JP2018077615A - Advertising image generation device, advertising image generation method and program for advertising image generation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To dynamically generate an advertising image of which the advertising effect has a high expected value.SOLUTION: An advertising effect database 51 for storing advertising effects of advertising images is provided in a storage unit, and an advertising image generation device 1 for dynamically generating an advertising image acquires a generation parameter group 101, generates an advertising image 102 from the generation parameter group 101, acquires an advertising effect 104 of the advertising image 102, and stores the generated advertising image 102, at least one of representations as a feature quantity of the advertising image, and the acquired advertising effect 104 in the advertising effect database 51 in association with each other.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an advertisement image generation apparatus, an advertisement image generation method, and an advertisement image generation apparatus program that generate an advertisement image based on parameters.

  Today, along with the development and popularization of information technology, it is possible to analyze a huge amount of data, and the sophistication in the use of analysis results is also progressing. In particular, a field called machine learning technology has seen remarkable improvement in accuracy in identification and regression due to the discovery of new calculation models, and there are increasing opportunities for use not only in the research field but also in the real world.

  As an example, there are an increasing number of scenes where tasks that have been performed by humans in the past are replaced by machine learning techniques. Initially, the focus was on reducing labor costs and standardizing standards for rather simple tasks, but nowadays, the call center handles customers on behalf of people, and automatically Businesses and research are being carried out that not only make humans more efficient than humans, but also efficiently perform complex tasks such as driving cars.

  In recent years, in particular, in the field of advertisement distribution, digital advertisement distribution has been activated in an advanced advertisement distribution network called an ad network as the use time of electronic terminals of general users has increased. In the ad network, since the numerical values related to the advertisement can be measured at high speed and with high accuracy, further improvement in the efficiency of advertisement distribution is required every day.

  In response to such a problem, an information distribution device that improves the estimation accuracy of the click rate and can appropriately select an advertisement to be distributed has been disclosed.

Japanese Patent Publication No. 2006-118918

  According to Patent Document 1, a learning device that calculates a feature value from an image is used to calculate a feature value of an image having a history selected by the user, and the calculated feature value and the user who has selected the image. By estimating the attribute information of the user who selected the image from the feature amount of the image based on the attribute information indicating the attribute, it is possible to improve the estimation accuracy of the click rate and appropriately select the advertisement to be distributed. An information distribution apparatus is disclosed.

  However, in the configuration disclosed in Patent Document 1, all of the contents of the advertisement to be distributed must be created manually, and if new advertisement contents are created, is it really improved? There remains a problem that the index of how much improvement is impossible to measure quantitatively unless it is actually distributed.

  Moreover, it is not the advertisement placement side but the platformer side that is the distributor that can effectively use the configuration disclosed in Patent Document 1. For users who place advertisements, higher advertising effectiveness is possible. The problem remains whether or not it depends on the algorithm on the platform side with the advertising space.

  Therefore, the inventor of the present invention pays attention to the fact that the design activity that has been performed by humans can be realized at low cost and at high speed using a computer by expressing the advertisement image with parameters and generating the advertisement image from the parameters. did.

  In addition, the inventor of the present invention performs optimal learning by performing learning using machine learning or the like in analyzing the relationship between the image or the feature amount of the image and the main KPI including the click rate. Focusing on the ability to dynamically calculate parameters and generate advertising images that use them to create a system that automatically replaces advertising images with more effective ones simply by continuing to deliver ads. .

  The present invention expresses an advertisement image with a generation parameter, generates an advertisement image from the generation parameter, associates the advertisement image with the advertisement effect when the advertisement image is published, and constructs an estimation model of the advertisement effect on the advertisement image, thereby An advertisement image generation apparatus, an advertisement image generation method, and an advertisement image generation apparatus program for generating an advertisement image with generation parameters optimized for the effect are provided.

  The present invention provides the following solutions.

The invention according to the first aspect is an advertisement image generation device that includes an advertisement effect database for storing the advertisement effect of an advertisement image in a storage unit, and dynamically generates an advertisement image,
Generation parameter group acquisition means for acquiring a generation parameter group;
Advertising image generating means for generating an advertising image from the generation parameter group;
Advertising effect acquisition means for acquiring the advertising effect of the advertising image;
An advertisement effect storage means for associating and storing the generated advertisement image, at least one of the advertisement image represented as a feature quantity, and the acquired advertisement effect in an advertisement effect database;
An advertisement image generating device is provided.

  According to the first aspect of the invention, the advertisement image generation device that dynamically includes the advertisement image generation apparatus that includes the advertisement effect database that stores the advertisement effect of the advertisement image in the storage unit, acquires the generation parameter group, and An advertisement image is generated from the generation parameter group, the advertisement effect of the advertisement image is acquired, and the generated advertisement image is associated with at least one of the advertisement images represented as a feature quantity and the acquired advertisement effect. And store it in the advertising effectiveness database.

  The invention according to the first feature is the category of the advertisement image generation device, but the same effect is obtained according to the category also in the category of the advertisement image generation method and the program for the advertisement image generation device.

