JP7319443B1 - Information processing system, computer program, and information processing method - Google Patents

Abstract

Kind Code: A1 To provide a technology capable of placing highly effective advertisements. A first advertisement amount and a second advertisement amount are determined, an advertisement is placed on a first medium based on the first advertisement amount, and a second advertisement is placed based on the second advertisement amount. Advertisement placement on the medium is instructed (S250). A first track record, which is a track record for exposure of reserved advertisements, is determined (S260). A second track record, which is a track record for exposure of non-reserved advertisements, is determined (S270). An exposure schedule for non-exposure advertisements waiting to be exposed is determined (S290). With respect to a plurality of time points, the first ad placement amount and the second ad placement amount are based on the possible remaining ad placement amount, the exposure schedule of the unexposed advertisement, the first performance, and the second performance for each time point. determined (S240). [Selection drawing] Fig. 4

Description

The present disclosure relates to information processing systems and methods.

Conventionally, there is known a technique for integrating broadcasting and network advertisements, in particular, a technique for automatically synchronizing the timing of advertisements, optimizing their display mode, and obtaining high advertising effectiveness (for example, Patent Document 1). reference).

WO2008/081596

The allocation of advertising budgets to selectable media is also important to achieve effective advertising. Examples of advertising include scheduled advertising and programmatic advertising.
A reserved advertisement is a type of advertisement that requires prior reservation for distribution, and is distributed according to a schedule determined at the time of reservation. The distribution here includes not only digital distribution but also broadcasting. Examples of scheduled advertising include television commercials, which are advertisements through television broadcasts.

Programmatic advertising is a type of advertising whose distribution conditions can be changed in real time, unlike reserved advertising. Examples of programmatic advertising include advertising through communication networks, particularly the Internet. Programmatic advertisements can be distributed at timings specified by advertisers without prior reservations.

In reserved advertising, while there are highly effective advertising slots, it is necessary to decide in advance where to place the advertisement. On the other hand, programmatic advertising can be delivered in real time, but the advertising effect may be lower than reserved advertising. As described above, there are merits and demerits in reserved advertisements and programmatic advertisements.

Therefore, according to one aspect of the present disclosure, under conditions where the total amount of possible advertisements is determined, through determination of the amount of advertisements for reserved advertisements and non-reserved advertisements over multiple times, allocation with high advertisement effect It is desirable to be able to provide technology capable of placing both reserved and non-reserved advertisements.

According to one aspect of the present disclosure, an information processing system is provided. The information processing system includes an advertisement placement instruction section, a first performance determination section, a second performance determination section, and a schedule determination section. The advertisement placement instruction unit determines a first placement amount and a second placement amount, and places an advertisement on the first medium based on the first placement amount and a second advertisement placement based on the second placement amount. It is configured to instruct advertisement placement on the medium of The first placement amount is the placement amount for reserved advertisements exposed through the first medium. The second placement amount is the placement amount for non-reserved advertisements exposed through the second medium.

The first performance determination unit is configured to determine a first performance that is a performance regarding exposure of the reserved advertisement exposed through the first medium. The second performance determination unit is configured to determine a second performance, which is a performance regarding exposure of the non-reserved advertisement exposed through the second medium. The schedule discriminating unit is configured to discriminate an exposure schedule of non-exposure advertisements waiting for exposure among reserved advertisements.

According to one aspect of the present disclosure, the total amount of possible placements for reserved advertisements and non-reserved advertisements is determined in advance. The placement instruction unit determines the first placement amount and the second placement amount based on the possible remaining placement amount, the exposure schedule of the unexposed advertisement, the first performance, and the second performance for each of the plurality of time points. It is configured to determine the amount of two advertisements.

It is possible to use a system that determines the amount of reserved and non-reserved advertisements to be displayed based on the exposure performance of reserved and non-reserved advertisements and also based on the exposure schedule of non-reserved advertisements. Under the condition that the total amount of advertisement is determined, it is possible to place reserved advertisement and non-reserved advertisement with high advertisement effect allocation by determining the advertisement amount of reserved advertisement and non-reserved advertisement over multiple times. .

According to one aspect of the present disclosure, the first amount of the total possible amount of advertisements can be predetermined as the amount of advertisements for reservation-type advertisements. A second amount of the total amount may be defined as a shared placement amount for scheduled and non-reserved advertisements. The placement instruction unit is given the power to determine the allocation of the second amount of reserved advertisements and non-reserved advertisements.

According to one aspect of the present disclosure, there may be cases where the placement of reserved advertisements corresponding to at least a portion of the first amount is completed at the first point in time that is earlier than the plurality of points in time. In this case, the above-mentioned exposure schedule may include an exposure schedule of non-exposure advertisements that are scheduled advertisements that have been published at the beginning and are waiting for exposure.

The placement instruction unit can determine the first placement amount and the second placement amount for each time point, taking into account the exposure schedule of the unexposed advertisements including the reserved advertisements that have already been placed at the first time point.

By using this information processing system, the user can understand the minimum amount of advertisements for reserved advertisements, and from the state where the initial advertisement of reserved advertisements has been completed, the information processing system can display highly effective reserved advertisements and non-reserved advertisements. can be calculated. That is, the information processing system controls the minimum amount of advertisements for reservation type advertisements, and determines the amount of advertisements for reservation type advertisements and non-reservation type advertisements over multiple times. It is possible to instruct the publication of

According to one aspect of the present disclosure, the first placement amount may be the placement amount of the reserved advertisement. The second placement amount may be the placement amount of the non-reserved advertisement. The total possible placement amount may be an advertisement placement budget including reserved advertisements and non-reserved advertisements. Therefore, by using the information processing system according to one aspect of the present disclosure, the user can realize posting of highly effective reserved advertisements and non-reserved advertisements within a limited budget.

According to one aspect of the present disclosure, the ad placement instruction unit can determine the first ad placement amount and the second ad placement amount for each time point according to the dynamic optimization algorithm. According to one aspect of the present disclosure, dynamic optimization algorithms may include at least one of reinforcement learning, contextual bandit algorithms, and Kalman filters.

According to one aspect of the present disclosure, the placement instruction unit can determine the first placement amount and the second placement amount for each time point according to reinforcement learning or a contextual bandit algorithm. Determining the placement amount through such an algorithm enables users to place highly effective reserved ads and non-reserved ads.

According to one aspect of the present disclosure, for each time point, the placement instruction unit sets the first placement amount and the second placement amount as actions related to advertising placement at the corresponding time point based on the state or context at the corresponding time point. can decide. A state or context may be defined with a possible remaining placement volume, an unexposed ad exposure schedule, a first performance, and a second performance.

Based on the first performance and the second performance for each time point, the advertisement placement instruction unit calculates the advertisement effect of the reserved advertisement and the non-reserved advertisement newly exposed by the advertisement placement up to the corresponding time point as a reward for the action. can be determined as

The advertisement placement instruction unit may update the state or context in consideration of the advertisement placement at the corresponding time and the first performance and the second performance at the corresponding time for each time. The placement instruction unit can update the policy regarding action selection based on the reward at each point in time.

According to one aspect of the present disclosure, the ad placement instruction unit uses reinforcement learning in which states, rewards, and actions are defined, for each time point, the first ad placement amount and the second is determined as behavior.

According to one aspect of the present disclosure, a state may be defined with a possible remaining placement volume, a first performance, a second performance, and a schedule of unexposed ads at the corresponding time. A reward may be defined using the advertising effectiveness at the corresponding time determined from the first performance and the second performance.

According to one aspect of the present disclosure, the placement instruction unit uses a contextual bandit algorithm in which context, reward, and behavior are defined, for each time point, the first placement amount and It can be configured to determine the second ad placement amount as an action.

A context may be defined with the remaining placement volume, first performance, second performance, and unexposed ad schedule at the corresponding time. A reward may be defined using the advertising effectiveness at the corresponding time determined from the first performance and the second performance.

According to one aspect of the present disclosure, the ad placement instruction unit uses, for example, a state space model such as a Kalman filter to determine, for each time point, the first ad placement amount and the second ad placement amount at the corresponding time point. can be configured.

