JP7660245B1 - Analysis server, advertisement distribution system, terminal device, analysis method and program - Google Patents

Abstract

[Problem] To provide a technology for improving the accuracy of extracting users to whom advertisements are to be delivered. [Solution] An analysis server includes a means for calculating, based on a user's feature amount and a difference between an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, the degree of contribution that the user's feature amount makes to the magnitude of the difference, and a means for extracting the feature amount whose degree of contribution satisfies a predetermined condition. [Selected Figure] Figure 1

Description

The present invention relates to an analysis server, an advertisement delivery system, a terminal device, an analysis method, and a program.

Advertising plays an important role in providing products and services to consumers. Patent Document 1 discloses an advertising effectiveness prediction system that builds a prediction model by learning training data that associates known advertisements with consumer response information to the advertisements, and inputs unknown advertisements into this prediction model to predict consumer response information in response to the unknown advertisements. This system can be used to predict the effectiveness of an advertisement before it is distributed, and by making corrections or other adjustments so as to achieve the desired effect, it is possible to create an advertisement that increases consumer purchasing motivation. In order to increase advertising effectiveness, in addition to improving the content of the advertisement, it is important to consider to what target the advertisement is distributed to.

JP 2019-144916 A

The problem that this disclosure aims to solve is to provide an analysis server, an ad delivery system, a terminal device, an analysis method, and a program that can improve the accuracy of extracting users who are the targets of ad delivery.

The present invention aims to provide an analysis server, an advertisement delivery system, a terminal device, an analysis method, and a program that solve at least part of the above-mentioned problems.

According to one aspect of the present invention, the analysis server includes a means for calculating the degree of contribution of the user's feature value to the magnitude of the difference between an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, based on the feature value of the user and the difference between the index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, and a means for extracting the feature value whose contribution degree satisfies a predetermined condition.

According to one aspect of the present invention, an advertisement delivery system includes the above-described analysis server, a delivery server including a means for receiving an advertisement delivery request, a means for identifying advertisement targets having at least one matching feature based on the feature extracted for each of the candidate targets of the advertisement to be delivered in response to the delivery request, the feature having a degree of contribution that satisfies a predetermined condition, as the advertisement targets to be delivered to the user, and a means for delivering the advertisement identified by the identifying means.

According to one aspect of the present invention, a terminal device receives an advertisement transmitted by the above-mentioned advertisement distribution system.

According to one aspect of the present invention, the analysis method is an analysis method executed by a computer, and includes a step of calculating the degree of contribution of the user's feature value to the magnitude of the difference between an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, based on the feature value of the user and the difference between the index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, and a step of extracting the feature value whose contribution degree satisfies a predetermined condition.

According to one aspect of the present invention, the program causes a computer to function as a means for calculating the degree of contribution of the user's feature to the magnitude of the difference between an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, based on the user's feature and a means for extracting the feature whose degree of contribution satisfies a predetermined condition.

The present invention can improve the accuracy of extracting users who are the targets of advertisement delivery.

1 is a block diagram illustrating an example of an advertisement delivery system according to an embodiment. FIG. 13 is a diagram illustrating an example of a purchase record table according to the embodiment. FIG. 2 is a diagram illustrating an example of a purchase prediction model according to the embodiment. FIG. 13 is a diagram showing an example of a calculation result of a purchase probability according to the embodiment. FIG. 11 is a diagram illustrating an example of a difference in purchase probability between a case where an advertisement is delivered and a case where an advertisement is not delivered according to the embodiment. FIG. 13 is a diagram illustrating an example of a contribution model according to the embodiment. 11 is a flowchart illustrating an example of a process for analyzing a feature amount that contributes to an advertising effect according to the embodiment. 11 is a flowchart illustrating an example of an advertisement delivery process according to the embodiment. 11 is a flowchart illustrating an example of a process for delivering a point-added advertisement according to the embodiment. FIG. 13 is a diagram illustrating an example of a purchase probability table according to the embodiment. FIG. 2 is a diagram illustrating an example of a hardware configuration of an advertisement delivery system according to an embodiment.

<Embodiment>
(System Configuration)
FIG. 1 is a block diagram illustrating an example of an advertisement delivery system according to an embodiment.
The advertisement delivery system 1 includes a user terminal 10, a DB server 20, an advertising media server 30, a purchase management server 40, and an analysis server 50. The user terminal 10, the DB server 20, the advertising media server 30, the purchase management server 40, and the analysis server 50 are communicatively connected via a network NW.

The user terminal 10 is a terminal device such as a smartphone or a PC (personal computer). The user terminal 10 is used by a user who can purchase products and services by using an online shopping mall provided by the purchase management server 40 or advertisements provided by the advertising media server 30. For example, when a user operates the user terminal 10 and logs into the advertising media server 30 or the purchase management server 40 using his or her account information, an advertisement suitable for the user is delivered to the user terminal 10. When a user views or clicks on an advertisement displayed on the user terminal 10, the history is recorded in the advertising media server 30 in association with the user ID of the user. In addition, when a user operates the user terminal 10 to purchase a product at an online shopping mall, the purchase history is recorded in the purchase management server 40 in association with the user ID.

The DB server 20 (DB is an abbreviation for database) includes a user information management DB 21. The DB server 20 writes various information to the user information management DB 21 and reads various information from the user information management DB 21. The DB server 20 transmits the read information to the analysis server 50, etc. The user information management DB 21 stores personal information such as the name, age, sex, address, email address, and telephone number of the user who uses the advertising media server 30 or the purchase management server 40, and various feature quantities that represent the characteristics of the user, in association with the user ID. The user feature quantities include demographic information such as the age and sex of each user, and user attribute information such as hobbies, preferences, and tendencies. The user attribute information includes factual attributes and estimated attributes estimated from other user attribute information. In addition, the user information management DB 21 may store the advertisement delivery history for each user ID and each product, the number of times an advertisement is displayed, the number of times an advertisement is clicked, and the purchase history of the product. In addition, different user IDs may be assigned to the same user for each of the advertising media server 30, the purchase management server 40, and each of the stores located in the shopping mall provided by the purchase management server 40, and the association between these multiple user IDs may be set in the user information management DB 21. In the following description, the user ID is used to refer to the identification information of one user collectively, rather than to each of these different IDs individually.