The invention according to the second aspect is characterized in that an advertisement image placement means for electronically placing the generated advertisement image as an advertisement,
An advertisement image generation device which is an invention according to a first feature is provided.

  According to the invention relating to the second feature, the advertisement image generating device which is the invention relating to the first feature electronically places the generated advertisement image as an advertisement.

The invention according to the third aspect includes an advertisement image acquisition means for acquiring one or more advertisement images;
Advertising effect estimation means for estimating the advertising effect of the acquired advertising image by performing a calculation based on the advertising effect stored in the advertising effect database;
An advertisement image generation device is provided which is an invention according to the first or second feature.

  According to the third aspect of the invention, the advertisement image generating device according to the first or second aspect of the invention acquires one or more advertisement images and uses the advertisement effect stored in the advertisement effect database as a basis. By performing the calculation, the advertising effect of the acquired advertising image is estimated.

  The invention according to the fourth feature is the advertisement image according to the third feature, wherein the advertisement image acquired by the advertisement image acquisition means is an advertisement image generated by the advertisement image generation means A generating device is provided.

  According to the fourth aspect of the invention, in the advertisement image generation device according to the third aspect of the invention, the acquired advertisement image is the generated advertisement image.

The invention according to a fifth feature uses an advertisement effect estimation model learning means for learning an advertisement effect estimation model, which is a machine learning model provided in a storage unit, using the acquired advertisement effect;
The advertisement effect is estimated according to the advertisement effect estimation model, and the advertisement image generation device according to the third or fourth feature is provided.

  According to the fifth aspect of the invention, the advertisement image generation device according to the third or fourth aspect of the invention uses the acquired advertisement effect to provide an advertisement effect that is a machine learning model provided in a storage unit. The estimation model is learned, and the advertisement effect is estimated by the advertisement effect estimation model.

The invention according to a sixth aspect includes a generation parameter group in which the estimated value of the advertisement effect of the generated advertisement image is maximized by performing a calculation based on the advertisement effect stored in the advertisement effect database. A generation parameter group optimizing means to be obtained by calculation;
An advertisement image generating apparatus is provided which is an invention according to any one of the third to fifth features.

  According to the sixth aspect of the invention, the advertisement image generation device according to any one of the third to fifth aspects of the present invention performs a calculation based on the advertisement effect stored in the advertisement effect database. A generation parameter group is calculated such that the estimated value of the advertising effect of the generated advertisement image is the highest value.

  The invention according to a seventh aspect includes time-series information in the acquired advertisement effect, and the estimation of the advertisement effect estimates an advertisement effect along a time series, and In calculating the generation parameter group that has the highest estimated value of advertising effectiveness, it is necessary to calculate one or more generation parameter groups together with the combination of the publication times that maximizes the total of the estimated value of advertising effectiveness. An advertisement image generation device which is an invention according to a sixth feature is provided.

  According to the seventh aspect of the invention, the advertisement image generating device according to the sixth aspect of the invention includes time-series information in the acquired advertisement effect, and the estimation of the advertisement effect is in time series In the calculation of the generation parameter group that estimates the advertising effect and the estimation value of the advertising effect of the generated advertisement image is the highest value, one or more generation parameter groups are set as the estimation value of the advertising effect. It is calculated with a combination of the publication times that maximizes the sum.

  In the invention according to an eighth feature, the generation parameter group includes at least one of content and position of an image arranged on the advertisement image, character information arranged on the advertisement image, and color arrangement information. There is provided an advertisement image generating device which is an invention according to any one of the first to seventh features.

  According to the eighth aspect of the invention, in the advertisement image generation device according to any one of the first to seventh aspects, the generation parameter group includes the content of the image arranged on the advertisement image, and It includes at least one of a position, character information arranged on the advertisement image, and color arrangement information.

The invention according to a ninth aspect includes an advertising propensity factor group input receiving unit that receives an input of an advertising propensity factor group that is a set of information indicating the propensity of an advertisement from a user of the advertising image generation device,
Advertising propensity factor group conversion means for converting the advertising propensity factor group that has received the input into a part or all of the generation parameter group;
An advertisement image generating device which is an invention according to any one of the first to eighth features is provided.

  According to the ninth aspect of the invention, the advertisement image generation device according to any one of the first to eighth aspects is provided by the user of the advertisement image generation device with information indicating the propensity of the advertisement. An input of an advertising propensity factor group that is a set is received, and the advertising propensity factor group that has received the input is converted into a part or all of the generation parameter group.

  According to a tenth aspect of the invention, there is provided the advertising image generating apparatus according to the ninth aspect, wherein the advertising propensity factor group includes at least an industry related to the advertising image and an advertising purpose. provide.

  According to the tenth aspect of the invention, in the advertisement image generating device according to the ninth aspect of the invention, the advertisement propensity factor group includes at least the industry related to the advertisement image and the purpose of the advertisement.

