JP6494679B2 - Analysis device, information providing device, analysis method, and program - Google Patents

Description

  The present invention relates to an analysis device, an information providing device, an analysis method, and a program.

  2. Description of the Related Art Conventionally, on an electronic page such as a web page published on the Internet, an image related to the content is posted along with content such as a news article. In general, there is a certain restriction on the size of an image posting area in an electronic page in view of the ease of viewing the electronic page and the access load. For this reason, an editor or the like exclusively cuts out a part of the image from the original image and places the cut out part of the image on the electronic page. In relation to this, a technique for automatically cutting out a part of an image from an original image is known (for example, see Patent Document 1).

Japanese Unexamined Patent Publication No. 2016-21671

  However, in the prior art, when the purpose is to increase the number of accesses to an electronic page, an automatically cut image is not necessarily an image suitable for achieving the purpose.

  The present invention has been made in view of the above problems, and an analysis device, an information providing device, an analysis method, and a program capable of automatically generating a more suitable image as an image to be posted on an electronic page are provided. It is intended to provide.

  One aspect of the present invention is an acquisition unit that acquires an evaluation value in which an action taken by a user who viewed an electronic page on which an cropped image cut out from an original image is posted; and the cropped image in the original image A determination unit that determines a preferable improvement parameter as the parameter by an optimization method of an unknown function based on a parameter related to the region from which the parameter is cut out and an evaluation value acquired by the acquisition unit .

  According to one embodiment of the present invention, an analysis device, an information providing device, an analysis method, and a program that can automatically generate a more suitable image as an image to be posted on an electronic page can be provided.

It is a figure which shows an example of the analysis system 1 containing the analysis apparatus 200 in 1st Embodiment. 4 is a diagram illustrating an example of an electronic page provided by a service providing apparatus 100. FIG. It is a figure which shows an example of a structure of the service provision apparatus 100 in 1st Embodiment. It is a figure which shows an example of the link origin page log information D1. It is a figure which shows an example of link destination page log information D2. It is a figure which shows an example of original image information D3. It is a figure which shows an example of the crop parameter information D4. It is a figure for demonstrating the cutting-out method of the cropping image CRP. It is a figure which shows an example of the evaluation information D5. It is a figure which shows an example of a structure of the analyzer 200 in 1st Embodiment. It is a flowchart which shows an example of the process performed by the analysis side control part 210 in 1st Embodiment. It is a figure which shows an example of parameter versus evaluation value information D6. It is a figure for demonstrating the determination method of the improvement parameter by a Bayesian optimization method. It is a figure for demonstrating the method of redeciding an improvement parameter. It is a figure which shows typically a mode that the modified link origin page of each generation is produced | generated. It is a figure which shows an example of the function information D7. It is a figure for demonstrating the clipping method of the cropping image CRP when a restriction | limiting is provided in the crop parameter used as the improvement parameter candidate. It is a figure for demonstrating the determination method of the improvement parameter when the restriction | limiting range of a parameter is provided. It is a figure for demonstrating the correction method of the confidence interval CB of the black box function F (X). It is a figure which shows an example of the hardware constitutions of the service provision apparatus 100 and analysis apparatus 200 of embodiment.

  Hereinafter, an analysis apparatus, an information providing apparatus, an analysis method, and a program to which the present invention is applied will be described with reference to the drawings.

[Overview]
The analysis device is realized by one or more processors. The analysis device acquires an evaluation value in which an action taken by a user who has viewed an electronic page on which a cropped image CRP cut out from the original image ORG is posted is evaluated. Then, the analysis apparatus preferably improves as a parameter of the region from which the cropped image CRP is to be cut out by an unknown function optimization method based on the parameter related to the region from which the cropped image CRP is cut out in the original image ORG and the acquired evaluation value. The parameter is determined and the determined improvement parameter is output. Thereby, a more suitable image can be automatically generated as an image to be posted on the electronic page by the own device or another device.

  The unknown function optimization method is a method for optimizing an unknown function (an unknown objective function) whose shape is not known in advance, such as a Bayesian optimization method, a genetic algorithm, a differential evolution method, an annealing method, For example, the grid search method. In the following embodiment, an example in which a Bayesian optimization method is applied as an optimization method for an unknown function will be described. However, other methods such as a genetic algorithm and a differential evolution method may be used. In this case, further, a function (distribution) according to a certain probability model based on parameters (search points) obtained in the process of search processing performed as an optimization technique using a distribution estimation algorithm (Estimation of Distribution Algorithms) or the like. You may ask for.

  As described above, the Bayesian optimization method is a method of optimizing an unknown function based on a posterior distribution of a function assuming a certain prior distribution when a certain unknown function is derived. In this embodiment, as an example, a Bayesian optimization method that assumes a prior distribution using a stochastic process (a set of random variables) called a Gaussian process will be described. A distribution may be assumed.

<First Embodiment>
[overall structure]
FIG. 1 is a diagram illustrating an example of an analysis system 1 including an analysis apparatus 200 according to the first embodiment. The analysis system 1 in the first embodiment includes, for example, one or more terminal devices 10, a service providing device 100, and an analysis device 200. These devices are connected via a network NW. Note that the service providing apparatus 100 may be integrated in the analysis apparatus 200. The service providing apparatus 100 is an example of an “information providing apparatus”.

  Each apparatus shown in FIG. 1 transmits and receives various information via the network NW. The network NW includes, for example, a radio base station, a Wi-Fi access point, a communication line, a provider, the Internet, and the like. Note that it is not necessary for all combinations of the devices shown in FIG. 1 to be able to communicate with each other, and the network NW may partially include a local network.

  The terminal device 10 is a device used by a user. The terminal device 10 is a computer device such as a mobile phone such as a smartphone, a tablet terminal, or a personal computer.

  Upon receiving a predetermined operation from the user, the terminal device 10 accesses a website provided by the service providing device 100 via a web browser. For example, the website provided by the service providing apparatus 100 may be an information providing site such as news or weather forecast, a search site, a shopping site, or the like, and enjoys SNS (Social Networking Service), mail service, and the like. It may be a possible website.

