JP5996747B1 - Generating device, generating method, and generating program - Google Patents

Abstract

A model for appropriately predicting a click rate of content is generated. A generation apparatus according to the present application includes an acquisition unit and a generation unit. The acquisition unit acquires an evaluation value related to the prediction target content and an evaluation value related to the peripheral content arranged around the prediction target content among the content continuously arranged on the predetermined display medium. The generation unit generates a model that predicts the probability that the prediction target content is selected based on the evaluation value related to the prediction target content acquired by the acquisition unit and the evaluation value related to the surrounding content. [Selection] Figure 2

Description

  The present invention relates to a generation device, a generation method, and a generation program.

  Conventionally, a technique for predicting whether or not a user clicks on content has been provided.

JP 2014-174753 A

  However, the above-described conventional technology cannot always generate a model that appropriately predicts the click rate of content. For example, a model based on a combination of advertisements (contents) clicked by the user does not always generate a model that appropriately predicts the click rate of the contents.

  The present application has been made in view of the above, and an object thereof is to provide a generation device, a generation method, and a generation program that can generate a model that appropriately predicts the click rate of content.

  The generating apparatus according to the present application acquires an evaluation value related to a prediction target content and an evaluation value related to peripheral content arranged around the prediction target content among content continuously arranged on a predetermined display medium. And a generation unit that generates a model for predicting a probability that the prediction target content is selected based on the evaluation value related to the prediction target content and the evaluation value related to the surrounding content acquired by the acquisition unit. And.

  According to one aspect of the embodiment, there is an effect that a model that appropriately predicts the click rate of content can be generated.

FIG. 1 is a diagram illustrating an example of a generation process according to the embodiment. FIG. 2 is a diagram illustrating a configuration example of the prediction device according to the embodiment. FIG. 3 is a diagram illustrating an example of a content information storage unit according to the embodiment. FIG. 4 is a diagram illustrating an example of a distribution log information storage unit according to the embodiment. FIG. 5 is a diagram illustrating an example of a learning information storage unit according to the embodiment. FIG. 6 is a diagram illustrating an example of a process for predicting a click probability based on the arrangement order according to the embodiment. FIG. 7 is a flowchart illustrating an example of a process procedure of model generation according to the embodiment. FIG. 8 is a flowchart illustrating an example of the processing procedure of the click probability prediction according to the embodiment. FIG. 9 is a diagram illustrating a configuration example of a prediction device according to a modification. FIG. 10 is a diagram illustrating an example of a content information storage unit according to a modification. FIG. 11 is a diagram illustrating an example of a learning information storage unit according to the embodiment. FIG. 12 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the prediction device.

  Hereinafter, a generation apparatus, a generation method, and a mode for executing a generation program (hereinafter referred to as “embodiment”) according to the present application will be described in detail with reference to the drawings. Note that the generation device, the generation method, and the generation program according to the present application are not limited by this embodiment. In the following embodiments, the same portions are denoted by the same reference numerals, and redundant description is omitted.

(Embodiment)
[1. Generation process)
First, an example of the generation process according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram illustrating an example of a generation process according to the embodiment. Note that the generation processing in the present embodiment includes processing for predicting the probability that content is selected (hereinafter, may be referred to as “click probability”) using the generated model. For example, the click probability here means a click rate based on prediction, that is, a predicted CTR (Click Through Rate). In the following, predicting the probability that content will be selected may be referred to as “predicting click probability” or “predicting click rate”. Further, in the following, the content for which the click probability is to be predicted may be referred to as “target content”.

  As illustrated in FIG. 1, the prediction system 1 includes a terminal device 10, a content distribution device 50, and a prediction device 100. The terminal device 10, the content distribution device 50, and the prediction device 100 are connected to be communicable by wire or wireless via a predetermined network (not shown). Note that the prediction system 1 illustrated in FIG. 1 may include a plurality of terminal devices 10, a plurality of content distribution devices 50, and a plurality of prediction devices 100.

  The terminal device 10 is an information processing device used by a user. The terminal device 10 is realized by, for example, a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. In the example illustrated in FIG. 1, the terminal device 10 is a smartphone having a touch panel function. Hereinafter, the terminal device 10 may be referred to as a user. That is, hereinafter, the user can be read as the terminal device 10.

  In addition, the terminal device 10 requests the content distribution device 50 to distribute the content. Further, the terminal device 10 displays the content distributed from the content distribution device 50 by a predetermined application. The predetermined application may be a browser.

  The content distribution device 50 is an information processing device that distributes content displayed on the terminal device 10. For example, in response to a distribution request from the terminal device 10, the content distribution device 50 distributes a plurality of contents (hereinafter sometimes referred to as “articles”) displayed in a predetermined arrangement order to the terminal device 10. The content distribution device 50 acquires log information from the terminal device 10 that has distributed the content. For example, the content distribution device 50 acquires information regarding which article is clicked among a plurality of articles displayed in a predetermined arrangement order.

  The prediction device 100 is an information processing device that provides a service for predicting a click probability of target content using a generated model. The prediction device 100 is a generation device that generates a model for predicting the click probability of the target content based on the evaluation value regarding the target content and the evaluation value regarding the surrounding content. FIG. 1 shows a case where the click rate of each content, that is, CTR is used as the evaluation value for each content. The peripheral content is content arranged around the target content, and details will be described later.

  As shown in FIG. 1, the content distribution device 50 distributes content to the terminal device 10 used by the user U1 (step S11-1). For example, when a content distribution request is received from the terminal device 10 used by the user U1, the content distribution device 50 distributes the content to the terminal device 10 used by the user U1.

  Then, the terminal device 10 used by the user U1 displays the received content. In FIG. 1, the terminal device 10 used by the user U1 displays an article A, an article B, an article C, and an article D, which are the first to fourth placement contents in the received content. Note that the user U1 who uses the terminal device 10 sequentially displays the articles below the article D by touching the screen of the terminal apparatus 10 and performing a scroll operation or the like. In FIG. 1, it is assumed that the user U1 selects, that is, clicks, the article A and the article D displayed on the terminal device 10.