The invention according to an eleventh aspect is an advertisement image generation method executed by an advertisement image generation device that includes an advertisement effect database that stores advertisement effects of advertisement images in a storage unit and dynamically generates an advertisement image,
Obtaining a generation parameter group;
Generating an advertisement image from the generation parameter group;
Obtaining an advertising effect of the published advertising image;
Storing the generated advertisement image, at least one of the advertisement image represented as a feature quantity, and the acquired advertisement effect in association with each other in an advertisement effect database;
An advertisement image generation method is provided.

The invention according to a twelfth aspect includes an advertisement effect database that stores an advertisement effect of an advertisement image in a storage unit, and an advertisement image generation device that dynamically generates an advertisement image,
Obtaining generation parameters,
Generating an advertisement image from the generation parameter group;
Obtaining an advertising effect of the published advertising image;
Storing the generated advertisement image, at least one of the advertisement image represented as a feature quantity, and the acquired advertisement effect in association with the advertisement effect database;
A program for an advertisement image generating device is provided.

  According to the present invention, an advertisement image is expressed by a generation parameter, an advertisement image is generated from the generation parameter, and is associated with the advertisement effect when the advertisement image is put out, and an advertisement effect estimation model for the advertisement image is constructed. It is possible to provide an advertisement image generation apparatus, an advertisement image generation method, and an advertisement image generation apparatus program that generate an advertisement image with a generation parameter optimized for the effect.

FIG. 1 is a diagram showing an outline of the advertisement image generating apparatus 1. FIG. 2 is a schematic diagram of a general neural network. FIG. 3 is a schematic diagram illustrating an example of a convolutional neural network. FIG. 4 is a functional block diagram of the advertisement image generating device 1. FIG. 5 is a flowchart of the advertisement image generation process executed by the advertisement image generation apparatus 1. FIG. 6 is a flowchart of the generation parameter group acquisition process executed by the advertisement image generation apparatus 1. FIG. 7 is a diagram illustrating an example of a generation parameter group and an advertisement image output based on the generation parameter group.

  Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings. This is merely an example, and the technical scope of the present invention is not limited to this.

[Outline of Advertising Image Generating Device 1]
FIG. 1 is a diagram showing an outline of the advertisement image generating apparatus 1.

  First, the outline of the advertisement image generating apparatus 1 will be described. In FIG. 1, the advertisement image generation device 1 acquires a generation parameter group 101, generates an advertisement image 102, places an advertisement in an advertisement frame 103, acquires an advertisement effect 104, records it in the advertisement effect database 51, The advertisement effect estimation model 52 is updated using the data, and the generation parameter group 101 is newly acquired again. Each step will be described in detail below.

  The advertisement image generation device 1 acquires the generation parameter group 101 (step S01). Here, “acquisition” refers to accepting an input from a user who is a user of this apparatus, or dynamically generating it in the advertisement image generating apparatus 1. For example, it is conceivable that an input is received from the user as an initial value for a part of the generation parameter group 101 and the remaining parameters are generated from the part.

  Next, the advertisement image generation device 1 generates the advertisement image 102 from the acquired generation parameter group 101 (step S02). In the present embodiment, the advertisement image constitutes the advertisement image by arranging each element in a predetermined layout. The generation parameter group 101 is referred to when selecting or generating the element, and determines the element.

  Note that the advertisement image generation device 1 does not need to be able to generate all possible advertisement images, and may limit the advertisement images that can be theoretically generated by the structure of the generation parameter group 101. For example, the layout of the elements on the advertisement image may be selected from predetermined ones given in advance, or the background image or product image may be selected from predetermined ones given in advance or those uploaded by the user .

  Next, the advertisement image generation device 1 places the generated advertisement image (step S03). Regardless of this place, it may be, for example, a leaflet or a signboard, but by optimizing the advertisement image described later, the advertisement image can be posted more quickly by placing it in an advertising space that can be posted at high speed and measuring the effect reliably. The effect of the present invention can be enhanced. For example, a system called an ad network completes all procedures from posting to obtaining advertisement effects for an advertising space 103 such as a personal site or a plurality of blogs by electronic procedures. It is assumed that such electronic placement is performed.

  Next, the advertisement image generating apparatus 1 acquires the advertisement effect 104 of the advertisement image 102 that has been put out (step S04). The advertising effect 104 mainly includes the posting cost, PV (page view, ie, the number of times of display), CTR (percentage of clicks, ie, the click rate, of the number of times the Internet advertisement is displayed), CVR (PV and user in the target period) It is an index that quantifies how effective the advertising image is, such as the ratio of the number of conversions to the number (ie, conversion rate).

  Next, the advertisement image generation device 1 stores the posted advertisement image 102 and the advertisement effect 104 in the advertisement effect database 51 in association with each other. Here, the advertisement image 102 may be stored as digital data, or may be expressed as a generation parameter group, or may be stored as a representation of the advertisement image as a feature amount by a machine learning model described later. It may be.