  Further, when receiving a predetermined operation from the user, the terminal device 10 communicates with the service providing device 100 via a preinstalled application, and acquires content to be displayed or reproduced on the application. The content is, for example, text data, image data, moving image data, audio data, and the like. Thereby, the terminal device 10 is provided with services similar to the services (information provision, search, shopping, etc.) provided by the various websites described above via the application.

  The service providing apparatus 100 may be a web server apparatus that provides a website such as a shopping site or a search site on the Internet, and communicates with the terminal device 10 on which the application is started to exchange various information. The application server apparatus to perform may be sufficient.

  Hereinafter, web pages or application pages provided as services by the service providing apparatus 100 will be collectively referred to as electronic pages.

  FIG. 2 is a diagram illustrating an example of an electronic page provided by the service providing apparatus 100. As shown in the figure, an electronic page introduces, for example, an article ATC such as news, a part of an original image ORG (a cropping image CRP described later) related to the contents of the article ATC, and detailed contents of the article ATC. The link (for example, hyperlink etc.) L to the electronic page where the article to be posted is posted. For example, when an article ATC posted on an electronic page is an article introducing a specific person such as a celebrity, the “original image ORG related to the content of the article ATC” is an entire image of the person (for example, head to toe). Image of the whole body). The part of the original image ORG is, for example, a cropped image CRP obtained by cutting out an area where only the upper body and the head are copied. Hereinafter, an electronic page accessible from the link L will be referred to as a “link destination page”, and an electronic page on which the article ATC, the cropping image CRP, and the link L are posted will be referred to as a “link source page”.

  The analysis apparatus 200 uses, for example, an improvement parameter preferable as a parameter (hereinafter referred to as a crop parameter) related to a region where the cropped image CRP posted on the link source page is cut out from the original image ORG, as unknown such as a Bayesian optimization method. Determined by function optimization method. “Preferred” means that it is recommended to change the current parameter to an improved parameter.

[Configuration of service providing device]
Hereinafter, each structure of the service provision apparatus 100 and the analysis apparatus 200 is demonstrated. FIG. 3 is a diagram illustrating an example of the configuration of the service providing apparatus 100 according to the first embodiment. As illustrated, the service providing apparatus 100 includes, for example, a service providing side communication unit 102, a service providing side control unit 110, and a service providing side storage unit 130.

  The service providing side communication unit 102 includes, for example, a communication interface such as a NIC (Network Interface Card) or a DMA (Direct Memory Access) controller. The service providing side communication unit 102 communicates with the terminal device 10 and the analysis device 200 via the network NW.

  The service providing side control unit 110 includes, for example, a service providing unit 112, an image processing unit 114, and an evaluation value deriving unit 116. The components of the service providing side control unit 110 are realized, for example, when a processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) executes a program stored in the service providing side storage unit 130. Also, some or all of the components of the service providing side control unit 110 are realized by hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), or FPGA (Field-Programmable Gate Array). Alternatively, it may be realized by cooperation of software and hardware.

  The service providing side storage unit 130 is realized by, for example, an HDD (Hard Disc Drive), a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), a RAM (Random Access Memory), or the like. The service providing storage unit 130 includes link source page log information D1, link destination page log information D2, original image information D3, crop parameter information D4, evaluation information D5, and the like in addition to various programs such as firmware and application programs. Remember. Such information will be described later.

  The service providing unit 112 provides an electronic page as a service to the terminal device 10 using the service providing side communication unit 102. For example, when the service providing device 100 is a web server device, the service providing unit 112 receives a HTTP (Hypertext Transfer Protocol) request from the terminal device 10 by the service providing side communication unit 102, and the web corresponding to the request. HTML page text data, style sheets such as CSS (Cascading Style Sheets), image data, moving image data, audio data, and the like are transmitted to the terminal device 10 that is the transmission source of the HTTP request. The text data includes, for example, an article ATC. Further, the image data includes, for example, a cropping image CRP. In response, a web page is drawn on the screen of the terminal device 10 by the function of the web browser.

  Further, for example, when the service providing apparatus 100 is an application server apparatus, the service providing unit 112 transmits a request when a predetermined request is received by the service providing side communication unit 102 from the terminal device 10 in which the application is activated. Text data including an article ATC and image data including a cropped image CRP are transmitted as contents to the original terminal device 10. When the terminal device 10 receives the content, the terminal device 10 arranges the received content on the screen based on, for example, a style sheet acquired at the time of application installation. As a result, the application page is drawn on the screen of the terminal device 10.

  Further, the service providing unit 112 collects an index value such as the number of impressions in a predetermined period of each page for the link source page on which the link L of the link destination page is posted, thereby obtaining the link source page log information D1. Generate. Impression means that, for example, there is access (visit) to the link source page, and the link source page is displayed on the screen of the terminal device 10 that is the access request source (for example, the transmission source of the HTTP request).

  FIG. 4 is a diagram illustrating an example of the link source page log information D1. As in the example shown in the figure, the link source page log information D1 includes, for each link source page, the ID of the content posted on the link source page and each link destination associated with the link L posted on the link source page. This is information in which the page ID of the page is associated with the number of impressions of the link source page. As illustrated, the content ID includes, for example, the ID of the article ATC and the ID of the cropped image CRP.

  Further, the service providing unit 112 generates link destination page log information D2 by collecting index values such as the number of accesses from the link source page, for example, for link destination pages accessible from the link L of the link source page. To do.

  FIG. 5 is a diagram illustrating an example of the link destination page log information D2. As in the illustrated example, the link destination page log information D2 is information in which the number of accesses from the link source page is associated with the page ID of each link destination page.

  The image processing unit 114 generates a cropped image CRP from the original image ORG related to the content of the article ATC posted on the link source page scheduled to be provided by the service providing unit 112. For example, the image processing unit 114 refers to the original image information D3 and specifies the original image ORG related to the content of the article ATC posted on the link source page scheduled to be provided by the service providing unit 112.

  FIG. 6 is a diagram illustrating an example of the original image information D3. As in the illustrated example, in the original image information D3, the image data of the original image ORG and the ID of the article ATC related to the content of the original image ORG are associated with the ID of the original image ORG. Information. It is assumed that such original image information D3 is stored in advance in the service providing side storage unit 130.