  Thereafter, the content distribution device 50 acquires log information related to the content distributed from the terminal device 10 used by the user U1 (step S11-2). For example, the content distribution device 50 acquires information indicating that the user U1 has clicked the article A and the article D from the terminal device 10 used by the user U1.

  The content distribution device 50 distributes content to the terminal device 10 used by the user U2 (step S12-1). For example, when a content distribution request is received from the terminal device 10 used by the user U2, the content distribution device 50 distributes the content to the terminal device 10 used by the user U2.

  Then, the terminal device 10 used by the user U2 displays the received content. In FIG. 1, the terminal device 10 used by the user U2 displays the article B, the article D, the article A, and the article C, which are the contents of the first to fourth arrangement order among the received contents. Note that the user U2 who uses the terminal device 10 sequentially displays the articles below the article C by touching the screen of the terminal apparatus 10 and performing a scroll operation or the like. Further, in FIG. 1, it is assumed that the user U2 selects, that is, clicks, the article A and the article C displayed on the terminal device 10.

  Thereafter, the content distribution device 50 acquires log information related to the content distributed from the terminal device 10 used by the user U2 (step S12-2). For example, the content distribution device 50 acquires information indicating that the user U2 has clicked the article A and the article C from the terminal device 10 used by the user U2.

  And the content delivery apparatus 50 transmits the log information acquired from the terminal device 10 to the prediction apparatus 100 (step S13). The prediction device 100 stores log information acquired from the content distribution device 50 (hereinafter sometimes referred to as “distribution log information”) in the distribution log information storage unit 122 (see FIG. 2). In the prediction system 1, when the prediction device 100 acquires the distribution log information in advance, the processing of step S11 to step S13 may not be performed.

  Next, the prediction device 100 uses the CTR of each article stored in the content information storage unit 121 (see FIG. 2), the distribution log information 122-1 and 122-2 acquired from the content distribution device 50, and the like. A model for predicting the click probability of the target content is generated. For example, the distribution log information 122-1 is distribution log information acquired from the terminal device 10 of the user U1, and the distribution log information 122-2 is distribution log information acquired from the terminal device 10 of the user U2.

  Here, in the example shown in FIG. 1, each content including the target content is displayed in a list in the terminal device 10 in the vertical direction. Therefore, a user who uses the terminal device 10 browses (views) the content on the upper side, that is, the content in the order of arrangement in order. Therefore, it can be assumed that what kind of content is arranged above the target content affects whether the target content is clicked by the user. Therefore, in the content displayed in a list in the vertical direction in FIG. 1, the click probability of the target content is calculated by the following formula (1).

  “P (Click (i))” in the left side of the above formula (1) corresponds to the click probability of the target content whose arrangement order is i-th. For example, “P (Click (i))” takes a value from 0 to 1.

The first term “α _{i, 0} · CTR (C (i))” in the right side of the above formula (1) is “CTR (C (i))” which is the CTR of the target content whose arrangement order is i. ”And a predetermined weight“ α _{i, 0} ”.

The second term in the right side of the above formula (1) corresponds to the click probability of the content arranged above the i-th target content in the arrangement order. For example, when K = 1, “α _{i, 1} · P (Click (i−1))” is “P (Click (i−1))”, which is the click probability of the content whose arrangement order is the (i−1) th. ”Multiplied by a predetermined weight“ α _{i, 1} ”. The same applies to K = 2 and later.

  In other words, the prediction device 100 determines the number of upwardly generated models taking into account the influence of content from the i-th target content in the arrangement order by changing the value of K. That is, the prediction device 100 determines the click probability considering the influence of up to the number of contents from the i-th target content in the arrangement order by changing the value of K.

  Here, when the right side in the above formula (1) is transformed, the following formula (2) is obtained.

The second term in the right side of the above formula (2) corresponds to the click probability of the content arranged in the (i−1) th order, that is, one content above the target content. For example, “α _i−1,0 · CTR (C (i−1))” in the second term in the right side of the formula (2) is “CTR” which is the CTR of the content whose arrangement order is the (i−1) th. (C (i−1)) ”is multiplied by a predetermined weight“ α _i−1,0 ”. In the other part of the second term in the right side of the formula (2), “K” in the second term in the right side of the formula (1) is set to “K−1”, and “i” is set to “i”. Except for “−1”, it is the same as the second term in the right side of the formula (1).

In addition, the third term in the right side of the above formula (2) is “k = 2”, and therefore corresponds to the click probability of the content arranged at two or more of the target contents of the i-th target content. . For example, when K = 2, “α _{i, 2} · P (Click (i−2))” is “P (Click (i−2))” which is the click probability of the content whose arrangement order is the (i−2) th. ”Multiplied by a predetermined weight“ α _{i, 2} ”. The same applies to K = 3 and thereafter.

  Further, when the right side in the formula (2) is further modified, the following formula (3) is obtained.

For example, the second term in the right side of the above equation (3) is replaced with “α _i−1,0 · CTR (C (i−1))” of the second term in the right side of the above equation (2). Correspond. Further, for example, the third term in the right side of the formula (3) is other than “α _i−1,0 · CTR (C (i−1))” in the second term in the right side of the formula (2). And the third term in the right side of the above formula (2) are combined.

  By transforming the above equation (3) into a form of multiplying the CTR of each content and its weight, the following equation (4) is obtained.

The right side of the above formula (4) is, for example, “β _{i, 0} · CTR (C (i−0))” when K = 0, and “P”, which is the click probability of the i-th target content in the arrangement order. “(Click (i))” is a value obtained by multiplying “CTR (C (i))”, which is the CTR of the target content, by a predetermined weight “β _{i, 0} ”. When K = 1, a value obtained by multiplying “CTR (C (i))” which is the CTR of the target content by a predetermined weight “β _{i, 0} ” and “β _{i, 1} · CTR (C (i -1)) ". Note that “β _{i, 1} · CTR (C (i−1))” is a predetermined weight “β” to “CTR (C (i−1))” which is the CTR of the content whose arrangement order is the (i−1) th. _{i, 1} ". The same applies to K = 2 and later.

