Advertising banner production system

Abstract

To provide an advertising banner production system which produces an advertising banner having a good advertising effect in a short time.SOLUTION: In an arithmetic processing unit 4 in an advertising banner production system, a foreground image production unit 11 produces a foreground image to be used in an advertising banner candidate, an advertising text production unit 12 produces an advertising text to be used in the advertising banner candidate, an advertising banner candidate production unit 13 combines the foreground image produced by the foreground image production unit 11 and the advertising text produced by the advertising text production unit 12 to thereby produce the advertising banner candidate, and a click rate prediction unit 14 predicts a click rate of the produced advertising banner candidate to be used.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an advertisement banner generating system. [Background technology]

[0002] One analysis device (a) collects log data indicating whether or not an advertising banner has been displayed on a user terminal and whether or not a target website corresponding to that advertising banner has been displayed on the user terminal, and (b) calculates a lift coefficient, which indicates advertising effectiveness, by analyzing the log data (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-133482 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the case of the above-mentioned analysis device, the advertising effect cannot be known unless the advertising banner is actually used for advertising. In other words, in the case of the above-mentioned analysis device, it is difficult to specify the advertising effect in advance, and the advertising banner with low advertising effect is improved while actually using the advertising banner, which takes time to generate an advertising banner with good advertising effect, and is inefficient.

[0005] The present invention has been made in consideration of the above problems, and has an object to provide an advertising banner generating system capable of generating advertising banners with good advertising effectiveness in a short period of time. [Means for solving the problem]

[0006] An advertising banner generation system according to the present invention includes a foreground image generation unit that generates a foreground image to be used for an advertising banner candidate, an advertising text generation unit that generates advertising text to be used for the advertising banner candidate, an advertising banner candidate generation unit that generates an advertising banner candidate by combining the foreground image generated by the foreground image generation unit and the advertising text generated by the advertising text generation unit, and a click rate prediction unit that predicts a click rate when the generated advertising banner candidate is used.

[0007] A click rate prediction device according to the present invention is a click rate prediction device that predicts a click rate when an advertising banner candidate is used, and includes a banner feature vector generation unit that generates a feature vector of the advertising banner candidate, and a click rate prediction unit that predicts a click rate based on the feature vector of the advertising banner candidate. The click rate prediction unit includes a machine-learned learner that uses combinations of feature vectors of past advertising banners and click rates of the advertising banners as training data, and derives a click rate prediction value corresponding to the feature vector of the advertising banner candidate using the learner.

[0008] An advertising banner generating method according to the present invention includes an image generating step of generating a foreground image to be used for an advertising banner candidate, a text generating step of generating advertising text to be used for the advertising banner candidate, an advertising banner candidate generating step of generating an advertising banner candidate by combining the foreground image generated in the image generating step and the advertising text generated in the text generating step, and a click rate predicting step of predicting a click rate when the generated advertising banner candidate is used. Effect of the Invention

[0009] According to the present invention, an advertising banner generating system is provided that can generate advertising banners with good advertising effectiveness in a short period of time.

[0010] The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. [Brief description of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram showing a configuration of an advertising banner generating system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a block diagram showing an example of foreground image generating unit 11 in FIG. [Diagram 3] FIG. 3 is a block diagram showing an example of the advertisement text generating unit 12 in FIG. [Figure 4] FIG. 4 is a diagram for explaining the layout setting. [Diagram 5] FIG. 5 is a diagram showing an example of an advertising banner candidate based on the layout setting shown in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0013] Embodiment 1

[0014] Fig. 1 is a block diagram showing a configuration of an advertising banner generation system according to an embodiment of the present invention. The advertising banner generation system shown in Fig. 1 is configured with one information processing device (such as a personal computer), but processing units and units described below may be distributed among multiple information processing devices capable of data communication with each other. Furthermore, such multiple information processing devices may include a GPU (Graphics Processing Unit) that processes specific calculations in parallel.

[0015] The advertising banner generation system shown in FIG. 1 is a system that generates advertising banners used for web advertising, etc. (here, banner images that are clickable and embedded in web pages with links to web pages of advertisers, etc.), and comprises a storage device 1, a communication device 2, a user interface 3, and a processing device 4.

[0016] The storage device 1 is a non-volatile storage device such as a flash memory or a hard disk, and stores various data (setting data, etc.) and programs.

[0017] The communication device 2 is a device capable of data communication, such as a network interface, a peripheral device interface, or a modem, and performs data communication with other devices as necessary.