A state-space model can be a model that defines the relationship between state quantities, observable quantities, and input quantities. The state-space model can be a model that contains information about the exposure schedule of unexposed advertisements. The input amount may be an amount related to advertisement placement defined using a first placement amount and a second placement amount.

The state quantity may be a state quantity that changes according to the placement of advertisements. The state quantity can be defined using the possible remaining ad placement amount, the first performance, and the second performance. The observable quantity may be a quantity that defines the advertising effectiveness of the posted scheduled and non-reserved advertisements determined based on the first performance and the second performance.

According to one aspect of the present disclosure, a computer program for causing a computer to implement functions as an advertisement placement instruction unit, a first performance determination unit, a second performance determination unit, and a schedule determination unit in the information processing system described above. may be provided. The computer program may be recorded on a computer-readable recording medium and provided.

According to one aspect of the present disclosure, the following information processing method may be provided. The information processing method determines a first ad placement amount and a second ad placement amount, places an advertisement on the first medium based on the first ad placement amount, and determines a second ad placement amount based on the second ad placement amount. may include directing the placement of advertisements on the media of The first ad placement amount may be an ad placement amount for reserved advertisements exposed through the first medium. The second placement amount may be the placement amount for non-reserved advertisements exposed through the second medium.

The information processing method may include determining a first performance that is a performance relating to exposure of the scheduled advertisements exposed through the first medium. The information processing method may include determining a second performance that is a performance relating to the exposure of non-reserved advertisements exposed through the second medium.

The information processing method may include determining an exposure schedule for non-exposure advertisements waiting to be exposed among the scheduled advertisements. The total amount of possible placements for reserved advertisements and non-reserved advertisements can be predetermined.

Instructing the placement of advertisements includes, for each of a plurality of time points, the first placement of advertisements based on the possible remaining amount of advertisements, the exposure schedule of unexposed advertisements, the first performance, and the second performance. Determining an amount and a second placement amount.

By using such a method, similarly to the information processing system, under the condition that the total amount of possible advertisements is determined, through determination of the amount of advertisements for reserved advertisements and non-reserved advertisements over multiple times, advertisement effectiveness is high It is possible to implement reserved advertisements and non-reserved advertisements.

According to one aspect of the present disclosure, the information processing method may be performed by a computer. According to one aspect of the present disclosure, the first ad spend and the second ad spend may be determined for each time point according to dynamic optimization algorithms such as reinforcement learning, contextual bandit algorithms, and Kalman filters. In addition, an information processing method corresponding to the information processing system described above may be provided.

1 is a block diagram showing the configuration of an information processing system; FIG. 4 is a flow chart showing a first advertisement-related process executed by a processor in the first embodiment; It is explanatory drawing regarding the use of a reserve budget. 10 is a flow chart showing second placement-related processing executed by the processor in the first embodiment. FIG. 10 is a flow chart showing a first advertisement-related process executed by a processor in the second embodiment; FIG. FIG. 11 is a flow chart showing a second advertisement-related process executed by a processor in the second embodiment; FIG.

Exemplary embodiments of the present disclosure are described below with reference to the drawings.
[First embodiment]
The information processing system 1 of this embodiment is configured by installing a computer program specific to this embodiment in a general-purpose computer system. The information processing system 1 shown in FIG. 1 includes a processor 11 , memory 12 , storage 13 , display 15 , input device 17 , media reader/writer 18 and communication device 19 .

Processor 11 is configured to execute processing according to a computer program stored in storage 13 . Memory 12 includes RAM. Memory 12 is used as a work area when processor 11 executes processing according to a computer program. The memory 12 temporarily stores computer programs and data read from the storage 13 .

The storage 13 stores computer programs and various data. One of the computer programs stored in the storage 13 is that the processor 11 maximizes the advertising effect within the specified advertising budget, by adjusting the amount of advertisements for each of the reserved advertisements and the non-reserved advertisements, A computer program is included for stepwise determination based on a dynamic optimization algorithm. Examples of storage 13 include hard disk drives and solid state drives.

The display 15 is configured to display various information to the user. The display 15 is, for example, a liquid crystal display. The input device 17 is configured to input operation signals from the user to the processor 11 . The input device 17 includes a user-operable keyboard and pointing device.

The media reader/writer 18 is configured to be able to read information recorded on a recording medium such as a memory card and write new information to the recording medium. Communication device 19 is controlled by processor 11 and configured to communicate with external systems within a local area network and/or within a wide area network.

The processor 11 executes advertisement-related processing based on computer programs stored in the storage 13 according to commands input by the user through the input device 17 . The advertisement related processing includes a first advertisement related processing (see FIG. 2) and a second advertisement related processing (see FIG. 4). After executing the first advertisement-related process, the processor 11 executes a second advertisement-related process according to a further command input by the user through the input device 17 .

In the first placement-related process, a minimum placement budget for advertisements for a plurality of media is determined under specified conditions, and a placement plan is output. In the second placement-related processing, the amount of placements on a plurality of media is determined based on the minimum placement budget determined in the placement plan, and the placement of advertisements is instructed.

Multiple media include multiple media that deliver reserved advertisements (hereinafter referred to as “reserved media”) and multiple media that deliver non-reserved advertisements (hereinafter referred to as “non-reserved media”). be The distribution of advertisements here means the distribution of advertisements, and should be broadly understood to include not only the distribution of advertisements through the Internet, but also the broadcasting of advertisements through television broadcasting, radio broadcasting, and the like. Advertisements are exposed to consumers through media. Advertisements may relate to, for example, goods sold by the advertiser or services provided by the advertiser.

Non-reserved advertisements, for example, are programmatic advertisements that are distributed almost simultaneously with the placement of advertisements, and are digital advertisements that are distributed over the Internet. A reserved advertisement as an example is an advertisement distributed in the form of a television commercial (CM) through television broadcasting. TV commercials include spot commercials. A reservation-type advertisement as a TV commercial is broadcast in a predetermined broadcast frame according to a schedule determined at the time the advertisement is placed. Below, an example will be described in which a television commercial is assumed as a reserved advertisement and a digital advertisement is assumed as a non-reserved advertisement.

In the first step of the advertisement related processing, the user gives the following information as execution conditions.
Advertisement budget Campaign period Advertisement effect estimation model Target information Penalty When the processor 11 starts the first advertisement placement related process (see FIG. 2), in S110, through the storage 13 or the input device 17 to obtain this information.

The advertising budget is the total advertising budget for advertisements including reserved advertisements and non-reserved advertisements that can be used in campaigns. The campaign period corresponds to the period during which advertising activities are performed. The advertising effectiveness estimation model is a mathematical model for estimating advertising effectiveness from an index related to advertising exposure results.

Ad effectiveness may be quantified as reach, which is the number of times an ad reaches a target consumer. Alternatively, advertising effectiveness can be quantified as a conversion rate. Alternatively, the advertising effect can be expressed by quantifying the magnitude of the brand lift. Advertisement effectiveness may be expressed numerically as a score that combines multiple indices such as the number of reach, conversion rate, and brand lift.

Target information is information that specifies a consumer segment for advertising targets. Advertising effectiveness is estimated as the effectiveness for the consumer segment specified by the targeting information. Consumer segments are specified, for example, by gender and age group.