  For example, the image processing unit 114 specifies the image data of the original image ORG associated with the ID of the article ATC posted on the link source page as the original image ORG related to the content of the article ATC in the original image information D3. To do.

  Then, the image processing unit 114 refers to the crop parameter information D4 and determines an image region (hereinafter referred to as a crop window) to be cut out as the cropped image CRP in the specified original image ORG. The crop parameter information D4 is information indicating various crop parameters for determining a crop window for each original image ORG.

  FIG. 7 is a diagram illustrating an example of the crop parameter information D4. As illustrated, the crop parameter information D4 is information in which crop parameters such as the position of the crop window, the size of the crop window, and the aspect ratio of the crop window are associated with each ID of the original image ORG. . For example, the position of the crop window is represented by the coordinate in the width direction and the coordinate in the height direction of the original image ORG when the predetermined position of the original image ORG is the origin O and the origin O is the reference. The size of the crop window is represented by, for example, the size A of one side and the size B of the other side that touches or intersects the one side when the window has a substantially square shape. In addition, the aspect ratio of the crop window is represented by, for example, the ratio between the size A of one side described above and the size B of the other side. The image processing unit 114 sets a crop window in the original image ORG in accordance with such parameters, and cuts out an image area that overlaps the set crop window as a cropped image CRP. Such crop parameters are changed when the improvement parameter is determined by the analysis apparatus 200. It is assumed that a predetermined default parameter is set as the crop parameter until the improvement parameter is determined.

  FIG. 8 is a diagram for explaining a cropping image CRP clipping method. In the figure, X represents the width direction in the original image ORG, and Y represents the height direction in the original image ORG. In the figure, W represents a crop window, and CP represents the center position of the crop window. For example, when the ID of the identified original image ORG is “ORG_i”, the image processing unit 114 refers to the crop parameter information D4 and changes the size as appropriate according to the crop parameter associated with the original image ORG_i. A crop window W is set on the original image ORG_i. Then, the image processing unit 114 cuts out an image area overlapping the crop window W set on the original image ORG_i as a cropped image CRP_i corresponding to the original image ORG_i. Thereby, the cropped image CRP_i is generated.

  Note that the crop parameters include, for example, parameters for specifying the shape of the crop window (eg, an elliptical shape or a polygonal shape including a square shape) instead of or in addition to the above-described crop window position, size, and aspect ratio. It may be.

  Further, the crop parameter may include a parameter for designating the characteristics of the image area in which the crop window is set. For example, when the crop parameter includes a parameter that specifies the presence / absence of a feature derived from eyes, nose, mouth, and the like, the image processing unit 114 includes features such as eyes, nose, and mouth on the original image ORG. Set the crop window to overlap the image area. As a result, an image region including features such as eyes, nose, and mouth, that is, an image region that captures a human head is cut out as a cropped image CRP_i.

  When the link source page on which the cropped image CRP generated by the image processing unit 114 is provided is provided by the service providing unit 112, the evaluation value deriving unit 116 takes action taken by the user who viewed the link source page. The evaluated evaluation value is derived for each link source page.

  For example, the evaluation value deriving unit 116 refers to the link source page log information D1 and the link destination page log information D2, and indicates the number of accesses of a certain link destination page and the link on which the link L to the link destination page is posted. CTR (Click Through Rate) divided by the number of impressions of the original page is derived as an evaluation value.

  Note that the evaluation value deriving unit 116 instead of or in addition to deriving the CTR as the evaluation value, CVR (Conversion Rate) based on the stay time on the linked page or the number of established conversions on the linked page ) Or the like may be derived as an evaluation value. The predetermined conversion is, for example, browsing an advertisement placed on the linked page or accessing another page from a link placed on the linked page.

  Then, the evaluation value deriving unit 116 associates the derived evaluation value and the link source page as the evaluation value derivation target with the evaluation value associated with the cropped image CRP posted on the link source page. The generated information is stored in the service providing storage unit 130 as evaluation information D5.

  FIG. 9 is a diagram illustrating an example of the evaluation information D5. Like the example of illustration, evaluation information D5 is information with which ID of the cropping image CRP published on the page, and the evaluation value of the user who browsed the page were matched for every ID of the link origin page. is there. For example, different evaluation values may be associated with each link source page, or the same evaluation value may be associated with each link source page. Each link source page may be associated with a plurality of different evaluation values ((CTR, stay time, CVR,...) In the example in the figure).

  When the link source page associated with the evaluation value is modified in the evaluation information D5, the evaluation value deriving unit 116 may derive the evaluation value of the modified link source page again. “Modification” of the link source page means, for example, that the cropping image CRP posted on the link source page is changed.

  For example, a designer of a link source page such as a programmer or a designer refers to the evaluation value derived by the evaluation value deriving unit 116 so that the link source page can be more easily accessed from the link source page (evaluation). It is envisaged to change the cropped image CRP posted on the linked page (to improve the value).

  Further, “modification” of the link source page may be performed by any component (for example, the image processing unit 114) of the service providing side control unit 110 instead of being performed by the designer.

  For example, when the link source page is modified, the service providing unit 112 uses the service providing side communication unit 102 to display a modified link source page (hereinafter referred to as a modified link source page) by a predetermined user. It transmits to the terminal device 10 to operate. The predetermined user is, for example, a user who participates in a crowdsourcing working group composed of an unspecified number of users. Crowdsourcing means seeking cooperation for the requested work. For example, the service providing unit 112 provides the modified link source page to the terminal devices 10 of users who participate in crowdsourcing. In response to this, a user who participates in crowdsourcing operates the terminal device 10 to access the provided modified link source page, and if there is interest or interest, selects the link L to the link destination page. And access.

  Then, when the modified link source page is provided to the terminal device 10 by the service providing unit 112 and the link destination page is accessed by the user who operates the terminal device 10, the evaluation value deriving unit 116 The evaluation value of the original page is derived. For example, the evaluation value deriving unit 116 may update the evaluation value associated with the page ID of the modified link source page in the evaluation information D5, or change the page ID of the modified link source page to a new page. By treating it as an ID, a log of evaluation values may be added to the evaluation information D5 as a record for each modified link source page.

