Advertisement creative effect prediction method and device, equipment and readable storage medium

Abstract

The invention discloses an advertisement creative effect prediction method and device, equipment and a readable storage medium, and relates to the technical field of artificial intelligence. Comprising the steps of firstly obtaining a target creative image, inputting the target creative image into a pre-trained advertisement creative effect prediction model, and extracting visual information of the target creative image; extracting text information of the target creative image, and fusing the visual information and the text information to obtain visual text information; then feature decoupling is carried out on the visual information to obtain decoupling features, and the decoupling features and the visual text information are spliced to obtain splicing features; and finally, based on the splicing features, determining an advertisement creative effect prediction result of the target creative image. By extracting text information and identification information, the model can penetrate through the complexity of a visual surface, and a significant visual feature which truly determines an advertisement effect and has a high signal-to-noise ratio is extracted, so that the accuracy and robustness of prediction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an advertisement creative effect prediction method, device and equipment and readable storage medium. BACKGROUND

[0002] In the current field of computing advertising, the prediction of advertisement creative effect (such as click rate, conversion rate) is the core link to improve the efficiency of delivery. The known technical solutions can be mainly classified into two categories: one category is the statistical or machine learning method based on historical delivery data, for example, using a logistic regression, gradient boosting decision tree or other model to predict according to the size, industry, advertiser and other metadata of the creative and its historical performance, but this kind of method is heavily dependent on historical data, and there is a "cold start" problem for new creatives without delivery records, and it cannot give an effective estimate. The other category is the end-to-end deep learning method based on content, which extracts features from creative images, videos and text content and performs end-to-end prediction by using convolutional neural networks or Transformer models, which to some extent alleviates the cold start problem, but usually treats creatives as an indivisible whole. This "holism" processing method leads to a lack of understanding of the internal combination structure of "creative style" and "promoted product" by the model, making it difficult to generalize to new creative combinations that have not been seen before. At the same time, the model is also easily disturbed by noise information such as irrelevant background and secondary elements in the creative, and cannot distinguish the primary and secondary importance of different elements, thereby affecting the accuracy and robustness of the prediction.

[0003] Therefore, there is an urgent need for an advertisement creative effect prediction method that can overcome the above-mentioned defects. SUMMARY

[0004] The purpose of the present application is to provide an advertisement creative effect prediction method, device, equipment and readable storage medium, which can penetrate the complexity of the visual surface by extracting text information and identifying information, extract significant visual features with high signal-to-noise ratio that really determine the effect of the advertisement, and improve the accuracy and robustness of the prediction.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: In a first aspect, the present application provides an advertisement creative effect prediction method, which comprises: obtaining a target creative image and inputting the target creative image into a pre-trained advertisement creative effect prediction model to extract visual information of the target creative image; extracting text information of the target creative image and fusing the visual information and the text information to obtain visual text information; decoupling the visual information to obtain decoupled features, and splicing the decoupled features with the visual text information to obtain spliced features; Based on the splicing feature, a creative effect prediction result of the target creative image is determined.

[0006] In some embodiments, visual information of the target creative image is extracted, including: The target creative image is input into a foreground segmentation model to obtain a binary foreground mask of the target creative image; Based on the binary foreground mask and the target creative image, a foreground image is generated; The foreground image is input into a visual encoder to obtain foreground features; The target creative image is input into a saliency detection model to obtain a grayscale saliency map of the target creative image; The grayscale saliency map is copied into three channels and then input into the visual encoder to obtain saliency features; The foreground features and the saliency features are fused to obtain visual information of the target creative image.

[0007] In some embodiments, text information of the target creative image is extracted, and the visual information and the text information are fused to obtain visual-text information, including: The target creative image is processed by a text encoder to obtain text information of the target creative image; The text information is processed based on the visual information to obtain visual-text information.

[0008] In some embodiments, the visual information is feature-decoupled to obtain decoupled features, including: The visual information is input into a first multi-layer perception network and a second multi-layer perception network of a decoupler respectively to obtain decoupled features; the first multi-layer perception network and the second multi-layer perception network have the same structure but different parameters; the decoupled features include attribute visual features and object visual features.