The penalty is a parameter for suppressing the placement-related processing from instructing placement of an advertisement exceeding a designated advertisement budget, which is the total possible amount of placement. The lower the advertiser's tolerance for over-budget ad placements, the greater the penalty is specified in order to discourage over-budget ad placement directives. As can be understood from this explanation, the total amount of “possible” ad placement means the total amount of adverts that can be placed at least up to the corresponding amount, and placing more than the “possible” amount is prohibited. Note that the expression does not imply that

In subsequent S120, the processor 11 selects, according to a reinforcement learning algorithm, an action regarding advertisement placement at time t=0 before the start of the campaign period. Specifically, based on a given initial state s (t=0), the processor 11 performs a plurality of reserved media and a plurality of non-reserved media as actions at time t=0 before the start of the campaign period. Minimum advertising budget for media b ₁ ^R (0), b ₂ ^R (0), ..., b _m ^R (0), ..., b _M ^R (0), b ₁ ^P (0), b ₂ ^P (0) _{, . . . , b n P} ⁽ 0), ^. The advertisement budget here is the budget for purchasing the advertising space.

b _m ^R (0) is the ad placement budget for the m-th reservation-type medium among the plurality of reservation-type media. M is the number of reservation-type media when the reservation-type media exist from the first reservation-type medium to the M-th reservation-type medium. M takes an integer value of 1 or more, and m takes an integer value from 1 to M.

b _n ^P (0) is the ad placement budget for the n-th non-reserved medium among the plurality of non-reserved mediums. N is the number of non-reserved media when non-reserved media exist from the first non-reserved medium to the Nth non-reserved medium. N takes an integer value of 1 or more, and n takes an integer value from 1 to N.

The initial state s(0) is defined by the advertising budget. In the reinforcement learning algorithm, based on the specified advertising budget information and further based on the pre-learned policy regarding action selection, the minimum advertisement budget b ₁ ^R ( 0), b ₂ ^R (0), . . . , b _{MR (} 0), ^b ₁ ^P (0), ^b ₂ ^P (0), .

^In subsequent S130, the processor 11 sets ^the minimum advertising budget _b1R (0), _b2R (0), ^... , _bMR ⁽ 0), _b1P ( ⁰ ) _, b ₂ ^P (0), .

The advertising budget is as described above. The reserve budget is calculated from the advertising budget as the minimum advertising budget b ₁ ^R (0), b ₂ ^R (0), ..., b _M ^R (0), b ₁ ^P (0), b ₂ ^P (0), ..., corresponds to the sum of b _N ^P (0) subtracted. When the advertising budget is assigned the variable B and the reserve budget is assigned the variable Br, the reserve budget is Br=B−(b ₁ ^R (0)+b ₂ ^R (0)+ . . . +b _M ^R (0)+b ₁ ^P (0 ⁾ + _b2P (0)+...+ _bNP (0)) ^. In other words, the reserve budget corresponds to the advertising budget that is not allocated to either the reserved medium or the non-reserved medium.

At S<b>130 , the processor 11 can output the advertisement placement plan by display through the display 15 . The processor 11 can output the publication plan to the storage 13 as a data file that can be viewed by the user. The processor 11 can store in the storage 13 the information of the advertisement plan necessary for the second advertisement related processing. After that, the processor 11 terminates the first publication-related processing.

Advertisement plans are presented to advertisers before the start of the campaign period. When the advertisement plan _is adopted by the advertiser, the minimum advertisement budget b ₁ ^R (0 ⁾ , b ₂ ^R (0), . The number of advertising spaces is purchased, and the work of placing advertisements for each reserved medium is completed. This advertisement work is performed manually by the person in charge of the advertising agency, or automatically through the media relay system 31 .

After that, the distribution schedule for the posted reservation-type advertisement is recorded in the storage 13 . Processor 11 can automatically acquire the delivery schedule through relay system 31 . Alternatively, the person in charge of the advertising agency can acquire the delivery schedule from the company on the medium side and input it to the information processing system 1 through the input device 17 or the media reader/writer 18 .

After the first advertisement-related processing ends, when the user inputs an instruction to execute processing based on the advertisement placement plan through the input device 17, the processor 11 starts the second advertisement-related processing (see FIG. 4). In the second advertisement related processing, the advertisement budget B according to the advertisement placement plan, the minimum advertisement budget of each medium b ₁ ^R (0), b ₂ ^R (0), . . . , b _M ^R (0), b ₁ ^P (0) , b ₂ ^P (0) ^, _.

At the start of the campaign period, which is the start of the second ^{advertisement} _- related processing, reservations based on the minimum advertisement budgets b ₁ ^R (0), b ₂ ^R (0), . The placement of advertisements on printed media has been completed. On the other hand, no advertisements are placed on the non-reservation type media based on the minimum advertisement budgets b ₁ ^P (0), b ₂ ^P (0), . . . , b _N ^P (0). Ad placements on non-reservation type media based on the minimum placement budgets b ₁ ^P (0), b ₂ ^P (0), . . . , b _N ^P (0) are performed after the campaign period starts.

In this embodiment, as shown in FIG. 3, the reserve budget Br is used step by step for placing reserved advertisements or non-reserved advertisements during the campaign period. According to FIG. 3, before the start of the campaign period, the advertisement budget for reservation type media is "40", the advertisement budget for non-reservation type media is "20", and the reserve budget is "10". Numerical values are examples, and their units are arbitrary.

In the middle of the campaign period, "4" of the reserve budget of "10" are allocated to reserved media and "2" are allocated to non-reserved media, and additional advertisements are placed on each medium. At a still later point, "2" of the remaining reserve budget of "4" will be allocated to reserved media, and a further "2" will be allocated to non-reserved media, with additional ad placements on each medium. done. This will consume the reserve budget. The explanation using FIG. 3 is made for the purpose of explaining that the reserve budget is used in stages, and it should be understood that the numerical values have no meaning and are merely examples.

This stepwise allocation of reserve budget is performed according to a reinforcement learning algorithm to maximize advertising effectiveness. In the second placement-related process (see FIG. 4), the processor 11 reads from the storage 13 the delivery schedule of the posted reservation-type advertisements placed in advance according to the placement plan (S210).

Further, the processor 11 sets the state s(1) at time t=1 on the first day of the campaign period, which is the start time of the campaign period, as follows, in order to determine the amount of advertisements to be placed on each medium according to the reinforcement learning algorithm. (S230).

A state variable consists of six terms. The first term “1/T” indicates the rate of progress of the campaign period on the first day of the campaign period. t/T indicates the rate of elapse of the campaign period at time t.

In the following, the point in time t means the point in time of the tth day of the campaign period. That is, t is a discrete time representing the time in days within the campaign period. T represents the length of the campaign period, specifically, the number of days of the campaign period.

The second term "B(1)/B" indicates the ratio of the budget balance B(1) to the advertising budget B at time t=1. That is, B(t)/B is the ratio of the remaining budget B(t) to the advertising budget B on the tth day of the campaign period. Budget balance B(1) is calculated by the following equation.

このように予算残高Ｂ（１）は、予備予算Ｂｒに対応する。

Budget balance B(1) thus corresponds to reserve budget Br.

The third term represents cost performance related to advertisement distribution of each reserved medium from the first reserved medium to the Mth reserved medium. Since the cost performance is unknown on the first day of the campaign period, that is, at time t=1, the third term at time t=1 indicates that the cost performance of each reserved medium is zero. That is, {0} _1≤m≤M is a one-dimensional array with M elements and the value of each element is zero.

The fourth term represents cost performance related to advertisement distribution for each non-reservation type medium from the first non-reservation type medium to the Nth non-reservation type medium. Since the cost performance is unknown on the first day of the campaign period, the fourth term at time t=1 indicates that the cost performance of each non-reserved medium is zero. That is, {0} _{1 ≤ n ≤ N} is a one-dimensional array with N elements and the value of each element is zero. However, both the third and fourth terms may be replaced with zero, and the cost performance measured in past similar advertisement placements may be used.

The fifth term is GRP ( Gross Rating Point).

G _m,τ (1) is the GRP (gross audience rating) estimated on the first day of the campaign period, and according to the distribution schedule, on the corresponding day τ, distributed on the corresponding reserved medium (m) It is the total estimated audience rating of undelivered scheduled advertisements that have been placed. That is, the fifth term is an array {G m _{, τ (1)} of G m, τ (} 1) for each combination of m and τ with 1 ≤ m ≤ M and _{1 ≤ τ ≤ T 1 ≤ m ≤ M, 1≤τ≤T} .

The processor 11 can generate the array {G _{m, τ} (1)} _{1≦m≦M, 1≦τ≦T} according to the distribution schedule obtained in S210. The estimated GRP can be calculated from the past track record of audience ratings for the same time period. Performance data of audience rating required for calculation can be stored in the storage 13 in advance.