  Next, the advertisement image generation device 1 constructs an advertisement effect estimation model 52 that is a calculation model for estimating the advertisement effect from the combination of the advertisement image 102 and the advertisement effect 104 (step S05). Various algorithms can be used for estimating the advertisement effect. In this embodiment, it is assumed that a convolutional neural network that is highly effective in the area of image recognition is used.

  Next, using the advertisement effect estimation model 52, an optimal generation parameter group 101 that generates the maximum advertisement effect estimation value is generated (step S06). Here, a general optimization method can be used to generate the optimal generation parameter group 101. However, since the full search or the like is almost difficult due to the limitation of calculation time, the optimum value here does not necessarily need to be an exact solution and may be an approximate solution.

  Here, a convolutional neural network, which is a representative model suitable for this purpose, will be described as an example of learning in machine learning. Prior to that, a general neural network will be described first.

  FIG. 2 is a schematic diagram of a general neural network. The neural network can be said to be an operation model that adds a predetermined operation learned in advance to input data. Here, as an example, a neural network having a single intermediate layer is illustrated.

  The neural network includes an input layer such as node 201, an intermediate layer such as node 202, and an output layer such as node 203. The intermediate layer may be a multilayer, and a plurality of nodes may be interposed between the node 202 and the node 203. However, as the number of layers increases, the degree of freedom of mathematical expressions that can be expressed is improved, but the required time and the amount of data tend to increase.

  Edges 204 connecting nodes have weights independently. The weight of the edge 204 is multiplied by the numerical value input to the node 201 that is the starting point of the edge 204 and is passed to the node 202 that is the ending point. Since the node 202 is the end point of a plurality of edges, the sum of these values is passed as the final input to the node 202.

  For example, when the input 207 to the input layer is 2 and the weight of the edge 204 is 1.5, 3 is delivered to the node 202. Similarly, the nodes 205 and 206 located in the input layer also have independent inputs and edge weights, so the sum of the three numbers becomes the input to the node 202.

  In this way, the input to each node in the intermediate layer can be calculated. In addition, since the nodes of the intermediate layer and the nodes of the output layer are also connected by edges having independent weights, the input to each of the nodes of the output layer depends on the input to the intermediate layer and the weight of the edge. Will be calculated. This result is nothing but the output of the entire neural network.

  Here, since the number of layers and the number of nodes are given at the time of designing the neural network, it can be said that the act of learning is an act of optimizing the weight of each edge.

  Next, learning will be described. There are various learning methods depending on the machine learning model to be used, but a neural network generally uses a method called backpropagation. In learning, first, input data and an ideal value when the input data is input are prepared. This ideal value is called teacher data. The input data is sometimes called learning data.

  In the back propagation method, first, input data is input to a neural network to be learned, and an output at that time is obtained. Then, for each output node, the difference between the output and the ideal value is determined, and as the cause that caused the difference, the edge having the largest weight among the previous edges is determined, and the weight of the edge is determined as the ideal value. Adjust the direction so that it comes out. The steepest descent method is used for adjustment. When this adjustment is completed, the same adjustment is performed before that. By doing so, the weight adjustment is back-propagated from the output layer to the input layer.

  By repeating this adjustment on the teacher data, the weight of each edge is continuously adjusted so as to produce a value close to the ideal value, and finally a neural network suitable for the purpose is obtained. Note that learning of a neural network does not necessarily converge, so that the completion of learning is practically determined by a threshold value based on accuracy or the amount of teacher data.

  When constructing a machine learning model using an image as an input as in the present case, a convolutional neural network is often used instead of a simple neural network. A convolutional neural network has a structure imitating a human visual cortex, and is known to contribute to improving accuracy particularly in the field of image processing.

  That is, in a simple neural network, an image is handled as one-dimensional data, and thus a parallel moved or rotated image is often separated as an element. The convolutional neural network can handle the input as it is in two dimensions, and thus has an advantage that it easily absorbs the image deformation.

  FIG. 3 is a schematic diagram illustrating an example of a convolutional neural network. Although the basic principle of the convolutional neural network is the same as that of the neural network, it is different in that a process called “convolution” is included in the calculation.

  In FIG. 3, as an example of the input layer 301, an image expressed by vertical 250, horizontal 250, and RGB three colors is input. That is, the calculation is started for the input image as a numerical group of 300 × 250 × 3.

  A process called “convolution” is performed on the input image as process 302. The convolution functions like a filter for a region of n × n and calculates the output to the next layer. A filter is also called a kernel. For example, in FIG. 3, a total value obtained by multiplying each value in a 5 × 5 region by a parameter is taken. Thereby, the 250 × 250 region is converted into a 246 × 246 convolved region. In addition, by having a plurality of these filters, it is possible to convert to a feature quantity having a three-dimensional thickness. Further, the size of the filter need not be constant, and may be a plurality of types of sizes.