[0009] In some embodiments, a training method of the creative effect prediction model includes: A sample creative image is obtained, and the sample creative image is cleaned; A structured label of the sample creative image is created; Based on the structured label, the creative effect prediction model is trained.

[0010] In some embodiments, the structured label of the sample creative image is created, including: Creative attributes and product objects of the sample creative image are obtained; For each sample creative image, the creative attributes and the product objects of the sample creative image are combined as a structured label of the sample creative image.

[0011] In a second aspect, the present application further provides an advertisement creative effect prediction device, which comprises: an information extraction module configured to obtain a target creative image, and input the target creative image into a pre-trained advertisement creative effect prediction model to extract visual information of the target creative image; an information fusion module configured to extract text information of the target creative image, and fuse the visual information and the text information to obtain visual-text information; a feature splicing module configured to decouple features of the visual information to obtain decoupled features, and splice the decoupled features with the visual-text information to obtain spliced features; an effect prediction unit configured to determine an advertisement creative effect prediction result of the target creative image based on the spliced features.

[0012] In a third aspect, the present application further provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the advertisement creative effect prediction method provided in the first aspect when executing the computer program.

[0013] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the advertisement creative effect prediction method provided in the first aspect.

[0014] In a fifth aspect, the present application further provides a computer program product, which comprises a computer program, and the computer program is executable on a processor to implement the advertisement creative effect prediction method provided in the first aspect.

[0015] The advertisement creative effect prediction method provided in the present application comprises the following steps: first, obtaining a target creative image, and inputting the target creative image into a pre-trained advertisement creative effect prediction model to extract visual information of the target creative image; second, extracting text information of the target creative image, and fusing the visual information and the text information to obtain visual-text information; third, decoupling features of the visual information to obtain decoupled features, and splicing the decoupled features with the visual-text information to obtain spliced features; and finally, determining an advertisement creative effect prediction result of the target creative image based on the spliced features. By extracting the text information and the recognition information, the model can penetrate the complexity of the visual surface, extract significant visual features with high signal-to-noise ratio which really determine the advertisement effect, so as to improve the accuracy and robustness of the prediction.

[0016] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, and to implement the content of the description, the following will describe the preferred embodiments of the present application in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 This is a flowchart illustrating a method for predicting the effectiveness of advertising creatives according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a training method for an advertising creative effect prediction model according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an advertising creative effect prediction device according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0018] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should be noted that references to "an embodiment," "embodiment," "example embodiment," etc., in this specification refer to the described embodiment including specific features, structures, or characteristics; however, not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge scope of those skilled in the art.

[0019] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0020] In some embodiments, such as Figure 1 As shown, a method for predicting the effectiveness of advertising creative is provided, the specific method including: S101, acquire the target creative image and input the target creative image into the pre-trained advertising creative effect prediction model to extract the visual information of the target creative image.

[0021] The target creative image is the image for which advertising creative effect prediction is required. The advertising creative effect prediction model includes a subsequent foreground segmentation model, a visual encoder, a saliency detection model, a text encoder, a decoupler, and an output layer.

[0022] Specifically, the target creative image can be input into a foreground segmentation model first to obtain a binary foreground mask of the target creative image; then a foreground image is generated based on the binary foreground mask and the target creative image; then the foreground image is input into a visual encoder to obtain foreground features; then the target creative image is input into a saliency detection model to obtain a grayscale saliency map of the target creative image; then the grayscale saliency map is copied into three channels and input into the visual encoder to obtain saliency features; finally, the foreground features and the saliency features are fused to obtain visual information of the target creative image.

[0023] For example, the foreground segmentation model can be a Segment Anything Model (SAM), the visual encoder can be a CLIPViT-L / 14 model, and the saliency detection model can be a U²-Net. A binary foreground mask can be generated according to the input target creative image, where 1 represents the foreground and 0 represents the background. The binary foreground mask is multiplied by the target creative image to obtain a foreground image of the target creative image. The foreground image is then processed by the visual encoder to obtain foreground features. At the same time, the target creative image is input into the saliency detection model to obtain a grayscale saliency map of the target creative image. The grayscale saliency map is then copied into three channels and input into the visual encoder to obtain saliency features. The foreground features and the saliency features are then input into a feature fusion module for linear projection, with the specific formula being: ; ; ; wherein, is a learnable projection matrix, is the foreground feature, is the saliency feature, and Q, K, and V obtained by calculation represent query vectors, key vectors, and value vectors, respectively, which are used to capture information patterns that need to be focused on in the target creative image, attention weights for matching and calculation with Q, and final feature representations output after attention weighting.