In the case of the above formula (4), for example, the weight “β _{i, 0} ” corresponds to “α _{i, 0} ”, and the weight “β _{i, 1} ” corresponds to “α _{i, 1} · α _i−1,0 ”. Corresponding to

  As described above, by changing the value of K, the prediction apparatus 100 determines how many models of the arrangement order are affected by the content up from the i-th target content. That is, the prediction device 100 learns by changing the values of “K = 1”, “K = 2”, “K = 3”, and “K” in various ways, and thereby a plurality of models corresponding to each “K”. Can be generated. In the following description, when the expression (4) is used, an appropriate value of “K” is selected to predict the click probability of the target content. For example, in the example shown in FIG. 1, “2” is selected for “K”.

In FIG. 1, the prediction device 100 derives the weights “β _{i, 0} ”, “β _{i, 1} ”, etc. by learning according to the value of “K” in the right side of the equation (4). For example, in the case of “K = 2” in the above formula (4), the prediction apparatus 100 takes into account the influence of content up to two levels above the target content. That is, when “K = 2”, the prediction device 100 derives the weights “β _{i, 0} ”, “β _{i, 1} ”, and “β _{i, 2} ” by learning. Thereby, the prediction device 100 generates a model for predicting the click probability of the target content (step S14). For example, the prediction device 100 stores the model (weight) generated in step S14 in the learning information storage unit 123 (see FIG. 2).

In FIG. 1, the prediction device 100 generates a model where “position 0”, that is, the weight “β _{i, 0} ” corresponding to the target content is “0.7”, as a model for predicting the click probability of the target content. In addition, the model generated by the prediction device 100 is “position 1”, that is, the weight “β _{i, 1} ” corresponding to the content arranged one above the target content is “0.15”, and “position 2”. That is, the weight “β _{i, 2} ” corresponding to the content arranged two above the target content is “0.05”. Note that the weight of the model generated by the prediction device 100 may be 0 or a negative value.

  Next, the prediction apparatus 100 predicts the click probability of the target content using the generated model. For example, the prediction device 100 uses the above equation (4), the CTR of each article stored in the content information storage unit 121, and the weight of each feature stored in the learning information storage unit 123, and the click probability of the target content. Predict.

  In FIG. 1, an article Z is arranged in the first place in the arrangement order, an article X is arranged in the second place in the arrangement order, an article Y is arranged in the third place in the arrangement order, an article W is arranged in the fourth place in the arrangement order, and the vertical direction A case will be described in which the prediction apparatus 100 performs prediction on the list display content LO11 that is displayed in a list. In FIG. 1, the prediction device 100 predicts the click probability with the article Y having the third placement order as the target content TG11 in the list display content LO11 (step S15).

  Specifically, the prediction apparatus 100 calculates the CTR of the article Y, the CTR of the article X, the CTR of the article Z, the weight of “position 0”, the weight of “position 1”, and the weight of “position 2” by the above formula (4). ), The click probability of the article Y that is the target content TG11 is predicted. For example, the prediction device 100 predicts (outputs) the prediction result value PV11 of the click probability of the article Y by the prediction in step S15. In FIG. 1, the prediction device 100 predicts that the click probability of the article Y is “0.06” when the articles W to Z are arranged as in the list display content LO11.

  As described above, the prediction device 100 predicts that the click probability of the article Y having the CTR of “0.04” is “0.06”. That is, the prediction device 100 predicts that the probability that the article Y is clicked is higher than the CTR when the articles W to Z are arranged as in the list display content LO11. That is, the prediction apparatus 100 uses the information such as CTR of the content arranged in the vicinity of the position where the target content is arranged (hereinafter, may be referred to as “peripheral content”) to increase the click probability of the target content. Can be predicted appropriately. As described above, the prediction device 100 can generate a model that appropriately predicts the click rate of content. Therefore, the prediction device 100 can appropriately predict the click rate of content based on the generated model.

  Note that the prediction device 100 may transmit the predicted click probability of the target content to an external information processing device. For example, the prediction device 100 may transmit the predicted click probability of the target content to the content distribution device 50. The prediction device 100 may be a content distribution device that distributes content. In this case, the prediction device 100 can predict the click probability of the target content and distribute the content (article) arranged in an appropriate arrangement order.

  In addition, the prediction device 100 may consider the influence of peripheral content continuously arranged below the target content arrangement position, not limited to the peripheral content above the target content arrangement position. For example, the prediction apparatus 100 sets the position immediately below the target content as “position-1” and sets the position two below the target content as “position-2” below the position where the target content is positioned. Peripheral content arranged in the site may also be included in the feature.

  Moreover, the prediction apparatus 100 may also consider the influence of the surrounding content arrange | positioned in the left-right direction of the object content, or the diagonal direction according to the arrangement | positioning mode of content. For example, when the prediction apparatus 100 is arranged side by side in the left-right direction, the position to the right of the target content is “position 1” and the position to the left of the target content is “position-1”. Peripheral content arranged continuously in the left-right direction of the content may also be included in the feature.

  In addition, for example, when the content is arranged vertically and horizontally, that is, in a planar shape (lattice shape), the prediction device 100 may set the arrangement position of the target content as “position (0, 0)”. In this case, the prediction device 100 sets the position one position above the target content as “position (0, 1)” and the position one position below the target content as “position (0, −1)”. Peripheral content arranged continuously in the vertical direction of the content may also be included in the feature. Also, the prediction device 100 sets the target content one position to the right as “position (1, 0)” and the one position to the left of the target content as “position (−1, 0)”. Peripheral content continuously arranged in the left-right direction may also be included in the feature. Also, the prediction device 100 sets “position (1, 1)” as the most recent arrangement position on the right diagonal of the target content, and “position (1, 1)” as the arrangement position on the right diagonal next to “position (1, 1)”. 2, 2) ", and peripheral content arranged continuously in an oblique direction of the target content may be included in the feature. Note that the prediction apparatus 100 may also include peripheral content arranged at a predetermined number position by associating a predetermined number with the arrangement position of each content based on the arrangement of the content.