[Configuration of analyzer]
FIG. 10 is a diagram illustrating an example of the configuration of the analysis apparatus 200 according to the first embodiment. As illustrated, the analysis device 200 includes, for example, an analysis-side communication unit 202, an analysis-side control unit 210, and an analysis-side storage unit 230.

  The analysis side communication unit 202 includes, for example, a communication interface such as a NIC or a DMA controller. For example, the analysis-side communication unit 202 communicates with the service providing apparatus 100 via the network NW.

  The analysis-side control unit 210 includes, for example, an analysis-side acquisition unit 212, a parameter extraction unit 214, a parameter determination unit 216, and an output unit 218. The components of the analysis-side control unit 210 are realized, for example, when a processor such as a CPU or GPU executes a program stored in the analysis-side storage unit 230. Also, some or all of the components of the analysis-side control unit 210 may be realized by hardware such as LSI, ASIC, or FPGA, or may be realized by cooperation of software and hardware.

  The analysis side storage unit 230 is realized by, for example, an HDD, a flash memory, an EEPROM, a ROM, a RAM, and the like. The analysis-side storage unit 230 stores the above-described crop parameter information D4 and evaluation information D5, parameter pair evaluation value information D6, and function information D7 in addition to various programs such as firmware and application programs. The parameter pair evaluation value information D6 and the function information D7 will be described later.

[Processing flow]
Hereinafter, each component of the analysis-side control unit 210 will be described with reference to a flowchart. FIG. 11 is a flowchart illustrating an example of processing executed by the analysis-side control unit 210 in the first embodiment. The processing of this flowchart is repeatedly performed at a predetermined cycle, for example.

  First, the analysis-side acquisition unit 212 acquires the crop parameter information D4 and the evaluation information D5 from the service providing apparatus 100 that is the communication partner of the analysis-side communication unit 202 via the analysis-side communication unit 202 (S100). Is stored in the analysis-side storage unit 230.

  Next, the parameter extraction unit 214 extracts the crop parameter from the crop parameter information D4 acquired by the analysis side acquisition unit 212 as a parameter used when the improvement parameter is determined by the Bayesian optimization method (S102). ).

  Next, the parameter extraction unit 214 refers to the evaluation information D5, extracts an evaluation value associated with the cropped image CRP cut out from the original image ORG using the extracted crop parameter, and extracts the cropped image CRP, Parameter pair evaluation value information D6 in which parameters and evaluation values are associated with each other is generated (S104), and this information is stored in the analysis-side storage unit 230.

  FIG. 12 is a diagram illustrating an example of the parameter pair evaluation value information D6. As illustrated, for example, the parameter pair evaluation value information D6 includes, for each cropping image CRP ID, a cropping parameter used when the cropping image CRP is generated, and an evaluation value of the cropping image CRP. Is associated information.

  At this time, the cropping parameter is normalized so that the value is in a numerical range of 0 to 1. For example, when the cropping parameter is “size in the width direction and height direction of the crop window”, the parameter extraction unit 214 sets the “size in the width direction and height direction of the crop window” extracted as the cropping parameter as the size. Is normalized by dividing by the maximum value (for example, the size of the original image ORG in the width direction and the height direction).

  As described above, the cropping parameters included in the parameter versus evaluation value information D6 include the position of the crop window, the size of the crop window, the aspect ratio of the crop window, the shape of the crop window, the characteristics of the image area in which the crop window is set, and the like. Among various parameters, it may be a one-dimensional parameter (one basis) having any one parameter, or a multi-dimensional (a plurality of bases) parameter having a plurality of parameters.

  Next, the parameter determination unit 216 refers to the parameter pair evaluation value information D6, and is an improvement parameter that is preferable (recommended to improve) as a crop parameter by a Bayesian optimization method based on the crop parameter and the evaluation value. Is determined (S106). The parameter determination unit 216 performs this improvement parameter determination processing for each cropped image CRP indicated by the parameter pair evaluation value information D6. Note that, as described above, the parameter determination unit 216 may determine the improvement parameter by using another optimization method such as a genetic algorithm or a differential evolution method instead of the Bayesian optimization method.

  FIG. 13 is a diagram for explaining a method for determining an improvement parameter by a Bayesian optimization method. For example, when the crop parameter is X, the evaluation value can be expressed as an unknown function (hereinafter referred to as a black box function F (X)) having the crop parameter X as an element.

In order to derive the black box function F (X), the parameter determination unit 216 uses the evaluation value of the parameter pair evaluation value information D6 as an initial value. In the illustrated example, evaluation values F (Xa) and F (Xa) are given as initial values. The crop parameters of the evaluation values F (Xa) and F (Xa) are Xa and Xb. In the figure, μ is a black box function F (X) estimated by a Gaussian process (a crop parameter X which is a random variable follows an N-dimensional Gaussian distribution N (μ, σ ² )) based on the initial values of these two points. Represents the average. In the figure, CB represents a confidence interval (μ ± σ) obtained by adding the standard deviation σ (variance σ ² ) to the average μ of the black box function F (X). For example, the standard deviation σ (variance σ ² ) may be obtained based on a Gaussian process, and may be arbitrarily determined as 1σ, 2σ, 3σ.

  For example, the parameter determination unit 216 determines the crop parameter X (uncertain) that has the largest acquisition function (hereinafter referred to as a (X)) indicating the size (absolute value) of the confidence interval CB of the black box function F (X). A highly probable parameter X) is determined as an improvement parameter. For example, a (X) may be obtained based on the following mathematical formula (1). In the equation, K represents a kernel function.

  a (X) = (μ (X) + Kσ (X)) (1)

  In the illustrated example, since a (X) is maximized when the improvement parameter is Xc, the parameter determination unit 216 determines the parameter Xc as an improvement parameter. Thus, the improvement parameter is determined according to the magnitude of a (X).

  Next, the output unit 218 outputs (transmits) the improvement parameter determined by the parameter determination unit 216 to the service providing apparatus 100 using the analysis-side communication unit 202 (S108).

