Model generation method and model generation apparatus

Abstract

We provide technology for providing effective incentives. [Solution] An information processing device relating to one aspect of the present disclosure acquires target samples of behavioral history related to the purchasing behavior of a target user, provides the acquired target samples to an immediate prediction model to predict the first degree to which purchasing behavior will be triggered in an immediate period, provides the acquired target samples to a long-term prediction model to predict the second degree to which purchasing behavior will be triggered in a long period, determines whether or not to provide an incentive to the target user based on the predicted first and second degrees, and if it is decided to provide an incentive to the target user, outputs incentive information relating to the provision of an incentive to the target user.

Description

Technical Field

[0001] The present disclosure relates to a model generation method and a model generation apparatus.

Background Art

[0002] In Patent Document 1, an information processing apparatus for promoting purchases by distributing coupons (benefits) is proposed. Specifically, the proposed information processing apparatus uses a machine learning model to estimate the probability of occurrence of a user's usage pattern of a predetermined service when a benefit is provided for the predetermined service, and the probability of occurrence of the user's usage pattern of the predetermined service when no benefit is provided. The information processing apparatus determines whether to provide a benefit related to the predetermined service to the user based on the difference in the estimated probability of occurrence of each usage pattern. When it is determined that a benefit related to the predetermined service is to be provided, the information processing apparatus provides a benefit related to the predetermined service to the user.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Conventional methods allow for efficient sales promotion by providing incentives such as coupons to users who are highly likely to make a purchase. However, the inventors of this case have found the following problems with conventional methods. Specifically, conventional methods evaluate whether a user makes a purchase on the day the incentive is given or within the immediate period of its validity. Based on this evaluation, it is possible to expect efficient incentive provision to users who make immediate purchases. However, a user's purchase is not always triggered immediately. There may be users who, while having a relatively low degree of immediate purchase triggered by the incentive, have a high degree of purchase triggered in the long term (i.e., whose purchase behavior is likely to be sustained). With conventional methods, because the degree of immediate purchase triggered is relatively low, it becomes difficult to provide incentives to such users, and as a result, the long-term sales promotion effect of providing incentives may not be expected. In other words, it may be difficult to provide effective incentives.

[0005] In one respect, this disclosure is made in light of these circumstances. One of the purposes of this disclosure is to provide technology for providing effective incentives. [Means for solving the problem]

[0006] This disclosure adopts the following configuration to solve the aforementioned problems. Note that the following configurations can be combined as appropriate.

[0007] An information processing device relating to one aspect of this disclosure includes a control unit. The control unit acquires target samples of behavioral history related to the purchasing behavior of a target user, provides the acquired target samples to an immediate prediction model to predict the first degree to which purchasing behavior will be induced in an immediate period by providing incentives, provides the acquired target samples to a long-term prediction model to predict the second degree to which purchasing behavior will be induced in a period longer than the immediate period by providing incentives, and decides whether or not to provide incentives to the target user based on the predicted first and second degrees. If it is decided to grant an incentive, the system is configured to output incentive information regarding the granting of incentives to the target users.

[0008] This configuration predicts not only the degree to which immediate purchasing behavior is triggered (degree 1), but also the degree to which long-term purchasing behavior is triggered (degree 2). Because the predicted degree 2 is included in the indicators used to decide whether or not to provide an incentive, it becomes easier to provide incentives to users who have a relatively low degree of immediate purchasing behavior but a high degree of long-term purchasing behavior. Therefore, this configuration can be expected to provide effective incentives.

[0009] In the information processing device relating to the above aspect, the control unit may be configured to further determine the amount of incentive such that the amount of incentive increases as the degree of second-degree purchase behavior increases. The incentive information may be configured to relate to the granting of the determined amount of incentive. With this configuration, by prioritizing the granting of incentives to users with a high degree of long-term purchasing behavior (degree second-degree purchase behavior), it can be expected that the sales promotion effect of granting incentives will be obtained in the long term.

[0010] In the information processing device relating to the above aspect, determining the amount of incentive may include determining the amount of incentive such that the lower the second degree and the higher the first degree, the smaller the amount of incentive. According to this configuration, the amount of incentive can be reduced for users who have a high degree of immediate purchasing behavior (first degree) but a low degree of long-term purchasing behavior (second degree). This makes it possible to prioritize the provision of incentives to users who have a high degree of long-term purchasing behavior (second degree). As a result, it can be expected that the sales promotion effect from the provision of incentives will be obtained over the long term.