[0018] The user interface 3 includes a display device 3a such as a display for displaying an operation screen and the like, and an input device 3b such as a keyboard or a touch panel for receiving user operations.

[0019] In FIG. 1, the user interface 3 is connected to the arithmetic processing device 4 via an internal bus or a peripheral device interface. Alternatively, for example, if the arithmetic processing device 4 is built into a server on a network (such as the Internet or a LAN (Local Area Network)), the user interface 3 serves as a user interface device of a terminal device capable of communicating with the server, and functions as a user interface for processing units and units realized in the arithmetic processing device 4 through data communication between the server and the terminal device.

[0020] The arithmetic processing device 4 is a computer equipped with a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc., and operates as various processing units by loading programs from the ROM, the storage device 1, etc. into the RAM and executing them on the CPU. The RAM is used as a work memory for the processing units.

[0021] Here, by executing a program in the storage device 1, the calculation processing device 4 operates as a foreground image generation unit 11, an advertising text generation unit 12, an advertising banner candidate generation unit 13, a click rate prediction unit 14, and an advertising banner selection unit 15.

[0022] Foreground image generating unit 11 generates a foreground image to be used for an advertising banner candidate (advertising banner candidate). The advertising banner candidate is a raster image (an image in a file format such as PNG or JPEG) generated by combining a foreground image (an image of a specific object such as a person, a landscape, or a product) and advertising text (an advertising copy such as a catch phrase) in a specific layout and a specific color scheme.

[0023] FIG. 2 is a block diagram showing an example of foreground image generating unit 11 in FIG.

[0024] In this embodiment, for example, as shown in FIG. 2, foreground image generation unit 11 includes ranking processing unit 21 and foreground image selection unit 22.

[0025] The ranking processing unit 21 derives a rank for each stock image in the stock image group. Here, the stock images are existing images such as image materials, and include foreground images of people, landscapes, specific objects, and the like.

[0026] For example, the stock image is acquired via a network from an image library that provides image material free of charge or for a fee. For example, the foreground image generating unit 11 acquires stock images related to a pre-specified keyword using a public API (Application Programming Interface) of such an image library. When it is desired to extract a stock image for a foreground image of an advertisement banner of a used car dealer, for example, the stock image is extracted and acquired using keywords such as "photo only", "Japanese", "female", "one person", "smile", "business", and "OK sign".

[0027] These keywords may be specified in whole or in part by the user, or may be extracted in whole or in part by image analysis from foreground images with click rates exceeding a predetermined value that have been used in past advertising banners for the company or its competitors (i.e., advertising targets of the same type as the current advertising target). In other words, the attribute values ​​of the foreground images obtained by image analysis are used as keywords. If the foreground image includes people, the attribute values ​​include the number of people, gender (if multiple people, gender ratio), age (if multiple people, ratio of each age), facial expression (smiling, surprised, afraid, disgust, anger, sadness, expressionless, etc.), race (Japanese, Asian, Caucasian, Black, etc.), etc.

[0028] Specifically, the ranking processing unit 21 derives the similarity (here, cosine similarity) between the feature vector of a past advertising banner and the feature vector of an existing stock image, and generates a rank for each stock image based on the similarity.

[0029] The past advertisement banners are, for example, advertisement banners that have actually been used as advertisements in the past, and are images that have obtained a click rate that exceeds a predetermined value. The feature vectors are generated from each image by a predetermined embedding process.

[0030] The foreground image selection unit 22 selects a foreground image from the stock images ranked as described above based on their ranks. For example, a predetermined number of top-ranked stock images are selected as the foreground images.

[0031] In this manner, a foreground image is selected from an existing stock of images based on the similarity measures described above.

[0032] In addition, foreground image generation unit 11 may use stable diffusion or the like to automatically generate a foreground image from words related to the advertising target of the candidate advertising banner (product name, service name, company name, related keywords, etc.) using a machine-learned learning device.

[0033] Returning to Fig. 1, the advertising text generation unit 12 generates advertising text to be used for the advertising banner candidate. Here, the advertising text generation unit 12 (a) automatically generates advertising text candidates from words related to the advertising target of the advertising banner candidate using, for example, GPT (Generative Pretrained Transformer), (b) derives the similarity between the feature vector of the text information obtained from the past advertising banner and the feature vector of the advertising text candidate, and (b) selects advertising text (such as a catchy copy) from the advertising text candidates based on the similarity.

[0034] FIG. 3 is a block diagram showing an example of the advertisement text generating unit 12 in FIG.