The sixth term is an array {W _n (1)} _1≦n≦N of the balance of the advertising budget for each non-reserved medium at time t=1. The balance W _n (1) is the balance of the ad placement budget for the n-th non-reserved medium at time t=1, and is the minimum ad budget b for the n-th non-reserved medium calculated in the first ad placement related process. _nP ⁽ 0) is set. That is, _{ W ₁ (1), W ₂ (1), . . . , W _N (1)}={b ₁ ^P (0), b ₂ ^P (0), ^. .

In subsequent S240, the processor 11, regarding the current time point t=1, based on the state s(t) at this time point t, determines whether the reserved medium and the non-reserved medium at time point t are acting as an action regarding the advertisement at time point t. The ad spend for each is determined using a reinforcement learning algorithm.

That is, the processor 11 calculates the ad placement amount (b ₁ ^R (t), b ₂ ^R (t), . . . , b _M ^R (t), b ₁ ^P (t), b ₂ ^P (t), . . . , b _N ^P (t)) is determined based on the state s(t) and the policy for action selection. Here, b _m ^R (t) is the ad placement amount for the m-th reservation-type medium at time t. b _n ^P (t) is the ad placement amount for the n-th non-reserved medium at time t. The amount of advertisement for each medium is specifically the amount of advertisement for each medium, in other words, the purchase amount of the advertisement space for each medium.

According to the reinforcement learning algorithm, the policy is updated with the reward r(t) described below. According to this embodiment, a plurality of ad placement volumes with different quantities are defined as action options. As an action selection according to the policy, the amount of advertisement is determined by selecting one quantity from the options.

In subsequent S250, the processor 11 calculates the ad placement amount ( _b1R ( ^t ), _b2R (t), ^... , _bMR ( ^t ), _b1P ( ^t ), _b2P ( ^t ),..., b _N ^P (t)) to direct the placement of advertisements on each medium.

In S250, the processor 11 can instruct the relay system 31 to place an advertisement, for example. The relay system 31 can automatically place an advertisement for each media system according to the contents of the instruction. The processor 11 may instruct the user to place advertisements on each medium by displaying the amount of advertisements on the display 15 . The user can at least partially manually place an advertisement for each medium according to the displayed content.

In subsequent S260, the processor 11 determines the average audience rating P _m ^R (t) of the reservation-type advertisements distributed through each reservation-type medium from the first reservation-type medium to the Mth reservation-type medium as the actual exposure performance. . The processor 11 is further based on the advertisement placement cost c _m ^R (t)=b _m ^R (t) for each reservation type medium and the average audience rating P _m ^R (t) of the reservation type advertisement for each reservation type medium. , the exposure amount I _m ^R (t)=c _m ^R (t)×P _m ^R (t) at time t for each reservation type medium is calculated.

I _m ^R (t) represents the amount of exposure at time t of the reserved advertisement delivered through the m-th reserved medium. P _m ^R (t) represents the average audience rating at time t on the m-th reservation-type medium, and c _m ^R (t) is the cost of advertising distributed through the m-th reservation-type medium at time t. , that is, represents the amount of money placed. The processor 11 acquires information on the average audience rating P _m ^R (t) of each reservation-type medium, for example, from the measurement system 35 that measures viewing behavior through the communication device 19, and calculates the average audience rating P _m ^R (t). can be determined.

In subsequent S270, the processor 11 calculates the CPM ( Cost per Mille): P _n ^P (t) is discriminated as the exposure performance, and the advertisement placement cost c _n ^P (t) for each non-reservation type medium and the CPM of non-reservation type advertisement in each non-reservation type medium: P _n Based on ^P (t) and Impression I _n ^P (t)=c _n ^P (t)/P _n ^P (t) at time t for each reserved medium is calculated.

I _n ^P (t) represents the impression at time t on the n th non-reserved medium. P _n ^P (t) represents the CPM of the non-reserved advertisement at time t on the n th non-reserved medium, and c _n ^P (t) is the ad placement at time t on the n th non-reserved medium represent the cost. _cnP (t) represents the amount of money actually required for ^placement based on _bnP (t) ^.

In subsequent S280, processor 11 calculates reward r(t) at time t. Specifically, the processor 11 calculates the exposure amount {I _m ^R (s)} _{1≦m≦M, 1≦s≦t} of the reserved advertisement through the reserved medium up to time t, and the non-exposure amount up to time t. Impression {I _n ^P (s)} _{1 ≤ n ≤ N, 1 ≤ s ≤ t} , which is the amount of exposure of non-reserved advertisements through reserved media, is input to the advertising effectiveness estimation model. Obtain the advertising effectiveness Z(t) at time t from the model. The processor 11 calculates the difference Z(t)-Z(t-1) between the advertising effect Z(t) at time t and the advertising effect Z(t-1) at the previous time as a reward r( t).

r(t)=Z(t)-Z(t-1)
However, if the budget balance B(t) is less than zero, ie, the reward r(t) in an over-budget situation, is defined as a negative reward -C. The value C is a positive value and the penalty mentioned above.

r(t) = -C
In subsequent S290, the processor 11 updates the state s(t) to the state of the current time point t when the next time point t arrives. That is, processor 11 sets state s(t) as follows. At this time, the latest delivery schedule at time t is obtained for setting the state s(t).

The first and second terms of state s(t) are as described above. If the current time t is the second day of the campaign period (that is, if t=2), then the first term is updated to 2/T and the second term is updated to B(2)/B. be. B(2) is the budget balance as of the second day of the campaign period.

The third term represents cost performance related to advertisement distribution for each reserved medium at time t. The cost performance of the m-th reservation-type media is the average audience rating P _m ^R (τ) of the m-th reservation-type media at each time τ from the first day of the campaign period to the time t-1 one year before the current time. It is quantified by the average value. The third term is an array whose elements are average values (1/(t−1))ΣP _m ^R (τ) in each medium from the first to the Mth reservation-type media.

The fourth term represents the cost performance regarding advertisement distribution for each non-reservation type medium at time t. The cost performance of the n-th non-reservation-type media is the CRM for the n-th non-reservation-type media at each time τ from the first day of the campaign period to the time t-1 one before the current time: P _n ^P (τ) is quantified by the average value of The fourth term is an array whose elements are CRM average values (1/(t−1))ΣP _n ^P (τ) in each medium from the first non-reserved medium to the Nth non-reserved medium. .

The fifth term is the GRP (Gross Rating Point) of τ for each day of the campaign period (1 ≤ τ ≤ T) in each reservation-type medium from the first reservation-type medium to the M-th reservation-type medium estimated at time t. is an array of That is, the fifth term is an array {G _{m, τ (t)} of G m, τ (} t) for each combination of m and τ in the ranges of 1 ≤ m ≤ M and _{1 ≤ τ ≤} T _{M, 1≤τ≤T} .

G _m,τ (t) is the estimated GRP on the t-th day of the campaign period, and according to the distribution schedule, on the corresponding day It is the total estimated viewership of undelivered reserved ads.

The calculation of G _m,τ (t) also takes into account the audience rating according to the distribution schedule of the reservation-type advertisements additionally posted after the start of the campaign period. On the other hand, it should be noted that viewership ratings for ads that have already been delivered are not used to calculate GRP. Therefore, the G _m,τ (t) corresponding to the period τ<t before the current time t are all zero. A distributed reserved advertisement is reflected in a reward r(t) as an advertising effect based on the amount of exposure I _m ^R (t) as described above at the time t when the advertisement is distributed.

The sixth term is an array {W _n (t)} of the balance of the advertising budget for each non-reserved medium at time t. The balance W _n (t) is the balance of the ad placement budget for the n-th non-reserved medium at time t. The balance W _n (t) is updated according to the following equation.

As described above, c _n ^P (t−1) is the placement cost for the advertisement delivered through the n-th non-reservation type medium at time t−1. In addition, the budget balance B(t) at time t is calculated by the following equation.