  In response to this, the image processing unit 114 of the service providing apparatus 100 changes the crop parameter of the crop parameter information D4 to the improvement parameter output by the analysis apparatus 200, so that a new cropped image CRP is generated from each original image ORG. Is generated. The service providing unit 112 of the service providing apparatus 100 is posted as a response to this request when, for example, an HTTP request is received from the terminal device 10 of a user participating in crowdsourcing by the service providing side communication unit 102. The link source page (modified link source page) that has been changed to the newly generated cropped image CRP is provided to the terminal device 10 that is the request transmission source. At this time, before and after the change of the cropping image CRP, the linked pages associated with the links L of these pages are the same page. Note that the image posted on the link source page may be changed by a designer or the like as described above.

  When the modified link source page is provided by the service providing unit 112, the evaluation value deriving unit 116 of the service providing apparatus 100 newly derives the evaluation value of the modified link source page.

  Next, the analysis side acquisition unit 212 of the analysis apparatus 200 acquires the evaluation value of the modified link source page from the service providing apparatus 100 using the analysis side communication unit 202 (S110).

  Next, the parameter determination unit 216 re-determines the improvement parameter by the Bayesian optimization method using the evaluation value (actual measurement value) of the modified link source page (S112).

  FIG. 14 is a diagram for explaining a method of redetermining improvement parameters. For example, as illustrated in FIG. 13 described above, the parameter Xc is determined as an improvement parameter, and the cropped image CRP newly generated according to the crop parameter is displayed on the link source page. Is modified. In this case, the parameter determination unit 216 uses the initial values F (Xa) and F (Xb) of the evaluation value and the evaluation value F (Xc) acquired as the actual measurement value of the parameter Xc determined as the previous improvement parameter as a fixed point. The average μ of the black box function F (X) is estimated by a Gaussian process using (the actual value of the event to be observed instead of the estimated value obtained stochastically). Then, the parameter determining unit 216 determines again the parameter X that maximizes a (X) as an improvement parameter. In the illustrated example, a (X) is maximized when the parameter is Xd, so the parameter determination unit 216 determines the parameter Xd as an improvement parameter.

  Next, the output unit 218 uses the analysis-side communication unit 202 to output the improvement parameter determined again by the parameter determination unit 216 to the service providing apparatus 100 (S114).

  Next, the analysis-side control unit 210 determines whether or not the number of calculation processes for determining an improvement parameter has reached a predetermined number (S116). If the number of calculations has not reached the predetermined number, the process proceeds to S110 described above. Move processing. As a result, the cropping image CRP is changed in the link source page according to the repetitive determination of the improvement parameter until the number of operations reaches the predetermined number of times, and a modified link source page is generated each time the change is made. Is done. The modified link source page generated according to the improvement parameter being repeatedly determined is an example of “electronic page of each generation”. For example, every time the calculation process for determining the improvement parameter is repeated once, the “generation” of the link source page increases by one generation.

  Each time the cropped image CRP is changed and a modified link source page is generated, the service providing unit 112 of the service providing device 100 sends the modified link source page of each generation to the terminal device 10 of the user who participates in crowdsourcing. Send. Then, the evaluation value deriving unit 116 derives an evaluation value for each modified link source page of each generation.

  FIG. 15 is a diagram schematically illustrating how a modified link source page of each generation is generated. For example, it is assumed that the cropping image CRP1 is generated by the image processing unit 114 as the first process. In this case, the service providing unit 112 provides the link source page on which the cropped image CRP1 is posted to the terminal device 10.

  When a link source page on which the cropping image CRP1 is posted is provided and an evaluation value corresponding to the behavior of the user who has viewed the link image is derived, the parameter determination unit 216 determines an improvement parameter using a Bayesian optimization method. The image processing unit 114 generates a new cropped image CRP from the original image ORG according to the improvement parameter. For example, when the improvement parameter is a parameter that designates a region that is shifted to the left side in the figure while maintaining the size of the crop window from the cropping position of the cropped image CRP1, the image processing unit 114 Cuts out the cropped image CRP2 from the original image ORG. The service providing unit 112 provides the terminal device 10 with the link source page on which the cropped image CRP2 newly cut (generated), that is, the first generation modified link source page is posted.

  When the link source page on which the cropping image CRP2 is posted is provided and an evaluation value corresponding to the behavior of the user who has viewed the link is derived, the parameter determination unit 216 newly determines an improvement parameter using a Bayesian optimization method. To do. The image processing unit 114 generates a new cropped image CRP from the original image ORG according to the improvement parameter. For example, when the improvement parameter is a parameter for designating a new cropped image cutout position as a new cropped image cutout position while maintaining the size of the crop window from the cutout position of the cropped image CRP2 114 crops the cropping image CRP3 from the original image ORG. The service providing unit 112 provides the terminal device 10 with the link source page on which the cropped image CRP3 newly cut (generated), that is, the second generation modified link source page is posted.

  When the link source page on which the cropping image CRP3 is posted is provided and an evaluation value corresponding to the behavior of the user who has viewed the link is derived, the parameter determination unit 216 newly determines an improvement parameter using a Bayesian optimization method. To do. The image processing unit 114 generates a new cropped image CRP from the original image ORG according to the improvement parameter. For example, when the improvement parameter is a parameter for designating a new cropped image cutout position as a new cropped image cutout position while maintaining the size of the crop window from the cutout position of the cropped image CRP2 114 crops the cropping image CRP3 from the original image ORG. The service providing unit 112 provides the terminal device 10 with the link source page on which the cropped image CRP3 newly cut (generated), that is, the second generation modified link source page is posted.

  In this way, a new generation of modified link source pages are successively generated according to the improvement parameter. The tendency of the black box function F (X) is determined by repeating the Bayesian optimization method in accordance with the evaluation value of each of these modified link source pages. As a result, even if it is a crop parameter for which the evaluation value of the link source page has not been obtained, how much evaluation value is obtained even if it is an unknown crop parameter that has not been tried by the derived black box function F (X). You can see what you can get.