  Here, in the above-described neural network, the connection of each layer is updated by learning, but in the convolutional neural network, the parameter of this filter is updated. In addition, the convolutional neural network includes an intermediate layer called a pooling layer, and the layer is reduced in the pooling layer. In the intermediate layer 303, the width of the image which was originally 250 is 50, which is due to pooling. In addition to extracting arbitrary elements, pooling includes maximum pooling for extracting a maximum value from a certain area, average pooling for extracting an average value, and the like. In a convolutional neural network, it is said that it has invariance to translation, rotation, and scaling by pooling, and learning accuracy is improved in image recognition.

  The above convolution and pooling are repeated, and the input layer 301 is converted into a 4096-dimensional intermediate layer 306 through the intermediate layer 303, the intermediate layer 304, and the intermediate layer 305 by calculation. Here, the intermediate layer 306 can be regarded as a 4096-dimensional feature amount representing an input image when a network in which learning has sufficiently progressed is used.

  Note that the intermediate layer 306, the intermediate layer 307, and the output layer 308 may be fully coupled as in a normal neural network. In the present embodiment, for example, if the output layer 308 is learned as representing ten different advertising effectiveness indexes, a model that estimates the advertising effectiveness for the input advertising image can be learned. Thus, the number and meaning of the output layers 308 can be changed according to the purpose of learning.

  The above is the outline of the neural network and the convolutional neural network.

  Returning to FIG. 1, when the generation parameter group 101 is given by the above machine learning model, the advertisement image 102 can be generated, and the estimated value of the advertisement effect 104 can be obtained before publication. As a result, the advertisement image 102 can obtain the generation parameter group 101 having a higher estimated value of the advertisement effect without actually placing the advertisement, and the generation parameter group optimized in accordance with the current advertisement effect estimation model 52 101 will be generated again. By repeating this, data is collected, the estimation accuracy of the advertisement effect estimation model 52 is improved, and the advertisement image 102 is converged to those that are highly likely to increase the advertisement effect. In addition, since the advertisement image 102 is renewed repeatedly, a secondary effect that it is possible to cope with a decrease in the interest of the user due to the same advertisement image being displayed for a long period of time can be obtained.

  The above is the outline of the advertisement image generating device 1.

[System Configuration of Advertising Image Generating Device 1]
The advertisement image generating apparatus 1 is a household or business appliance having a function described later. The advertisement image generation device 1 is, for example, an information home appliance such as a smartphone, a tablet terminal, a netbook terminal, a slate terminal, an electronic book terminal, and a portable music player in addition to a personal computer, a server device, a mobile phone, and a portable information terminal. It's okay.

[Description of each function]
The configuration of each device will be described with reference to FIG.

  The advertisement image generation apparatus 1 includes a central processing unit (CPU), a random access memory (RAM), a read only memory (ROM), and the like as the control unit 11, and data and images controlled by the control unit as the input / output unit 12. In addition to a touch panel, a keyboard, a mouse, and the like that accept input from a user or supporter, an input / output device that captures an image captured by an external device is provided. In addition, the advertisement image generating apparatus 1 can be used as the communication unit 13, for example, a WiFi (Wireless Fidelity) compatible device compliant with IEEE 802.11, a wireless device compliant with the IMT-2000 standard such as a third generation mobile communication system, or the like. Equipped with a device that supports communication standards.

  In addition, the advertisement image generating apparatus 1 includes a data storage unit such as a hard disk, a semiconductor memory, a recording medium, or a memory card as the storage unit 14 that stores data and files. The advertisement image generation device 1 includes an advertisement effect database 51 and an advertisement effect estimation model 52 in the storage unit 14.

  In the advertisement image generation device 1, the control unit 11 reads a predetermined program to cooperate with the input / output unit 12 to generate a generation parameter group acquisition module 15, an advertisement image generation module 16, an advertisement image acquisition module 17, an advertisement tendency The factor group input receiving module 18 is realized. In the advertisement image generation device 1, the control unit 11 reads a predetermined program, thereby realizing the advertisement effect acquisition module 19 and the advertisement image placement module 20 in cooperation with the communication unit 13. Further, in the advertisement image generating apparatus 1, the control unit 11 reads a predetermined program, so that in cooperation with the storage unit 14, the advertisement effect storage module 21, the advertisement effect estimation module 22, the advertisement effect estimation model learning module 23, An advertisement propensity factor group conversion module 24 is realized.

[Advertising image output processing]
FIG. 5 is a flowchart of the advertisement image generation process executed by the advertisement image generation apparatus 1. The processing performed by the module of each device described above will be described together with this processing.

  First, the generation parameter group acquisition module 15 of the advertisement image generation device 1 executes a generation parameter group acquisition process to acquire a generation parameter.

[Generation parameter group acquisition processing]
FIG. 6 is a flowchart of the generation parameter group acquisition process executed by the advertisement image generation apparatus 1. The processing performed by the module of each device described above will be described together with this processing. In addition, acquisition of a value here means accepting input from the user, referring to a predetermined value, generating other values from some values, generating random values, etc. It may be a method or a combination of these methods.