[0035] In this embodiment, as shown in FIG. 3, for example, advertising text generation unit 12 includes an automatic text generation processing unit 31, embedding processing units 32 and 33, a ranking processing unit , and an advertising text selection unit .

[0036] The automatic text generation processing unit 31 automatically generates advertisement text candidates from words related to the advertising target of the advertisement banner candidates, for example, by using GPT.

[0037] When GPT is used, the learning device of the automatic text generation processing unit 31 is pre-trained based on a predetermined corpus, and then fine-tuned using pairs of advertising text (catchphrase) and advertising target (product name, service name, company name, etc.), related keywords (those indicating temporal characteristics such as "fast", those indicating economic characteristics such as "cheap", those indicating psychological characteristics such as "peace of mind", etc.), as training data, and the automatic text generation processing unit 31 uses the learning device after machine learning to automatically generate advertising text candidates from words related to the advertising target of the advertising banner candidates (product name, service name, company name, related keywords, etc.).

[0038] The embedding processing unit 32 performs embedding processing to generate a feature vector of the automatically generated advertisement text candidate.

[0039] The embedding processor 33 performs an embedding process similar to that of the embedding processor 32, and generates a feature vector of text information obtained from a past advertisement banner. Note that this text information is obtained by extracting text entities included in the past advertisement banner by image analysis, and executing OCR processing on the text entities.

[0040] The ranking processing unit 34 derives the similarity (here, cosine similarity) between the feature vector of the text information obtained from the past advertising banner and the feature vector of the advertising text candidate, and generates a rank for each advertising text candidate based on the similarity.

[0041] The advertisement text selection unit 35 selects advertisement text from the advertisement text candidates ranked as described above based on their ranks. For example, a predetermined number of top-ranked advertisement text candidates are selected for the foreground image.

[0042] In this manner, advertising text is selected from the advertising text candidates based on the above-mentioned similarity.

[0043] Returning to FIG. 1, advertising banner candidate generating unit 13 generates advertising banner candidates by combining the foreground image generated by foreground image generating unit 11 and the advertising text generated by advertising text generating unit 12.

[0044] Specifically, the advertisement banner candidate generating unit 13 generates a plurality of combinations of foreground images and advertisement texts from a plurality of foreground images and a plurality of advertisement texts, and generates a plurality of advertisement banner candidates by applying a plurality of banner setting data to each of the plurality of combinations, thereby generating a large number of advertisement banner candidates that combine different foreground images, different advertisement texts, and different banner setting data.

[0045] Here, each banner setting data includes a layout setting, a font setting, a color setting, etc. The layout setting indicates the respective positions and sizes of the foreground image and the advertising text in the size of the advertising banner. Specifically, the layout setting includes, for one or more foreground images and one or more pieces of text, the vertical and horizontal sizes of the bounding box of the foreground image or the advertising text, and the pixel coordinate values ​​of a specific vertex (such as the upper left vertex) of the bounding box. The font setting indicates the typeface of the advertising text. The color setting indicates the color of the advertising text, the background color (the color of the parts other than the advertising text and the foreground image), the color tone of the entire advertising banner candidate (cool colors, warm colors, etc.), etc.

[0046] For example, with regard to banner setting data (layout settings, font settings, color settings, etc.), banner setting data of past advertising banners (advertising banners that have actually been used in advertisements in the past) for the same type of advertising target (product, company, etc.) for which an advertising banner candidate is to be generated is collected, and of the collected banner setting data, that whose average click rate exceeds a predetermined value is used.

[0047] The advertisement text is text data (character code string), but is embedded as a raster image in the advertisement banner candidate.

[0048] Fig. 4 is a diagram for explaining the layout setting, and Fig. 5 is a diagram showing an example of an advertising banner candidate based on the layout setting shown in Fig. 4.

[0049] 4, foreground image region 111 and text regions 112, 113, and 114 are specified in a specified predetermined banner region 101, with a foreground image being assigned to foreground image region 111 and text being assigned to text regions 112, 113, and 114. Here, text regions 112, 113, and 114 are each assigned an attribute indicating the type of corresponding text, with the attribute of text region 112 indicating a catchy slogan as the text type, the attribute of text region 113 indicating a product name (company name) as the text type, and the attribute of text region 114 indicating an appealing point as the text type.

[0050] 5, text 213, 214 (product name (company name) and selling points) to be assigned to text areas 113, 114 are specified in advance and assigned fixedly, a generated foreground image 211 is assigned to foreground image area 111, and a generated advertisement text 212 is assigned to text area 112. In this way, one advertisement banner candidate 201 is generated.