The advertising medium server 30 is, for example, a web server that provides a place where advertisements for various products and services are posted. The place where advertisements are posted is, for example, a web page such as news or blog, which is called an advertising medium. The advertising medium server 30 transmits the ID of a user who logs in to the advertising medium server 30 and is viewing the medium and various advertisements contained in the medium to the analysis server 50, and requests the distribution of the advertisement. In response to this request, the analysis server 50 selects an advertisement to be distributed to the user ID and distributes the selected advertisement. Then, the advertising medium server 30 displays the distributed advertisement at a predetermined position on the advertising medium. In addition, the advertising medium server 30 may store log information (advertising distribution history) such as when an advertisement was distributed and how many times it was distributed, and advertising impression information such as whether the user viewed or clicked the advertisement, for each user ID. Note that among the advertisements, there are advertisements with points that are awarded to users who view the advertisement, and the advertising medium server 30 can display the advertisements with points on the advertising medium. For example, when a user logs in to the advertising media server 30, the analysis server 50 judges whether an advertisement with points is effective for that user, and if it is judged to be effective, delivers the advertisement with points. When the user views the advertisement with points, the advertising media server 30 may perform point management, such as awarding points to the user.

The purchase management server 40 manages an online shopping mall. For example, the purchase management server 40 transmits web pages on which products are posted and processes the purchase of products in response to requests from the user terminal 10. The purchase management server 40 manages log information (purchase history) that records which web pages of products the user has accessed and the purchases of those products. Based on this log information, the purchase management server 40 can also analyze the number of times a product web page was accessed for each user ID and the number of times the product was purchased. The purchase management server 40 also logs in to the purchase management server 40 and transmits the user ID of a user who is considering purchasing a product to the analysis server 50, requesting the delivery of an advertisement. In response to this request, the analysis server 50 selects an advertisement to be delivered to the user ID and delivers the selected advertisement. The purchase management server 40 then displays the delivered advertisement in a specified position on the web page on which the product is posted.

The analysis server 50 analyzes the features of the user that contribute to the magnitude of the advertising effect. The advertising effect refers to the fact that the user who received the advertisement purchases the advertised product or accesses the product purchase page, etc., and the advertising effect refers to the degree of increase in interest in the product. In the following explanation, as an example, the difference between the purchase probability when the advertisement is delivered and the purchase probability when the advertisement is not delivered is used as an index value of the advertising effect. For example, the analysis server 50 analyzes what features a user has in an advertisement for a certain product will purchase the product or become interested in the product through the advertisement. In addition, based on the analysis result of the features of the user that contributes to the magnitude of the advertising effect, the analysis server 50 determines (selects) a product that matches the features of the user who sent the request for advertisement delivery by logging in to the advertisement management server 30, etc., and delivers the advertisement for the product. In other words, the analysis server 50 determines the user to whom the advertisement is to be delivered based on the analysis result. The analysis server 50 includes an input receiving unit 51, a control unit 52, and a memory unit 53.

The input reception unit 51 receives information input to the analysis server 50 using an input device such as a keyboard or mouse, or information transmitted from another device through the communication unit 54. The input reception unit 51 outputs the received information to the control unit 52 or records it in the memory unit 53. For example, the input reception unit 51 receives designation of a product or the like for which user features that contribute to the magnitude of advertising effectiveness are to be analyzed. The input reception unit 51 also receives a request for advertisement delivery from the advertising media server 30. The input reception unit 51 also receives user IDs of users to whom advertisements are to be delivered from the advertising media server 30 or the purchase management server 40.

The control unit 52 controls the analysis of user features that contribute to the magnitude of advertising effectiveness and the distribution of advertisements based on the analysis results. The control unit 52 includes a purchase probability prediction unit 521, a contribution analysis unit 522, a distribution decision unit 523, and an advertisement distribution unit 524.

The purchase probability prediction unit 521 predicts the probability that a user will purchase a product. The purchase probability prediction unit 521 predicts the probability that a user will purchase a product based on the past purchase history of a product of multiple users when an advertisement for the product was delivered and when an advertisement was not delivered. An example of a purchase history table for a product is shown in FIG. 2. As shown in the figure, the purchase history table 200 has items such as a user ID, multiple feature amounts such as age, whether or not an advertisement was delivered, and whether or not a product was purchased. For example, a user with a user ID of 001 is 21 years old, has an advertisement delivered, and has purchased a product. The feature amounts include age, sex, area of residence, and what products have been purchased in the past. The purchase history table 200 is stored in the memory unit 53. For example, the purchase probability prediction unit 521 may request the feature amounts, the number of times advertisements for the target product were delivered, and the product purchase history for all users stored in the user information management DB 21 from the DB server 20, and store this information in the purchase history table 200 of the memory unit 53. Alternatively, the purchase probability prediction unit 521 may obtain information from the advertising media server 30 that can determine whether or not an advertisement has been delivered for the target product for each user ID (for example, an advertisement delivery history, the number of deliveries, advertisement impression information, etc.), obtain information on the purchase history of the target product for each user ID from the purchase management server 40, determine whether or not an advertisement has been delivered and whether or not a purchase record has been made based on the obtained information, and store the determination result in the purchase record table 200. At this time, a user who does not have advertisement impression information or the like may be classified as having no advertisement delivery. The purchase probability prediction unit 521 uses the information stored in the purchase record table 200 as learning data, and creates a purchase prediction model in which the user's features and the presence or absence of advertisement delivery are explanatory variables and the product purchase probability is an objective variable. An example of the purchase prediction model is shown in FIG. 3. The purchase probability prediction unit 521 constructs a purchase prediction model 531 that outputs the probability of purchasing the product when the user's features and the presence or absence of advertisement delivery are inputted, by machine learning or the like. The purchase probability prediction unit 521 calculates, for example, the purchase probability with and without ad delivery for all users stored in the purchase history table 200 using the purchase prediction model 531. An example of the calculation result is shown in FIG. 4. FIG. 4(a) shows the purchase probability with ad delivery for a certain product, and FIG. 4(b) shows the purchase probability without ad delivery. For example, for a user with user ID 101, the purchase probability with ad delivery is 0.81, and the purchase probability without ad delivery is 0.54.