Here, the function E(c _n ^P (t−1)) is such that the cost of placing an advertisement on the n-th non-reserved medium at time t−1, c _n ^P (t−1), is the n-th non-reserved medium at time t−1. It is a function that outputs a value of 1 when it is greater than the budget balance W _n (t−1) of the type medium, and outputs a value of 0 otherwise.

That is, the budget balance B(t) at time t is the reserve budget B(t-1) at time t-1 that has been consumed for the placement of reserved advertisements and non-reserved advertisements at time t-1. Corresponds to the value minus the total budget.

After that, the processor 11 returns to S240, and regarding the current time t=2 and later, based on the state s(t) at the current time t set in S290, Advertisement volume is determined using a reinforcement learning algorithm.

Further, in S250, the processor 11 calculates the amount of advertisement (b ₁ ^R (t), b ₂ ^R (t), . . . , b _M ^R (t), b ₁ ^P (t), b ₂ ^P (t), . . . , b _N ^P (t)) to direct the placement of advertisements on each medium. In S260, the processor 11 determines the average audience rating P _m ^R (t) of each reserved medium, and the exposure amount I _m ^R (t) at time t of the reserved advertisement delivered through each reserved medium = c Calculate _mR ⁽ t) x _PmR ⁽ t).

In subsequent S270, the processor 11 determines the CPM at time t for each non-reserved medium: P _n ^P (t), and the impression I _n ^P (t) at time t for each reserved medium = c _n ^P (t)/P _n ^P (t) is calculated. In subsequent S280, the processor 11 calculates the reward r(t). In subsequent S290, the state s(t) is updated.

In this way, the processor 11 selects an action based on the state s(t) (that is, decides the amount of ad placement) based on the state s(t), calculates the reward r(t) based on the action, and updates the state s(t) in steps S240 to S290. By repeating this, according to the reinforcement learning algorithm, the amount of advertisements for reserved media and non-reserved media at each point in time in the direction of maximizing the advertising effect, which is the reward r(t), is determined using the reserve budget Br. do. A policy for action selection is adjusted based on the reward r(t) according to a reinforcement learning algorithm.

When the end condition is satisfied by the end of the campaign period (Yes in S300), the processor 11 ends the second placement-related process.

According to the information processing system 1 of the present embodiment described above, according to the reinforcement learning algorithm, advertisements on a plurality of media including reserved media and non-reserved media are maximized with a predetermined advertising budget. decide to

With reserved advertising, there is a delay from the time the ad is placed until the advertising effect is achieved. Due to this delay, a simple reinforcement learning algorithm cannot select an appropriate action. In this embodiment, this problem is solved by adding GRP information of undelivered advertisements to the state s(t). .

In this embodiment, in order to hold information on the time lag from placement to delivery of reserved advertisements as a state s(t) in the form of GRP of undelivered advertisements, the effects of delays are suppressed and advertisement effects are maximized. Advertisement amounts can be determined for a plurality of media, including reserved media and non-reserved media, so as to be more flexible.

Therefore, according to the present embodiment, it is possible to provide an information processing system 1 that is meaningful for placing advertisements across different media including reserved media and non-reserved media.

In the above example, the reserved advertisement is a TV commercial and the non-reserved advertisement is a programmatic advertisement that is digitally distributed. It may be an advertisement that

In addition, at time t=0 before the _start of the campaign period, the minimum placement budget b ₁ ^P (0), b ₂ ^P (0), ^. To avoid, a negative reward r1(t) may be added to the reward r(t).

According to this reward r1(t), the smaller b _n ^P (0) is, the larger negative reward r1(t) is calculated. The amount of advertisements to be placed will be determined positively for the placement of ads early in the period.

The processor 11 can acquire information on the tuning parameter α of this reward r1(t) through the storage 13 or through the input device 17 in the first step (S110) of the advertisement placement related process. α is a positive real number.

[Second embodiment]
Next, details of the information processing system 1 of the second embodiment will be described with reference to FIGS. 5 and 6. FIG. The information processing system 1 of the second embodiment differs from the first embodiment in that it uses a contextual bandit algorithm instead of a reinforcement learning algorithm for stepwise determination of the amount of advertisements. It is basically configured in the same manner as the information processing system 1 of the first embodiment. The hardware configuration of the information processing system 1 is the same as that of the first embodiment. Therefore, the contents of the processing related to the contextual bandit algorithm will be selectively described below, and other descriptions will be omitted.

As is known, reinforcement learning algorithms select actions to maximize rewards based on defined states, actions, and rewards. According to the first embodiment, the action is determination of the amount of advertisements to be placed on each medium.

In contrast, contextual bandit algorithms are algorithms that select actions to maximize reward based on defined context, actions, and rewards. Actions and rewards in the second embodiment are defined in the same way as in the first embodiment. In other words, action is the determination of the amount of advertisements to be placed on each medium. Remuneration is the advertising effect obtained by distributing advertisements through each medium.

In the present embodiment, the processor 11, based on a command from the user, performs first advertisement-related processing (see FIG. 5) and second advertisement-related processing (see FIG. 6) using a contextual bandit algorithm as advertisement-related processing. ).

Variables and the like are defined as follows when explaining the processing contents. The number of days in the campaign period is T. The number of reserved advertisement media is M, and the M reserved media include the m-th reserved medium. m takes an integer value ranging from 1, 2, . . . The number of non-reserved advertisement media is N, and the N non-reserved media include the n-th non-reserved media. n takes an integer value ranging from 1, 2, . . . The action space A is an (M+N)-dimensional discrete action space in which each dimension has a plurality of elements corresponding to options for the amount of advertisements.

At time t, an average audience rating P _m ^R (t) is observed on the m-th reservation-type medium. At time t, CPM:P _n ^P (t) is observed on the nth non-reserved medium. The “GRP of the reserved advertisement delivered on the τ day of the campaign period through the m-th reserved medium” estimated at time t is G _m,τ (t).
The context x _t at time t>1 and the context x ₀ at time t=0 are represented by the following equations.

In the first ad placement-related process (see FIG. 5), the processor 11 first acquires information on the specified condition in the same manner as in the process of S110 (S310). In subsequent S320, the processor 11 selects an action regarding advertisement placement at time t=0 before the start of the campaign period according to the contextual bandit algorithm.

Specifically, the processor 11 calculates the UCB (Upper Confidence Bound) score of each action that can be selected at time t=0 based on the given context x ₀ instead of the state s(0), and the UCB score a ^* ( ₀ )=( _b1R (0), ^{b2R(0),...,bmR} ₍ ⁰ ₎ ^, ..., _bMR ( ⁰ ), ^b1P (0) , b ₂ ^P (0 _{), . . . , b n} ^P ₍ ⁰ ), .

As a result, the ^minimum ad budgets b ₁ ^R (0), b ₂ ^R (0), . . . , _{b m R} ⁽ 0), . ⁰ ), b ₁ ^P (0), b ₂ ^P (0 _{), . . . , b n P} ⁽ 0), .

In subsequent S330, the processor 11 sets the minimum advertising budgets _{b 1} ^R (0), b ₂ ^R ( ⁰ ), . (0), b ₁ ^P (0), b ₂ ^P (0), . . . , b _N ^P (0). After that, the processor 11 terminates the first publication-related processing.

After finishing the first advertisement related process, the processor 11 starts the second advertisement related process shown in FIG. In the second placement-related process, the processor 11 reads from the storage 13 the distribution schedule of the reservation-type advertisement placed in advance according to the placement plan (S410).

Furthermore _, the processor 11 determines the amount of advertisements to be placed on each medium according to the contextual bandit algorithm. is set as follows (S430).

The processor 11 can generate the array {G _{m, τ} (1)} _{1≦m≦M, 1≦τ≦T} according to the distribution schedule acquired in S410, as in the first embodiment.

At S440, the processor 11 calculates the UCB score of each action that can be selected based on the context x _t at this time t for the current time t=1, and the action a ^* (t) with the maximum UCB score =( _b1R (t), ^b2R (t), ^... , ^bmR (t) _, ^... , _bMR (t), _b1P ( ^t ), _b2P ( ^t ), _... , Choose _bnP ⁽ t),..., _bNP ⁽ t)).