[0011] The form of the information processing device described herein is not limited to the above-described information processing device. As another form of the information processing device relating to each of the above aspects, one aspect of the disclosure may be an information processing method that implements all or part of the above-described configurations, a program, or a machine-readable storage medium that stores such a program. Here, a machine-readable storage medium may be a non-temporary medium that stores information such as programs by electrical, magnetic, optical, mechanical, or chemical action. A non-temporary storage medium may include storage media (CDs, DVDs, semiconductor memory, etc.), auxiliary storage devices of a computer, external storage devices connected to a computer, etc.

[0012] For example, an information processing method relating to one aspect of this disclosure may be performed by a computer. The information processing method may include: obtaining a target sample of behavioral history related to the purchasing behavior of a target user; feeding the obtained target sample to an immediate prediction model to predict the first degree to which purchasing behavior will be induced in an immediate period by providing an incentive; feeding the obtained target sample to a long-term prediction model to predict the second degree to which purchasing behavior will be induced in a period longer than the immediate period by providing an incentive; deciding whether or not to provide an incentive to the target user based on the predicted first and second degrees; and, if it is decided to provide an incentive to the target user, outputting incentive information relating to the provision of an incentive to the target user.

[0013] Furthermore, for example, a program relating to one aspect of this disclosure may be a program that causes a computer to execute an information processing method. The information processing method acquires target samples of behavioral history related to the purchasing behavior of target users, provides the acquired target samples to an immediate prediction model to predict the degree to which purchasing behavior will be triggered in an immediate period by providing incentives, and provides the acquired target samples to a long-term prediction model to predict the degree to which purchasing behavior will be triggered in an immediate period. By assigning incentives, the system may predict a second degree of likelihood of triggering purchasing behavior over a longer period than the immediate period, determine whether or not to provide incentives to target users based on the predicted first and second degrees, and, if it is decided to provide incentives to target users, output incentive information regarding the provision of incentives to target users.

[0014] Furthermore, the embodiments of the present invention are not limited to the inference (prediction) stage described above. Embodiments of the present invention may also focus on the model generation stage used for the above prediction. The model generation stage may be, for example, a machine learning stage. For example, one aspect of the present invention may be a model generation method executed by a computer, which is a model generation method for generating a trained machine learning model (immediate prediction model, long-term prediction model). For example, one aspect of the present invention may be a model generation method executed by a computer, which is a model generation method for generating a trained machine learning model (transformation model) for generating true values ​​used in machine learning of the above-mentioned long-term prediction model. Each model generation method is an example of a method for producing a trained machine learning model.

[0015] For example, a model generation method relating to one aspect of this disclosure may be performed by a computer. The model generation method may include controlling the machine learning of a long-term prediction model and outputting the results of the machine learning. The long-term prediction model may be configured to predict the degree to which incentives will induce purchasing behavior over a longer period than the immediate period, based on the user's behavioral history related to purchasing behavior. For a training sample of a sample user's behavioral history, the true value of the degree to which purchasing behavior will be induced over a long period may be derived using a transformation model from an actual value of the degree to which purchasing behavior was induced by incentives, which was measured over a shorter period than the long period. The machine learning may include training the long-term prediction model by providing training samples to the long-term prediction model so that the predicted value of the degree derived by the long-term prediction model approaches the true value corresponding to the training sample.

[0016] This configuration allows for the generation of a long-term prediction model (trained machine learning model) that acquires the ability to predict the likelihood of long-term purchasing behavior from a user's behavioral history related to their purchasing actions. By using this long-term prediction model to identify users with a high likelihood of long-term purchasing behavior, it is possible to expect the provision of effective incentives, as described above. Furthermore, in order to obtain the true values ​​(labels, training signals) used in the machine learning of the long-term prediction model from actual measured values, it is necessary to measure the likelihood of users' purchasing behavior over a long period. This measurement is time-consuming and costly. In contrast, with this configuration, by using a transformation model, predicted values ​​are derived from actual measured values ​​taken over a shorter period than the long period, and these derived predicted values ​​are used as substitutes for the true values. At least some of the true values ​​may be derived substitutes using this transformation model. This makes it possible to shorten the period for measuring the likelihood of purchasing behavior, thereby reducing the effort required to collect the true values ​​used in the machine learning of the long-term prediction model.