By creating the purchase prediction model 531 illustrated in FIG. 3, it is possible to predict the purchase probability of a user who has not purchased a product in the past based on the feature amount of the user. For a user who has purchased a product, the purchase probability may be predicted as follows. For example, the purchase probability prediction unit 521 acquires information on the number of times an advertisement for the product was delivered to the user and the number of times the advertisement was delivered that the user purchased the product. The purchase probability prediction unit 521 acquires the number of times the advertisement was delivered from the advertising media server 30, compares it with the information on the purchase date and time of the product acquired from the purchase management server 40, and, for example, if the product was purchased before a predetermined time has elapsed since the advertisement delivery and the advertisement delivery is the third advertisement delivery, it is determined that the user purchased the product at the third advertisement delivery. Alternatively, if there is log information indicating that the advertisement was clicked at the third advertisement delivery and a purchase was made thereafter, it is determined that the user purchased the product at the third advertisement delivery. In this case, the purchase probability prediction unit 521 predicts the purchase probability at the time of advertisement delivery to be 1/3. In addition, the purchase probability prediction unit 521 acquires log information stored by the purchase management server 40, and calculates the purchase probability in the absence of advertisement delivery based on information regarding the number of times the user logged in to the purchase management server 40 and accessed the product in question at the online shopping mall and the number of accesses at which the user purchased the product. For example, if the user purchased the product on the fifth access, the purchase probability prediction unit 521 predicts that the purchase probability in the absence of advertisement delivery will be 1/5.

The contribution analysis unit 522 analyzes the user's feature values that contribute to the magnitude of the advertising effect. The contribution analysis unit 522 analyzes the user's feature values that affect the magnitude of the difference between the purchase probability with and without advertising delivery (difference = "purchase probability with advertising delivery" - "purchase probability without advertising delivery"), and extracts feature values that contribute greatly to the difference. The reason why the difference between the purchase probability with and without advertising delivery is large is that the advertisement promotes purchases. Therefore, the user's feature values that contribute to the increase in the difference can be said to be the user's feature values that contribute to the magnitude of the advertising effect. For example, the contribution analysis unit 522 calculates the data of the table illustrated in FIG. 5, creates a learning model by machine learning using the user's feature values as explanatory variables and the purchase probability difference between the case with and without advertising delivery predicted by the purchase probability prediction unit 521 as the objective variable, and calculates the contribution (contribution degree) given to the magnitude of the difference for each user's feature value by performing a SHAP analysis on the created learned model. A model that indicates the user's feature values that contribute to the magnitude of the advertising effect is called a contribution model. An example of a contribution model is shown in FIG. 6. The vertical axis of FIG. 6 indicates the user feature amount, and the horizontal axis indicates the absolute value of the SHAP value. The contribution analysis unit 522 extracts feature amounts whose contribution value is equal to or greater than a predetermined value, and registers the extracted feature amounts in the storage unit 53. Alternatively, the contribution analysis unit 522 may arrange the feature amounts in order of the greatest contribution amount, extract a predetermined number of feature amounts from the top, and register them in the storage unit 53. The user feature amounts that contribute to the magnitude of the advertising effect extracted by the contribution analysis unit 522 may be registered in the user information management DB 21 for each product. In the case of the contribution model 532 illustrated in FIG. 6, the contribution analysis unit 522 extracts feature amount 1 and feature amount 2 as the user feature amounts that contribute to the magnitude of the advertising effect.

The method of calculating the contribution is not limited to SHAP analysis. For example, the correlation coefficient between each user feature and the difference may be calculated, and features with a correlation coefficient value equal to or greater than a threshold value may be extracted. Alternatively, data with a difference magnitude equal to or greater than a predetermined value may be extracted from the difference data of each user illustrated in FIG. 5, information on the user features corresponding to the extracted data may be tallied, and the frequency of appearance of the user features may be counted as the contribution. For example, in a data group with a difference magnitude equal to or greater than a predetermined value for a certain product α, if the frequency of appearance of data with the feature "age in 20s" is equal to or greater than a predetermined number of times, or if the frequency of appearance of data with the feature "has purchased related product β in the past" is equal to or greater than a predetermined number of times, the contribution analysis unit 522 may extract the features "age in 20s" and "has purchased related product A in the past" as features with high contribution.

Note that, here, we have calculated the difference in purchase probability between cases with and without ad delivery. However, instead of calculating the purchase probability, it is also possible to use other actions that can confirm that a user has become interested in a product due to an advertisement, such as the probability of accessing a web page on which the product is posted, or the probability of downloading a catalog related to the product, requesting information or making an inquiry, as index values to calculate the difference in the probability of accessing a web page for purchasing the product between cases with and without ad delivery, and analyze the feature quantities that have a strong influence on the magnitude of this difference.

