementation method, each classification label can be combined according to different label types, and at least two classification labels of different classification dimensions can be combined to obtain multiple design solutions. For example, assuming that the classification labels of two advertising pictures are [“color 1”, “font 1”, “background 1”, “style 1”], [“color 2”, “font 1”, “background 1”, “style 1”], then the classification labels of these two advertising pictures are combined to obtain label matching solutions including [“color 1+font 1”], [“color 1+background 1”], [“color 1+style 1”], [“color 2+font 1”], [“color 2+background 1”], [“color 2+style 1”], [“font 1+background 1"], ["Font 1 + Style 1"], ["Background 1 + Style 1"], ["Color 1 + Font 1 + Background 1"], ["Color 2 + Font 1 + Background 1"], ["Color 1 + Background 1 + Style 1"], ["Color 2 + Background 1 + Style 1"], ["Color 1 + Font 1 + Style 1"], ["Color 1 + Font 1 + Style 1"], ["Font 1 + Background 1 + Style 1"], ["Color 2 + Font 1 + Background 1 + Style 1"].

[0112] As another implementation method, in order to reduce the number of label matching schemes, and thereby reduce the number of co-occurrences that need to be counted and the number of label matching schemes that need to be stored, so as to reduce the demand for computing power resources and storage resources, the classification labels can also be combined to obtain multiple label matching schemes, wherein each label matching scheme only contains classification labels of two different classification dimensions or contains classification labels of all classification dimensions. For example, in the above classification label examples of ["color 1", "font 1", "background 1", "style 1"], ["color 2", "font 1", "background 1", "style 1", the label matching solutions that can be obtained according to this combination method are ["color 1 + font 1"], ["color 1 + background 1"], ["color 1 + style 1"], ["color 2 + font 1"], ["color 2 + background 1"], ["color 2 + style 1"], ["font 1 + background 1"], ["font 1 + style 1"], ["background 1 + style 1"], ["color 1 + font 1 + background 1 + style 1"], ["color 2 + font 1 + background 1 + style 1"]. This not only reduces the demand for computing power resources and storage resources, but also counts the overall matching solutions of advertising images and the combination and matching solutions of each label with other individual labels, which can meet the requirements of recommendation breadth and accuracy of subsequent design solutions.

[0113] Step C20, for each of the label collocation schemes, determining the total number of times the classification labels included in the label collocation scheme co-occur in the co-occurrence matrix, and determining the total number as the number of co-occurrences of the label collocation scheme;

[0114] It is easy to understand that, taking the co-occurrence matrix shown in Table 2 above as an example, if the classification labels included in the label matching scheme have values ​​of 1 in the same row of the co-occurrence matrix, then it is determined that the classification labels included in the label matching scheme co-occur in the co-occurrence matrix. That is, the number of co-occurrences can be specifically the cumulative number of times that all the classification labels included in the label matching scheme have values ​​of 1 in the same row of the above co-occurrence matrix.

[0115] Step C30: combining each of the tag collocation schemes and the co-occurrence times corresponding to each of the tag collocation schemes to obtain multi-tag co-occurrence information.

[0116] In this embodiment, the co-occurrence matrix corresponding to all classification labels is determined, and each classification label is combined to obtain at least one label matching scheme. For each label matching scheme, the total number of times the classification labels included in the label matching scheme co-occur in the co-occurrence matrix is ​​determined, and the total number is determined as the co-occurrence number of the label matching scheme. Each label matching scheme is combined with the co-occurrence number corresponding to each label matching scheme to obtain multi-label co-occurrence information, thereby realizing intelligent and automatic counting of the co-occurrence number of each classification label in the advertising image to obtain multi-label co-occurrence information. There is no need to spend a lot of time manually to analyze the possible label matching schemes of the combined image and the co-occurrence numbers of various label matching schemes, thereby realizing the intelligent setting of the label matching scheme and reducing the setting and implementation cost of the multi-label co-occurrence information.

[0117] Based on the above-mentioned first, second and / or third embodiments, a fourth embodiment of the design scheme recommendation method of the present application is proposed. In this embodiment, the same or similar contents as those of the above-mentioned first, second and third embodiments can be referred to the above introduction, and will not be repeated later. After the step of querying the candidate co-occurrence counts of each of the candidate label collocation schemes, the method also includes:

[0118] Step D10, obtaining the candidate classification labels included in each of the candidate label matching schemes, and grouping each of the candidate label matching schemes according to the candidate classification labels to obtain at least one candidate label matching scheme group;

[0119] The candidate classification label is used to represent the classification label included in the candidate label matching scheme. For example, assuming that a candidate label matching scheme is "big character poster + street + orange + regular font", the candidate classification labels included in the candidate label matching scheme are "big character poster", "street", "orange" and "regular font".