[0017] Furthermore, for example, a model generation method relating to one aspect of this disclosure may be performed by a computer. The model generation method may include controlling the machine learning of the transformation model and outputting the results of the machine learning. The transformation model may be configured to predict a degree from a third degree, which induces purchasing behavior in a short period, to a second degree, which induces purchasing behavior in a long period, by providing incentives. The machine learning may include training the transformation model by providing training values ​​of the third degree to the transformation model so that the predicted value of the second degree derived by the transformation model approaches the true value of the second degree corresponding to the training values. According to this configuration, it is possible to generate a transformation model (trained machine learning model) that has acquired the ability to predict the degree of inducing purchasing behavior in a long period from the degree of inducing purchasing behavior in a short period. By obtaining at least a portion of the true values ​​used in machine learning for long-term prediction models through a transformation model, the effort required to collect those true values ​​can be reduced, as described above.

[0018] Note that the embodiments of the present disclosure are not necessarily limited to the above model generation method (information processing method). As another aspect of the information processing method according to each of the above aspects, one aspect of the present disclosure may be an information processing apparatus (model generation apparatus) that realizes all or a part of each of the above configurations, or may be a program (model generation program), or may be a machine-readable storage medium such as a computer that stores such a program. A machine-readable storage medium may be a non-temporary medium that accumulates information such as a program by an electrical, magnetic, optical, mechanical, or chemical action.

Advantages of the Invention

[0019] According to one aspect of the present disclosure, it is possible to provide a technology for attempting to effectively provide incentives.

Brief Description of the Drawings

[0020] [Figure 1] FIG. 1 schematically shows an example of a scene to which the present disclosure is applied. [Figure 2] FIG. 2 schematically shows the degree of induction of purchase behavior for each type in an immediate period and a long-term period. <° [Figure 3] FIG. 3 schematically shows an example of a scene for determining the amount of incentives. [Figure 4] FIG. 4 schematically shows an example of a scene for generating a long-term prediction model. [Figure 5] FIG. 5 schematically shows an example of a scene for generating a conversion model. [Figure 6] FIG. 6 schematically shows an example of a scene for generating an immediate prediction model. [Figure 7] FIG. 7 schematically shows an example of the hardware configuration of an information processing apparatus. [Figure 8] FIG. 8 schematically shows an example of the hardware configuration of a model generation apparatus. [Figure 9] FIG. 9 schematically shows an example of the software configuration of an information processing apparatus. [Figure 10] FIG. 10 schematically shows an example of the software configuration of a model generation apparatus. [Figure 11] Figure 11 is a flowchart showing an example of the processing steps for generating a transformation model. [Figure 12] Figure 12 is a flowchart showing an example of the processing procedure for generating a long-term prediction model. [Figure 13] Figure 13 is a flowchart showing an example of the processing procedure for generating a rapid-response prediction model. [Figure 14] Figure 14 is a flowchart showing an example of the processing procedure for granting incentives. [Modes for carrying out the invention]

[0021] Hereinafter, embodiments relating to one aspect of this disclosure will be described with reference to the drawings. However, the embodiments described below are merely illustrative in all respects of this disclosure. Various improvements or modifications may be made without departing from the scope of this disclosure. In implementing this disclosure, specific configurations may be adopted as appropriate depending on the embodiment. In this embodiment, the data appearing is described in natural language, but more specifically, it is specified in pseudo-language, commands, parameters, machine code, electrical signals, etc., that can be recognized by machines such as computers.

[0022] §1 Examples of Application Figure 1 schematically shows an example of a scenario to which this disclosure applies. The information processing device 1 according to this embodiment is one or more computers configured to predict the degree of effect of providing an incentive (degree to which purchasing behavior is induced) and to decide whether or not to provide an incentive based on the prediction result.