  On the other hand, when the number of calculations reaches a predetermined number, the analysis-side control unit 210 stores information associating the link source page with the black box function F (X) in the analysis-side storage unit 230 as function information D7. Store (S118).

  FIG. 16 is a diagram illustrating an example of the function information D7. For example, the function information D7 includes, for each page ID of the link source page, a black parameter derived by calculating a crop parameter X when the cropped image CRP posted on the link source page is generated and a predetermined number of times. This is information associated with the box function F (X). The crop parameter X is a one-dimensional parameter or a multi-dimensional parameter used when deriving the black box function F (X). If a multi-dimensional parameter, what element is used as a basis? , And so on.

  Next, the output unit 218 uses the analysis-side communication unit 202 to determine the crop parameter X () in which the evaluation value F (X) is equal to or greater than the threshold in the black box function F (X) derived by performing a predetermined number of operations. (Hereinafter referred to as optimal parameters) is output to the service providing apparatus 100 (S120).

  In response to this, the image processing unit 114 of the service providing apparatus 100 generates a cropped image CRP from the original image ORG in accordance with the optimum parameter. The service providing unit 112 generates a modified link source page by changing the image to be posted to the cropped image CRP generated based on the optimum parameters, and participates in the crowdsourcing for the modified link source page. This is provided not only to the terminal device 10 operated by the user but also to all the terminal devices 10 that have received the request. Thereby, the processing of this flowchart is completed.

  In the first embodiment described above, the original image information D3 illustrated in FIG. 6 has been described as information in which one original image ORG is associated with one article ATC. However, the present invention is not limited to this. Absent. For example, the original image information D3 may be information in which a plurality of original images ORG are associated with one article ATC. At this time, it is preferable that the plurality of original images ORG is a set of images belonging to the same category. For example, when an article ATC is an article that conveys a game result of “baseball”, among the plurality of original images ORG, an original image ORG is an image of “baseball player” who participated in the game, and another original image ORG. The image ORG is an image of a “ballpark” where the game was played.

  In the first embodiment described above, the image processing unit 114 is described as generating the cropped image CRP according to the improvement parameter every time the improvement parameter is determined by the parameter determination unit 216. However, the present invention is not limited to this. A plurality of cropped images CRP may be generated in advance before the improvement parameter is determined. For example, the image processing unit 114 generates a plurality of cropped images CRP from one original image ORG while changing the cutout position, aspect ratio, and the like. At this time, the image processing unit 114 changes the crop parameter according to a predetermined rule, and generates the cropped image CRP with the cutout positions and the like dispersed to some extent. Then, the service providing unit 112 provides a link source page on which any of the cropped images CRP among the plurality of cropped images CRP generated by the image processing unit 114 is posted.

  When the improvement parameter is determined by the parameter determination unit 216, the image processing unit 114 selects a cropped image CRP generated using a crop parameter that is the same as or close to the improvement parameter among a plurality of cropped images CRP generated in advance. select. For example, assume that a plurality of cropped images CRP are generated while changing the aspect ratio of the crop window to 0.1, 0.2, 0.3,. At this time, when the improvement parameter for designating the aspect ratio of the crop window to 0.28 is determined, the image processing unit 114 generates an aspect ratio of 0.3 among a plurality of cropped images CRP generated in advance. A cropping image CRP is selected. As a result, the generation process of the cropped image CRP performed in accordance with the determination of the improvement parameter can be omitted, so that the processing time required to determine the black box function can be shortened.

  According to the first embodiment described above, the analysis-side acquisition unit 212 that acquires an evaluation value that evaluates an action taken by a user who has viewed an electronic page on which a cropped image CRP cut out from the original image ORG is posted. Then, based on the crop parameter related to the region where the cropped image CRP is cut out in the original image ORG and the evaluation value acquired by the analysis side acquisition unit 212, an optimization method of an unknown function such as a Bayesian optimization method, By providing the parameter determination unit 216 that determines a preferable improvement parameter as a crop parameter, the service providing apparatus 100 and the like can automatically generate a more suitable image as an image to be posted on the electronic page.

  Further, according to the first embodiment described above, since the improvement parameter is determined by the optimization method of the unknown function such as the Bayesian optimization method, the region that the designer has not tried before is used as the cropped image CRP. It can be cut out, and an innovative electronic page can be generated that overturns conventional common sense.

  In addition, according to the first embodiment described above, for example, when a Bayesian optimization method is used as an optimization method for an unknown function, a black box function (continuous function) based on a Gaussian process is assumed, In order to determine a parameter with high uncertainty in the black box function as an improvement parameter, as described above, a series of trial processing (live test) on what action the user has taken in response to electronic page modification Therefore, it is possible to optimize the black box function that is the objective function faster than the optimization method of other unknown functions such as a genetic algorithm that requires a large amount of trial processing results.

  Further, according to the first embodiment described above, the derived black box function is stored in association with the crop parameter used when generating the cropped image CRP, so that an unknown function optimization method such as a Bayesian optimization method is used. As a result obtained by the above, it is possible to share among the designers what crop parameter the cropping image CRP can be used to improve the evaluation value. As a result, for example, an electronic page capable of obtaining a high evaluation value can be designed even by a designer with poor design knowledge of the electronic page.

Second Embodiment
Hereinafter, a second embodiment will be described. In the first embodiment described above, the image processing unit 114 sets the crop window as the original image ORG in accordance with the improvement parameter determined by the analysis apparatus 200, so that the image area overlapping the crop window is changed to the new cropped image CRP. As described as what to cut out. The second embodiment is different from the first embodiment described above in that a limitation is set in advance on crop parameters that are candidates for improvement parameters, and a new cropped image CRP is generated in consideration of this limitation. Hereinafter, differences from the first embodiment will be mainly described, and description of points common to the first embodiment will be omitted. In the description of the second embodiment, the same parts as those in the first embodiment will be described with the same reference numerals.