The distribution decision unit 523 decides the product to be advertised. For example, when the distribution decision unit 523 receives an advertisement distribution request from the advertising media server 30, etc., it compares the user feature amount associated with the user ID with the feature amount of the user that contributes to the magnitude of the advertisement effect extracted by the contribution analysis unit 522, selects one or more products for which the feature amount corresponding to the user ID matches the feature amount extracted by the contribution analysis unit 522 as the advertisement distribution target, and decides to distribute an advertisement for a specific product from among the selected products. In the above example, if user A is in his 20s and has purchased product β in the past, the distribution decision unit 523 decides to distribute an advertisement for product α to user A. If user A is in his 30s and has not purchased product β in the past, the distribution decision unit 523 decides not to distribute an advertisement for product α to user A. Note that, in cases where there are multiple contribution amounts extracted by the contribution analysis unit 522, the decision on whether to distribute the advertisement can be arbitrarily set according to the magnitude of the contribution amount of each feature amount. For example, when the contribution of "has purchased related product A in the past" is particularly high (large SHAP value, large correlation coefficient, frequent appearance) among the features "age 20s" and "has purchased related product A in the past" and the contribution of "has purchased related product A in the past" is not so high in comparison, the advertisement for product α may be delivered to a user who has the feature of "has purchased related product A in the past" even if the user is not in his/her 20s, or the advertisement for product α may be delivered, for example, once in three times to a user who is in his/her 20s and does not have the feature of "has purchased related product A in the past." In addition, when there are multiple products that are candidates for delivery in response to a request for advertisement delivery received from the advertising medium server 30 or the like, and there are user features that match the multiple products, the delivery determination unit 523 determines to deliver an advertisement for the product from which a user feature with the highest or higher contribution has been extracted. For example, suppose that feature 1 is registered in the storage unit 53 as a feature with a high degree of contribution for product α1, feature 2 is registered as a feature with a high degree of contribution for product α2, and feature 3 is registered as a feature with a high degree of contribution for product α3, and the relationship of the magnitude of each degree of contribution (e.g., SHAP value) is contribution of feature 1 > contribution of feature 2 > contribution of feature 3. If products α1 to α3 are targets for advertisement distribution and user A1 has all of features 1 to 3, the distribution decision unit 523 decides to distribute an advertisement for product α1 corresponding to feature 1, which has the highest degree of contribution. Also, if user A2 has features 2 to 3, the distribution decision unit 523 decides to distribute an advertisement for product α2 corresponding to feature 2 to user A2.

The advertisement distribution unit 524 distributes advertisements for the products that the distribution decision unit 523 has decided to distribute to the user terminal 10.

The memory unit 53 stores various types of information. For example, the memory unit 53 stores a purchase prediction model 531 and a contribution model 532.

(Operation)
(1) Analysis of Features Contributing to Advertising Effectiveness Next, the operation of the advertisement delivery system 1 will be described.
FIG. 7 is a flowchart illustrating an example of a process for analyzing a feature amount that contributes to an advertising effect according to the embodiment.

The input reception unit 51 receives the designation of a product (step S1). The entity that inputs the designation information of the product to be analyzed to the analysis server 50 may be an advertiser, a product provider, a requester for product analysis, etc., or may be an analyst of advertising effectiveness, an administrator of the analysis server, etc. Furthermore, multiple products may be designated. If multiple products are designated, the processes of steps S2 to S5 are executed for each product.

Next, the purchase probability prediction unit 521 creates a purchase prediction model (step S2). The purchase probability prediction unit 521 creates a purchase prediction model 531 (Figure 3) that predicts the purchase probability for the product specified in step S1 based on the user's features and the presence or absence of advertisement delivery.

Next, the purchase probability prediction unit 521 performs purchase prediction (step S3). The purchase probability prediction unit 521 inputs the feature quantities of many users and the presence or absence of ad delivery to the purchase prediction model 531 created in step S2, and predicts the purchase probability with and without ad delivery for each user (Figure 4).

Next, the contribution analysis unit 522 creates a contribution model (step S4). From the prediction results of step S3, the contribution analysis unit 522 calculates the difference between the purchase probability for each user when advertising is delivered and when advertising is not delivered, and calculates the data shown in FIG. 5. The contribution analysis unit 522 analyzes the features that contribute to the magnitude of the calculated difference using SHAP analysis or the like, and creates a contribution model 532 (FIG. 6).

Next, the contribution analysis unit 522 extracts and registers features that contribute highly to the advertising effectiveness (step S5). For example, the contribution analysis unit 522 refers to the contribution model 532, extracts features whose contribution (for example, absolute value of SHAP value) is equal to or greater than a predetermined threshold, and registers the extracted features in the storage unit 53, user information management DB 21, etc. in association with the product specified in step S1. Note that information on user features that contribute highly to the advertising effectiveness extracted for each product may be provided to advertisers, product providers, product analysis clients, etc. For example, the contribution analysis unit 522 may output the extraction results of step S5 to an electronic file, etc., and transmit this electronic file to advertisers, product providers, product analysis clients, etc.

(2) Advertisement Delivery Operation FIG. 8A is a flowchart illustrating an example of an advertisement delivery process according to the embodiment.
It is assumed that a user to whom an advertisement is to be delivered is logged in to the advertising medium server 30 or the purchase management server 40. In addition, it is assumed that, for an advertisement to be delivered, a user feature amount that contributes highly to the advertising effect is analyzed in advance by the process illustrated in FIG. 7, and the result is registered in the storage unit 53.
The input reception unit 51 receives a request for advertisement delivery from the advertisement media server 30 or the like (step S11). The request for advertisement delivery may be transmitted, for example, when the user logs in to the advertisement media server 30 or the purchase management server 40, or when an advertisement space is displayed on a web page accessed by the user operating the user terminal 10. The request for advertisement delivery includes a user ID and a type of advertisement medium. The input reception unit 51 outputs to the control unit 52 that the advertisement delivery request has been received. In the control unit 52, the delivery determination unit 523 determines a product to be advertisement delivered (step S12). The delivery determination unit 523 refers to the user information management DB 21 using the user ID included in the request for advertisement delivery, and acquires various features registered in association with the user ID. In addition, the delivery determination unit 523 reads out from the storage unit 53 user features that have a high contribution to the advertisement effect and are registered in association with the product to be advertised. Each product registered in the storage unit 53 in association with a user feature that has a high contribution to the advertisement effect is a candidate for an advertisement delivery target. The distribution decision unit 523 compares the feature acquired from the user information management DB 21 with the feature read from the storage unit 53, and if there is a feature that matches the two, it decides to distribute an advertisement for the product. If there is no advertisement for a product with a matching feature, the distribution decision unit 523 decides not to distribute the advertisement. If it is decided not to distribute the advertisement (there is no product for which advertisement is distributed) (step S13; No), the processing of FIG. 8A is terminated. If it is decided to distribute the advertisement (step S13; Yes), the distribution decision unit 523 instructs the advertisement distribution unit 524 to distribute an advertisement for the determined product. If there is one product whose feature matches the user's feature and the feature extracted for the product to be advertised, the distribution decision unit 523 instructs the advertisement distribution unit 524 to distribute an advertisement for that product. When there are multiple products whose feature values of the user and the feature values extracted for the products match, the distribution decision unit 523 specifies a product for which an advertisement is to be distributed from among the multiple products, and instructs the advertisement distribution unit 524 to distribute the advertisement for the specified product. For example, the distribution decision unit 523 may specify a product corresponding to the feature value with the highest contribution as the product for which advertisement is to be distributed, or may specify a product having the most features that match the feature values of the user (a product having the most features that match when comparing multiple features with high advertising effectiveness associated with the products and multiple features owned by the logged-in user) as the product for which advertisement is to be distributed. The advertisement distribution unit 524 distributes the advertisement (step S14). The advertisement distribution unit 524 distributes the advertisement for the product specified by the distribution decision unit 523. The user terminal 10 receives the transmitted advertisement and outputs it to the display unit.