  In addition, the prediction device 100 may determine the peripheral content to be added to the model generation based on the distance from the arrangement position of the target content. For example, the prediction apparatus 100 may generate a model using content located within a distance L from the arrangement position of the target content as peripheral content. In this case, the prediction device 100 generates a model by changing the value of K according to the size of the content. For example, in the case where the content positioned in the range of the distance L upward from the target content is set as the peripheral content, the prediction apparatus 100 has the content position one level higher than the target content within the distance L, and two of the target content When the position of the upper content is far from the distance L, the model is generated with “K = 1”. In addition, for example, when the prediction apparatus 100 uses peripheral content located within the range of the distance L from the target content as the peripheral content, the prediction apparatus 100 determines that “K” = 2 ", the model is generated. Note that various content positions may be used as the content position here, depending on the purpose. For example, the position of the content may be the center position of the content, the position of the part closest to the target content, or the position of the part farthest from the target content. Thus, the prediction apparatus 100 can include any peripheral content in the feature according to the content arrangement mode. That is, the prediction apparatus 100 can generate a model that appropriately predicts the click rate of content according to the arrangement mode of content. Therefore, the prediction device 100 can appropriately predict the click rate of content based on the generated model.

[2. Configuration of prediction device]
Next, the configuration of the prediction apparatus 100 according to the embodiment will be described with reference to FIG. FIG. 2 is a diagram illustrating a configuration example of the prediction device 100 according to the embodiment. As illustrated in FIG. 2, the prediction device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The prediction device 100 includes an input unit (for example, a keyboard and a mouse) that acquires various operations from an administrator of the prediction device 100 and a display unit (for example, a liquid crystal display) for displaying various types of information. May be.

(Communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card). The communication unit 110 is connected to the network by wire or wireless, and transmits and receives information to and from the terminal device 10 and the content distribution device 50.

(Storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. As illustrated in FIG. 2, the storage unit 120 according to the embodiment includes a content information storage unit 121, a distribution log information storage unit 122, and a learning information storage unit 123.

(Content information storage unit 121)
The content information storage unit 121 according to the embodiment stores various types of information related to content (articles). FIG. 3 shows an example of the content information storage unit 121 according to the embodiment. The content information storage unit 121 illustrated in FIG. 3 includes items such as “content ID”, “content”, “CTR”, and “provider ID”.

  “Content ID” indicates identification information for identifying the content. “Content” indicates an article that is content acquired from a provider such as a so-called content provider. FIG. 3 shows an example in which conceptual information such as “article A” is stored in “content”, but in reality, a file indicating character information, a combination of character information and an image, or a storage location thereof. Stores path names and so on. “CTR” indicates the CTR of the corresponding content, that is, the click rate. The “provider ID” indicates identification information for identifying the content provider.

  For example, in the example illustrated in FIG. 3, the article A identified by the content ID “CT11” is content acquired from the provider identified by the provider ID “CP11”. Further, the article A identified by the content ID “CT11” indicates that the CTR is “0.25”.

  The content information storage unit 121 is not limited to the above, and may store various types of information according to the purpose. For example, the content information storage unit 121 may store information regarding the date and time when the content was acquired and the date and time when the content was created. The content information storage unit 121 may store information related to the click probability when each content is optimally arranged.

(Distribution log information storage unit 122)
The distribution log information storage unit 122 according to the embodiment stores various types of information related to the distribution log. FIG. 4 shows an example of the distribution log information storage unit 122 according to the embodiment. In FIG. 4, the distribution log information acquired from each terminal device 10 is stored. For example, the distribution log information storage unit 122 includes distribution log information 122-1, 122-2, and the like. For example, the distribution log information 122-1 is distribution log information acquired from the terminal device 10 of the user U1, and the distribution log information 122-2 is distribution log information acquired from the terminal device 10 of the user U2. Each distribution log information 122-1 and 122-2 shown in FIG. 4 includes items such as “arrangement order”, “content”, and “click”.

  “Arrangement order” indicates the arrangement order of contents distributed to the terminal device 10. “Content” indicates content arranged in a corresponding arrangement order. “Click” indicates whether the corresponding content is clicked by the user.

  For example, in the example illustrated in FIG. 4, in the distribution log information 122-1, the first arrangement order of the contents distributed to the terminal device 10 of the user U <b> 1 is the article A, the second is the article B, and the third is the article. C, indicating that the fourth place is article D. Moreover, it shows that the article A and the article D were clicked in the terminal device 10 of the user U1 as the distribution destination.

  The distribution log information storage unit 122 is not limited to the above, and may store various information according to the purpose. For example, the distribution log information storage unit 122 may store information regarding the time when each content is displayed and the order in which the content is clicked.

(Learning information storage unit 123)
The learning information storage unit 123 according to the embodiment stores various types of information related to learning. For example, the learning information storage unit 123 stores various pieces of information related to the weight corresponding to each feature (arrangement position) derived by learning. For example, the learning information storage unit 123 stores a weight corresponding to “β _{i, 0} ” or the like in the above equation (4). FIG. 5 shows an example of the learning information storage unit 123 according to the embodiment. The learning information storage unit 123 illustrated in FIG. 5 includes items such as “feature” and “weight”.

  “Feature” indicates an object (information) from which a weight for learning is derived. In FIG. 5, “feature” is a content arrangement position. Specifically, “feature” corresponds to the relative position of other content when the position of the target content is “position 0”. For example, “position 1” corresponds to a position one position higher than the arrangement position of the target content, and “position 2” corresponds to a position two positions higher than the arrangement position of the target content. In the example shown in FIG. 5, the weight of “position 0” is “0.7”, the weight of “position 1” is “0.15”, and the weight of “position 2” is “0.05”. Indicates that there is.