  FIG. 17 is a diagram for explaining a cropping image CRP extraction method when restrictions are set on crop parameters that are candidates for improvement parameters. For example, it is assumed that a certain numerical range related to the height direction Y is set to a parameter limit range (a numerical range that prohibits setting of the crop window). At this time, when the center coordinate of the crop window W is designated as CP1 as the improvement parameter, the size in the width direction X of the crop window W is designated as A1, and the size in the height direction Y is designated as B1, as shown in the figure, Some position coordinates of the crop window W are included in the parameter limit range. In this case, the image processing unit 114 corrects the improvement parameter so that the position coordinates of the crop window W are not included in the restriction range. As illustrated, the image processing unit 114 changes, for example, the center coordinates of the crop window W designated as the improvement parameter from CP1 to CP1 #. Thus, for example, as a conventional knowledge, when it is known in advance that the evaluation value of the link source page is lowered when an image of just the foot is posted on the link source page, the coordinates at the bottom of the image are set. If the parameter limit range is set, when the original image ORG is the entire image of the person, only the foot image is not generated as the cropped image CRP. As a result, the crop parameter, which has already been clarified what evaluation value can be obtained, is not determined as an improvement parameter. Therefore, the improvement parameter is determined, and the cropped image is determined accordingly. It is possible to reduce the number of iterations in which a CRP is generated, a newly modified link source page is provided, an evaluation value of this page is derived, and an improvement parameter is determined again by a Bayesian optimization method. it can. As a result, the processing time until the black box function F (X) is determined can be shortened.

  Further, the above-described parameter limit range may be determined in advance for each category of the original image ORG. For example, when the original image ORG belongs to the category of “politics”, the limit range of parameters provided in the original image ORG is such that a person's face is easily cut out as a cropped image CRP, and the lower part of the original image ORG and both It may be set to the side region coordinates. In addition, when the original image ORG belongs to the “sports” category, the parameter limitation range provided in the original image ORG is set to the “politics” category so that the entire movement of the player can be easily extracted as the cropped image CRP. It may be narrower than the parameter limit range when belonging.

  In the above-described example, the parameter limit range has been described as being considered when the cropped image CRP is cut out in the service providing apparatus 100. However, the present invention is not limited to this, and the analysis apparatus 200 is improved by a Bayesian optimization method. It may be taken into account when determining the parameters.

  FIG. 18 is a diagram for explaining a method for determining an improvement parameter when a parameter restriction range is provided. For example, in a crop parameter X that is a candidate for an improvement parameter, when a certain parameter range is set as a limit range, the parameter determination unit 216 sets a (X) derived in the limit range to, for example, zero. For example, before the limit range is provided, the parameter determination unit 216 determines the parameter Xd that maximizes a (X) as the improvement parameter. On the other hand, when the limit range is provided, the parameter determination unit 216 sets a (X) of the region to zero, and therefore does not determine the parameter Xd as an improvement parameter, but the parameter X in a range other than the limit range. Is determined as an improvement parameter. In the illustrated example, since a (X) is maximized when the parameter is Xe, the parameter determination unit 216 determines the parameter Xe as an improvement parameter. As a result, the crop parameters included in the limited range are not treated as uncertain parameters, and the improvement parameters can be searched for more effectively.

  According to the second embodiment described above, the improvement parameter can be searched for more effectively by providing a limitation on the crop parameter that has already been determined what value the evaluation value can be obtained. Or the processing time until the black box function F (X) is determined can be shortened.

<Third Embodiment>
Hereinafter, the third embodiment will be described. In the third embodiment, the image processing unit 114 further performs object recognition processing on the original image ORG, and the parameter determination unit 216 uses the crop parameters that are candidates for improvement parameters based on the recognition result by the image processing unit 114. Is different from the above-described first and second embodiments in that the improvement parameter is determined within the determined crop parameter range. Hereinafter, description will be made centering on differences from the first and second embodiments, and description of points common to the first and second embodiments will be omitted. In the description of the third embodiment, the same parts as those in the first and second embodiments will be described with the same reference numerals.

  The image processing unit 114 in the third embodiment extracts a predetermined image feature from the original image ORG. The predetermined image feature includes, for example, a feature based on a luminance difference between pixels (for example, Haar-Like feature), a feature related to a luminance gradient in a local region of the image (for example, HOG (Histograms of Oriented Gradients) and EOH (Edge of Orientation Histograms)). Etc. Then, the image processing unit 114 recognizes the position of the predetermined object OB on the original image ORG from the extracted features. The predetermined object OB is an object that is desired to be easily cut out from the original image ORG as a cropped image CRP. For example, the predetermined object OB is an object such as a person, an animal such as a dog or a cat. The image processing unit 114 uses the service providing side communication unit 102 to transmit information regarding the position of the predetermined object OB on the original image ORG to the analysis apparatus 200.

  The parameter determination unit 216 according to the third embodiment, when determining an improvement parameter using a Bayesian optimization method, is a crop parameter that is a candidate for an improvement parameter based on the position of the predetermined object OB transmitted by the service providing apparatus 100. Determine the range. For example, the parameter determination unit 216 corrects the confidence interval CB of the black box function F (X) based on the position of the predetermined object OB, and determines a range of crop parameters that are candidates for improvement parameters.

  FIG. 19 is a diagram for explaining a method of correcting the confidence interval CB of the black box function F (X). For example, when cropping images CRP_a and CRP_b are cut out in advance in a certain original image ORG_i, as described above, evaluation values Fa and Fb of the link source pages on which these images are posted are given as initial values. It is done. When the position of the predetermined object OB is specified by the image processing unit 114 in the original image ORG_i, the parameter determination unit 216 increases the section width of the confidence interval CB according to the position of the predetermined object OB with respect to the original image ORG_i. .

For example, when the crop parameter, which is the horizontal axis of the black box function F (X), is the coordinate in the width direction with respect to the original image ORG, the reliability of the crop parameter included in the predetermined area A _OB with respect to the position of the predetermined object OB. The confidence interval CB is corrected by increasing the interval width of the interval CB. More specifically, when the standard deviation σ of the probability distribution assumed by the Gaussian process is 1σ, the parameter determination unit 216 changes the standard deviation σ of the predetermined area A _OB to 2σ or 3σ, thereby providing a confidence interval. Increase the section width of CB. In the figure, CB # represents a confidence interval after correction. Thus, since the crop parameters included in the predetermined region A _OB a (X) is greater than before the correction, likely range of crop parameters included in the predetermined region A _OB is determined as the improved parameter. That is, the predetermined area A _OB is more easily searched. In the illustrated example, the parameter Xf is newly determined as an improvement parameter by correcting the confidence interval CB.

