Information processing device, information processing method, and information processing program

Abstract

A technique is provided for calculating an uplift score that indicates purchasing propensity. [Solution] An information processing device divides a user group consisting of multiple users into a first user group in which an intervention was made to stimulate a predetermined conversion, estimates a first probability which is the probability that the users who achieved the conversion are included in the first user group, and a second probability which is the probability that the users who achieved the conversion are included in the second user group, estimates a third probability which is the probability that the entire user group achieved the conversion, and calculates an uplift score which represents the effect of the intervention on the user group using the first probability, the second probability, and the third probability.

Description

[Technical Field]

[0001] The present invention relates to a technique for calculating an uplift score. [Background technology]

[0002] In the field of advertising, it is difficult to know whether a user's purchases following a given intervention (e.g., the implementation of a marketing measure such as the distribution of coupons) can truly be considered the effect of the intervention. However, gaining such knowledge can improve the efficiency of marketing activities. For example, when distributing advertisements online that include coupons, it is possible to select as targets users who will make a purchase if they receive the coupon, and distribute the advertisements only to these targets. In this way, by narrowing down the targets to whom the advertisements are expected to be effective, it is expected that marketing activities will become more effective and efficient.

[0003] Uplift modeling is known as a method for selecting targets that are expected to improve the effectiveness of an intervention, as described above. Uplift modeling is a method for estimating which users should be targeted for intervention in order to improve the effectiveness of the intervention. In uplift modeling, an uplift score, which is an index for target selection, is calculated, and targets are estimated based on the calculated uplift score.

[0004] The uplift score can be calculated as the difference between the probability of purchase when an intervention is performed and the probability of purchase when no intervention is performed, as described in Patent Document 1, for example. Since it can be said that the higher the uplift score, the greater the effect of the intervention, by selecting a group of users with higher uplift scores as targets, it is possible to carry out marketing activities that can be expected to have a high effect from the intervention. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2024-131197 [Non-patent literature]

[0006] [Non-Patent Document 1] Dmitri Goldenberg et al., “Free Lunch! Retrospective Uplift Modeling for Dynamic Promotions Recommendation within ROI Constraints”, August 2020, arXiv: 2008.06293. Summary of the Invention [Problem to be solved by the invention]

[0007] As disclosed in Patent Document 1, according to a conventional uplift calculation method, the uplift score can be calculated by the difference between the probability of purchase when an intervention is performed and the probability of purchase when no intervention is performed. However, this calculation method does not take into account the probability that all users actually purchase (i.e., the probability of conversion). By taking into account the probability that all users purchase, the uplift score, which is an index for target selection, can become an index that better represents purchase propensity.

[0008] In view of the above-mentioned problems, the present disclosure aims to provide a technique for calculating an uplift score that indicates purchasing tendency. [Means for solving the problem]

[0009] In order to solve the above problem, one aspect of an information processing device according to the present invention includes a grouping unit that groups a user group consisting of multiple users into a first user group that has intervened to induce a predetermined conversion and a second user group that has not intervened; a first estimation unit that estimates a first probability that is the probability that a user who has achieved the conversion is included in the first user group and a second probability that is the probability that a user who has achieved the conversion is included in the second user group; a second estimation unit that estimates a third probability that is the probability that the entire user group has achieved the conversion; and a calculation unit that calculates an uplift score that represents the effect of the intervention on the user group using the first probability, the second probability, and the third probability.

[0010] In order to solve the above problem, one aspect of the information processing method according to the present invention includes grouping a user group consisting of multiple users into a first user group that has intervened to induce a predetermined conversion and a second user group that has not intervened; estimating a first probability that is the probability that a user who has achieved the conversion is included in the first user group, and a second probability that is the probability that a user who has achieved the conversion is included in the second user group; estimating a third probability that is the probability that the entire user group has achieved the conversion; and calculating an uplift score that represents the effect of the intervention on the user group using the first probability, the second probability, and the third probability. include.

[0011] In order to solve the above problem, one aspect of the information processing program of the present invention is an information processing program for causing a computer to execute information processing, the program causing the computer to execute processes including: a grouping process for grouping a user group consisting of multiple users into a first user group that has intervened to stimulate a predetermined conversion and a second user group that has not intervened; a first estimation process for estimating a first probability that the user who achieved the conversion is included in the first user group and a second probability that the user who achieved the conversion is included in the second user group; a second estimation process for estimating a third probability that the entire user group achieved the conversion; and a calculation process for calculating an uplift score that represents the effect of the intervention on the user group using the first probability, the second probability, and the third probability. [Effects of the Invention]