[0023] The information processing device 1 according to this embodiment acquires target samples 30 of the behavioral history related to the purchasing behavior of target user T1. The information processing device 1 provides the acquired target samples 30 to a rapid prediction model 5 to predict a first degree 31 that will induce purchasing behavior in a rapid period by providing incentives. That is, the information processing device 1 uses the rapid prediction model 5 to predict the first degree 31 from the target samples 30. The predicted first degree 31 may also be referred to as the predicted value of the first degree 31. Furthermore, the information processing device 1 provides the acquired target samples 30 to a long-term prediction model 6 to predict a second degree 32 that will induce purchasing behavior in a period longer than the rapid period by providing incentives. That is, the information processing device 1 uses the long-term prediction model 6 to predict the second degree 32 from the target samples 30. The predicted second degree 32 may also be referred to as the predicted value of the second degree 32. The information processing device 1 decides whether or not to grant an incentive to the target user T1 based on the predicted first degree 31 and second degree 32. If the information processing device 1 decides to grant an incentive to the target user T1, it outputs incentive information 35 regarding the granting of an incentive to the target user T1. On the other hand, if it decides not to grant an incentive to the target user T1, the information processing device 1 omits outputting the incentive information 35.

[0024] Figure 2 schematically illustrates the degree to which incentives induce purchasing behavior for each type of incentive, both in the immediate and long term. Incentives may include, for example, coupons. Purchasing behavior may include, for example, the purchase of goods (products, services, etc.).

[0025] Type A users are an example of users who are highly motivated to purchase in the immediate period in response to incentives, but less so in the long term. Examples of Type A users include CherryPicka and point-collecting users. For example, when a Type A user receives a coupon, they will use it to purchase a large quantity of goods, but will hardly purchase any goods at other times. As a result, their motivation for purchasing is high in the immediate period, but low in the long term. For Type A users, the presence or absence of an incentive (such as a lower price) may carry more weight in determining whether or not to purchase a product than in the quality (such as the store or the goods). In other words, Type A users are more likely to decide whether or not to purchase a product based on the presence or absence of an incentive than on the quality of the store or the goods. For example, if detergent products AA and BB are sold at stores C and D, and store C is offering a coupon for product AA, making product AA the cheapest at store C, then a Type A user is more likely to buy product AA at store C. On the other hand, if store D is offering a coupon for product BB, making product BB the cheapest at store D, then a Type A user is more likely to buy product BB at store D. In other words, Type A users are more likely to switch stores and products in response to incentives, and are less likely to stay with a particular store or product. Therefore, when incentives are offered to Type A users, the immediate effect on stimulating purchasing behavior may be high, but the long-term effect tends to be low.

[0026] On the other hand, Type B users are an example of users who, while relatively less likely to be motivated to purchase in the immediate period by incentives, are more likely to be motivated to purchase in the long term. For example, Type B users may perceive a store as a good store that regularly provides incentives if the incentives are consistently offered, and may continue purchasing from that store even during periods without incentives. This increases the likelihood of long-term purchase activity. For Type B users, the quality of the store and merchandise may carry more weight than the presence or absence of incentives in influencing their purchasing behavior. In other words, Type B users are more likely to decide whether or not to purchase based on the quality of the store and merchandise, not just the presence or absence of incentives. Type B users are more likely to continue purchasing from stores and merchandise they like (become fans, etc.). As a specific example, If a user of type B likes product AA from the detergent products mentioned above, and also likes store C as the store that sells product AA, then type B users are likely to continue purchasing product AA at store C. In other words, while type B users may switch stores and products in response to incentives, once they settle into a particular store and product, they are likely to continue their purchasing behavior with that store and product. Therefore, while providing incentives to type B users may have a low immediate effect on stimulating purchasing behavior, it tends to have a high long-term effect. However, if the period without incentives is long, type B users may drift away from the target store (i.e., stop making purchases at the target store).

[0027] In the immediate future period following the granting of incentives, Type A users will purchase more goods (total purchase amount, etc.) than Type B users. On the other hand, in the long term, Type B users may purchase more goods than Type A users. If a store that implements incentives appeals only to Type A users, profits in the immediate period may improve. However, Type B users may leave the store if they do not receive incentives. As a result, improvements in long term profits may not be expected. Therefore, in order to improve overall profits at the store, it is desirable to appeal not only to Type A users but also to Type B users. In the specific example above, store C can expect to improve overall profits by providing incentives to some extent to Type B users as well as Type A users. In other words, it can expect to provide effective incentives.