  Note that the learning information storage unit 123 is not limited to the above, and may store various types of information according to the purpose. For example, when the influence of the content arranged below the target content placement position is also considered by the prediction apparatus 100, the target content placement position is lower than the target content placement position, such as “position-1” or “position-2”. The weight of the content that is located may also be stored.

(Control unit 130)
Returning to the description of FIG. 2, the control unit 130, for example, various programs (an example of a generation program) stored in a storage device inside the prediction device 100 by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. Is realized by using the RAM as a work area. The control unit 130 is realized by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

  As illustrated in FIG. 2, the control unit 130 includes an acquisition unit 131, a generation unit 132, a prediction unit 133, and a transmission unit 134, and realizes or executes information processing functions and operations described below. . The internal configuration of the control unit 130 is not limited to the configuration illustrated in FIG. 2, and may be another configuration as long as the information processing described later is performed. In addition, the connection relationship between the processing units included in the control unit 130 is not limited to the connection relationship illustrated in FIG. 2, and may be another connection relationship.

(Acquisition part 131)
The acquisition unit 131 acquires an evaluation value related to the prediction target content and an evaluation value related to the peripheral content arranged around the prediction target content among the content continuously arranged on a predetermined display medium. The display medium here may be, for example, the display screen of the terminal device 10. For example, the acquisition unit 131 acquires an evaluation value related to the prediction target content and an evaluation value related to the peripheral content arranged around the prediction target content among the content continuously arranged on a predetermined display screen. In FIG. 1, the acquisition unit 131 acquires from the content information storage unit 121 the CTR of the article Y that is the target content and the CTR of the article Z and the article X that are the peripheral content arranged around (upward) the target content.

  For example, the acquisition unit 131 acquires an evaluation value related to peripheral content continuously arranged in the prediction target content among the content arranged in a predetermined direction. For example, the acquisition unit 131 acquires evaluation values related to peripheral content continuously arranged above the content to be predicted among the content arranged in a predetermined direction. For example, the acquisition unit 131 acquires a click rate (CTR) related to the content to be predicted and a click rate (CTR) related to the surrounding content.

  Further, the acquisition unit 131 acquires log information from the content distribution device 50. For example, the acquisition unit 131 may acquire the CTR of each content from an external information processing apparatus. In this case, the acquisition unit 131 stores the CTR of each content acquired from the external information processing apparatus in the content information storage unit 121.

(Generator 132)
The generation unit 132 generates a model that predicts the probability that the prediction target content is selected based on the evaluation value related to the prediction target content acquired by the acquisition unit 131 and the evaluation value related to the surrounding content. For example, the generation unit 132 uses the CTR of each article stored in the content information storage unit 121, the distribution log information 122-1 and 122-2 acquired from the content distribution device 50, and the like to determine the click probability of the target content. Generate a model to predict. For example, the generation unit 132 uses the above formula (4), the CTR of each article stored in the content information storage unit 121, the distribution log information 122-1 and 122-2 acquired from the content distribution device 50, and the like. The weight of each feature (position) is derived. For example, the generation unit 132 generates a model having “position 0”, that is, a weight “β _{i, 0} ” corresponding to the target content of “0.7” as a model for predicting the click probability of the target content.

(Prediction unit 133)
The prediction unit 133 predicts the click probability of the target content using the model generated by the generation unit 132. For example, the prediction unit 133 uses the above equation (4), the CTR of each article stored in the content information storage unit 121, and the weight of each feature stored in the learning information storage unit 123, and the click probability of the target content. Predict. In FIG. 1, the prediction unit 133 sets the CTR of the article Y, the CTR of the article X, the CTR of the article Z, the weight of “position 0”, the weight of “position 1”, and the weight of “position 2” in the above formula (4). By substituting into the corresponding variable, the click probability of the article Y that is the target content is predicted. In this case, the prediction unit 133 predicts that the click probability of the article Y is “0.06” when the articles W to Z are arranged as in the list display content LO11.

(Transmitter 134)
The transmission unit 134 transmits various information to an external information processing apparatus. For example, the transmission unit 134 may transmit a distribution log information acquisition request to the content distribution device 50. The transmission unit 134 may transmit the click probability of the target content predicted by the prediction unit 133 to an external information processing apparatus. For example, the transmission unit 134 may transmit the click probability of the target content predicted by the prediction unit 133 to the content distribution device 50.

[3. Click probability prediction)
Next, prediction of the click probability based on the arrangement order by the prediction apparatus 100 according to the embodiment will be described with reference to FIG. FIG. 6 is a diagram illustrating an example of a process for predicting a click probability based on the arrangement order according to the embodiment.

  In FIG. 6, a case will be described as an example in which the click probability that varies according to the arrangement of other contents is compared for the list display content LO20 in which the arrangement order of the article D that is the target content TG20 is determined to be fourth. In FIG. 6, the prediction apparatus 100 uses the CTR of each content stored in the content information storage unit 121 illustrated in FIG. 3 and the model (weight of each feature) stored in the learning information storage unit 123. Moreover, the prediction apparatus 100 uses said Formula (4) for prediction of a click probability.

  First, the prediction device 100 determines the arrangement order of articles other than the article D that is the target content TG20 like the list display content LO21 (step S21). In FIG. 6, the prediction device 100 determines that the first place in the list display content LO21 is article A, the second place is article B, the third place is article C, and the fifth place is article W.

  Then, the prediction device 100 predicts the click probability of the article D that is the target content TG20 in the arrangement of the list display content LO21 (step S22). For example, the prediction device 100 predicts (outputs) the prediction result value PV21 of the click probability of the article D by the prediction in step S22. In FIG. 6, the prediction device 100 predicts that the click probability of the article D is “0.05” when each article is arranged as in the list display content LO21.