As a result, the processor 11 calculates the amount of advertisements (b ₁ ^R (t), b ₂ ^R (t), . . . , b _MR (t), b ₁ ^P (t), b ₂ ^P (t), . . . , b _NP (t)) are ^determined using the contextual bandit ^algorithm . b _m ^R (t) is the placement amount of reserved advertisements for the m-th reserved medium at time t. b _n ^P (t) is the amount of non-reserved advertisements placed on the n-th non-reserved medium at time t.

In subsequent S450, the processor 11 calculates the ad placement amount ( _b1R ( ^t ), _b2R (t), ^... , _bMR ( ^t ), _b1P ( ^t ), _b2P ( ^t ),..., b _N ^P (t)) to direct the placement of advertisements on each medium.

In subsequent S460, the processor 11 calculates the average audience rating P _m ^R (t) of the reservation-type advertisement distributed through each reservation-type medium from the first reservation-type medium to the M-th reservation-type medium to the corresponding information. The amount of exposure I _m ^R (t)=c _m ^R (t)×P _m ^R (t) at time t for each reservation-type medium is calculated in the same manner as in the processing in S260.

In subsequent S470, the processor 11 determines the CPM of non-reserved advertisements distributed through the non-reserved media at time t for each non-reserved medium from the first non-reserved medium to the Nth non-reserved medium: _PnP (t) is determined by obtaining the corresponding ^information , and ^similar to the processing at S270, _{impressions InP(t) at time t for each reserved medium = cnP (} ^t )/ _Pn Calculate ^P (t).

In subsequent S480, the processor 11 calculates the reward r(t) at time t according to the formula r(t)=Z(t)-Z(t-1), as in the processing in S280. The reward r(t) when the budget balance B(t) is less than zero is the negative reward −C (r(t)=−C).

In subsequent S490, the processor 11 waits for the arrival of the next time point t and then updates the context _xt . That is, processor 11 sets context x _t as follows.

The calculation of G _m,τ (t) is the same as the processing in S290 in the first embodiment. The budget balance B(t) at time t is calculated as follows.

After that, the processor 11 returns the process to S440, and regarding the current time t=2 or later, based on the context x _t of the current time t set in S490, the reservation type Advertisement volumes for each medium and non-reserved medium are determined using a contextual bandit algorithm. Further, in S450, the ad placement amount (b ₁ ^R (t), b ₂ ^R (t), . . . , b _M ^R (t), b ₁ ^P (t), b ₂ ^P (t), . . . , b _N ^P (t)) to direct the placement of advertisements on each medium.

In this way, the processor 11 repeatedly selects an action based on the context x _t (that is, determines the amount of advertisement), calculates the reward r(t) based on the action, and updates the context x _t in steps S440 to S490. , according to the contextual bandit algorithm, determines the amount of advertisements to be placed on reserved media and non-reserved media at each point in the direction of maximizing the advertisement effect, which is the reward r(t).

When the end condition is satisfied by the end of the campaign period (Yes in S500), the processor 11 ends the second placement-related process.

According to the information processing system 1 of the present embodiment described above, the reservation-type Appropriate decisions can be made regarding placement on a plurality of media, including media and non-reserved media. Therefore, the information processing system 1 is significant for placing advertisements across different media including reserved media and non-reserved media.

[Other embodiments]
The present disclosure is not limited to the above embodiments, and can take various forms.
Above, as dynamic optimization algorithms, reinforcement learning algorithm (first embodiment) and contextual bandit algorithm (second embodiment) were used to cross different media including reserved media and non-reserved media. An example of the information processing system 1 that realizes optimal advertisement placement has been described.

However, similar functionality may be achieved using other algorithms as dynamic optimization algorithms. For example, the information processing system 1 may be configured to use a state space model, particularly a Kalman filter, to determine the amount of advertisements placed on reservation-type media and non-reservation-type media at each point in the campaign period.

A Kalman filter is a kind of dynamic control technique that consists of a state equation and an observation equation. A state equation and an observation equation in a linear model are represented by the following equations, and define the relationship between the state quantity _xt , the observation quantity _yt , and the control input _ut .

The state equation and observation equation of the generalized model including the nonlinear model are as follows.

When the technical idea of the present disclosure is implemented using a Kalman filter, it is possible to allocate the ad placement amount for each medium at time t to the control input u _t . That is, control inputs u _t can be mapped to actions in a reinforcement learning algorithm.

The budget balance, CPM, audience rating, etc. at time t can be assigned to the state quantity _xt . That is, the state quantity _xt can be associated with the state s(t) in the reinforcement learning algorithm. For the observable _yt , the advertising effectiveness at time t can be assigned. That is, the observable _yt can be mapped to the reward r(t) in the reinforcement learning algorithm.

When using the Kalman _filter , the delay from placing an advertisement to distribution in reserved advertisements is an observation equation y _t = _h (x _t , u ₀ , u _t , v _t ). That is, the delivery schedule can be incorporated into the observation equation by formulating the relationship between the control input u _t and the observable y _t . This observation equation indicates that the control input u ₀ regarding the placement of the reserved advertisement at time t=0 affects the observable quantity y _t .

The technical idea of the present disclosure may be realized using a PID control technique. As the reward index, in addition to an advertisement exposure index such as reach and frequency, an index related to consumer awareness may be used. Consumer awareness indicators may include awareness, interest, interest, comprehension, purchase intent, first thought, and repeat purchase intent.

Reward indicators include indicators obtained by the advertiser, specifically sales amount and number of sales of the advertised product, number of accesses to the product introduction site, number of uses of the advertised service, and related application software. , the number of monthly active users (MAU), and the number of inquiries related to the advertisement may be used.

Indicators related to the state that should be considered when making action choices regarding advertisement placement include indicators related to reach and the unit price required for acquisition, specifically CPM, CPC (Cost Per Click), CPA (Cost per Acquisition), and , ROI (Return on investment), etc. may be included.

In addition, indicators related to the state to be considered include indicators related to advertising performance, specifically, CTR (Click Through Rate), CVR (Conversion Rate), VTR (View Through Rate), and TARP (Target Audience Rating Point). ) etc. may be included.

In addition, the preferred state index differs depending on the index used as the reward for reinforcement learning. For example, when reach is adopted as a reward for reinforcement learning, and when the reserved advertisement is distributed by television broadcasting, examples of indicators that are preferable for the status indicator are audience rating (or GRP) and attention rate. Yes, and if the reserved advertisement is a digital delivery, examples of preferred metrics for status metrics are CPM and vCPM (viewable Cost Per Mille). The same is true for non-reserved advertisements.

When the number of sales is used as a reward for reinforcement learning, and when the reserved advertisement is distributed by television broadcasting, examples of indicators that are preferable for the status indicator are the audience rating (or GRP) and the attention rate. , if the scheduled ad is digital delivery, examples of preferred metrics for the state metrics are CPM, conversions, CPA, and CPC.

If the non-reserved advertisement is distributed by television broadcasting, examples of preferable indicators for the status indicator are the number of conversions and CPA. Examples are conversions, CPA, and CPC.

In the above-described embodiment, advertisement placement on a plurality of reservation-type media and a plurality of non-reservation-type media was taken as an example. may Alternatively, a common ad placement amount may be determined for a plurality of reserved media, and similarly a shared ad placement amount may be determined for a plurality of non-reserved media.

Although the above embodiments include the concept of additional placement of reserved advertisements, the concept of canceling reservation of reserved advertisements (in other words, canceling the placement of advertisements) may be further included. That is, the action options in the dynamic optimization algorithm may include the action of canceling the posted reservation-type advertisement.

This cancel action corresponds to receiving a refund of the purchase price of the advertising space by canceling the reservation. Therefore, the concept of reservation cancellation can be introduced into the dynamic optimization algorithm by introducing negative ad placement. In this manner, the present disclosure may determine the ad placement amount in consideration of reservation cancellation. According to the reinforcement learning algorithm, when determining a negative amount of advertisements, it is possible to reduce the advertisement distribution schedule held as a state and increase the budget balance B(t).