[0028] Conventional methods evaluate whether or not a user's purchasing behavior is triggered in an immediate period. Therefore, conventional methods make it difficult to provide incentives to Type B users, and as a result, the long-term sales promotion effect of providing incentives may not be expected. In contrast, this embodiment predicts not only the degree to which purchasing behavior is triggered in an immediate period (first degree 31), but also the degree to which purchasing behavior is triggered in the long term (second degree 32). Not only the predicted first degree 31, but also the predicted second degree 32 is included as an indicator for deciding whether or not to provide an incentive. This makes it easier to provide incentives to users who have a relatively low degree to which immediate purchasing behavior is triggered (first degree 31), but a high degree to which long-term purchasing behavior is triggered (second degree 32). In other words, it becomes easier to appeal not only to Type A users, but also to Type B users. Therefore, according to this embodiment, effective incentive provision can be expected.

[0029] [Purchasing behavior] Purchasing behavior may include the purchase of goods. Goods may include any object that can be the subject of a contract (and for which consideration may be incurred), such as products and services. Goods may be physical items that are the subject of a sale. Goods may include not only tangible objects of all kinds but also intangible objects such as applications (software). Services may be the provision of services. Services may include contracts such as rentals, leases, and subscriptions. Services may include, for example, tickets to entertainment facilities, e-book subscriptions, regular use of applications, in-app items, and cloud service usage contracts. Entertainment facilities may include, for example, movie theaters, concert halls, stadiums, amusement parks, aquariums, art museums, museums, zoos, etc. E-books may include, for example, magazines and newspapers.

[0030] The types of behaviors related to purchasing behavior are not particularly limited and may be appropriately selected depending on the embodiment, as long as they are related to the degree to which purchasing behavior is induced by the provision of incentives. For example, behaviors related to purchasing behavior may include at least one of the purchasing behavior itself and behaviors that lead to purchasing. Behaviors that lead to purchasing may include, for example, accessing a website related to the product. This may include preliminary actions taken before making a purchase, such as browsing information about the product. Information about the product may include, for example, product information, advertisements, etc. If information about the product is provided as web information, browsing information about the product may include accessing the website. Accessing the website may include, for example, accessing the product on an e-commerce (electronic commerce) site. This may include accessing product pages, accessing the websites of companies that provide the products, and searching for products on search engines.

[0031] The history of purchasing behavior may include, at least partially, a history of when incentives were previously provided. This allows the history of purchasing behavior to show trends in the degree to which purchasing behavior is triggered by the provision of incentives. On the other hand, even if the history of when incentives were provided is not included, the history of purchasing behavior may show trends in purchasing behavior during periods without incentives, such as trends in purchasing behavior during periods other than the immediate response period in Figure 2. Trends in purchasing behavior during periods without incentives may be related to the degree to which purchasing behavior is triggered by the provision of incentives, as shown in Type A, Type B, etc. in Figure 2. Therefore, regardless of whether or not the history of when incentives were previously provided is included, purchasing behavior may be related to the degree to which purchasing behavior is triggered by the provision of incentives. Furthermore, providing incentives can be an act that stimulates purchasing behavior. By providing incentives related to products that the customer has a high level of interest in (willingness to buy, etc.), the likelihood of purchasing those products may increase. Actions that lead to purchases may indicate the level of interest in purchases. Therefore, actions that lead to purchases may also be related to the degree to which purchasing behavior is triggered by the provision of incentives.

[0032] Behavioral history related to purchasing behavior may be structured to show the actual results of behavior related to purchasing behavior. Behavioral results (behavioral history) may be measured for each of one or more items. The measurement items are not particularly limited and may be appropriately selected depending on the embodiment, as long as the degree of behavior related to purchasing behavior can be evaluated. Measurement items may include, for example, the number of times a product is purchased, whether or not the product is a repeat purchase, the purchase quantity, the purchase amount, the number of times the website related to the product is accessed, the number of times information about the product is viewed, etc. The number of times a product is purchased may include whether or not the product was purchased. Whether or not it is a repeat purchase may include whether or not it is a regular purchase. The number of accesses may include whether or not the website was accessed. The number of views may include whether or not the website was viewed. Each item of the behavioral history may be measured by any method. Each item of the behavioral history may be measured by known methods such as the use of purchase data or tracking of web browser browsing history. Purchase data may be, for example, POS (Point of Sales) data.