  Next, the prediction apparatus 100 determines the arrangement order of articles other than the article D that is the target content TG20 as in the list display content LO22 (step S23). In FIG. 6, the prediction apparatus 100 determines that the first place in the list display content LO22 is the article W, the second place is the article B, the third place is the article Y, and the fifth place is the article A.

  Then, the prediction device 100 predicts the click probability of the article D that is the target content TG20 in the arrangement of the list display content LO22 (step S24). For example, the prediction device 100 predicts (outputs) the prediction result value PV22 of the click probability of the article D by the prediction in step S24. In FIG. 6, the prediction device 100 predicts that the click probability of the article D is “0.04” when each article is arranged as in the list display content LO22.

  Further, the prediction device 100 determines the arrangement order of articles other than the article D that is the target content TG20 as in the list display content LO23 (step S25). In FIG. 6, the prediction device 100 determines the first place in the list display content LO23 as the article C, the second place is the article X, the third place is the article A, the fifth place is the article Y, and the like.

  Then, the prediction device 100 predicts the click probability of the article D that is the target content TG20 in the arrangement of the list display content LO23 (step S26). For example, the prediction device 100 predicts (outputs) the prediction result value PV23 of the click probability of the article D by the prediction in step S26. In FIG. 6, the prediction device 100 predicts that the click probability of the article D is “0.07” when each article is arranged as in the list display content LO23.

  Then, the prediction device 100 compares the prediction result values PV21 to PV23 to increase the click probability of the article D that is the target content TG20 among the list display contents LO21 to LO23. Can be determined to be most appropriate.

[4. (Model generation)
Next, generation of a model in the prediction device 100 according to the embodiment will be described with reference to FIG. FIG. 7 is a flowchart illustrating an example of a process procedure of model generation according to the embodiment.

  First, in the example illustrated in FIG. 7, the prediction device 100 acquires the CTR of each content (step S <b> 101). For example, the prediction device 100 acquires the CTR of each content from the content information storage unit 121. Moreover, the prediction apparatus 100 acquires distribution log information (step S102). For example, the prediction device 100 acquires distribution log information from the content distribution device 50 or the distribution log information storage unit 122.

  And the prediction apparatus 100 produces | generates the model based on arrangement | positioning order and CTR of each content (step S103). For example, the prediction device 100 generates a model based on the click results of content in each arrangement order and the CTR of each content. For example, the prediction device 100 generates the weight of each feature as a model using the above equation (4).

[5. (Click probability prediction of target content)
Next, prediction of the click probability in the prediction device 100 according to the embodiment will be described using FIG. FIG. 8 is a flowchart illustrating an example of the processing procedure of the click probability prediction according to the embodiment.

  As shown in FIG. 8, the prediction apparatus 100 acquires CTR of each content (step S201). For example, the prediction device 100 acquires the CTR of each content from the content information storage unit 121. Further, the prediction device 100 acquires the content arrangement order (step S202). Further, the prediction device 100 acquires the designation of the content to be predicted (step S203). Moreover, the prediction apparatus 100 acquires a model (step S204). For example, the prediction device 100 acquires a model (weight) from the learning information storage unit 123.

  Thereafter, the prediction device 100 predicts the probability (click probability) that the content to be predicted is clicked based on the CTR of each content, the content arrangement order, and the model (step S205).

[6. (Modification)
The prediction device 100 according to the above-described embodiment may be implemented in various different forms other than the above-described embodiment. For example, the prediction device 100 may generate a model in which information related to the content content such as the degree of association with the target content is added to the content evaluation value. This point will be described with reference to FIGS. Further, in the following modified examples, the same portions as those in the embodiment are denoted by the same reference numerals, and redundant description is omitted.

[6-1. Configuration of prediction device]
Next, the configuration of the prediction apparatus 200 according to the modification will be described with reference to FIG. FIG. 9 is a diagram illustrating a configuration example of a prediction device according to a modification. In addition, the same structure as the prediction apparatus 100 in the prediction apparatus 200 attaches | subjects the same number, and abbreviate | omits description. As illustrated in FIG. 9, the prediction device 200 includes a communication unit 110, a storage unit 220, and a control unit 230.

(Storage unit 220)
The storage unit 220 is realized by, for example, a semiconductor memory element such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk. As illustrated in FIG. 9, the storage unit 220 according to the modification includes a content information storage unit 221, a distribution log information storage unit 122, and a learning information storage unit 223.

(Content information storage unit 221)
The content information storage unit 221 according to the modification stores various types of information related to content (articles). FIG. 10 shows an example of the content information storage unit 221 according to the modification. The content information storage unit 221 illustrated in FIG. 10 includes items such as “content ID”, “content”, “CTR”, “provider ID”, and “category”.

  “Category” indicates content classification based on the content of the corresponding content. For example, in the example illustrated in FIG. 10, the article A identified by the content ID “CT11” is classified into the category “sports”.

  The content information storage unit 221 is not limited to the above, and may store various types of information according to the purpose. For example, the content information storage unit 221 may store information on feature words (important keywords) included in each content. The content information storage unit 221 may store a score for each category of content. For example, for the article A identified by the content ID “CT11”, the content information storage unit 221 has a category of “80” for sports, “30” for entertainment, “40” for region, “5” for economy, etc. The score may be stored.

(Learning information storage unit 223)
The learning information storage unit 223 according to the modification stores various types of information related to learning. For example, the learning information storage unit 223 stores various pieces of information related to the weights corresponding to the features (arrangement positions) derived by learning. FIG. 11 shows an example of the learning information storage unit 223 according to the modification. The learning information storage unit 223 illustrated in FIG. 11 includes items such as “feature”, “first weight”, and “second weight”.

  “Feature” indicates an object (information) from which a weight for learning is derived. In FIG. 11, “feature” is a content arrangement position. Specifically, “feature” corresponds to the relative position of other content when the position of the target content is “position 0”. For example, “position 1” corresponds to a position one position higher than the arrangement position of the target content, and “position 2” corresponds to a position two positions higher than the arrangement position of the target content.