(2') Processing for distributing advertisements with points In the processing in Fig. 8A, only a determination is made as to whether or not to transmit an advertisement, but if it is determined that an advertisement is to be distributed, a further determination may be made as to whether an advertisement with points or an advertisement without points is to be distributed. The processing in this case will be described with reference to Fig. 8B.
It is assumed that users to whom advertisements are to be delivered are logged in to the advertising media server 30 or the purchase management server 40. For the advertisements to be delivered, user features that contribute highly to the effectiveness of the advertisements are analyzed in advance by the process illustrated in Fig. 7, and the results are registered in the storage unit 53. It is further assumed that a purchase probability table 201 (Fig. 9), which will be described later, is registered in the storage unit 53. Note that steps S11 to S13 are similar to the process described in Fig. 8A, and will be described briefly.

The input reception unit 51 receives a request for advertisement distribution from the advertising media server 30 (step S11). Next, the distribution decision unit 523 decides the product for which the advertisement is to be distributed (step S12). If it is decided not to distribute the advertisement (there is no product for which the advertisement is to be distributed) (step S13; No), the processing of FIG. 8B is terminated. If it is decided to distribute the advertisement (step S13; Yes), the distribution decision unit 523 decides whether to distribute an advertisement with points or an advertisement without points (step S141). For example, the memory unit 53 stores a purchase probability table 201 illustrated in FIG. 9. As shown in the figure, the purchase probability table 201 registers, for a certain product, the purchase probability without advertisement distribution, the purchase probability with advertisement distribution with points, and the purchase probability with advertisement distribution without points for each user ID. For example, instead of the purchase history table 200 illustrated in FIG. 2 , in the column for whether or not advertisement is delivered in the purchase history table 200, an option of "2: Ad delivery with points" is added in addition to "0: No advertisement delivery" and "1: Ad delivery", and a table is prepared in which, for each user, whether the product was purchased in the case of "0: No advertisement delivery", whether the product was purchased in the case of "1: Ad delivery" (equivalent to "Ad delivery without points"), or whether the product was purchased in the case of "2: Ad delivery with points", and the purchase probability prediction unit 521 creates a purchase prediction model 531' with the user's features and whether or not advertisement is delivered (the above flags 0 to 2) as explanatory variables and the purchase probability as a target variable. The purchase probability prediction unit 521 inputs the characteristic amount of the user with user ID=101 and the above flag "0" into this purchase prediction model 531' to calculate the purchase probability (1.0%) when no advertisement is delivered, inputs the characteristic amount of the user with user ID=101 and the above flag "1" to calculate the purchase probability (1.5%) when no point advertisement is delivered, inputs the characteristic amount of the user with user ID=101 and the above flag "2" to calculate the purchase probability (2.5%) when point advertisement is delivered, and registers these values in the purchase probability table 201. Similarly, the purchase probability prediction unit 521 predicts the purchase probability of other users with user ID=102, 103, etc. when no advertisement is delivered, when no point advertisement is delivered, and when point advertisement is delivered, using the purchase prediction model 531', and registers the predicted purchase rate in the purchase probability table 201. The purchase probability table 201 illustrated in FIG. 9 is a table created in advance in this way.

The distribution decision unit 523 refers to the purchase probability table 201 and determines whether to distribute an advertisement with points or without points (step S141). For example, if the purchase probability when an advertisement with points is distributed is greater than the purchase probability when an advertisement without points is distributed by a predetermined value or more, the distribution decision unit 523 decides to distribute an advertisement with points, and otherwise decides to distribute an advertisement without points. For example, in the case of the example of FIG. 9, if the user to whom the advertisement is distributed is a user with user ID = 101, the distribution decision unit 523 decides to distribute an advertisement with points, and if the user to whom the advertisement is distributed is a user with user ID = 102, the distribution decision unit 523 decides to distribute an advertisement without points. When it is decided to distribute an advertisement with points (step S141; Yes), the distribution decision unit 523 instructs the advertisement distribution unit 524 to distribute an advertisement with points for the product determined in step S13. The advertisement distribution unit 524 distributes an advertisement with points for the instructed product (step S15). If it is decided to distribute an advertisement without points (step S141; No), the distribution decision unit 523 instructs the advertisement distribution unit 524 to distribute an advertisement without points for the product decided in step S13. The advertisement distribution unit 524 distributes the advertisement without points for the instructed product (step S16). The user terminal 10 receives the transmitted advertisement and outputs it on the display unit.