[0024] Further calculation is performed based on multi-head attention: ; wherein, T represents a transposed matrix, K represents the dimension size of the key vector, which is used to scale the inner product result to prevent the gradient of the softmax function from being too small due to a too large numerical value, thereby ensuring the stability and effectiveness of the attention calculation. Then, after residual connection and layer normalization, the visual information of the target creative image is obtained.

[0025] S102, extract the text information of the target creative image, and fuse the visual and text information to obtain visual text information.

[0026] Optionally, a text encoder can be used to process the target creative image to obtain the text information of the target creative image; then, based on the visual information, the text information can be further processed to obtain visual text information.

[0027] Specifically, for the creative combination in the target creative image, a structured text cue can be constructed, such as: "An advertisement image about 'sneakers.' Creative features: live model display, prominent discount." This text cue is then encoded into an initial text feature vector, i.e., text information, using CLIP's text encoder. Subsequently, the text and visual information are input into the context alignment module, which internally employs a cross-modal multi-head attention mechanism. Its linear projection is defined as: ; ; ; in, For text information, This is visual information. Here... Emphasizing the current need for attention to textual semantics in cross-modal attention, This measures the correlation between semantic and visual features. To provide the final visual representation for cross-modal interaction after attention weighting.

[0028] Attention is calculated as follows: ; The final output is visual text information: .

[0029] S103, decouple the visual information to obtain decoupled features, and then concatenate the decoupled features with the visual text information to obtain concatenated features.

[0030] Optionally, visual information is input into the first and second multilayer perceptual networks of the decoupler to obtain decoupling features; then, the decoupling features are concatenated with visual text information to obtain concatenated features.

[0031] The first and second multilayer perceptrons have the same structure but different parameters; the decoupling features include attribute visual features and object visual features.

[0032] For example, the decoupler comprises two structurally identical but parameter-independent multilayer perceptrons. and That is, the first layer sensing network and the second layer sensing network, which are defined as: ; ; The input visual information is decoupled using a decoupler. Decoupling to obtain attribute visual features separately and object visual features Then, the attribute visual features, object visual features, and visual text information are concatenated: The final output is the spliced ​​feature: ,in, It is a multi-layer MLP network.

[0033] S104, Based on the splicing features, determine the predicted advertising creative effect of the target creative image.

[0034] Specifically, the pre-trained advertising creative effect prediction model can output the advertising creative effect prediction result of the target creative image based on the splicing features.

[0035] The advertising creative effect prediction method in the above embodiments first acquires the target creative image and inputs it into a pre-trained advertising creative effect prediction model to extract the visual information of the target creative image; then, it extracts the textual information of the target creative image and fuses the visual and textual information to obtain visual-textual information; next, it decouples the visual information to obtain decoupled features, and then concatenates the decoupled features with the visual-textual information to obtain concatenated features; finally, based on the concatenated features, it determines the advertising creative effect prediction result of the target creative image. By extracting textual and recognition information, the model can penetrate the complexity of the visual surface and extract the truly decisive, high signal-to-noise ratio, significant visual features, thereby improving the accuracy and robustness of the prediction.

[0036] Optionally, in another embodiment, such as Figure 2 As shown, a training method for an advertising creative effect prediction model is also provided, the specific method including: S201, Obtain the sample creative image and clean the sample creative image.

[0037] Specifically, historical creative images are obtained as sample creative images. However, these sample creative images may have too few exposures and inaccurate click-through rates (CTRs). Therefore, sample creative images with low exposures, as well as those with excessively low or high CTRs, need to be removed. Additionally, sample creative images with low image quality also need to be removed. Then, the original CTR for each sample creative image is calculated, and Bayesian smoothing is used to correct the original CTR. ; in, This indicates the total number of clicks on the creative image in the history. This indicates the total number of times the creative image has been displayed in the historical record. and These are prior distribution parameters determined based on the average clicks and impressions of all ads across the entire platform. For example, You can set it as the average number of clicks across all creatives. Set as the average number of unclicked items. Smoothed This represents the smoothed click-through rate, used to reduce the bias caused by low-exposure samples or extreme click-through rate samples, and will serve as the final regression target for training the advertising creative effect prediction model.