[0012] According to the present invention, a technique is provided for calculating an uplift score that indicates a propensity to purchase. The above-mentioned objects, aspects, and advantages of the present invention, as well as other objects, aspects, and advantages of the present invention not described above, will be understood by those skilled in the art from the following detailed description of the invention by referring to the accompanying drawings and the claims. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 shows an example of the configuration of an information processing system according to an embodiment. [Figure 2] FIG. 2 shows a conceptual diagram of an uplift score calculation process according to one embodiment. [Figure 3] FIG. 3 shows an example of the functional configuration of an e-commerce server according to an embodiment. [Figure 4] FIG. 4 shows an example of the functional configuration of an information processing device according to an embodiment. [Figure 5] FIG. 5 shows an example of the hardware configuration of an information processing device according to an embodiment. [Figure 6]FIG. 6 shows a flowchart of a process executed by an information processing system according to one embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, with reference to the accompanying drawings, an embodiment for carrying out the present invention will be described in detail. Among the components disclosed below, those having the same function will be given the same reference numerals, and their description will be omitted. Note that the embodiment disclosed below is an example of a means for realizing the present invention, and should be appropriately modified or changed depending on the configuration of the device to which the present invention is applied and various conditions. The present invention is not limited to the following embodiment. Furthermore, not all of the combinations of features described in the present embodiment are necessarily essential to the solution of the present invention.

[0015] [Configuration of information processing system] FIG. 1 shows an example configuration of an information processing system 1 according to this embodiment. The information processing system 1 includes an information processing device 10, an e-commerce server 11, and a user device 12. The information processing device 10, the e-commerce server 11, and the user device 12 are configured to be able to communicate with each other via a network 13. The network 13 may include the Internet, an intranet, a local area network (LAN), a wide area network (WAN), a mobile communication network, and the like. Although FIG. 1 illustrates one user device 12, the information processing system 1 may include multiple user devices having similar functions to the user device 12. In the present disclosure, the multiple user devices are collectively referred to as the user device 12. The user device 12 is operated by a user 14. In the present disclosure, the terms "user device" and "user" may be understood to be synonymous.

[0016] The e-commerce server 11 is a server device capable of providing services for e-commerce. In this disclosure, the e-commerce server 11 is described as a server device that operates an e-commerce mall (mall-type e-commerce site) in which multiple stores operate an online shopping mall. Specifically, the e-commerce server 11 operates an e-commerce mall that operates a shopping mall via sales pages (web pages) of products sold by multiple stores (merchants). Items including tangible and / or intangible products and services are sold in the e-commerce mall. The following description focuses on products, but similar descriptions can be applied to other items.

[0017] The e-commerce server 11 can accept access from a user 14 from a user device 12 via a network 13 and provide the user 14 with various services related to shopping at an e-commerce mall. For example, when the user 14 accesses the e-commerce mall and performs an action such as a purchase or a click on a sales page for a product, the e-commerce server 11 provides the user 14 with a service related to the product. The e-commerce server 11 can also acquire (collect) and manage information regarding the user 14's clicks and purchases on sales pages for products sold at the e-commerce mall. In the present disclosure, an action such as the purchase of an item provided in an e-commerce service provided by the e-commerce server 11, which is expected of a user using the service, is referred to as a conversion. For example, a conversion is an action by the user 14, such as a click on a sales page or a purchase, and includes an action that the operator of the e-commerce server 11 expects the user 14 to take on the e-commerce mall.

[0018] The e-commerce server 11 can perform any intervention to encourage conversions by users who use e-commerce services. For example, the e-commerce server 11 performs the intervention by delivering advertisements related to the e-commerce service (an e-commerce mall in this embodiment). The advertisements may include information such as coupons and discount vouchers that can be used for the service. The e-commerce server 11 may deliver the advertisements, for example, by banners in the e-commerce mall accessed by the user, or may deliver or provide the advertisements by email or other means.

[0019] In the present disclosure, a user who performs any (e.g., predetermined) intervention (treat) is also referred to as a "treated user." On the other hand, in the present disclosure, controlling a service or the like but not performing any predetermined intervention is referred to as "control," and a user who does not perform any intervention is also referred to as a "control user." The e-commerce server 11 can determine which user to treat as the treated user based on the uplift score calculated by the information processing device 10, as will be described later.

[0020] The user 14 can operate the user device 12 to access the e-commerce server 11 and receive services provided by the e-commerce server 11. In the present disclosure, the user 14 can receive various services at an e-commerce mall provided by the e-commerce server 11. For example, the user 14 can operate the user device 12 to access the e-commerce mall, browse sales pages for various products offered at the e-commerce mall, and purchase products sold on the sales pages. The user 14 can purchase a product by adding the product to a cart on the product sales page, entering predetermined information, and making a payment.

[0021] In order to use services at an e-commerce mall, the user 14 registers information about the user 14 (hereinafter also referred to as user attributes). In one embodiment, the user 14 sets up an account in which a user ID (user identification information) that identifies the user 14 and is linked to the user attributes of the user 14 is set up. The user 14 then logs in to the e-commerce mall provided by the e-commerce server 11 using the user 14's account and uses services at the e-commerce mall. By setting the user ID, the user 14 can use services at the e-commerce mall even from a user device other than the user device 12 connected to the network 13. The types and number of user attributes that the user 14 should register can be determined in advance by the e-commerce server 11.