[0051] Returning to Fig. 1, the click rate prediction unit 14 predicts a click rate when the generated advertising banner candidate is used. In this embodiment, the click rate prediction unit 14 includes a machine-learned learner that uses a combination of feature vectors of past advertising banners and click rates of the advertising banners as training data, and derives a click rate prediction value corresponding to the feature vector of the advertising banner candidate with the learner. Here, the click rate is CTR (Click Through Rate), but may be another similar index.

[0052] Here, the feature vector of the past advertising banner includes, for example, a feature vector that combines the feature vector of the past advertising banner itself (image) and the feature vectors of one or more entities (such as a foreground image or natural language text) detected in the past advertising banner.

[0053] The feature vector of the advertising banner candidate, like the feature vectors of past advertising banners, may include, for example, a feature vector that combines the feature vector of the advertising banner candidate (image) and the feature vectors of one or more entities (such as a foreground image or natural language text) detected in the advertising banner candidate.

[0054] These feature vectors are generated, for example, by a CLIP (Contrastive Language-Image Pre-training) model.

[0055] Furthermore, in this embodiment, the click rate prediction unit 14 is equipped with a machine-learned learner that uses the feature vectors of past advertising banners and combinations of campaign settings of those past advertising banners and the click rates of those advertising banners as training data, and derives a click rate prediction value corresponding to the feature vectors of candidate advertising banners and the campaign settings of those candidate advertising banners using the learner.

[0056] Here, the campaign settings are campaign settings for web advertising, etc. For example, the values ​​(settings) of the setting items (bid price, budget, region, schedule such as advertisement display time) of Google Ads campaign settings are used as the campaign settings. In addition, the input of the training data and the learning device may include layout settings (size and position of the foreground image and advertisement text). The campaign settings and layout settings are normalized as appropriate before use.

[0057] The advertisement banner selection unit 15 selects an advertisement banner to be used for a web advertisement from among the advertisement banner candidates based on the generated click rate predicted value. For example, an advertisement banner candidate whose click rate predicted value exceeds a predetermined value is selected as the advertisement banner. Alternatively, for example, a predetermined number of advertisement banner candidates in descending order of click rate predicted value are selected as the advertisement banner. The advertisement banner candidates and the selected advertisement banner are displayed on, for example, the display device 3a.

[0058] Note that a machine learning processing unit 16 may be provided to perform machine learning for each of the above-mentioned learning devices.

[0059] Next, the operation of the above system will be described.

[0060] First, as described above, foreground image generating unit 11 generates a plurality of foreground images, and advertising text generating unit 12 generates a plurality of advertising texts.

[0061] Next, the advertising banner candidate generation unit 13 combines each of the generated foreground images and each of the generated advertising texts, and applies one or more banner settings (layout settings, font settings, color settings, etc.) to each combination to generate multiple advertising banner candidates (images).

[0062] Then, the click rate prediction unit 14 derives a click rate predicted value for each of the generated plurality of advertising banner candidates in the manner described above, and the advertising banner selection unit 15 selects an advertising banner from the generated plurality of advertising banner candidates based on the click rate predicted value, and displays the selected advertising banner on the display device 3a.

[0063] As described above, according to the first embodiment, foreground image generation unit 11 generates a foreground image to be used for the advertising banner candidate, advertising text generation unit 12 generates advertising text to be used for the advertising banner candidate, and advertising banner candidate generation unit 13 generates advertising banner candidates by combining the foreground image generated by foreground image generation unit 11 and the advertising text generated by advertising text generation unit 12. Then, click rate prediction unit 14 predicts the click rate when the generated advertising banner candidate is used.

[0064] In this way, an advertisement banner with a good click rate is generated in advance without actually using the advertisement banner in an advertisement, so that an advertisement banner with a good advertisement effect is generated in a short time. Also, since an advertisement banner with a good advertisement effect is generated without actually using the advertisement banner in an advertisement, costs for a test advertisement are not required, and the cost for obtaining an advertisement banner with a good advertisement effect is reduced.

[0065] Embodiment 2

[0066] The click rate prediction device of embodiment 2 is a device that predicts a click rate when an advertising banner candidate is used, and includes a banner feature vector generation unit that generates a feature vector of the above-mentioned advertising banner candidate, and a click rate prediction unit that predicts a click rate based on the feature vector of the advertising banner candidate.

[0067] The banner characteristic vector generating unit in the second embodiment generates characteristic vectors of advertisement banner candidates in the same manner as the banner candidate generating unit 13 in the first embodiment.