The “first weight” indicates a weight used for a term related to CTR. For example, in the following equation (5), this corresponds to the weight “γ _{i, k} ” of “CTR (C (i−k))”. The “second weight” indicates a weight used for a term related to the degree of association of the content with the target content. For example, it corresponds to the weight “ω _{i, k} ” of “REV (C (i−k))” in the following equation (5).

  In the example illustrated in FIG. 11, the first weight of “position 0” is “0.6”, and the second weight is “−”. In this case, the content at “position 0” is the target content itself, and the relevance of the content is not considered. For example, the second weight of “position 0” may be “0”. Note that the second weight of “position 0” may be calculated as a weight related to the importance of the target content. In addition, the first weight of “position 1” is “0.05”, and the second weight is “0.5”. The first weight of “position 2” is “0.01” and the second weight is “0.2”.

  Note that the learning information storage unit 223 is not limited to the above, and may store various types of information according to the purpose. For example, when the influence of the content arranged below the target content placement position is also considered by the prediction apparatus 200, the target content placement position is lower than the target content placement position such as “position-1” or “position-2”. The weight of the content that is located may also be stored.

(Control unit 230)
Returning to the description of FIG. 9, the control unit 230 executes various programs (corresponding to an example of a generation program) stored in the storage device inside the prediction device 200 using the RAM as a work area, for example, by a CPU, MPU, or the like. Is realized. The control unit 230 is realized by an integrated circuit such as ASIC or FPGA, for example.

  As illustrated in FIG. 9, the control unit 230 includes an acquisition unit 131, a generation unit 232, a prediction unit 233, and a transmission unit 134, and implements or executes information processing functions and operations described below. . The internal configuration of the control unit 230 is not limited to the configuration illustrated in FIG. 9, and may be another configuration as long as the information processing described below is performed. Further, the connection relationship between the processing units included in the control unit 230 is not limited to the connection relationship illustrated in FIG. 9, and may be another connection relationship.

(Generator 232)
The generation unit 232 generates a model that predicts the probability that the prediction target content is selected based on the degree of relevance of the peripheral content acquired by the acquisition unit 131 with the prediction target content. In this case, the generation unit 232 may calculate the degree of association between the target content and the surrounding content using various techniques as appropriate. For example, the generation unit 232 may calculate the degree of association between the target content and the surrounding content based on the category, feature word, or the like of each content stored in the content information storage unit 221. For example, the generation unit 232 may calculate the degree of association between the target content and the surrounding content by calculating the cosine similarity between the target content and the surrounding content.

  For example, the generation unit 232 uses the CTR of each article stored in the content information storage unit 221, the degree of association between the calculated target content and the surrounding content, the distribution log information 122-1, 122-2, and the like. Generate a model that predicts the click probability of content. For example, the generation unit 232 derives the weight of each feature (position) using the following equation (5) or the like.

“P (Click (i))” in the left side of the above equation (5) corresponds to the click probability of the target content whose arrangement order is i-th. For example, “P (Click (i))” takes a value from 0 to 1. The first term in “Σ” on the right side of the above equation (5) is the same as that in the above equation (4) except for the difference in the weight symbol between the weight “γ _{i, k} ” and the weight “β _{i, k} ”. It is equivalent to the right side.

In addition, the second term in “Σ” on the right side of the above equation (5) is “REV (C (ik))” indicating the degree of association with the target content of the ikth content. It is obtained by multiplying a predetermined weight “ω _{i, k} ”.

  The generation unit 232 learns by changing the values of “K = 1”, “K = 2”, “K = 3”, and “K” in various ways, and thereby a plurality of models corresponding to each “K”. Can be generated. As described above, the generation unit 232 generates a model that predicts the click probability of the target content based on the above formula (5) and the degree of relevance of the surrounding content to the prediction target content.

(Prediction unit 233)
The prediction unit 233 predicts the click probability of the target content using the model generated by the generation unit 232. For example, the prediction unit 233 uses the above formula (5), the CTR of each article stored in the content information storage unit 221, the degree of association between the target content and the surrounding content, and each stored in the learning information storage unit 223. The click probability of the target content is predicted using the feature weight.

[7. effect〕
As described above, the prediction devices 100 and 200 according to the embodiment and the modification include the acquisition unit 131 and the generation units 132 and 232. The acquisition unit 131 is arranged around the evaluation value regarding the content to be predicted (in the embodiment, “target content”; the same applies hereinafter) among the content continuously arranged on a predetermined display medium and the content to be predicted. Get evaluation values for peripheral content. The generation units 132 and 232 generate a model that predicts the probability that the prediction target content is selected based on the evaluation value related to the prediction target content acquired by the acquisition unit 131 and the evaluation value related to the surrounding content.

  As a result, the prediction devices 100 and 200 according to the embodiment and the modified example generate a model in consideration of the evaluation value of the content arranged around the target content, thereby appropriately predicting the click rate of the content. Can be generated. Therefore, the prediction devices 100 and 200 can appropriately predict the click rate of content based on the generated model.

  Further, in the prediction devices 100 and 200 according to the embodiment and the modification, the acquisition unit 131 calculates evaluation values related to peripheral content continuously arranged in the prediction target content among the content arranged in a predetermined direction. get.

  As a result, the prediction devices 100 and 200 according to the embodiment and the modified example appropriately predict the click rate of the content by generating the model in consideration of the evaluation value of the content continuously arranged in the target content. Can be generated. Therefore, the prediction devices 100 and 200 can appropriately predict the click rate of content based on the generated model.

  Further, in the prediction devices 100 and 200 according to the embodiment and the modification, the acquisition unit 131 evaluates the peripheral content continuously arranged above the prediction target content among the content arranged side by side in a predetermined direction. Get the value.

  As a result, the prediction devices 100 and 200 according to the embodiment and the modified example appropriately predict the click rate of the content by generating the model in consideration of the evaluation value of the content continuously arranged above the target content. A model can be generated. Therefore, the prediction devices 100 and 200 can appropriately predict the click rate of content based on the generated model.