  In addition, this process is a process of acquiring a generation parameter group when generating an advertisement, and one of the effects of the present invention is to generate an advertisement image and learn the advertisement effect when the advertisement image is distributed, Considering that the advertisement image is regenerated by estimating the advertisement effect more improved, there is no problem even if the generation parameter group is randomly given each time.

  However, in light of the usage in the real world, when a company or the like makes some kind of advertisement, it is not preferable to distribute a randomly generated image as it is in the initial state. Further, at each stage of generating the advertisement image, it is possible to promote the improvement of the advertisement effect by generating an image having as high an advertisement effect as possible. Therefore, the following describes a method for acquiring a generation parameter group that is expected to have a higher advertising effect by receiving input from the user regarding the purpose of the advertisement and the propensity of the advertisement desired by the user.

  First, the advertisement propensity factor group input receiving module 18 of the advertisement image generating apparatus 1 receives an input of the industry related to the advertisement and the purpose of the advertisement from the user as the user, or performs analysis by data analysis (step S21). ). As an analysis method, for example, by using a learning model designed for classification, the word input by the user and the page content described in the URL input by the user are acquired, and the industry and purpose are estimated. To do.

  Here, the industry related to the advertisement and the purpose of the advertisement are categorized and classified in advance for the industry related to the contents of the advertisement and the purpose of the advertisement distribution, respectively, when the user distributes the advertisement. It is a thing. Examples of the industry include EC sites, personal computers, wine, and qualifications, but are not limited to this, including the granularity of classification. Examples of the purpose include sale, introduction of e-mail registration, seminar application, request for materials, etc., but the classification granularity is not limited to this as in the industry. One of the reasons for accepting these inputs is that, as wine advertisements and bargain advertisements look completely different, advertisements can be determined by their industry and purpose, and the content can be greatly restricted. Because there is.

  Next, the advertising propensity factor group input receiving module 18 of the advertising image generating apparatus 1 receives an input of an advertising propensity factor group other than the industry and purpose from the user (step S22). Here, the advertising propensity factor group is a set of one or more factors representing the propensity of the advertisement, and refers to, for example, warmth, luxury, and masculine. As an advertising propensity factor group representing the opposite meaning, it is also possible to input things such as cool, cheap, feminine.

  Next, the advertising propensity factor group conversion module 24 of the advertising image generating device 1 converts the advertising propensity factor group that has received the input into a generation parameter group (step S23). Here, the generation parameter group is a parameter group that is referred to when generating an advertisement image, which will be described later, and functions to specify elements constituting the advertisement image. For example, the generation parameter group includes character color and gradation, font type, position, size, inclination, shadow, boundary line, presence / absence and size of embossing, and text content. In addition, the type, position, and size of the product image, the type and repetition of the background image, the overlapping order of these elements, and the like are included, but the present invention is not limited to this. The generation parameter group only needs to be able to generate an advertisement image based on the generation parameter group, and the advertisement image generated from one generation parameter group may not necessarily be specified as one.

  Note that the above-mentioned conversion rules for the advertising propensity factor group may be predetermined, for example, conversion rules such as changing the character color to red and orange if “warm” is input as the propensity. If “high quality” is input, a conversion rule may be set in advance such that the font of the character is Mincho and the background image of the sealing wax is used.

  To what degree of freedom these ad propensity factor groups can be input differ depending on the embodiment. For example, if the input propensity factors that can be input are limited and enumerated, it is sufficient to set conversion rules in advance for all possible combinations of advertising propensity factor groups. On the other hand, if the advertising propensity is expressed in a vector in a multidimensional space and a regression model is prepared that can estimate which vector expression the user has entered in free words, a wide range of advertising propensity factors It is possible to convert into a generation parameter group corresponding to the expression.

  Also, how far these advertising propensity factor groups can be specified may differ depending on the embodiment. As an example, a more specific advertising propensity factor such as “used a lot of red” may be designated. In this case, a conversion rule in which most of the generation parameters relating to color are fixed to red can be considered. In general, the more an abstract advertisement propensity factor can be interpreted, the lower the burden on the user's input and the larger the range of possible parameter groups, so that more various advertisement images can be generated.

  Finally, the generation parameter group acquisition module 15 of the advertisement image generation device 1 accepts or generates an input if there is an unacquired portion of the generation parameter group (step S24). For example, text to be displayed may be received from the user as an input, or a dictionary corresponding to the purpose of the advertisement may be referred to and text such as “big sale” or “free shipping” may be set as a parameter.

  Note that it is not necessary to completely eliminate randomness in the acquisition and conversion of the above parameters. This is because a variety of advertisement images are generated. Note that the randomness may be given in a step of generating an advertisement image described later.

  The above is the processing procedure of the generation parameter group acquisition processing. Returning to the advertisement image generation process, the advertisement image generation module 16 of the advertisement image generation apparatus 1 generates an advertisement image from the acquired generation parameter group (step S12). Specifically, an image or font specified by the generation parameter group is read, and each element is arranged or a character is drawn.