[0022] User attributes will be described using the example of user attributes of user 14. User attributes of user 14 may include the address, name, and demographic information of user 14 (demographic information about user 14, such as gender, age, residential area, occupation, and family composition). User attributes of user 14 may also include a registration number and registered name when using web services, including e-commerce malls, provided by e-commerce server 11. User attributes of user 14 may also include information about the usage history, search history, product purchase history, and points that can be accumulated by using services, of web services provided by e-commerce server 11, including e-commerce malls provided by e-commerce server 11. Thus, user attributes of user 14 may include any attribute information, including information related to user 14 himself or herself and information about the usage history of web services, including e-commerce malls. Attributes related to such registrations by user 14 and actual usage history of web services may also be referred to as actual user attributes of user 14.

[0023] The user attributes of the user 14 may include the actual user attributes of the user 14 as well as the estimated user attributes of the user 14. The estimated user attributes of the user 14 may be estimated, for example, by a trained user attribute estimation model, based on the actual user attributes of the user 14. The estimated attributes of the user may include products that the user 14 is estimated to be interested in, preferences, and / or a lifestyle estimated for the user 14.

[0024] The user device 12 is an information processing device such as a smartphone, a mobile phone, or a PDA (Personal Digital Assistant) tablet terminal. The user device 12 is configured to be able to communicate with the e-commerce server 11 and the information processing device 10 via a network 13. The user device 12 has a display unit (display surface) such as a liquid crystal display, and a user 14 can perform various operations using a GUI (Graphical User Interface) provided on the display unit. These operations include various operations on content such as images displayed on the screen, such as tapping, sliding, and scrolling using a finger or a stylus. The user device 12 may have a separate display unit, or may be a notebook PC (Personal Computer) or a desktop PC.

[0025] The e-commerce server 11 can acquire user attributes of users 14 who access the provided e-commerce mall in association with the user ID of the users 14. The e-commerce server 11 can also acquire conversion information by the users 14 (for example, a conversion history with time information attached or a history of whether or not conversions occurred over a certain period of time) in association with the user ID. The e-commerce server 11 provides the user attributes and conversion information associated with the user ID to the information processing device 10.

[0026] The information processing device 10 calculates an uplift score based on user attributes and conversion information associated with the user ID, which are acquired from the e-commerce server 11. The uplift score corresponds to an index for selecting targets for increasing the intervention effect. The calculation procedure for the uplift score according to this embodiment will be described later. Note that in FIG. 1, the information processing device 10 and the e-commerce server 11 are configured as separate devices, but the two devices may be configured as a single device, and for example, the information processing device 10 may be configured to include the functions of the e-commerce server 11.

[0027] [How the Uplift Score is calculated] A method for calculating the uplift score will be described. In the present disclosure, the uplift score is calculated for each user group, which is a group of one or more users having one or more common (identical) user attributes. In the present disclosure, one or more user groups having one or more common user attributes are referred to as a user group having user feature = X. When different user attributes are represented as x1, x2, ..., the user feature = X may be expressed as a vector of one or more user attributes (x1, x2, ...). For the sake of explanation, the presence or absence of intervention is represented by a variable t, with intervention represented as t = 1 and no intervention represented as t = 0. Furthermore, the user group with intervention (t = 1) is referred to as the intervention user group, and the user group without intervention (t = 0) is referred to as the control user group. Furthermore, the presence or absence of conversion (i.e., whether conversion was achieved or not) is represented by a variable y, with conversion represented as y = 1 and no conversion represented as y = 0.

[0028] (A) Conventional method for calculating Uplift Score First, we will explain the conventional method for calculating the uplift score. The probability of conversion (y=1) by the intervention user group (t=1) among the user group with user characteristic = X is expressed as p(y=1|t=1,X). Also, the probability of conversion (y=1) by the control user group (t=0) among the user group with user attribute = X is expressed as p(y=1|t=0,X). The conventional uplift score can be calculated using the difference between these two probabilities using equation (1).

number

[0029] The probabilities p(y=1|t=1,X) and p(y=1|t=0,X) can be estimated by machine learning. For this purpose, the information processing device 10 may use a certain amount of data of user attributes and conversion information associated with a user ID acquired from the e-commerce server 11 for learning, and may use the remaining data for estimation (prediction).

[0030] To estimate the probability p(y=1|t=1,X) and the probability p(y=1|t=0,X) and calculate the uplift score, for example, the following approaches can be used: S (Single)-Learner, which is a single ML (machine learning) model; T (Two)-Leaner, which is two separate ML models; or Class Transformation. Below, we will explain (A-1) S-Learner, (A-2) T-Learner, and (A-3) Class Transformation.