[0120] The candidate label matching schemes are grouped according to the candidate classification labels, and specifically, the candidate label matching schemes are grouped according to the classification dimensions. In each candidate label matching scheme group obtained by grouping, the classification dimensions of the candidate classification labels included in each candidate label matching scheme in the same candidate label matching scheme group are consistent, and any two candidate label matching schemes belonging to different candidate label matching scheme groups contain at least one candidate classification label with inconsistent classification dimensions. For example, the candidate label matching schemes include "big-character posters + buildings", "big-character posters + streets", "big-character posters + orange", "big-character posters + black", "big-character posters + bold font", "big-character posters + regular font", "big-character posters + buildings + orange", "big-character posters + streets + orange", "big-character posters + buildings + orange + bold font", "big-character posters + streets + orange + regular font", then grouping the candidate label matching schemes according to the classification dimension can obtain 5 candidate label matching scheme groups, namely: ["big-character posters + buildings", "big-character posters + streets"], ["big-character posters + orange", "big-character posters + black"], ["big-character posters + bold font", "big-character posters + regular font"], ["big-character posters + buildings + orange", "big-character posters + streets + orange"], ["big-character posters + buildings + orange + bold font", "big-character posters + streets + orange + regular font"].

[0121] Step D20 , for each of the candidate tag collocation scheme groups, based on the candidate co-occurrence counts of each of the candidate tag collocation schemes included in the candidate tag collocation scheme group, performing the step of selecting a target tag collocation scheme from each of the candidate tag collocation schemes based on the candidate co-occurrence counts.

[0122] For each candidate label matching scheme group, the target label matching scheme corresponding to the candidate label matching scheme group is selected based on the number of co-occurrences of the candidate label matching schemes in the candidate label matching scheme group, that is, for each candidate label matching scheme group, the target label matching scheme is selected from the candidate label matching schemes included in the candidate label matching scheme group. Exemplarily, assuming that the candidate label matching scheme corresponding to the highest number of co-occurrences is selected as the target label matching scheme, and the multi-label co-occurrence information is the associated information shown in Table 1 of the first embodiment, then for the candidate label matching scheme group [“big character poster + building”, “big character poster + street”], the number of co-occurrences of the two candidate label matching schemes are [100, 105] respectively, and the candidate label matching scheme “big character poster + street” has the highest number of co-occurrences, so “big character poster + street” is selected as the target label matching scheme corresponding to this candidate label matching scheme group. Similarly, for There are four candidate label combination scheme groups: [“big-character poster + orange”, “big-character poster + black”], [“big-character poster + Founder bold”, “big-character poster + regular script”], [“big-character poster + building + orange”, “big-character poster + street + orange”], [“big-character poster + building + orange + Founder bold”, “big-character poster + street + orange + regular script”], and the target label combination schemes selected in each candidate label combination scheme group are: “big-character poster + orange”, “big-character poster + Founder bold”, “big-character poster + building + orange”, and “big-character poster + building + orange + Founder bold”.

[0123] In this embodiment, the candidate label matching schemes are grouped according to the candidate classification labels contained in each candidate label matching scheme to obtain at least one candidate label matching scheme group, and the target label matching scheme is screened in each candidate label matching scheme group, so that the corresponding target design schemes can be screened according to different labels, so that the corresponding target design schemes can be recommended to users from different matching methods, thereby improving the recommendation breadth of design schemes.

[0124] Based on the first, second, third and / or fourth embodiments described above, a fifth embodiment of the design scheme recommendation method of this application is proposed. In this embodiment, the same or similar contents as those of the first, second, third and fourth embodiments described above can be referred to the above description and will not be repeated later. In this embodiment, before the step of obtaining multiple advertisement images with high frequency of clicks, the method further includes steps E10 to E50:

[0125] Step E10, obtaining a preset multi-label extraction model, wherein the multi-label extraction model includes a deep learning model and a multi-classifier, and the multi-classifier includes a plurality of classifiers for different classification dimensions;

[0126] The number of classifiers can be consistent with the total classification dimensions of the advertising images, that is, n classification dimensions can correspond to n classifiers, and each classifier is used to classify the training images from one classification dimension. For example, when the advertising images are classified from the four classification dimensions of color, style, background and font, the number of classifiers can also be four, and the training images are classified from the four aspects of color, style, background and font respectively.