[0033] In one example, behavioral history may be measured for each individual store. An individual store may be defined as, for example, one store, a market, etc. A market may consist of multiple stores. A market may include, for example, a supermarket, a department store, a shopping mall, an e-commerce site, etc. In another example, behavioral history may be measured independently of individual stores. Each item of behavioral history may be measured sequentially, or it may be measured aggregated by one or more statistics within a predetermined range. The statistics may be arbitrarily selected from, for example, the mean, variance, standard deviation, median, maximum value, minimum value, Nth percentile value, sum, cumulative value, etc. The predetermined range for measuring the statistics may be appropriately defined by a period, individual stores, etc. Each item of behavioral history may be measured as a continuous value, or as a discrete value (level, etc.).

[0034] The target sample 30 may consist of behavioral history measured within an arbitrary period prior to the target time. The target time may be, for example, the time at which the degree of purchase behavior is predicted (e.g., the present time). The arbitrary period may be the entire period during which the behavioral history was measured, or it may be only a portion of that period. For example, the target sample 30 may consist of behavioral history measured within a certain period starting from a time predetermined time before the target time and ending at the target time.

[0035] At least a portion of the behavioral history (each item) may be measured by the information processing device 1. The data may be measured by a computer other than the information processing device 1. If at least a portion of the behavioral history is measured by another computer, the information processing device 1 may appropriately acquire at least a portion of the target sample 30, which is composed of the behavioral history measured by the other computer, via a network, storage medium, etc.

[0036] The stores that collect the target samples 30 and the stores that decide whether or not to provide incentives using the collected target samples 30 may be the same, or may be at least partially different in some cases. For example, target samples 30 collected at a target store may be used to decide whether or not to provide incentives at that store. That is, when deciding whether or not to provide incentives to a target user T1 at a target store, the information processing device 1 may acquire target samples 30 measured at the target store. For example, if there are no problems in terms of data utilization, target samples 30 collected at one or more other stores besides the target store may be used to decide whether or not to provide incentives at the target store. Also, for example, target samples 30 collected in a market that includes the target store and one or more other stores may be used to decide whether or not to provide incentives at the target store belonging to the market. For example, target samples 30 collected in a market may be used to decide whether or not to provide incentives to the company operating the market.

[0037] Furthermore, the behavioral history data elements provided to the rapid prediction model 5 and the long-term prediction model 6 as the target sample 30 may be exactly the same or at least partially different. Providing the target sample 30 to the rapid prediction model 5 and the long-term prediction model 6 may include at least partially different data elements (e.g., items, periods, etc.) of the target sample 30 provided between the rapid prediction model 5 and the long-term prediction model 6, as long as the data belonging to the target sample 30 is provided. Providing the target sample 30 to the rapid prediction model 5 may consist of providing at least a portion of the target sample 30 to the rapid prediction model 5. Similarly, providing the target sample 30 to the long-term prediction model 6 may consist of providing at least a portion of the target sample 30 to the long-term prediction model 6.

[0038] [period] The immediate response period may be the period during which purchasing behavior is triggered by the provided incentive. For example, the immediate response period may be defined as the period from the time the incentive is granted until the time when the first predetermined time has elapsed in the future (the first predetermined period). Specifically, the immediate response period may be defined as the period from the day the incentive is granted (day 1) to within about one month (for example, day 1, 2-3 days, one week, one month, etc.). The immediate response period may also be defined according to the expiration date of the incentive being granted. The expiration date of the incentive may be, for example, the expiration date of a coupon, the expiration date of points, the expiration date of a prize, etc. The immediate response period may also be called the first period.

[0039] Furthermore, the long term may be defined to include a period that extends beyond the immediate period during which purchasing behavior occurs due to the provided incentive, and includes a period during which purchasing behavior would occur even without an incentive. For example, the long term may be defined as the period from the time the incentive is provided to the time when a second predetermined period has elapsed in the future (the second predetermined period). The second predetermined period may be defined to be longer than the first predetermined period. Specifically, the long term may be defined as a period of six months or more (e.g., six months, one year, two years, etc.). The long term may also be defined to evaluate the user's customer lifetime value. The period in between may also be called the second period.

[0040] [Degree of inducing purchasing behavior] The degree to which purchasing behavior is stimulated (first degree 31, second degree 32) may be appropriately defined to indicate the extent to which the goods are purchased over the target period (immediate period, long-term period). The degree to which a behavior is triggered may be evaluated (measured) using items such as the number of purchases, the quantity purchased, and the purchase amount. Similar to behavioral history, in one example, the degree to which a purchasing behavior is triggered may be evaluated for each individual store. In another example, the degree to which a purchasing behavior is triggered may be evaluated independently of individual stores. The degree to which a purchasing behavior is triggered may be evaluated sequentially, or it may be evaluated aggregated using one or more statistics within a predetermined range. Each item of the degree to which a purchasing behavior is triggered may be evaluated as a continuous value, or as a discrete value (level, etc.). The first degree 31 may be called the degree of immediate effect. The second degree 32 may be called the degree of long-term effect.