[0031] (A-1) S-Learner S-Learner is an approach that uses a single model with the presence or absence of intervention as a feature. Using a variable t that represents the presence or absence of intervention, the probability of conversion (y=1) is estimated for cases with intervention (t=1) and without intervention (t=0) as shown in equation (2).

number

number

[0032] (A-2) T-Learner T-Learner is an approach that uses two models, one for intervention (t=1) and one for no intervention (t=0). Using separate models for intervention (t=1) and no intervention (t=0), the probability of conversion is estimated as shown in equation (4).

number

number

[0033] (A-3) Class Conversion In class conversion, the objective variable Z, which represents the expected result, is defined as in the following equation (6).

number

number

[0034] The conventional uplift score will be explained using a numerical example. Here, we assume a case where the conversion is "purchase" and there is a first user group with user characteristic = X1 and a second user group with user characteristic = X2. For the first user group and the second user group, the probability of purchase with intervention and the probability of purchase without intervention are as follows: First group of users: Probability of purchasing with intervention = 0.7 Probability of purchasing without intervention = 0.6 Secondary user group: Probability of purchasing with intervention = 0.3 Probability of purchasing without intervention = 0.2

[0035] According to the conventional method for calculating the uplift score, the uplift score is 0.7-0.6=0.1 for the first user group and 0.3-0.2=0.1 for the second user group, so the uplift score is 0.1 for both user groups. In other words, the uplift scores for the first user group and the second user group are the same value.

[0036] On the other hand, the overall average purchase probability for the first user group and the second user group is ((0.7 + 0.6) / 2) = 0.65 for the first user group and ((0.3 + 0.2) / 2) = 0.25 for the second user group, which is higher for the first user group. This means that the first user group is more likely to make a purchase regardless of whether or not there is intervention. What can be seen from these numerical results is that even if the uplift score, which can be calculated from the difference between the probability of purchase with intervention and the probability of purchase without intervention, is the same value, there can be a difference in the average purchase probability. Since a user group with a higher average purchase probability is more likely to make a purchase as a whole, regardless of whether or not there is intervention, in the above example, it can be said that it is preferable to select the first user group as the target.

[0037] As described above, the conventional uplift score calculation method uses the difference between the probability of purchase with intervention and the probability of purchase without intervention, but the calculated uplift score does not reflect the purchase probability of the entire user group with and without intervention. As a result, there is a possibility that the target user group cannot be appropriately selected.

[0038] (B) Uplift Calculation Method According to the Present Embodiment An uplift calculation method according to the present embodiment, which is configured to solve the above problem, will be described. In this embodiment, a method for calculating an uplift score using the probability that the entire user group achieves conversion is proposed. For the sake of explanation, hereinafter, the intervention user group, which is a user group with intervention (t=1), will be represented as t, and the control user group, which is a user group without intervention (t=0), will be represented as c.

[0039] First, for a group of users with user characteristics = X, the probability of conversion (y = 1) among the intervention user group t, p(y = 1 | t, X), is expressed as equation (8) based on Bayes' theorem.

number

[0040] Next, for a group of users with user characteristics = X, the probability of conversion (y = 1) among the control user group c, p(y = 1|c, X), is expressed as equation (9) based on Bayes' theorem.

number

[0041] Next, the uplift score is calculated using the probability p(t|y=1,X), probability p(c|y=1,X), probability p(y=1|X), probability p(t|X), and probability p(c|X) using equation (10).

number

[0042] The probabilities p(y=1|t(=t=1),X) and p(y=1|c(=t=0),X) can be estimated using the S-Learner and T-Learner approaches as described above. Furthermore, the probability p(y=1|X) can be estimated based on the purchase history of a user group whose user characteristic is X. In the present disclosure, the probabilities p(t|X) and p(c|X) sum to 1 and are each a preset constant. That is, a user group whose user characteristic is X is divided into a user group with intervention and a user group without intervention in advance. For example, if the probabilities p(t|X) and p(c|X) are each 0.5, this means that, of the multiple users in the target population, half (or almost half) are users with intervention and a user group without intervention. Therefore, the probabilities p(t|y=1,X) and p(c|y=1,X) can be estimated from equations (8) and (9).

[0043] As can be seen from equation (10), the uplift score according to the present disclosure is calculated using the purchase probability p(y=1|X) of all users with user characteristic=X, that is, all users with and without intervention who have user characteristic=X. Next, two methods for calculating the uplift score expressed by equation (10) will be described.

[0044] (B-1) Retrospective method We will explain a method for calculating the uplift score expressed by equation (10) using the retrospective method described in Non-Patent Document 1. This method uses two ML models, a first ML model and a second ML model.

[0045] The first ML model is an ML model trained to estimate probability p(t|y=1,X) and probability p(c|y=1,X) according to the retrospective uplift model described in Non-Patent Document 1. The probability p(t|y=1,X) and probability p(c|y=1,X) are the probabilities that a user with conversion among a user group with user feature = X belongs to the intervention user group t or the control user group c, respectively, and the information processing device 10 can train the first ML model using only data with conversion (y=1). This makes it possible to reduce the amount of data used for estimation. The trained first ML model is used to estimate probability p(t|y=1,X) and probability p(c|y=1,X). The second ML model is an ML model trained to estimate the probability p(y=1|X). The trained second ML model is used to estimate the probability p(y=1|X). The information processing device 10 calculates the uplift score using the probability p(t|y=1,X), probability p(c|y=1,X), and probability p(y=1|X) estimated as described above, and the constant probabilities p(t|X) and p(c|X), using equation (10).