[0127] It should be noted that each classifier may be LR (Logistic Regression), SVM (Support Vector Machine), etc., and this embodiment does not impose any specific limitation on this.

[0128] Step E20, obtaining a training data set, wherein the training data set includes training images and a plurality of image labels corresponding to the training images;

[0129] The training picture may specifically be an advertisement picture collected in advance, and further, a frequently clicked advertisement picture may be collected as the training picture.

[0130] Furthermore, the collected advertising pictures can be preprocessed, and the preprocessed advertising pictures can be used as training pictures to improve the picture quality of the training pictures, and thus the training efficiency and training effect of the model. The picture preprocessing includes but is not limited to unifying the picture resolution, normalization, image denoising, etc., and this embodiment does not specifically limit this.

[0131] The image labels of the training images can be defined by relevant personnel from multiple classification dimensions (such as four classification dimensions of color, style, background and font) for the training images, and the relevant personnel can annotate the training images to obtain the image labels corresponding to each training image.

[0132] Step E30, inputting the training image into the deep learning model of the multi-label extraction model to obtain a target feature map;

[0133] Step E40, inputting the target feature map into each classifier in the multi-classifier of the multi-label extraction model to obtain the classification result output by each classifier,

[0134] Step E50, adjust the model parameters of the deep learning model and each classifier according to each of the classification results and each of the image labels, return to execute the step of inputting the training image into the deep learning model of the multi-label extraction model, and obtain the target feature map, until the preset iteration termination condition is met, and the trained deep learning model and the trained multi-classifier are obtained.

[0135] The preset multi-label extraction model is trained according to the training database. Specifically, the model parameters of the multi-label extraction model, such as model parameters such as max_depth and n_estunators, can be adjusted during the training process until the multi-label extraction model that meets the preset iteration end conditions is trained. Preferably, the model parameters of the multi-label extraction model can be set to the model parameters when the classification accuracy is the highest during the training process to obtain a trained multi-label extraction model. Among them, the preset iteration end condition is a condition for ending model training set in advance, such as reaching the maximum number of iterations, model convergence, etc., and this embodiment does not impose specific restrictions on this.

[0136] In a possible implementation, the step of inputting the training image into the deep learning model of the multi-label extraction model to obtain the target feature map includes steps F10 to F20:

[0137] Step F10, extracting features from the training image through the regularization module, and standardizing the extracted features to obtain an initial feature map;

[0138] Step F20, adjusting the initial feature map based on the channel weight of the initial feature map through the squeeze and excitation module to obtain a target feature map.

[0139] The deep learning model may specifically be a Resnet model, and may be a Resnet model that references a regularization module and an SE module.

[0140] The basic architecture of the Resnet model is based on repeated residual blocks. The core idea of ​​the residual block is to introduce skip connections. Taking ResNet-50 as an example, the basic architecture of the Resnet model is as follows:

[0141] Initial layers: usually include a 7x7 convolutional layer with a stride of 2, followed by a max pooling layer with a stride of 2.

[0142] Residual block stacking: ResNet-50 contains 4 residual block stacks, each stack contains multiple residual units. Each residual unit usually contains two 3x3 convolutional layers, and there may be a 1x1 convolutional layer in the middle for dimensionality reduction or dimensionality increase.

[0143] Skip connections: Each residual unit has a skip connection. If the input and output dimensions are the same, they are added directly; if the dimensions are different, they are passed through a 1x1 convolutional layer to match the dimensions and then added.

[0144] Activation function: Batch normalization and ReLU activation function are usually used after the convolutional layer.

[0145] Global Average Pooling: After all residual blocks, there is usually a global average pooling layer to reduce the spatial dimension to 1x1.

[0146] Fully connected layer and softmax: The last layer is a fully connected layer to output the final classification result, followed by a softmax activation function.

[0147] Specifically, the regularization module can be a BatchNormalization (BN) layer added after the convolutional layer of the Resnet model, which can improve training speed and stability. The BN layer normalizes each feature of each mini-batch of data, and then scales and translates it through learnable parameters.

[0148] The squeeze and excitation module (also known as the SE module) can be added to each residual block of ResNet. The module first "squeezes" the feature map, that is, obtains a channel descriptor through global average pooling. Then it "excites" and learns the channel descriptor through two fully connected layers to obtain the weight of each channel. Finally, these weights are recalibrated to the original feature map, that is, "recalibrated" (Re-scale), to enhance important features and suppress unimportant features.