  FIG. 7 is a diagram illustrating an example of a generation parameter group and an advertisement image output based on the generation parameter group. In FIG. 7, two generation parameter groups 701 and 705 are shown, and advertisement images 702 and 706 generated from them are shown.

  In FIG. 7, the generation parameter groups 701 and 705 are the same in the industry related to the advertisement, but are different depending on the purpose of the advertisement and other propensity for advertisement. In the generation parameter group 701, “cheap / warmth” is designated as the advertising tendency, so the characters 703 are friendly fonts and have an impression that the bargain feeling is emphasized. On the other hand, in the generation parameter group 705, since “advanced / cool” is designated as the advertising tendency, the character 707 is a sophisticated font and is an expression that makes the brand feel conscious. The product images 704 and 708 are the same, and the overall layout is the same, but the advertisement images 702 and 706 give different impressions. As a result, the click rate and the purchase rate of the user after clicking are also affected. It is thought that the advertising effect is also different.

  Next, the advertisement image placement module 20 of the advertisement image generation device 1 places the generated advertisement image (step S13). In so-called ad networks, there are many things that can be put out electronically by sending information such as images and advertising fees to an access point called an API. It is possible to publish without intervention. For example, the advertisement image generating device 1 may be connected to a direct mail distribution system and may be submitted for printing on a paper medium. Not limited.

  Next, the advertisement effect acquisition module 19 of the advertisement image generation apparatus 1 acquires what advertisement effect has been delivered and the advertisement effect that has been posted (step S14). The advertising effect includes, but is not limited to, the number of advertisements distributed, the number of distributions, the click rate for the number of unique users, and the conversion rate. If it is an ad network, these numerical values can be obtained with high accuracy in real time. In the case of a physical medium such as a direct mail or a signboard, it is difficult to accurately measure only the value related to the advertisement image, but it may be connected to these management systems to acquire the advertisement effect.

  Next, the advertisement effect storage module 21 of the advertisement image generation device 1 stores the advertisement effect acquired in the advertisement effect database 51 provided in the storage unit in association with the advertisement image (step S15). Here, the advertisement image stored in the advertisement effect database 51 is in the form of an image itself, in the form of a generation parameter group, or in the form of a feature amount obtained by converting the image by the above-described convolution neural network or the like. Also good.

  Next, the advertisement effect estimation model learning module 23 of the advertisement image generating device 1 learns the advertisement effect estimation model 52 provided in the storage unit using the advertisement image and the advertisement effect described above (step S16). Here, the estimation model is not limited as long as it is a model that can estimate an advertisement effect as an output by using an advertisement image as an input. One example is a convolutional neural network model called so-called deep learning, and it is considered that the estimation accuracy is further improved by giving a combination of a known advertisement image and advertisement effect as teacher data.

  The estimation model may be a combination of a plurality of models. For example, a convolutional neural network is used for a part that converts an advertisement image into a feature quantity, and SVM or linear regression is used for the estimation part of the feature quantity and advertisement effect. Learning may be performed with only some models using other models. In addition, the learning may be performed by learning the model from scratch each time, or so-called online learning in which only the difference is learned.

  In addition, the estimation model is designed as a recurrent neural network that supports time series, and the advertisement effect of the advertisement image is estimated as a change in time series by acquiring, inputting, and learning the advertisement effect in time series. Is possible. In general, the effect of advertising tends to decrease with long-term distribution.Therefore, it is better to prepare multiple advertising images in advance and submit them at the appropriate timing, rather than continue to distribute a single advertising image. It is possible to increase the advertising effect.

  Note that the estimation of the advertisement effect of such an advertisement image is not limited to the advertisement image generated by the advertisement image generation module 16, and an advertisement image created by another process is acquired by the advertisement image acquisition module 17. The effect may be estimated.

  Next, the advertisement image generating device 1 determines whether or not the advertisement image has been updated (step S17). As the update frequency, a predetermined value is given in advance, the user directly inputs it, it is determined by judging whether the latest advertisement effect is below a certain value, or calculated by the above time series analysis You may refer to the timing. When the advertisement image is not updated (step S17: “NO”), the advertisement image generation device 1 stops until the next advertisement image update, and the advertisement image generation process ends. On the other hand, when the advertisement image is updated (step S17: “YES”), the generation parameter group is subsequently optimized (step S18).

  In the optimization of the generation parameter group, the advertisement effect estimation module 23 of the advertisement image generation apparatus 1 uses the above-described advertisement effect estimation model 52 to acquire a generation parameter group that has a higher advertisement effect estimation value ( Step S18). Note that the optimization here is not necessarily an exact solution, but may be an approximate solution, as is generally the case with optimization performed.

  As a specific method of optimization, for example, by randomly generating parameters of a generation parameter group that can be randomly generated, and repeatedly estimating the advertising effect of the generated advertisement image, A method can be used in which an advertisement image with a high estimated value of advertisement effect is stored, and the advertisement image is regarded as an optimum value when the update has not been performed more than a certain number of times.