In the processing of step S141 in FIG. 8B, a decision is made as to whether to deliver an advertisement with points or an advertisement without points based on the purchase probability table 201 illustrated in FIG. 9. However, for example, a difference 1 (1.5%) obtained by subtracting the purchase probability when an advertisement with points is delivered in the purchase probability table 201 (2.5% for user ID = 101) from the purchase probability when no advertisement is delivered (1.0% for user ID = 101) and a difference 2 (0.5%) obtained by subtracting the purchase probability when an advertisement without points is delivered (1.5% for user ID = 101) from the purchase probability when no advertisement is delivered (1.0% for user ID = 101) may be calculated for all users, and the contribution analysis unit 522 may analyze the user features that contribute to the magnitude of difference 1 and the user features that contribute to the magnitude of difference 2 using SHAP analysis or the like, and determine whether to deliver an advertisement with points or an advertisement without points based on the analysis results. For example, the contribution analysis unit 522 has determined through analysis that the features that contribute to the magnitude of difference 1 for the product (advertising effectiveness of the ad with points) are feature 1 and feature 2, and the features that contribute to the magnitude of difference 2 (advertising effectiveness of the ad without points) are feature 1 and feature 3. If a user to whom the ad is to be delivered has feature 1 and feature 2, the delivery decision unit 523 decides to deliver an ad with points.

(effect)
As described above, according to this embodiment, it is possible to calculate the characteristic amounts of a user that contribute to the magnitude of advertising effectiveness. This makes it possible to improve the accuracy of extracting users to whom advertisements are to be delivered when delivering advertisements. Specifically, users having the characteristic amounts of a user that contribute to the magnitude of advertising effectiveness are extracted as users to whom advertisements are to be delivered. This makes it possible to deliver advertisements to users for whom the advertising effect is high. Since advertisements are matched with users based on the user's characteristic amounts, it is possible to provide advertisements that are suitable for new users even for whom there is no data such as past product purchase history.

In the above embodiment, the feature amount of the user that contributes to the magnitude of the advertising effect for each product or service is analyzed, but the feature amount of the user that contributes to the magnitude of the advertising effect for each product or service category may be calculated. For example, the feature amount of the user that contributes to the magnitude of the advertising effect when an advertisement related to food is delivered may be analyzed from the data of the purchase record when an advertisement for a product belonging to the food category is delivered and the purchase record when no advertisement is delivered. Alternatively, the feature amount of the user that contributes to the magnitude of the advertising effect may be analyzed on a store or manufacturer basis that provides the product or service. In the above embodiment, when an advertisement delivery request is received from the advertising medium server 30, etc., the analysis server 50 determines a specific product from at least one or more candidates of products to be delivered to the medium based on the received medium and user information, and delivers an advertisement for that product. However, the advertising medium server 30 may specify a product, request the analysis server 50 to deliver the advertisement for that product, and the analysis server 50 may determine whether or not to deliver the advertisement for that product. More specifically, the delivery decision unit 523 refers to the user information management DB 21 using the user ID included in the request for advertisement delivery, acquires various features registered in association with the user ID, reads from the storage unit 53 user features that contribute highly to the advertising effect and are registered in association with the product for which the advertisement is delivered, compares the features acquired from the user information management DB 21 with the features read from the storage unit 53, and if the two match (for example, if at least one feature matches, or if a predetermined number or more features match), decides to deliver the advertisement, and may decide not to deliver the advertisement for the requested product if the features do not match. If it is decided to deliver the advertisement, the delivery decision unit 523 instructs the advertisement delivery unit 524 to deliver the advertisement for the requested product.

In the above embodiment, the feature of the user that contributes to the magnitude of the advertising effect for each product or service is analyzed. However, if there are multiple types of advertisements for one product, the feature of the user that contributes to the magnitude of the advertising effect of each advertisement may be analyzed, and an advertisement whose feature of the advertisement delivery target user matches the feature of the advertisement with a large advertising effect may be delivered. For example, assume that there are three types of advertisements γ1 to γ3 for product α, and that the feature of advertisement γ1 with a large advertising effect is feature 1, the feature of advertisement γ2 with a large advertising effect is feature 2, and the feature of advertisement γ3 with a large advertising effect is feature 3. If the feature of advertisement delivery target user A1 is feature 1 and the feature of advertisement delivery target user A2 is feature 2, the delivery decision unit 523 decides to deliver advertisement γ1 to user A1 and advertisement γ2 to user A2.

In addition, in the configuration example of the advertisement delivery system 1 illustrated in FIG. 1, the analysis server 50 is provided with the advertisement delivery unit 524, but the advertisement delivery server may be provided as a server device separate from the analysis server 50. In this configuration, the function of the delivery decision unit 523 may be implemented in the advertisement delivery server, and the user features with a high degree of contribution to the advertising effect extracted for each product extracted by the contribution analysis unit 522 may be registered in the advertisement delivery server.

FIG. 10 is a diagram showing an example of the hardware configuration of an advertising delivery system according to an embodiment. The computer 900 includes a CPU 901, a main memory device 902, an auxiliary memory device 903, an input/output interface 904, and a communication interface 905. The above-mentioned user terminal 10, DB server 20, advertising media server 30, purchase management server 40, and analysis server 50 are implemented in the computer 900. Each of the above-mentioned processes is stored in the auxiliary memory device 903 in the form of a program. The CPU 901 reads the program from the auxiliary memory device 903, expands it in the main memory device 902, and executes the above-mentioned process according to the program. The CPU 901 also secures a memory area in the main memory device 902 according to the program. The CPU 901 also secures a memory area in the auxiliary memory device 903 for storing data being processed according to the program.

In at least one embodiment, the auxiliary storage device 903 is an example of a non-transient tangible medium. Other examples of non-transient tangible media include a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, and a semiconductor memory connected via the input/output interface 904. In addition, when this program is distributed to the computer 900 via a communication line, the computer 900 that receives the program may expand the program in the main storage device 902 and execute the above-mentioned processing. In addition, the program may be for realizing part of the above-mentioned functions. Furthermore, the program may be a so-called difference file (difference program) that realizes the above-mentioned functions in combination with another program already stored in the auxiliary storage device 903.