[0046] (B-2) Direct classification method In the direct classification method, one ML model is trained using the following three class labels, which are defined as follows:

number

[0047] Using these three class labels, the information processing device 10 trains an ML model to estimate the following three probabilities.

number

[0048] The uplift score using the direct classification method will be explained using a numerical example. Assume a case where the conversion is "purchase" and there is a first user group with user attribute = X1 and a second user group with user attribute = X2. Assume that the probabilities of p(0), p(1), and p(2) for the first and second user groups are estimated as follows: First group of users: p(0)=0.5, p(1)=0.4, p(2)=0.1 Secondary user group: p(0)=0.2, p(1)=0.1, p(2)=0.7

[0049] When the probabilities p(t|X) and p(c|X) are each 0.5, the uplift score calculated using equation (10) is 0.045 for the first user group and 0.015 for the second user group. According to the uplift score calculation method of this embodiment, it is possible to calculate uplift scores that reflect the purchase probabilities for the first user group and the second user group.

[0050] Although two methods have been described above, the information processing device 10 may estimate the probability p(t|y=1,X), the probability p(c|y=1,X), and the probability p(y=1|X) using other methods. Then, the information processing device 10 may calculate the uplift score by equation (10) using the estimated probability p(t|y=1,X), the probability p(c|y=1,X), and the probability p(y=1|X) and the constant probabilities p(t|X) and p(c|X).

[0051] [Conceptual diagram of the Uplift Score calculation process] Fig. 2 shows a conceptual diagram of the uplift score calculation process according to this embodiment. As described above, the uplift score is calculated as shown in formula (10). Here, as shown in Fig. 2, estimated values ​​are used for the probabilities p(t|y=1,X), p(c|y=1,X), and p(y=1|X), and preset constants are used for the probabilities p(t|X) and p(c|X). Then, using these values, the uplift score is calculated according to formula (10).

[0052] [E-commerce server configuration] 3 shows an example of the functional configuration of the e-commerce server 11 according to this embodiment. Here, only the functional configuration related to the processing of this embodiment is shown, and other functional configurations, including the functional configuration for providing services related to e-commerce, are omitted. The e-commerce server 11 has, as an example of its functional configuration, a user attribute acquisition unit 301, a conversion information acquisition unit 302, a user information provision unit 303, an intervening user determination unit 304, and an advertisement distribution unit 305. For the sake of explanation, reference will be made to a user device 12 and a user 14, but similar explanations can be applied to other user devices and users that can communicate with the e-commerce server 11.

[0053] The user attribute acquisition unit 301 acquires one or more user attributes of the user 14 associated with the user ID of the user 14 from the user device 12. The conversion information acquisition unit 302 acquires conversion information of the user 14 in the e-commerce mall provided by the e-commerce server 11 associated with the user ID of the user 14 from the user device 12 (for example, a conversion history with time information attached or whether or not a conversion occurred within a certain period of time). The user information provision unit 303 provides one or more user attributes of the user 14 associated with the user ID of the user 14 and the conversion information to the information processing device 10. The user information provision unit 303 may provide the user attributes and conversion information of the user 14 to the information processing device 10 each time it acquires them, or may provide the user attributes and conversion information of the user 14 accumulated over a certain period of time to the information processing device 10 each time it acquires them. The intervention user determination unit 304 acquires the uplift score for each user group calculated by the information processing device 10, and determines which users to set as intervention users, i.e., the intervention user group (also referred to as target users). Specifically, the intervention user determination unit 304 determines a group of intervention user IDs to set as the intervention user group. The advertisement distribution unit 305 functions as an intervention unit that performs intervention, and distributes advertisements that encourage conversions, such as purchases, to the intervention user group. Specifically, the advertisement distribution unit 305 distributes advertisements to user IDs identified by the intervention user ID group. Note that the advertisements may be generated by the advertisement distribution unit 305, may be generated by a predetermined function within the e-commerce server 11, or may be generated and acquired by another device. In the present embodiment, the advertisement distribution unit 305 performs intervention by distributing advertisements, but may be configured to perform other forms of intervention.

[0054] [Configuration of information processing device] 4 shows an example of the functional configuration of the information processing device 10 according to this embodiment. As an example of its functional configuration, the information processing device 10 has a user information acquisition unit 401, a user group generation unit 402, an uplift score calculation unit 403, an uplift score provision unit 404, a user information storage unit 410, and a learning model storage unit 420. The learning model storage unit 420 is configured to be able to store one or more machine learning models (e.g., data representing architectures and various parameters) used to calculate the uplift score. For the sake of explanation, the user device 12 and the user 14 will be referred to, but similar explanations can be applied to other user devices and users.