  According to the third embodiment described above, when a predetermined object OB predetermined as an object to be included in the cropped image CRP is recognized by the object recognition process on the original image ORG, the parameter determination unit 216 performs the original determination. Based on the position of the predetermined object OB on the image ORG, the range of the crop parameter that is a candidate for the improvement parameter is determined, so that the image area near the predetermined object OB in the original image ORG can be more easily cut out as the cropped image CRP. . As a result, the cropped image CRP can be generated more efficiently.

  Note that in the third embodiment described above, the crop parameter is, for example, an object included in a region to be cut out as the cropped image CRP, such as only a person, a combination of a person and a dog, a combination of a dog and a cat, or the like. It may include a parameter that specifies a combination.

<Hardware configuration>
Of the plurality of devices included in the analysis system 1 of the above-described embodiment, the service providing device 100 and the analysis device 200 are realized by a hardware configuration as illustrated in FIG. 20, for example. FIG. 20 is a diagram illustrating an example of a hardware configuration of the service providing apparatus 100 and the analysis apparatus 200 according to the embodiment.

  The service providing apparatus 100 includes an NIC 100-1, a CPU 100-2, a RAM 100-3, a ROM 100-4, a secondary storage device 100-5 such as a flash memory and an HDD, and a drive device 100-6. Are connected to each other. The drive device 100-6 is loaded with a portable storage medium such as an optical disk. A program stored in a portable storage medium attached to the secondary storage device 100-5 or the drive device 100-6 is expanded in the RAM 100-3 by a DMA controller (not shown) or the like and executed by the CPU 100-2. Thus, the service providing side control unit 110 is realized. The program referred to by the CPU 100-2 may be downloaded from another device via the network NW.

  The analysis device 200 includes a NIC 200-1, a CPU 200-2, a RAM 200-3, a ROM 200-4, a secondary storage device 200-5 such as a flash memory and an HDD, and a drive device 200-6 via an internal bus or a dedicated communication line. They are connected to each other. The drive device 200-6 is loaded with a portable storage medium such as an optical disk. A program stored in a portable storage medium attached to the secondary storage device 200-5 or the drive device 200-6 is expanded in the RAM 200-3 by a DMA controller (not shown) or the like and executed by the CPU 200-2. Thus, the analysis-side control unit 210 is realized. The program referred to by the CPU 200-2 may be downloaded from another device via the network NW.

  As mentioned above, although the form for implementing this invention was demonstrated using embodiment, this invention is not limited to such embodiment at all, In the range which does not deviate from the summary of this invention, various deformation | transformation and substitution Can be added.

  DESCRIPTION OF SYMBOLS 1 ... Analysis system, 10 ... Terminal device, 100 ... Service provision apparatus, 102 ... Service provision side communication part, 110 ... Service provision side control part, 112 ... Service provision part, 114 ... Image processing part, 116 ... Evaluation value derivation part , 130 ... service providing side storage unit, 200 ... analysis device, 202 ... analysis side communication unit, 210 ... analysis side control unit, 212 ... analysis side acquisition unit, 214 ... parameter extraction unit, 216 ... parameter determination unit, 218 ... output , 230 ... Analysis side storage unit, NW ... Network

Claims (7)

An acquisition unit that acquires an evaluation value in which an action taken by a user who viewed an electronic page on which a cropped image cut out from an original image is posted;
Based on a parameter related to a region where the cropped image is cut out in the original image and an evaluation value acquired by the acquisition unit, an improvement function preferable as the parameter is determined by an optimization method of an unknown function based on a stochastic process. A decision unit to
An analysis apparatus comprising: An image processing unit for cutting out the cropped image from the original image according to the improvement parameter;
The acquisition unit acquires an evaluation value obtained by evaluating an action taken by a user who has viewed an electronic page on which the cropped image cut out by the image processing unit is posted;
The determination unit newly determines the improvement parameter by the optimization method of the unknown function, with the evaluation value regarding each generation of electronic pages generated according to the determination of the improvement parameter being repeated, as a fixed point. ,
The analysis device according to claim 1. The image processing unit recognizes a predetermined object by extracting features from the original image,
The determination unit
Based on the position of the predetermined object recognized by the image processing unit, determine a parameter that is a candidate for the improvement parameter,
Among the parameters that are candidates for the determined improvement parameter, the improvement parameter is determined by an optimization method for the unknown function.
The analysis device according to claim 2. The image processing unit cuts out the cropped image according to the improvement parameter from the original image every time the improvement parameter is repeatedly determined by the determination unit,
A provision unit that provides an electronic page on which the cropped image is posted to a terminal device used by the user each time the cropped image is cut out by the image processing unit;
The analysis device according to claim 2 or 3. An image processing unit for cutting out the cropped image from the original image;
A provision unit for providing an electronic page on which a cropped image cut out by the image processing unit is provided to a terminal device used by a user;
The image processing unit is based on an evaluation value obtained by evaluating an action taken by the user who viewed the electronic page provided by the providing unit, and a parameter relating to a region where the cropped image is cut out in the original image. Then, a new cropping image is cut out from the original image by an unknown function optimization method based on a stochastic process .
Information providing device. Computer
Obtain an evaluation value that evaluates the action taken by the user who viewed the electronic page on which the cropped image cut out from the original image was posted,
A preferable improvement parameter is determined as the parameter by an unknown function optimization method based on a stochastic process based on a parameter related to a region where the cropped image is cut out in the original image and the acquired evaluation value.
analysis method. On the computer,
Obtain an evaluation value that evaluates the action taken by the user who viewed the electronic page on which the cropped image cut out from the original image was posted,
Based on a parameter related to a region where the cropped image is cut out in the original image and the obtained evaluation value, an optimization function of an unknown function based on a stochastic process is used to determine a preferable improvement parameter as the parameter.
program.

Images