In addition, the components in the above-described embodiments can be replaced with well-known components as appropriate without departing from the spirit of the present invention. Furthermore, the technical scope of this invention is not limited to the above-described embodiments, and various modifications can be made without departing from the spirit of the present invention.

Some or all of the above embodiments may be described as follows, but are not limited to the following:

(Appendix 1)
The analysis server is provided with: a means (contribution analysis unit 522) for calculating the contribution that the user's features make to the magnitude of a difference between an index value indicating the level of interest the user has in the target of the advertisement when an advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, based on the user's features and a means (contribution analysis unit 522) for extracting the features whose contribution satisfies a predetermined condition.

(Appendix 2)
The analysis server described in Appendix (1) further includes a means for receiving an advertisement delivery request (input receiving unit 51), a means for identifying a target of the advertisement to be delivered to the user based on characteristics of the user to be the target of the advertisement based on the delivery request and the characteristics extracted for each of the candidate targets of the advertisement to be delivered in response to the delivery request, the contribution level of which satisfies a predetermined condition (delivery decision unit 523), and a means for delivering the advertisement identified by the identification means (advertisement delivery unit 524).
The means for identifying the advertisement target may identify the advertisement target having at least one matching feature amount as the advertisement target to be delivered to the user.

(Appendix 3)
When the identification means identifies multiple targets of the advertisement having at least one feature that matches the feature of the user, the means for delivering the advertisement (advertising delivery unit 524) is the analysis server described in Appendix (2), which delivers an advertisement related to the target of the advertisement from which the feature having the greatest contribution among the matching features was extracted.

(Appendix 4)
The analysis server further includes a means (purchase probability prediction unit 521) for acquiring, in the case where an advertisement with points is provided for awarding points to a user who has received the advertisement, and an advertisement without points is provided, a first index value indicating the level of interest in the target of the advertisement held by a user to whom the advertisement is to be delivered when the advertisement with points is delivered, and a second index value indicating the level of interest in the target of the advertisement held by a user to whom the advertisement is to be delivered when the advertisement without points is delivered, and when a target of the advertisement to be delivered to the user is identified by the identification means, the delivery means (advertisement delivery unit 524) delivers the advertisement with points if the first index value is greater than the second index value, and delivers the advertisement without points if the second index value is greater than the first index value.

(Appendix 5)
The analysis server according to appendix (1) to appendix (4) further comprises: a means for accepting designation of a target of the advertisement (input accepting unit 51); a means for acquiring purchase history information of the target for each user as an index value indicating the level of interest in the target (purchase probability prediction unit 521); a means for acquiring information on whether an advertisement for the target has been delivered for each user (purchase probability prediction unit 521); and a means for calculating, based on the purchase history information and the delivery presence/absence information, a purchase probability of the target for each user when the advertisement is delivered as the index value when the advertisement is delivered, and a means for calculating a purchase probability of the target when the advertisement is not delivered as the index value when the advertisement is not delivered (purchase probability prediction unit 521).

(Appendix 6)
The analysis server described in Supplementary Notes (1) to (5) further includes a means (purchase probability prediction unit 521) for calculating, based on a prediction model in which the user's features and whether or not the advertisement is delivered are explanatory variables and the index value is a target variable, an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered.

(Appendix 7)
The means for calculating the contribution degree (contribution degree analysis unit 522) is an analysis server described in Supplementary Notes (1) to (6), which calculates the contribution degree by performing SHAP analysis on a machine learning model in which the user's features are used as explanatory variables and the difference is used as a target variable.

(Appendix 8)
The means for calculating the degree of contribution (contribution analysis unit 522) is an analysis server described in Supplementary Notes (1) to (7), which extracts the users whose difference is equal to or greater than a predetermined value, aggregates the extracted user features, and regards the frequency of appearance of the user features in the aggregated result as the degree of contribution.

(Appendix 9)
The means for calculating the degree of contribution (contribution analysis unit 522) is an analysis server described in Supplementary Notes (1) to (8), which extracts features whose contribution degree is equal to or greater than a predetermined value, or arranges the features in order of greatest contribution degree and extracts a predetermined number of the features from the top, and regards the extracted features as the features of the user that contribute to the magnitude of the advertising effect.

(Appendix 10)
The means for calculating the contribution level (contribution level analysis unit 522) is an analysis server described in Supplementary Notes (1) to (9) that calculates the features for each product, service, product or service category, and company that provides a product or service.

(Appendix 11)
The analysis server described in appendix (1) to appendix (10) is one of the following: the probability of delivering the target of the advertisement, the probability of accessing a web page in which the target of the advertisement is described, and the probability of requesting information or making an inquiry about the target of the advertisement.

(Appendix 12)
An advertisement delivery system comprising: an analysis server as described in Supplementary Notes (1) to (11); a delivery server comprising: a means for receiving an advertisement delivery request; a means for identifying a target of the advertisement to be delivered to the user based on characteristics of a user to be a target of the advertisement based on the delivery request and the characteristics extracted for each of the candidate targets of the advertisement to be delivered in response to the delivery request, the contribution level of which satisfies a predetermined condition; and a means for delivering the advertisement identified by the identification means.

(Appendix 13)
A terminal device that receives an advertisement transmitted by the analysis server citing supplementary note (2) or the advertisement distribution system described in supplementary note 12.

(Appendix 14)
An analysis method executed by a computer, comprising: a step of calculating, based on a user's feature values and a difference between an index value indicating the level of interest the user has in the target of the advertisement when an advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, a degree of contribution that the user's feature values make to the magnitude of the difference; and a step of extracting the feature values whose degree of contribution satisfies a predetermined condition.

(Appendix 15)
This program causes a computer to function as a means for calculating the contribution that a user's feature values makes to the magnitude of a difference between an index value indicating the level of interest the user has in the target of the advertisement when an advertisement is delivered and an index value indicating the level of interest the user has in the target of the advertisement when the advertisement is not delivered, based on the user's feature values and a means for extracting the feature values whose contribution level satisfies a predetermined condition.