[0149] In this embodiment, by introducing a regularization module in the deep learning model, the risk of overfitting is reduced while reducing the amount of calculation. By introducing the SE module in the deep learning model, the SE module adaptively recalibrates the feature response of the channel by explicitly modeling the interdependence between channels. Specifically, the importance of each channel is automatically obtained through learning, and then useful features are promoted according to this importance and features that are not very useful for the current task are suppressed, thereby improving the performance and generalization ability of the network.

[0150] In order to help understand the above embodiments of the design solution recommendation method of the present application, an example of the design solution recommendation method combined with the above embodiments is proposed. The specific steps are as follows:

[0151] 1. Advertising image data collection: Collect a large amount of frequently clicked advertising image data and pre-process these images, such as normalization and resizing, to obtain training images.

[0152] 2. Label definition and annotation: Define labels for high-frequency click advertising images from multiple classification dimensions such as color, style, background and font, and have relevant personnel annotate the images to obtain multiple image labels corresponding to the training images to form a training data set.

[0153] 3. Feature learning: Use the deep learning model Resnet to extract features from training images to obtain target feature maps, and learn the feature representation of images in different classification dimensions.

[0154] 4. Multi-classifier training: The extracted target feature maps and corresponding image labels are input into multiple classifiers for training respectively, the best model parameters are selected according to the classification accuracy, and the classifier with the highest classification accuracy is saved as the trained classifier.

[0155] 5. Tag classification and statistics: We will continue to classify the high-frequency clicked advertising images provided by relevant personnel and obtain classification labels of multiple classification dimensions. We will analyze the co-occurrence of multiple labels for each classification label and form a co-occurrence matrix. For example, the horizontal column is the label name, and the vertical column is the advertising images of different products. The content is filled with 0 or 1, where 0 means it is not the label and 1 means it is the label.

[0156] 6. Design reference formation: Based on the co-occurrence matrix, generate systematic multi-label co-occurrence information, recommend target design solutions based on the multi-label co-occurrence information, and output the recommended target design solutions to advertising designers to provide design guidance. First, store the co-linear data of the co-occurrence matrix in the database to obtain multi-label co-occurrence information, and then the designer initiates a design recommendation request through retrieval. Assuming that the designer searches for "big-character poster style", the target design solutions retrieved in the multi-label co-occurrence information are "big-character poster + building + orange + Founder bold" and "big-character poster + Founder bold", then the output terminal displays "the most commonly used styles for big-character posters are building background, orange color and Founder bold font", and "the most commonly used font for big-character poster style is Founder bold font".

[0157] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the recommended method of the design scheme of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0158] An embodiment of the present application provides a design scheme recommendation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the design scheme recommendation method in the above-mentioned embodiment one.

[0159] Reference below Figure 5, which shows a schematic diagram of the structure of a design solution recommended device suitable for implementing the embodiment of the present application. The design solution recommended device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The design scheme recommended device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0160] like Figure 5 As shown, the design solution recommendation device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the design solution recommendation device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the design solution recommendation device to communicate with other devices wirelessly or wired to exchange data. Although the figure shows a design solution recommendation device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided instead.

[0161] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0162] Compared with the prior art, the beneficial effects of the design scheme recommendation device provided in the embodiment of the present application are the same as the beneficial effects of the design scheme recommendation method provided in the above embodiment, and other technical features in the design scheme recommendation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0163] It should be understood that the various parts disclosed in the embodiments of the present application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0164] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0165] An embodiment of the present application provides a computer-readable storage medium having computer-readable program instructions (ie, a computer program) stored thereon, wherein the computer-readable program instructions are used to execute the design solution recommendation method in the above-mentioned embodiment.

[0166] The computer-readable storage medium provided in the embodiment of the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0167] The computer-readable storage medium may be included in the design solution recommendation device; or may exist independently without being assembled into the design solution recommendation device.

[0168] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the design solution recommendation device, the design solution recommendation device performs the functions defined in the method of the embodiment disclosed in the present application.

[0169] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0170] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0171] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0172] The readable storage medium provided in the embodiment of the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned design scheme recommendation method. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the embodiment of the present application are the same as the beneficial effects of the design scheme recommendation method provided in the above-mentioned embodiment, which will not be repeated here.

[0173] An embodiment of the present application also provides a computer program product, including a computer program, which implements the steps of the design solution recommendation method as described above when executed by a processor.

[0174] Compared with the prior art, the beneficial effects of the computer program product provided in the embodiment of the present application are the same as the beneficial effects of the design scheme recommendation method provided in the above embodiment, and will not be elaborated here.