[0041] [Incentives] Incentives may consist of optional promotions offered with the aim of increasing purchasing intent and promoting purchasing behavior. Promotions may include, for example, coupons, discounts, points, prizes, and invitations to events. Prizes may be any merchandise offered free of charge. Coupons may be sales promotion tools that provide benefits to users (customers) when purchasing merchandise. Benefits may include, for example, discounts, prizes, and points. Incentives may have usage conditions attached. Usage conditions may include, for example, an expiration date, eligible merchandise, and a minimum purchase amount. In order to evaluate the effect of providing incentives in an immediate period (degree of inducing purchasing behavior), incentives may be structured so that usage history can be tracked.

[0042] (Grant criteria) Whether or not to grant an incentive may be determined as appropriate using the predicted first degree 31 and second degree 32. The criteria for granting an incentive may be defined as appropriate depending on the embodiment so that incentives are more likely to be granted to users with high first degree 31 and second degree 32. In a typical example, the criteria for granting an incentive may include meeting a threshold criterion. That is, the information processing device 1 may compare each predicted degree (31, 32) with a threshold and decide whether or not to grant an incentive according to the result of the comparison.

[0043] For example, each degree (31, 32) may be defined such that a higher value for each degree (31, 32) indicates a higher likelihood of purchasing the product. Accordingly, the information processing device 1 may decide to provide an incentive if either the predicted first degree 31 or second degree 32 exceeds a threshold. The information processing device 1 may decide not to provide an incentive if both the predicted first degree 31 and second degree 32 are below the threshold. Alternatively, the information processing device 1 may decide to provide an incentive if both the predicted first degree 31 and second degree 32 exceed a threshold. The information processing device 1 may decide not to provide an incentive if either the predicted first degree 31 or second degree 32 is below the threshold. The predicted degree (31, 32) being equal to the threshold may be treated as either each degree (31, 32) exceeding the threshold or each degree (31, 32) being below the threshold.

[0044] Thresholds may be given individually to each degree (31, 32), or they may be given collectively to the combined result of each degree (31, 32). For example, the information processing device 1 may compare the predicted first degree 31 (predicted value) with the first threshold, and the predicted second degree 32 (predicted value) with the second threshold. Alternatively, for example, the information processing device 1 may combine the predicted first degree 31 and second degree 32 and compare the combined result (combined value) with the threshold. Combining may involve calculating statistics such as the average or sum. Each threshold may be arbitrarily defined.

[0045] In one example, the predicted first degree 31 and second degree 32 may be treated equally. In another example, a hierarchy (weight, priority, etc.) may be assigned between the predicted first degree 31 and second degree 32. For example, as a first step, the information processing device 1 compares the predicted first degree 31 (predicted value) with a first threshold and provides an incentive according to the result of the comparison. The information processing device 1 may make a provisional decision on whether or not to provide an incentive. In the second step, for users who have been provisionally determined not to receive an incentive, the information processing device 1 may compare the predicted second degree 32 (predicted value) with the second threshold and, depending on the result of the comparison, may re-determine whether or not to provide an incentive. The information processing device 1 may decide to provide an incentive for users who have been determined to receive an incentive in either the first or second step, and decide not to provide an incentive for users who have not been determined to receive an incentive. Alternatively, for example, the information processing device 1 may weight the predicted first degree 31 and second degree 32, calculate statistics (weighted average, weighted sum, etc.) of the weighted first degree 31 and second degree 32, and compare the calculated statistics (integrated value) with the threshold. The information processing device 1 may decide whether or not to provide an incentive depending on the result of the comparison between the statistics (integrated value) and the threshold.

[0046] (Incentive information) The configuration of the incentive information 35 is not particularly limited and may be determined appropriately depending on the embodiment, as long as it is possible to provide an incentive to the user (target user T1). Furthermore, the form in which the incentive information 35 is output is not particularly limited and may be determined appropriately depending on the embodiment, as long as the user is ultimately able to receive an incentive.