1...Advertisement distribution system 10...User terminal 20...DB server 21...User information management DB
30: Advertising medium server 40: Purchase management server 50: Analysis server 51: Input reception unit 52: Control unit 521: Purchase probability prediction unit 522: Contribution degree analysis unit 523: Distribution decision unit 524: Advertisement distribution unit 53: Storage unit 531: Purchase prediction model 532: Contribution degree model 900: Computer 901: CPU
902: Main memory device 903: Auxiliary memory device 904: Input/output interface 905: Communication interface

Claims (14)

a means for calculating a contribution of the user's feature amount to the magnitude of a difference between an index value indicating a probability that the user will take an action on the target of the advertisement when the advertisement is delivered and an index value indicating a probability that the user will take an action on the target of the advertisement when the advertisement is not delivered , based on the feature amount including demographic information of the user and user attribute information indicating a hobby, preference or tendency of the user; and
means for extracting the feature amount whose contribution degree satisfies a predetermined condition;
An analysis server comprising: A means for accepting requests to deliver advertisements;
a means for identifying, as a target of the advertisement to be delivered to the user, a target of the advertisement having at least one feature that matches the feature of the user, based on a feature of a user to which the advertisement related to the delivery request is to be delivered and the feature extracted for each candidate target of the advertisement to be delivered in response to the delivery request, the contribution degree of which satisfies a predetermined condition;
a means for delivering the advertisement identified by the identifying means;
The analysis server of claim 1 , further comprising: When the specifying means specifies a plurality of targets of the advertisement having at least one feature value matching the feature value of the user,
the means for delivering the advertisement delivers an advertisement related to the advertisement target from which the feature quantity having the greatest contribution degree was extracted among the matching feature quantities.
The analysis server according to claim 2 . The advertisement further comprises a means for acquiring, when a point-bearing advertisement for awarding points to a user to whom the advertisement is delivered and a non-point-bearing advertisement are present, a first index value indicating the probability that a user to whom the advertisement is delivered will take an action on the target of the advertisement when the point-bearing advertisement is delivered, and a second index value indicating the probability that a user to whom the advertisement is delivered will take an action on the target of the advertisement when the non -point-bearing advertisement is delivered,
When a target of the advertisement to be delivered to the user is identified by the identifying means,
the distribution means distributes the point-added advertisement when the first index value is greater than the second index value;
When the second index value is greater than the first index value, the no-point advertisement is delivered.
The analysis server according to claim 2 . A means for accepting designation of a target of the advertisement;
A means for acquiring purchase history information of the target of the advertisement for each user;
A means for acquiring information on whether or not an advertisement is delivered to the target of the advertisement for each user;
a means for calculating, for each user, a purchase probability of the target of the advertisement when the advertisement is delivered as the index value when the advertisement is delivered, based on the purchase record information and the delivery/non-delivery information, and calculating a purchase probability of the target of the advertisement when the advertisement is not delivered as the index value when the advertisement is not delivered;
The analysis server according to claim 1 or 2, further comprising: a means for calculating the index value that the user has when the advertisement is delivered and the index value that the user has when the advertisement is not delivered based on a prediction model in which the feature amount of the user and whether or not the advertisement is delivered are explanatory variables and the index value is a response variable;
The analysis server according to claim 1 or 2, further comprising: The means for calculating the degree of contribution calculates the degree of contribution by performing a SHAP analysis on a machine learning model in which the feature amount of the user is an explanatory variable and the difference is an objective variable.
The analysis server according to claim 1 or 2. the means for calculating the degree of contribution extracts the users whose difference is equal to or greater than a predetermined value, aggregates the extracted features of the users, and determines the frequency of appearance of the features of the users in the aggregated result as the degree of contribution.
The analysis server according to claim 1 or 2. the means for calculating the degree of contribution extracts a feature quantity whose degree of contribution is equal to or greater than a predetermined value, or arranges the feature quantities in order of the degree of contribution, extracts a predetermined number of the feature quantities from the top, and regards the extracted feature quantities as the feature quantities of the user that contribute to the magnitude of the advertising effect.
The analysis server according to claim 1 or 2. the means for calculating the degree of contribution calculates the feature amount for each product, service, product or service category, and company providing the product or service;
The analysis server according to claim 1 or 2. The index value is any one of the following: a probability of delivering the target of the advertisement, a probability of accessing a web page on which the target of the advertisement is posted, and a probability of making a request for information or an inquiry about the target of the advertisement.
The analysis server according to claim 1 or 2. The analysis server according to claim 1 ;
means for receiving an advertisement delivery request; means for identifying, as an advertisement target to be delivered to the user, an advertisement target having at least one feature that matches the user's feature, based on a feature of a user to be a delivery target of the advertisement related to the delivery request and the feature extracted for each candidate target of the advertisement to be delivered in response to the delivery request, the contribution degree of which satisfies a predetermined condition; and means for delivering the advertisement identified by the identifying means;
A distribution server comprising:
An advertisement delivery system comprising: 1. A computer-implemented analysis method comprising:
a step of calculating a degree of contribution of the user's feature amount to the magnitude of the difference based on a feature amount including demographic information of the user and user attribute information indicating a user's hobbies, preferences, or tendencies , and an index value indicating a probability that the user will take an action on the target of the advertisement when the advertisement is delivered and a difference between the index value indicating a probability that the user will take an action on the target of the advertisement when the advertisement is not delivered;
extracting the feature amount whose contribution degree satisfies a predetermined condition;
The analytical method has the following features: Computer,
a means for calculating a contribution of the user's feature amount to the magnitude of a difference between an index value indicating a probability that the user will take an action on the target of the advertisement when the advertisement is delivered and an index value indicating a probability that the user will take an action on the target of the advertisement when the advertisement is not delivered , based on the feature amount including the user's demographic information and user attribute information indicating a hobby, preference or tendency;
means for extracting the feature quantity whose contribution degree satisfies a predetermined condition;
A program to function as a

Images