[0055] The user information acquisition unit 401 acquires user information including one or more user attributes of the user 14 associated with the user ID of the user 14 and conversion information from the e-commerce server 11. The user information acquisition unit 401 stores the acquired user information in the user information storage unit 410. The user group generation unit 402 generates one or more user groups by grouping one or more users having one or more common (identical) user attributes based on the user attributes stored in the user information storage unit 410. The uplift score calculation unit 403 calculates an uplift score for each user group generated by the user group generation unit 402. The method for calculating the uplift score is as described above. The uplift score calculation unit 403 can calculate the uplift score using one or more machine learning models stored in the learning model storage unit 420. The uplift score provision unit 404 provides the calculated uplift score for each user group to the e-commerce server 11.

[0056] [Hardware configuration of information processing device] Next, a description will be given of an example of the hardware configuration of the information processing device 10. Fig. 5 is a block diagram showing an example of the hardware configuration of the information processing device 10 according to this embodiment. The hardware configuration of the e-commerce server 11 is also similar. The information processing device 10 according to this embodiment can be implemented on a single or multiple computers, mobile devices, or any other processing platform. 5, the information processing device 10 is illustrated as being implemented in a single computer, but the information processing device 10 according to this embodiment may be implemented in a computer system including multiple computers. The multiple computers may be connected to each other via a wired or wireless network so as to be able to communicate with each other.

[0057] 5, the information processing device 10 may include a CPU (Central Processing Unit) 501, a ROM (Read Only Memory) 502, a RAM (Random Access Memory) 503, a HDD (Hard Disk Drive) 504, an input unit 505, a display unit 506, a communication I / F (communication unit) (interface) 507, and a system bus 508. The information processing device 10 may also include an external memory. The CPU 501 controls the overall operation of the information processing device 10, and controls each component (502 to 507) via a system bus 508, which is a data transmission path.

[0058] The ROM 502 is a non-volatile memory that stores control programs and the like necessary for the CPU 501 to execute processes. The programs include instructions (codes) for executing the processes according to the above-described embodiments. The programs may be stored in a non-volatile memory such as the HDD 504 or an SSD (Solid State Drive), or in an external memory such as a removable storage medium (not shown). The RAM 503 is a volatile memory and functions as the main memory, work area, etc. of the CPU 501. That is, when executing a process, the CPU 501 loads necessary programs, etc. from the ROM 502 into the RAM 503 and executes the programs, etc. to realize various functional operations. The RAM 503 may include the user information storage unit 410 and the learning model storage unit 420 shown in FIG. 4.

[0059] The HDD 504 stores, for example, various data and information required when the CPU 501 performs processing using a program. The HDD 504 also stores, for example, various data and information obtained when the CPU 501 performs processing using a program. The input unit 505 is configured with a keyboard and a pointing device such as a mouse. The display unit 506 is configured by a monitor such as a liquid crystal display (LCD), etc. The display unit 506 may be configured in combination with the input unit 505 to function as a GUI (Graphical User Interface).

[0060] The communication I / F 507 is an interface that controls communication between the information processing device 10 and an external device. The communication I / F 507 provides an interface with a network and executes communication with the external device via the network. Various data, parameters, and the like are transmitted and received between the information processing device 10 and the external device via the communication I / F 507. In this embodiment, the communication I / F 507 may execute communication via a wired LAN (Local Area Network) or a dedicated line that conforms to a communication standard such as Ethernet (registered trademark). However, the network that can be used in this embodiment is not limited to this and may be configured as a wireless network. This wireless network includes wireless PANs (Personal Area Networks) such as Bluetooth (registered trademark), ZigBee (registered trademark), and UWB (Ultra Wide Band). It also includes wireless LANs (Local Area Networks) such as Wi-Fi (Wireless Fidelity) (registered trademark) and wireless MANs (Metropolitan Area Networks) such as WiMAX (registered trademark). It also includes wireless WANs (Wide Area Networks) such as 4G and 5G. The network may be any network that connects devices to each other so that they can communicate with each other, and the communication standard, scale, and configuration are not limited to those described above.

[0061] At least some of the functions of each element of the information processing device 10 shown in Fig. 4 can be realized by the CPU 501 executing a program. However, at least some of the functions of each element of the information processing device 10 shown in Fig. 4 may be configured to operate as dedicated hardware. In this case, the dedicated hardware operates under the control of the CPU 501.

[0062] [Processing flow in information processing systems] 6 shows a flowchart of the processing executed in the information processing system according to this embodiment. It is assumed that the processing of S61, S62, S66, and S67 is executed by the e-commerce server 11, and the processing of S63 to S65 is executed by the information processing device 10. However, if the information processing device 10 and the e-commerce server 11 are configured as a single device, the processing may be executed by that device.

[0063] In S61, the user attribute acquisition unit 301 of the e-commerce server 11 acquires one or more user attributes of multiple users at any timing, and the conversion information acquisition unit 302 acquires conversion information of the multiple users at any timing. In S62, the user information provision unit 303 of the e-commerce server 11 provides the user attributes and conversion information as user information to the information processing device 10. The user information is acquired by the user information acquisition unit 401 of the information processing device 10 and stored in the user information storage unit 410.