A function possessed by one component in the above embodiment may be distributed to a plurality of components. Functions possessed by multiple components may be integrated into one component. A part of the configuration of the above embodiment may be omitted. At least part of the configurations of the above embodiments may be added or replaced with respect to the configurations of other above embodiments. All aspects included in the technical ideas specified from the language in the claims are embodiments of the present disclosure.

[Technical concept disclosed in this specification]
It can be understood that the following technical ideas are disclosed in this specification.
[Item 1]
Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, an advertisement placement instruction unit configured to instruct advertisement placement on the first medium based on the first placement amount and advertisement placement on the second medium based on the second placement amount;
a first performance determination unit configured to determine a first performance, which is a performance related to exposure of the reserved advertisement exposed through the first medium;
a second performance discriminating unit configured to discriminate a second performance, which is a performance relating to exposure of the non-reserved advertisement exposed through the second medium;
a schedule discriminating unit configured to discriminate an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
The advertisement placement instruction unit, for each of a plurality of time points, determines the first and the second amount of advertisements.
[Item 2]
Of the total amount, a first amount is predetermined as an ad placement amount for the reserved advertisement, a second amount is set as an ad placement amount shared by the reserved advertisement and the non-reserved advertisement, and The instruction unit is given the right to decide allocation of the second amount of the reserved advertisement and the non-reserved advertisement,
Placement of the scheduled advertisement corresponding to at least a portion of the first amount is completed at an initial point in time that is earlier than the plurality of points in time, and the exposure schedule is set at the initial point in time. including an exposure schedule for an unexposed advertisement waiting for exposure, which is the reserved advertisement that has been posted in
The placement instruction unit adjusts the first placement amount and the second placement amount for each time point, taking into account the exposure schedule of the unexposed advertisement including the reserved advertisement that has been placed at the first time point. The information processing system according to item 1 to be determined.
[Item 3]
The first ad placement amount is the ad placement amount of the reserved advertisement,
The second ad placement amount is the ad placement amount of the non-reserved advertisement,
3. The information processing system according to item 1 or 2, wherein the total amount is an advertisement budget for advertisements including the reserved advertisement and the non-reserved advertisement.
[Item 4]
The information processing system according to any one of items 1 to 3, wherein the advertisement placement instruction unit determines the first advertisement amount and the second advertisement amount for each of the time points according to a dynamic optimization algorithm.
[Item 5]
5. Information processing system according to item 4, wherein the dynamic optimization algorithm includes at least one of reinforcement learning, contextual bandit algorithm, and Kalman filter.
[Item 6]
The publication instruction unit, at each time point,
Based on the state or context at the corresponding point in time, determining the first ad placement amount and the second ad placement amount as actions related to ad placement at the corresponding point in time,
Based on the first performance and the second performance, the advertisement effect of the reserved advertisement and the non-reserved advertisement newly exposed by the advertisement placement up to the corresponding time is determined as a reward for the action. ,
updating the state or context in consideration of the advertisement placement at the corresponding time and the first performance and the second performance at the corresponding time;
configured to update a policy regarding selection of said action based on said reward;
The state or context is any of items 1 to 5 defined using the possible remaining ad placement amount, the exposure schedule of the unexposed advertisement, the first performance, and the second performance. or the information processing system according to item 1.
[Item 7]
The placement instruction unit determines the first placement amount and the second placement amount at the corresponding point in time as the action by reinforcement learning in which states, rewards, and actions are defined. configured to
the states are defined using the possible remaining ad placement volume, the first performance, the second performance, and the schedule of unexposed ads at the corresponding time points;
6. The information processing system according to any one of items 1 to 5, wherein the remuneration is defined using the advertising effect at the corresponding time determined from the first performance and the second performance.
[Item 8]
The placement instruction unit determines the first placement amount and the second placement amount at the corresponding point in time as the action by a contextual bandit algorithm in which context, reward, and action are defined. is configured to
the context is defined using the possible remaining ad placement volume, the first performance, the second performance, and the unexposed ad schedule at the corresponding time points;
6. The information processing system according to any one of items 1 to 5, wherein the remuneration is defined using the advertising effect at the corresponding time determined from the first performance and the second performance.
[Item 9]
The ad placement instruction unit is configured to determine the first ad placement amount and the second ad placement amount at a corresponding time point for each of the time points using a state space model such as a Kalman filter, for example,
The state space model is a model that includes information on the exposure schedule of the unexposed advertisement, and defines relationships between state quantities, observable quantities, and input quantities;
The input amount is an amount related to advertisement placement defined using the first placement amount and the second placement amount,
The state quantity is a state quantity that changes according to the placement of the advertisement, and is defined using the possible remaining placement amount, the first performance, and the second performance,
The observable amount is an amount that defines the advertising effectiveness of the posted reserved advertisement and the non-reserved advertisement determined based on the first performance and the second performance. Information processing system according to any one of the preceding items.
[Item 10]
A computer program for causing a computer to implement functions as an advertisement placement instruction unit, a first performance determination unit, a second performance determination unit, and a schedule determination unit in the information processing system according to any one of items 1 to 9. .
[Item 11]
A computer-implemented information processing method comprising:
Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, instructing the placement of advertisements on the first medium based on the first amount of advertisements and the placement of advertisements on the second medium based on the second amount of advertisements;
Determining a first performance that is a performance related to exposure of the reserved advertisement exposed through the first medium;
Determining a second performance that is a performance related to exposure of the non-reserved advertisement exposed through the second medium;
Determining an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
Instructing the placement of the advertisement is based on the possible remaining placement amount, the exposure schedule of the unexposed advertisement, the first performance, and the second performance for each time point, and An information processing method including determining the first ad placement amount and the second ad placement amount.

DESCRIPTION OF SYMBOLS 1... Information processing system, 11... Processor, 12... Memory, 13... Storage, 15... Display, 17... Input device, 18... Media reader/writer, 19... Communication device, 31... Relay system, 35... Measurement system.

Claims (11)

Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, an advertisement placement instruction unit configured to instruct advertisement placement on the first medium based on the first placement amount and advertisement placement on the second medium based on the second placement amount;
a first performance determination unit configured to determine a first performance, which is a performance related to exposure of the reserved advertisement exposed through the first medium;
a second performance discriminating unit configured to discriminate a second performance, which is a performance relating to exposure of the non-reserved advertisement exposed through the second medium;
a schedule discriminating unit configured to discriminate an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
According to at least one of a reinforcement learning and a contextual bandit algorithm, the advertisement placement instruction unit, for each time point with respect to a plurality of time points ,
Based on the state or context at the corresponding point in time, determining the first ad placement amount and the second ad placement amount as actions related to ad placement at the corresponding point in time,
Based on the first performance and the second performance, the advertisement effect of the reserved advertisement and the non-reserved advertisement newly exposed by the advertisement placement up to the corresponding time is determined as a reward for the action. ,
updating the state or context in consideration of the advertisement placement at the corresponding time and the first performance and the second performance at the corresponding time;
update a policy for selection of said action based on said reward
configured as
The information processing system, wherein the state or context is defined using a possible remaining placement amount, an exposure schedule of the unexposed advertisement, the first performance, and the second performance. Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, an advertisement placement instruction unit configured to instruct advertisement placement on the first medium based on the first placement amount and advertisement placement on the second medium based on the second placement amount;
a first performance determination unit configured to determine a first performance, which is a performance related to exposure of the reserved advertisement exposed through the first medium;
a second performance discriminating unit configured to discriminate a second performance, which is a performance relating to exposure of the non-reserved advertisement exposed through the second medium;
a schedule discriminating unit configured to discriminate an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
The placement instruction unit calculates the first placement amount and the second placement amount at each corresponding time point with respect to a plurality of time points by reinforcement learning in which states, rewards, and actions are defined. is configured to determine as
the states are defined using the remaining ad placement volume, the first performance, the second performance, and the schedule of unexposed ads at the corresponding time points;
The information processing system, wherein the reward is defined using advertising effectiveness at the corresponding time determined from the first performance and the second performance. Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, an advertisement placement instruction unit configured to instruct advertisement placement on the first medium based on the first placement amount and advertisement placement on the second medium based on the second placement amount;
a first performance determination unit configured to determine a first performance, which is a performance related to exposure of the reserved advertisement exposed through the first medium;
a second performance discriminating unit configured to discriminate a second performance, which is a performance relating to exposure of the non-reserved advertisement exposed through the second medium;
a schedule discriminating unit configured to discriminate an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
The placement instruction unit calculates the first placement amount and the second placement amount at each corresponding time point with respect to a plurality of time points by a contextual bandit algorithm in which context, reward, and behavior are defined. configured to determine as said action ;
the context is defined using the remaining ad placement volume, the first performance, the second performance, and the unexposed advertisement schedule at the corresponding time points;
The information processing system, wherein the reward is defined using advertising effectiveness at the corresponding time determined from the first performance and the second performance. Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, an advertisement placement instruction unit configured to instruct advertisement placement on the first medium based on the first placement amount and advertisement placement on the second medium based on the second placement amount;
a first performance determination unit configured to determine a first performance, which is a performance related to exposure of the reserved advertisement exposed through the first medium;
a second performance discriminating unit configured to discriminate a second performance, which is a performance relating to exposure of the non-reserved advertisement exposed through the second medium;
a schedule discriminating unit configured to discriminate an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
The advertisement placement instruction unit uses a state space model such as a Kalman filter, for example, for a plurality of time points, for each time point, the possible remaining ad placement amount, the exposure schedule of the unexposed advertisement, the first performance, the configured to determine the first ad placement amount and the second ad placement amount at corresponding points in time based on a second performance,
The state space model is a model that includes information about the exposure schedule of the non-exposed advertisement, defines a relationship between a state quantity, an observable quantity, and an input quantity, and defines the information of the exposure schedule of the non-exposed advertisement. is incorporated into the state-space model by formulating the relationship between the input quantities and the observables according to the exposure schedule;
The input amount is an amount related to advertisement placement defined using the first placement amount and the second placement amount,
The state quantity is a state quantity that changes according to the placement of the advertisement, and is defined using the possible remaining placement amount, the first performance, and the second performance,
The information processing system, wherein the observable amount is an amount that defines advertising effectiveness of the posted reserved advertisement and the posted non-reserved advertisement determined based on the first performance and the second performance. Of the total amount, a first amount is predetermined as an ad placement amount for the reserved advertisement, a second amount is set as an ad placement amount shared by the reserved advertisement and the non-reserved advertisement, and The instruction unit is given the right to decide allocation of the second amount of the reserved advertisement and the non-reserved advertisement,
Placement of the scheduled advertisement corresponding to at least a portion of the first amount is completed at an initial point in time that is earlier than the plurality of points in time, and the exposure schedule is set at the initial point in time. including an exposure schedule for an unexposed advertisement waiting for exposure, which is the reserved advertisement that has been posted in
The placement instruction unit considers, as the state or the context, an exposure schedule of an unexposed advertisement including the reserved advertisement that has already been placed at the first time point, and calculates the first placement amount and the 2. The information processing system according to claim 1, wherein the second advertisement amount is determined. The first ad placement amount is the ad placement amount of the reserved advertisement,
The second ad placement amount is the ad placement amount of the non-reserved advertisement,
6. The information processing system according to any one of claims 1 to 5, wherein the total amount is an advertisement budget for advertisements including the reserved advertisement and the non-reserved advertisement. A method for causing a computer to realize the functions of the publication instruction unit, the first performance determination unit, the second performance determination unit, and the schedule determination unit in the information processing system according to any one of claims 1 to 5 computer program. A computer-implemented information processing method comprising:
Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, instructing the placement of advertisements on the first medium based on the first amount of advertisements and the placement of advertisements on the second medium based on the second amount of advertisements;
Determining a first performance that is a performance related to exposure of the reserved advertisement exposed through the first medium;
Determining a second performance that is a performance related to exposure of the non-reserved advertisement exposed through the second medium;
Determining an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
Instructing the placement of advertisements is performed according to at least one of a reinforcement learning and a contextual bandit algorithm, with respect to a plurality of time points, at each time point :
Based on the state or context at the corresponding point in time, determining the first ad placement amount and the second ad placement amount as actions related to ad placement at the corresponding point in time,
Based on the first performance and the second performance, the advertisement effect of the reserved advertisement and the non-reserved advertisement newly exposed by the advertisement placement up to the corresponding time is determined as a reward for the action. ,
updating the state or context in consideration of the advertisement placement at the corresponding time and the first performance and the second performance at the corresponding time;
Updating a policy for selecting the action based on the reward
including
The information processing method, wherein the state or context is defined using a possible remaining placement amount, an exposure schedule of the unexposed advertisement, the first performance, and the second performance. A computer-implemented information processing method comprising:
Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, instructing the placement of advertisements on the first medium based on the first amount of advertisements and the placement of advertisements on the second medium based on the second amount of advertisements;
Determining a first performance that is a performance related to exposure of the reserved advertisement exposed through the first medium;
Determining a second performance that is a performance related to exposure of the non-reserved advertisement exposed through the second medium;
Determining an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
Instructing the advertisement placement is performed by reinforcement learning in which states, rewards, and actions are defined, and for each time point, the first ad placement amount and the second ad placement amount at corresponding points in time. as said action ;
the states are defined using the remaining ad placement volume, the first performance, the second performance, and the schedule of unexposed ads at the corresponding time points;
The information processing method, wherein the reward is defined using advertising effectiveness at the corresponding time determined from the first performance and the second performance. A computer-implemented information processing method comprising:
Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, instructing the placement of advertisements on the first medium based on the first amount of advertisements and the placement of advertisements on the second medium based on the second amount of advertisements;
Determining a first performance that is a performance related to exposure of the reserved advertisement exposed through the first medium;
Determining a second performance that is a performance related to exposure of the non-reserved advertisement exposed through the second medium;
Determining an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
By a contextual bandit algorithm in which a context, a reward, and an action are defined, for each time point, the first ad placement amount and the second ad placement amount at a corresponding time point for a plurality of time points configured to determine the amount of advertisement as the action ,
the context is defined using the remaining ad placement volume, the first performance, the second performance, and the unexposed advertisement schedule at the corresponding time points;
The information processing method, wherein the reward is defined using advertising effectiveness at the corresponding time determined from the first performance and the second performance. A computer-implemented information processing method comprising:
Determining a first advertisement amount, which is an advertisement amount for reserved advertisements exposed through the first medium, and a second advertisement amount, which is an advertisement amount for non-reserved advertisements exposed through the second medium, instructing the placement of advertisements on the first medium based on the first amount of advertisements and the placement of advertisements on the second medium based on the second amount of advertisements;
Determining a first performance that is a performance related to exposure of the reserved advertisement exposed through the first medium;
Determining a second performance that is a performance related to exposure of the non-reserved advertisement exposed through the second medium;
Determining an exposure schedule of non-exposure advertisements waiting for exposure among the reserved advertisements;
with
The total amount of possible placements for the reserved advertisement and the non-reserved advertisement is predetermined,
Instructing the advertisement placement includes, for example, using a state space model such as a Kalman filter, with respect to a plurality of time points, for each time point, the possible remaining ad placement amount, the exposure schedule of the unexposed advertisement, and the first performance and determining the first ad placement amount and the second ad placement amount at the corresponding point in time based on the second performance ,
The state space model is a model that includes information about the exposure schedule of the non-exposed advertisement, defines a relationship between a state quantity, an observable quantity, and an input quantity, and defines the information of the exposure schedule of the non-exposed advertisement. is incorporated into the state-space model by formulating the relationship between the input quantities and the observables according to the exposure schedule;
The input amount is an amount related to advertisement placement defined using the first placement amount and the second placement amount,
The state quantity is a state quantity that changes according to the placement of the advertisement, and is defined using the possible remaining placement amount, the first performance, and the second performance,
The information processing method, wherein the observable amount is an amount that defines advertising effectiveness of the posted reserved advertisement and the non-reserved advertisement determined based on the first performance and the second performance.

Images