[0038] S202, Create structured labels for sample creative images.

[0039] Optionally, you can first obtain the creative attributes and product objects of the sample creative images; then, for each sample creative image, combine the creative attributes and product objects of the sample creative image into the structured tags of the sample creative image.

[0040] For example, creative attributes can be: 1. Visual style: {pure white / solid color background, scene background, hand-drawn / illustration style, poster / large character poster style, ...} 2. Subject: {product flat lay / collective display, product single close-up, live model display (full body / half body), celebrity / KOL endorsement, ...} 3. Marketing information: {discount / price highlighting, free gifts / buy-one-get-one-free information, function / selling point listing, brand logo highlighting, ...} 4. Emotional atmosphere: {warm home, business elite, youthful vitality, technological future, ...}; Product objects can be: {mobile phone, laptop, sneakers, lipstick, facial cleanser, ...}. Automated pre-labeling can be performed using a multimodal large model (such as GPT-4V) to obtain pre-labeled tags, which are then manually reviewed and corrected to obtain structured tags for the sample creative images.

[0041] S203 is a model for predicting the effectiveness of advertising creatives, trained based on structured tags.

[0042] Specifically, a dataset is first constructed based on structured labels and sample creative images. 80% of this dataset is used as the training set, 10% as the test set, and 10% as the validation set. This dataset is then used to train the advertising creative effect prediction model. The final regression objective of the training model is the smoothed... A weighted binary cross-entropy loss can be used, which is equivalent to the loss function of logistic regression and is more suitable for click-through rate prediction tasks. Specifically: ; in, and The weights are calculated based on the ratio of positive to negative samples. This represents the regression loss function, used to measure the difference between the model's predicted click-through rate and the actual click-through rate. This represents the actual click-through rate of the sample (the actual tag, which typically takes the value of 0 or 1). This represents the click-through rate predicted by the model.

[0043] Combined alignment loss To explicitly supervise the decoupling process, an alignment loss based on contrastive learning is introduced. Similarly, for... Also obtained using a decoupler and , representing attribute text features and object text features, respectively.

[0044] Within a batch Should be corresponding The cosine similarity is highest for (positive samples), while it is highest for other samples within the batch. The cosine similarity of (negative samples) is minimized. This can be achieved using the information-noise contrastive loss function.

[0045] ; in, The information-noise contrastive loss function is used to maximize the similarity of positive sample pairs and minimize the similarity of negative sample pairs. The calculated result is... This represents the cross-modal alignment loss, used to enhance the consistency between visual and textual features. Finally, the total loss is calculated: , This represents the regression loss, used for fitting click-through rate prediction tasks. It is a hyperparameter used to balance regression tasks and representation learning tasks.

[0046] The training method for the aforementioned advertising creative effect prediction model first acquires and cleans sample creative images; then, it creates structured labels for the sample creative images; finally, based on these structured labels, it trains the advertising creative effect prediction model. This increases the prediction accuracy of the advertising creative effect prediction model.

[0047] In another embodiment, the trained advertising creative performance prediction model can also be converted into a format supported by a high-performance inference engine (such as TensorRT) through intermediate representations such as ONNX, thereby achieving the following functions: Function 1: New Creative Prediction: Operations personnel upload a newly created advertising creative sketch, manually select its corresponding "product object" and "creative attribute," and the system calls the backend service to return the predicted CTR score and star rating within seconds. Function 2: Virtual A / B Testing: Operations personnel can upload multiple sketches with different creative directions (for example, the same product, using three styles: "celebrity endorsement," "promotional poster," and "scenario story"). The system will simultaneously predict their performance and provide a ranking of their merits. Operations personnel can directly select the best-performing scheme for refined production, greatly saving testing costs and time. Function 3: Creative Insight Analysis: The backend periodically aggregates and analyzes all prediction results to generate insight reports. For example, "Recently, in the 'clothing' category, the CTR performance of the 'real model street photography' style is generally higher than that of the 'product flat lay' style." This kind of insight can directly guide future creative directions.