[0064] In S63, the user group generation unit 402 of the information processing device 10 generates one or more user groups by grouping one or more users having one or more common (identical) user attributes based on the user attributes stored in the user information storage unit 410. For the sake of explanation, it is assumed that the user group generation unit 402 generates a user group with user characteristic = X, and for the sake of explanation of FIG. 6, this user group is also referred to as user group X. The user group generation unit 402 groups the user group X into a first user group t (i.e., an intervention user group) that has performed an intervention to stimulate a predetermined conversion, and a second user group c (i.e., a control user group) that has not performed the intervention. In this embodiment, the user groups are grouped so that the first user group t and the second user group c account for half (or approximately half). As a result, the probabilities p(t|X) and p(c|X) are each 0.5.

[0065] In S64, the uplift score calculation unit 403 calculates an uplift score for the user group X generated by the user group generation unit 402. In this embodiment, the uplift score calculation unit 403 uses a means such as a machine learning model to estimate the probability p(t|y=1,X) that a user who has achieved conversion is included in the first user group t and the probability p(c|y=1,X) that a user who has achieved conversion is included in the second user group c, as described above. The uplift score calculation unit 403 also uses a means such as a machine learning model to estimate the probability p(y=1|X) that the entire user group X has achieved conversion. For convenience of explanation, the probabilities p(t|y=1,X), p(c|y=1,X), and p(y=1|X) are referred to as the first probability, the second probability, and the third probability, respectively. The uplift score calculation unit 403 then calculates an uplift score based on the first probability, the second probability, and the third probability. Specifically, the uplift score calculation unit 403 calculates an uplift score for the user group X according to equation (10) using the first probability, the second probability, the third probability, the probabilities p(t|X) and p(c|X). The information processing device 10 calculates uplift scores for one or more user groups through the processes of S63 to S65, and provides the calculated uplift scores for each user group to the e-commerce server 11. The provided uplift scores for the one or more user groups are acquired by the intervening user determination unit 304 of the e-commerce server 11.

[0066] In S66, the intervention user determination unit 304 of the e-commerce server 11 determines an intervention user group based on the uplift scores for one or more user groups. For example, the intervention user determination unit determines a user group whose uplift score is higher than a predetermined value as the intervention user group. The predetermined value is, for example, 0.5. Subsequently, in S67, the advertisement distribution unit 305 of the e-commerce server 11 performs an intervention on the intervention user group. For example, the advertisement distribution unit 305 distributes an advertisement that encourages conversion, such as a purchase, to the intervention user group. Furthermore, the advertisement distribution unit 305 may perform an intervention using user features of the intervention user group. For example, the advertisement distribution unit 305 may use the user features of the intervention user group to distribute an advertisement that encourages conversion for a product related to the user features. Furthermore, the advertisement distribution unit 305 may distribute an advertisement that encourages conversion for a product related to similar user features similar to the user features of the intervention user group. The similar user features may be acquired by machine learning.

[0067] In this way, the uplift score is calculated by taking into account the probability that all users have made a purchase, so the calculated uplift score can be an indicator of purchasing trends. Furthermore, because the calculated uplift score represents purchasing trends, intervention using the uplift score can be expected to improve ROI (Return on Investment).

[0068] Although specific embodiments have been described above, these embodiments are merely examples and are not intended to limit the scope of the present invention. The devices and methods described herein may be embodied in forms other than those described above. Furthermore, appropriate omissions, substitutions, and modifications may be made to the above-described embodiments without departing from the scope of the present invention. Such omissions, substitutions, and modifications are included within the scope of the claims and their equivalents, and belong to the technical scope of the present invention.

[0069] The disclosure of this embodiment includes the following configuration. [1] The system includes: a grouping unit that groups a user group consisting of multiple users into a first user group that has received an intervention to stimulate a predetermined conversion and a second user group that has not received the intervention; a first estimation unit that estimates a first probability that the first user group will achieve the conversion and a second probability that the second user group will achieve the conversion; a second estimation unit that estimates a third probability that the entire user group will achieve the conversion; and a calculation unit that calculates an uplift score that represents the effect of the intervention on the user group using the first probability, the second probability, and the third probability.

[0070] [2] The information processing device according to [1], wherein the user group has one or more common user attributes.

[0071] [3] The information processing device described in [1] or [2], further comprising an intervention unit that performs the intervention on the user group when the uplift score is higher than a predetermined value.

[0072] [4] The information processing device described in [3], wherein the user group has one or more common user attributes, and the intervention unit performs the intervention by utilizing the one or more common user attributes of the user group.

[0073] [5] An information processing device according to any one of [1] to [4], wherein the conversion is the purchase of an item offered in an e-commerce service.

[0074] [6] The information processing device according to any one of [1] to [5], wherein the intervention is to deliver an advertisement related to the service.

[0075] [7] An information processing device described in any of [1] to [6], wherein the first estimation unit estimates the first probability and the second probability using a first machine learning model, and the second estimation unit estimates the third probability using a second machine learning model.