  Another example is the use of a machine learning model called a generation model. The generation model is a network model that decodes the input again. As an example of the generation model, DCGAN (Deep Convolutional General Networks) has been announced. Here, by using the generation model, it is possible to generate an advertisement image as an original input using the advertisement effect as a new input. In DCGAN, there is a problem that an output image is blurred at present, but in this embodiment, an advertisement image is not generated directly, but a generation parameter group is generated. By interpreting the generation parameter group, an unblurred advertisement image can be generated.

  The above is the processing procedure of the advertisement image generation process.

  The means and functions described above are realized by a computer (including a CPU, an information processing apparatus, and various terminals) reading and executing a predetermined program. The program is provided in a form recorded on a computer-readable recording medium such as a flexible disk, CD (CD-ROM, etc.), DVD (DVD-ROM, DVD-RAM, etc.), for example. In this case, the computer reads the program from the recording medium, transfers it to the internal storage device or the external storage device, stores it, and executes it. The program may be recorded in advance in a storage device (recording medium) such as a magnetic disk, an optical disk, or a magneto-optical disk, and provided from the storage device to a computer via a communication line.

  As mentioned above, although embodiment of this invention was described, this invention is not limited to these embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

  1 advertising image generation device, 51 advertising effect database, 52 advertising effect estimation model

Claims (12)

An advertisement image generation apparatus that includes an advertisement effect database that stores advertisement effects of advertisement images in a storage unit and dynamically generates advertisement images,
Generation parameter group acquisition means for acquiring a generation parameter group;
Advertising image generating means for generating an advertising image from the generation parameter group;
Advertising effect acquisition means for acquiring the advertising effect of the advertising image;
An advertisement effect storage means for associating and storing the generated advertisement image, at least one of the advertisement image represented as a feature quantity, and the acquired advertisement effect in an advertisement effect database;
An advertisement image generating apparatus comprising: Advertising image placement means for electronically placing the generated advertisement image as an advertisement;
The advertisement image generating apparatus according to claim 1, further comprising: Advertising image acquisition means for acquiring one or more advertising images;
Advertising effect estimation means for estimating the advertising effect of the acquired advertising image by performing a calculation based on the advertising effect stored in the advertising effect database;
The advertisement image generation device according to claim 1, further comprising:   The advertisement image generating apparatus according to claim 3, wherein the advertisement image acquired by the advertisement image acquiring unit is an advertisement image generated by the advertisement image generating unit. Using the acquired advertisement effect, an advertisement effect estimation model learning unit that learns an advertisement effect estimation model that is a machine learning model provided in the storage unit;
The advertisement image generation apparatus according to claim 3, wherein the advertisement effect is estimated using the advertisement effect estimation model. A generation parameter group obtained by calculating a generation parameter group in which the estimated value of the advertisement effect of the generated advertisement image is the highest value by performing a calculation based on the advertisement effect stored in the advertisement effect database. Optimization means;
The advertisement image generating apparatus according to claim 3, comprising:   The acquired advertisement effect includes time series information, and the estimation of the advertisement effect is to estimate the advertisement effect along the time series, and the estimated value of the advertisement effect of the generated advertisement image is a maximum value. The calculation of the generation parameter group as described above, wherein one or more generation parameter groups are calculated together with a combination of the publication times that maximizes the sum of the estimated value of the advertisement effect. Advertising image generation device.   The generation parameter group includes at least one of content and position of an image arranged on the advertisement image, character information arranged on the advertisement image, and color arrangement information. The advertising image generation device according to any one of claims 7 to 9. An advertising propensity factor group input receiving means for receiving an input of an advertising propensity factor group, which is a set of information indicating the propensity of the advertisement, from a user of the advertising image generating device;
Advertising propensity factor group conversion means for converting the advertising propensity factor group that has received the input into a part or all of the generation parameter group;
The advertisement image generating apparatus according to claim 1, further comprising:   The advertisement image generation apparatus according to claim 9, wherein the advertisement propensity factor group includes at least an industry related to the advertisement image and an advertisement purpose. An advertisement image generation method executed by an advertisement image generation apparatus that includes an advertisement effect database that stores an advertisement effect of an advertisement image in a storage unit and dynamically generates an advertisement image,
Obtaining a generation parameter group;
Generating an advertisement image from the generation parameter group;
Obtaining an advertising effect of the published advertising image;
Storing the generated advertisement image, at least one of the advertisement image represented as a feature quantity, and the acquired advertisement effect in association with each other in an advertisement effect database;
An advertisement image generation method comprising: An advertisement image generating apparatus that dynamically stores an advertisement image, including an advertisement effect database that stores advertisement effects of advertisement images in a storage unit,
Obtaining generation parameters,
Generating an advertisement image from the generation parameter group;
Obtaining an advertising effect of the published advertising image;
Storing the generated advertisement image, at least one of the advertisement image represented as a feature quantity, and the acquired advertisement effect in association with the advertisement effect database;
The program for advertising image generation apparatuses characterized by performing these.

Images