[0048] To more comprehensively demonstrate this solution, this embodiment presents an optional method for predicting the effectiveness of advertising creative content: S301, acquire sample creative images and clean the sample creative images.

[0049] S302, obtain the creative attributes and product object of the sample creative image.

[0050] S303, For each sample creative image, combine the creative attributes and product objects of the sample creative image into a structured tag for the sample creative image.

[0051] S304, a model for predicting the effectiveness of advertising creatives, is trained based on structured tags.

[0052] S305, acquire the target creative image and input the target creative image into the pre-trained advertising creative effect prediction model.

[0053] S306, Input the target creative image into the foreground segmentation model to obtain the binary foreground mask of the target creative image.

[0054] S307, Generate a foreground image based on a binary foreground mask and a target creative image.

[0055] S308: Input the foreground image into the visual encoder to obtain foreground features.

[0056] S309, Input the target creative image into the saliency detection model to obtain the grayscale saliency map of the target creative image.

[0057] S310: The grayscale saliency image is copied to three channels and then input into the visual encoder to obtain saliency features.

[0058] S311, feature fusion is performed on foreground features and salient features to obtain visual information of the target creative image.

[0059] S312, use a text encoder to process the target creative image to obtain the text information of the target creative image.

[0060] S313 performs concretization processing on text information based on visual information to obtain visual text information.

[0061] S314, visual information is input into the first and second multilayer perceptual networks of the decoupler to obtain decoupling features.

[0062] The first and second multilayer perceptrons have the same structure but different parameters; the decoupling features include attribute visual features and object visual features.

[0063] S315, the decoupled features are concatenated with the visual text information to obtain the concatenated features.

[0064] S316, Based on splicing features, determine the predicted advertising creative effect of the target creative image.

[0065] The specific processes of S301-S316 described above can be found in the description of the above method embodiments. Their implementation principles and technical effects are similar, and will not be repeated here.

[0066] Based on the same inventive concept, this application also provides an advertising creative effect prediction device for implementing the advertising creative effect prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the advertising creative effect prediction device provided below can be found in the limitations of the advertising creative effect prediction method described above, and will not be repeated here.

[0067] In one embodiment, such as Figure 3 As shown, a device for predicting the effectiveness of advertising creative is provided, the device comprising: The information extraction module 30 is used to acquire the target creative image and input the target creative image into the pre-trained advertising creative effect prediction model to extract the visual information of the target creative image; The information fusion module 31 is used to extract text information from the target creative image and fuse visual and text information to obtain visual text information. The feature splicing module 32 is used to decouple the visual information to obtain decoupled features, and then splice the decoupled features with the visual text information to obtain spliced ​​features. The effect prediction unit 33 is used to determine the advertising creative effect prediction result of the target creative image based on the splicing features.

[0068] In another embodiment, the above Figure 3 The information extraction module 30 is specifically used to: input the target creative image into the foreground segmentation model to obtain the binary foreground mask of the target creative image; Based on a binary foreground mask and a target creative image, a foreground image is generated; the foreground image is input into a visual encoder to obtain foreground features; the target creative image is input into a saliency detection model to obtain a grayscale saliency map of the target creative image; the grayscale saliency map is copied to three channels and then input into a visual encoder to obtain saliency features; the foreground features and saliency features are fused to obtain the visual information of the target creative image.

[0069] In another embodiment, the above Figure 3 The information fusion module 31 is specifically used for: processing the target creative image using a text encoder to obtain the text information of the target creative image; and performing specific processing on the text information based on visual information to obtain visual text information.

[0070] In another embodiment, the above Figure 4 The feature splicing module 32 is specifically used to: input visual information into the first multilayer perceptron and the second multilayer perceptron of the decoupler to obtain decoupled features; the first multilayer perceptron and the second multilayer perceptron have the same structure but different parameters; the decoupled features include attribute visual features and object visual features.

[0071] In another embodiment, the above Figure 3 The device for predicting the effectiveness of advertising creatives is also specifically used for: acquiring sample creative images and cleaning them; creating structured labels for the sample creative images; and training an advertising creative effectiveness prediction model based on the structured labels.