[0068] Moreover, the click rate prediction unit in the second embodiment is the same as the click rate prediction unit 14 in the first embodiment.

[0069] That is, in the second embodiment, it is possible to predict click rates not only for advertising banner candidates generated in the manner as in the first embodiment, but also for advertising banner candidates generated by other generation methods.

[0070] Other configurations (such as hardware configuration) of the click rate prediction device according to the second embodiment are similar to those of the first embodiment, so a description thereof will be omitted.

[0071] As described above, according to the second embodiment, an advertising banner with a good click rate can be extracted in advance without actually using the advertising banner in an advertisement, so that an advertising banner with a good advertising effect can be obtained in a short period of time.

[0072] It should be noted that various changes and modifications to the above-described embodiments will be apparent to those skilled in the art. Such changes and modifications may be made without departing from the spirit and scope of the subject matter and without diminishing its intended advantages. In other words, such changes and modifications are intended to be included within the scope of the claims. [Industrial Applicability]

[0073] The present invention is applicable, for example, to the automatic generation of advertising banners for use in web advertising. [Explanation of symbols]

[0074] 11 Foreground image generation unit 12 Advertising text generation section 13 Advertising banner candidate generation unit 14 Click rate prediction section

Claims

1. a foreground image generating unit that generates a foreground image to be used for an advertisement banner candidate; an advertisement text generating unit that generates advertisement text to be used for the advertisement banner candidate; an advertisement banner candidate generation unit that generates the advertisement banner candidate by combining the foreground image generated by the foreground image generation unit and the advertisement text generated by the advertisement text generation unit; a click rate prediction unit that predicts a click rate when the generated advertisement banner candidate is used; An advertising banner generating system comprising:

2. The advertising banner generation system of claim 1, characterized in that the foreground image generation unit (a) derives a similarity between a feature vector of a past advertising banner and a feature vector of an existing stock image, and (b) selects the foreground image from the existing stock image based on the similarity.

3. 2. The advertising banner generating system according to claim 1, wherein the foreground image generating unit automatically generates the foreground image from words related to an advertising target of the advertising banner candidate.

4. 2. The advertising banner generation system according to claim 1, wherein the advertising text generation unit (a) automatically generates advertising text candidates from words related to the advertising target of the advertising banner candidate, (b) derives a similarity between a feature vector of text information obtained from a past advertising banner and a feature vector of the advertising text candidate, and (b) selects the advertising text from the advertising text candidates based on the similarity.

5. the advertisement banner candidate generation unit generates the advertisement banner candidate by combining the foreground image and the advertisement text in accordance with one or more banner setting data; The banner setting data includes a layout setting, a font setting, and a color setting, the layout settings indicate the respective positions and sizes of the foreground image and the advertisement text; the font setting indicates a typeface for the advertisement text; the color setting indicates at least a color of the advertisement text and a background color; 2. The advertising banner generating system according to claim 1,

6. 6. The advertising banner generation system according to claim 1, wherein the click rate prediction unit includes a machine-learned learner that uses a combination of feature vectors of past advertising banners and click rates of the advertising banners as training data, and the learner derives a click rate prediction value corresponding to the feature vector of the advertising banner candidate.

7. A click rate prediction device for predicting a click rate when an advertisement banner candidate is used, comprising: a banner feature vector generation unit for generating a feature vector of the advertisement banner candidate; a click rate prediction unit that predicts the click rate based on a feature vector of the advertisement banner candidate, the click rate prediction unit includes a machine-learned learning device that uses a combination of feature vectors of past advertising banners and click rates of the advertising banners as training data, and derives a click rate prediction value corresponding to the feature vector of the advertising banner candidate using the learning device; A click rate prediction device comprising:

8. The click rate prediction device according to claim 7, characterized in that the click rate prediction unit includes a machine-learned learner that uses feature vectors of past advertising banners and combinations of campaign settings of the past advertising banners and click rates of the advertising banners as training data, and derives a click rate prediction value corresponding to the feature vectors of the advertising banner candidates and the campaign settings of the advertising banner candidates using the learner.

9. an image generation step for generating a foreground image for use in a candidate advertising banner; a text generating step of generating advertisement text to be used for the advertisement banner candidate; an advertisement banner candidate generating step of generating the advertisement banner candidate by combining the foreground image generated in the image generating step and the advertisement text generated in the text generating step; a click rate prediction step of predicting a click rate when the generated advertisement banner candidate is used; 13. An advertising banner generating method comprising:

Images