  In the prediction devices 100 and 200 according to the embodiment and the modification, the acquisition unit 131 acquires a click rate related to the prediction target content as the evaluation value related to the prediction target content, and relates to the peripheral content as the evaluation value related to the peripheral content. Get click rate.

  Accordingly, the prediction devices 100 and 200 according to the embodiment and the modification can appropriately predict the click probability of the target content by using the CTR of the target content and the surrounding content. That is, the prediction devices 100 and 200 can generate a model that appropriately predicts the click rate of content. Therefore, the prediction devices 100 and 200 can appropriately predict the click rate of content based on the generated model.

  In the prediction device 200 according to the modification, the acquisition unit 131 acquires the degree of relevance of the surrounding content with the content to be predicted. Further, the generation unit 232 generates a model that predicts the probability that the prediction target content is selected based on the degree of relevance of the surrounding content acquired by the acquisition unit 131 with the prediction target content.

  Thereby, the prediction apparatus 200 which concerns on a modification can produce | generate the model which considered the relevance degree with object content. That is, the prediction device 200 can generate a model that appropriately predicts the click rate of content. Therefore, the prediction device 200 can appropriately predict the click rate of content based on the generated model.

[8. Hardware configuration)
The prediction apparatus 100 according to the embodiment described above is realized by a computer 1000 having a configuration as shown in FIG. 12, for example. FIG. 12 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the prediction device. The computer 1000 includes a CPU 1100, RAM 1200, ROM 1300, HDD 1400, communication interface (I / F) 1500, input / output interface (I / F) 1600, and media interface (I / F) 1700.

  The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400 and controls each unit. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started up, a program depending on the hardware of the computer 1000, and the like.

  The HDD 1400 stores programs executed by the CPU 1100, data used by the programs, and the like. The communication interface 1500 receives data from other devices via the network N and sends the data to the CPU 1100, and transmits data determined by the CPU 1100 to other devices via the network N.

  The CPU 1100 controls an output device such as a display and a printer and an input device such as a keyboard and a mouse via the input / output interface 1600. The CPU 1100 acquires data from the input device via the input / output interface 1600. Further, the CPU 1100 outputs the determined data to the output device via the input / output interface 1600.

  The media interface 1700 reads a program or data stored in the recording medium 1800 and provides it to the CPU 1100 via the RAM 1200. The CPU 1100 loads such a program from the recording medium 1800 onto the RAM 1200 via the media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase change rewritable disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. Etc.

  For example, when the computer 1000 functions as the prediction device 100 according to the embodiment, the CPU 1100 of the computer 1000 implements the function of the control unit 130 by executing a program loaded on the RAM 1200. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them. However, as another example, these programs may be acquired from other devices via the network N.

  As described above, some of the embodiments of the present application have been described in detail with reference to the drawings. It is possible to implement the present invention in other forms with improvements.

[9. Others]
In addition, among the processes described in the above embodiment, all or part of the processes described as being automatically performed can be performed manually, or the processes described as being performed manually can be performed. All or a part can be automatically performed by a known method. In addition, the processing procedures, specific names, and information including various data and parameters shown in the document and drawings can be arbitrarily changed unless otherwise specified. For example, the various types of information illustrated in each drawing is not limited to the illustrated information.

  Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured.

  Further, the above-described embodiments can be appropriately combined within a range in which processing contents are not contradictory.

  In addition, the above-mentioned “section (module, unit)” can be read as “means” or “circuit”. For example, the acquisition unit can be read as acquisition means or an acquisition circuit.

1 Prediction System 100 Prediction Device (Generation Device)
121 content information storage unit 122 delivery log information storage unit 123 learning information storage unit 130 control unit 131 acquisition unit 132 generation unit 133 prediction unit 134 transmission unit

Claims (7)

An obtaining unit that obtains an evaluation value relating to a prediction target content and an evaluation value relating to peripheral content arranged around the prediction target content among the content continuously arranged in a predetermined display medium;
A generating unit that generates a model for predicting a probability that the prediction target content is selected based on the evaluation value related to the prediction target content acquired by the acquisition unit and the evaluation value related to the surrounding content;
A generating apparatus comprising: The acquisition unit
The generation apparatus according to claim 1, wherein an evaluation value related to the peripheral content continuously arranged in the prediction target content among the content arranged in a predetermined direction is acquired. The acquisition unit
The generation apparatus according to claim 2, wherein evaluation values related to the peripheral content continuously arranged above the prediction target content among the content arranged side by side in a predetermined direction are acquired. The acquisition unit
The click rate related to the prediction target content is acquired as the evaluation value related to the prediction target content, and the click rate related to the peripheral content is acquired as the evaluation value related to the peripheral content. The production | generation apparatus of any one. The acquisition unit
Obtaining a degree of relevance of the surrounding content with the content to be predicted;
The generator is
The model for predicting the probability that the content to be predicted is selected is generated based on the degree of relevance of the content to be predicted of the surrounding content acquired by the acquisition unit. 5. The generating apparatus according to any one of 4. A generation method executed by a computer,
An obtaining step of obtaining an evaluation value relating to a prediction target content and an evaluation value relating to peripheral content arranged around the prediction target content among the content continuously arranged in a predetermined display medium;
A generation step of generating a model for predicting a probability that the prediction target content is selected based on the evaluation value related to the prediction target content acquired in the acquisition step and the evaluation value related to the surrounding content;
A generation method comprising: An acquisition procedure for acquiring an evaluation value related to a prediction target content and an evaluation value related to peripheral content arranged around the prediction target content among the content continuously arranged in a predetermined display medium;
A generation procedure for generating a model for predicting a probability that the prediction target content is selected based on the evaluation value related to the prediction target content acquired by the acquisition procedure and the evaluation value related to the surrounding content;
A program for causing a computer to execute.

Images