[0072] In another embodiment, creating structured tags for sample creative images specifically includes: obtaining the creative attributes and product objects of the sample creative images; and for each sample creative image, combining the creative attributes and product objects of the sample creative image into a structured tag for the sample creative image.

[0073] This application also provides an electronic device, in some embodiments, referring to... Figure 4As shown, the electronic device 700 includes an input unit 710, a memory 720, a processor 730, and an output unit 740. The memory 720 stores program instructions that can be executed on the processor 730. The processor 730 can execute the prediction method and / or technical solution based on the advertising creative effect in the foregoing embodiments by calling the program instructions. The electronic device 700 can be a mobile terminal device such as a mobile phone or a computer.

[0074] Furthermore, embodiments of this application also provide a computer-readable storage medium for storing a computer program that performs a method for predicting the effectiveness of advertising creatives. For example, computer program instructions, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions that invoke the methods of this application may be stored in a fixed or removable storage medium, and / or transmitted via data streams in broadcast or other signal carrying media, and / or stored in a storage medium that operates according to the program instructions.

[0075] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0076] The technical features of the above embodiments can be arbitrarily integrated. For the sake of brevity, not all possible integrations of the technical features in the above embodiments are described. However, as long as the integration of these technical features does not contradict each other, they should be considered to be within the scope of this specification.

[0077] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for predicting the effectiveness of advertising creative ideas, characterized in that, The method includes: Acquire a target creative image and input the target creative image into a pre-trained advertising creative effect prediction model to extract the visual information of the target creative image; Extract the text information of the target creative image, and fuse the visual information and the text information to obtain visual text information; The visual information is decoupled to obtain decoupled features, and the decoupled features are concatenated with the visual text information to obtain concatenated features; Based on the splicing features, the advertising creative effect prediction result of the target creative image is determined.

2. The method for predicting the effect of advertising creativity as described in claim 1, characterized in that, Extracting the visual information of the target creative image, including: The target creative image is input into the foreground segmentation model to obtain the binary foreground mask of the target creative image; Based on the binary foreground mask and the target creative image, a foreground image is generated; The foreground image is input into a visual encoder to obtain foreground features; The target creative image is input into the saliency detection model to obtain the grayscale saliency map of the target creative image; The grayscale saliency map is copied into three channels and then input into the visual encoder to obtain saliency features; The foreground features and the saliency features are fused to obtain the visual information of the target creative image.

3. The method for predicting the effect of advertising creativity as described in claim 2, characterized in that, Extract the text information from the target creative image, and fuse the visual information and the text information to obtain visual text information, including: The target creative image is processed using a text encoder to obtain the text information of the target creative image; The text information is further processed based on the visual information to obtain visual text information.

4. The method for predicting the effect of advertising creativity as described in claim 3, characterized in that, The visual information is decoupled to obtain decoupled features, including: The visual information is input into the first and second multilayer perceptrons of the decoupler to obtain decoupling features. The first and second multilayer perceptrons have the same structure but different parameters. The decoupling features include attribute visual features and object visual features.

5. The method for predicting the effect of advertising creativity as described in claim 1, characterized in that, The training method for the advertising creative effect prediction model includes: Acquire sample creative images and clean the sample creative images; Create structured labels for the sample creative images; The advertising creative effect prediction model is trained based on the structured tags.

6. The method for predicting the effect of advertising creativity as described in claim 5, characterized in that, Creating structured labels for the sample creative images includes: Obtain the creative attributes and product objects of the sample creative image; For each sample creative image, the creative attributes and product objects of the sample creative image are combined into a structured tag for the sample creative image.

7. A device for predicting the effectiveness of advertising creative ideas, characterized in that, The device includes: The information extraction module is used to acquire the target creative image and input the target creative image into a pre-trained advertising creative effect prediction model to extract the visual information of the target creative image; The information fusion module is used to extract the text information of the target creative image and fuse the visual information and the text information to obtain visual text information; The feature splicing module is used to decouple the visual information to obtain decoupled features, and splice the decoupled features with the visual text information to obtain spliced ​​features; The effect prediction unit is used to determine the advertising creative effect prediction result of the target creative image based on the splicing features.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting the advertising creative effect as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for predicting the effect of advertising creatives as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for predicting the effect of advertising creatives as described in any one of claims 1 to 6.

Images