[0076] [8] An information processing device described in any of [1] to [6], wherein the first estimation unit estimates the first probability and the second probability using a first machine learning model, and the second estimation unit estimates the third probability using the first machine learning model. [Explanation of symbols]

[0077] 10: Information processing device, 11: E-commerce server, 12: User device, 13: Network, 14: User, 301: User attribute acquisition unit, 302: Conversion information acquisition unit, 303: User information provision unit, 304: Intervening user determination unit, 305: Advertisement distribution unit, 401: User information acquisition unit, 402: User group generation unit, 403: Uplift score calculation unit, 404: Uplift score provision unit, 410: User information storage unit, 420: Learning model storage unit

Claims

1. An acquisition unit that acquires, from each of a plurality of user devices operated by a plurality of users, one or more user attributes of the plurality of users and conversion information including whether or not the plurality of users have performed a predetermined conversion within a certain period of time; a grouping unit that groups a user group consisting of a plurality of users having one or more common user attributes (X) into a first user group (t) that has performed an intervention to induce the conversion (y) and a second user group (c) that has not performed the intervention; a first estimation unit that estimates a first probability (p(t|y=1,X)) that is the probability that the user who has achieved the conversion (y=1) is included in the first user group, and a second probability (p(c|y=1,X)) that is the probability that the user who has achieved the conversion is included in the second user group; a second estimation unit that estimates a third probability (p(y=1|X)) that is the probability that the entire user group achieves the conversion; a calculation unit that calculates an uplift score representing the effect of the intervention on the user group using the first probability (p(t|y=1,X)), the second probability (p(c|y=1,X)), and the third probability (p(y=1|X)) using the following formula: and In the above formula, p(t|X) is the probability of the first user group relative to the user group, and p(c|X) is the probability of the second user group relative to the user group.

2. An intervention unit that performs the intervention on the user group when the uplift score is higher than a predetermined value. The information processing device according to claim 1 .

3. The intervention unit performs the intervention by utilizing the one or more common user attributes (X) of the user group. The information processing device according to claim 2 .

4. the conversion is a purchase of an item offered by the e-commerce service; The information processing device according to claim 1 .

5. The intervention is to deliver an advertisement related to the service. The information processing device according to claim 1 .

6. the first estimation unit estimates the first probability (p(t|y=1,X)) and the second probability (p(c|y=1,X)) using a first machine learning model; The second estimation unit estimates the third probability (p(y=1|X)) using a second machine learning model. The information processing device according to claim 1 .

7. the first estimation unit estimates the first probability (p(t|y=1,X)) and the second probability (p(c|y=1,X)) using a first machine learning model; the second estimation unit estimates the third probability (p(y=1|X)) using the first machine learning model; The information processing device according to claim 1 .

8. An information processing method executed by an information processing device, Acquiring, from each of a plurality of user devices operated by a plurality of users, conversion information including one or more user attributes of the plurality of users and whether or not a predetermined conversion has been performed by the plurality of users within a certain period of time; Grouping a user group consisting of a plurality of users having one or more common user attributes (X) into a first user group (t) that has performed an intervention to induce the conversion (y) and a second user group (c) that has not performed the intervention; Estimating a first probability (p(t|y=1,X)) that is the probability that the user who has achieved the conversion (y=1) is included in the first user group, and a second probability (p(c|y=1,X)) that is the probability that the user who has achieved the conversion is included in the second user group; Estimating a third probability (p(y=1|X)), which is the probability that the entire user group achieves the conversion; calculating an uplift score representing the effect of the intervention on the user group using the first probability (p(t|y=1,X)), the second probability (p(c|y=1,X)), and the third probability (p(y=1|X)) using the following formula: Including, In the above formula, p(t|X) is the probability of the first user group relative to the user group, and p(c|X) is the probability of the second user group relative to the user group.

9. An information processing program for causing a computer to execute information processing, the program including: an acquisition process for acquiring, from each of a plurality of user devices operated by a plurality of users, one or more user attributes of the plurality of users and conversion information including whether or not the plurality of users have performed a predetermined conversion within a certain period of time; a grouping process for grouping a user group consisting of a plurality of users having one or more common user attributes (X) into a first user group (t) that has intervened to induce the conversion (y) and a second user group (c) that has not intervened; a first estimation process for estimating a first probability (p(t|y=1,X)) that is the probability that the user who has achieved the conversion (y=1) is included in the first user group, and a second probability (p(c|y=1,X)) that is the probability that the user who has achieved the conversion is included in the second user group; a second estimation process for estimating a third probability (p(y=1|X)), which is the probability that the entire user group achieves the conversion; a calculation process of calculating an uplift score representing the effect of the intervention on the user group using the first probability (p(t|y=1,X)), the second probability (p(c|y=1,X)), and the third probability (p(y=1|X)) using the following formula: and In the above formula, p(t|X) is the probability of the first user group relative to the user group, and p(c|X) is the probability of the second user group relative to the user group.

Images