[0175] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A design solution recommendation method, characterized in that: The design scheme recommendation method includes: Acquire multiple advertisement images with high frequency of clicks, input each of the advertisement images into a pre-trained multi-label extraction model, and obtain multiple classification labels of each of the advertisement images under different classification dimensions, wherein the multi-label extraction model is used to perform feature extraction and feature classification of different classification dimensions on the advertisement images; Performing label co-occurrence analysis on each of the classification labels to obtain multi-label co-occurrence information, wherein the multi-label co-occurrence information includes at least one label matching scheme, and each of the label matching schemes includes at least two of the classification labels; If a design recommendation request input by a user is received, a target classification label corresponding to the design recommendation request is determined, and a target design solution is generated according to the target classification label and the multi-label co-occurrence information, and the target design solution is used as a recommendation result.

2. The method according to claim 1, characterized in that The multi-label extraction model includes a deep learning model and a multi-classifier. The step of inputting each of the advertising images into the pre-trained multi-label extraction model to obtain multiple classification labels of different classification dimensions corresponding to each of the advertising images includes: For each advertisement picture, extract features of the advertisement picture based on the pre-trained deep learning model to obtain an original feature map; The original feature map is input into the pre-trained multi-classifier to obtain multiple classification labels of the advertisement image under different classification dimensions.

3. The method according to claim 1, characterized in that The step of performing label co-occurrence analysis on each of the classification labels to obtain multi-label co-occurrence information includes: Determine the co-occurrence matrix corresponding to all the classification labels, and perform label combination on each of the classification labels to obtain at least one label matching scheme, wherein the co-occurrence matrix is ​​a matrix used to characterize whether each of the classification labels co-appears in the same advertisement image; For each of the label collocation schemes, determining the total number of times the classification labels included in the label collocation scheme co-occur in the co-occurrence matrix, and determining the total number as the number of co-occurrences of the label collocation scheme; The label collocation schemes are combined with the co-occurrence times corresponding to the label collocation schemes to obtain multi-label co-occurrence information.

4. The method according to claim 1, characterized in that The multi-label co-occurrence information also includes the co-occurrence times of each label matching scheme. The step of generating a target design scheme based on the target classification label and the multi-label co-occurrence information includes: Querying the candidate tag collocation schemes including the target classification tag in the multi-tag co-occurrence information, and querying the candidate co-occurrence counts of each of the candidate tag collocation schemes; A target label collocation scheme is selected from each of the candidate label collocation schemes based on the number of co-occurrences of each of the candidate labels, and the target label collocation scheme is determined as a target design scheme.

5. The method according to claim 4, characterized in that After the step of querying the candidate co-occurrence counts of each of the candidate tag collocation solutions, the method further includes: Acquire the candidate classification labels included in each of the candidate label matching schemes, and group each of the candidate label matching schemes according to the candidate classification labels to obtain at least one candidate label matching scheme group; For each of the candidate tag collocation scheme groups, based on the candidate co-occurrence counts of each of the candidate tag collocation schemes included in the candidate tag collocation scheme group, the step of selecting a target tag collocation scheme from each of the candidate tag collocation schemes based on the candidate co-occurrence counts is performed.

6. The method according to any one of claims 1 to 5, characterized in that: Before the step of obtaining a plurality of advertisement images with a high frequency of clicks, the method further includes: Acquire a multi-label extraction model including a deep learning model and a multi-classifier, wherein the multi-classifier includes a plurality of classifiers for different classification dimensions; Obtain a training data set, wherein the training data set includes training images and a plurality of image labels corresponding to the training images; Inputting the training image into the deep learning model of the multi-label extraction model to obtain a target feature map; Inputting the target feature map into each classifier in the multi-classifier of the multi-label extraction model to obtain a classification result output by each classifier; The model parameters of the deep learning model and each classifier are adjusted according to each of the classification results and each of the image labels, and the step of inputting the training image into the deep learning model of the multi-label extraction model to obtain the target feature map is returned to the step until the preset iteration termination condition is met, thereby obtaining the trained deep learning model and the trained multi-classifier.

7. The method according to claim 6, characterized in that The deep learning model includes a regularization module and a squeezing and excitation module. The step of inputting the training image into the deep learning model of the multi-label extraction model to obtain a target feature map includes: Extracting features from the training image using the regularization module, and standardizing the extracted features to obtain an initial feature map; The initial feature map is adjusted based on the channel weight of the initial feature map by the squeeze and excitation module to obtain a target feature map.

8. A design solution recommendation device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the design solution recommendation method according to any one of claims 1 to 7.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the design solution recommendation method according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the design solution recommendation method according to any one of claims 1 to 7 are implemented.

Images