[0047] For example, when the information processing device 1 issues an incentive, the incentive information 35 may be configured to indicate the incentive. In this case, the incentive information 35 may be, for example, coupon information, discount information, point information, prize information, invitation information, etc. The information processing device 1 may notify any destination (the terminal or account of target user T1) of the incentive information 35 directly or indirectly via another computer, etc. That is, outputting the incentive information 35 may include notifying the destination of the target user T1 of the incentive information 35.

[0048] In another example, if a computer other than the information processing device 1 issues an incentive, the incentive information 35 may include a command instructing the granting of the incentive (such as permission to issue an incentive or an issuance command). The information processing device 1 may directly or indirectly provide the incentive information 35 to the other computer, thereby causing the other computer to issue an incentive to the target user T1. In other words, outputting the incentive information 35 may include providing the incentive information 35 to the other computer. After receiving the incentive information 35, the other computer may, at any time, directly or indirectly notify the target user T1 of information indicating the incentive.

[0049] Furthermore, after deciding to grant an incentive, the information processing device 1 may output incentive information 35 to the target user T1 at any time. For example, the information processing device 1 may output incentive information 35 immediately in response to the decision to grant an incentive. Alternatively, for example, after deciding to grant an incentive, the information processing device 1 may output incentive information 35 in response to the fulfillment of predetermined conditions, such as receiving a trigger operation from an operator.

[0050] (Amount of incentive) The amount of incentive to be given to target user T1 may be determined as appropriate depending on the embodiment. The amount of incentive to be given may be a fixed value or a variable value. The amount of incentive may be given in advance. The amount of incentive may be determined manually by an operator or may be determined automatically by computer processing, at least in part. In one example, the amount of incentive to be given may be constant among the users to whom the incentive is given. For example, the amount of incentive may be derived by dividing the total amount of incentive by the number of users to whom the incentive is given. In another example, the amount of incentive to be given may be constant among at least a portion of the users to whom the incentive is given. They may be different.

[0051] Figure 3 schematically shows an example of a scenario in which the amount of incentive 33 according to this embodiment is determined. In one example, the information processing device 1 may further determine the amount of incentive 33 according to the predicted first degree 31 and second degree 32. The incentive information 35 may be configured to relate to the granting of the determined amount 33 of incentive.

[0052] The correspondence between the predicted degrees (31, 32) and the amount of incentive 33 is not particularly limited and may be defined as appropriate depending on the embodiment. For example, determining the amount of incentive 33 may include determining the amount of incentive 33 such that the higher the predicted second degree 32, the greater the amount of incentive 33. According to this example embodiment, by prioritizing the provision of incentives to users with a high degree of long-term purchasing behavior (second degree 32), it is possible to expect that the sales promotion effect of providing incentives will be obtained in the long term.

[0053] Furthermore, in one example, determining the amount of incentive 33 may include determining the amount of incentive 33 such that the lower the second degree 32 and the higher the first degree 31, the smaller the amount of incentive 33 becomes. According to this example, the amount of incentive 33 can be reduced for users who have a high degree of immediate purchasing behavior (first degree 31) but a low degree of long-term purchasing behavior (second degree 32). This makes it possible to prioritize the provision of incentives to users who have a high degree of long-term purchasing behavior (second degree 32). As a result, it can be expected that the sales promotion effect from the provision of incentives will be obtained over the long term.

[0054] The amount of incentive 33 may be defined as appropriate depending on the embodiment, such as the type of incentive. The amount of incentive 33 may be defined as, for example, a discount rate, a discount amount, a point amount, a prize value, a prize quantity, the rarity of the invitation event, or the value of the invitation event. In one example, determining the amount of incentive 33 may include setting the amount of incentive 33 to 0 (i.e., excluding from the target of the incentive). The amount of incentive 33 may be determined as a continuous value or as a discrete value. In another example, determining the amount of incentive 33 may consist of selecting an applicable incentive from among several candidates.

[0055] [Immediate-response prediction model / Long-term prediction model] The immediate prediction model 5 may be appropriately configured to predict the degree to which providing incentives will induce purchasing behavior in an immediate period (degree 1) based on the user's behavioral history related to purchasing behavior. The long-term prediction model 6 may be appropriately configured to predict the degree to which providing incentives will induce purchasing behavior in a period longer than the immediate period (degree 2) based on the user's behavioral history related to purchasing behavior.

[0056] Each model (5, 6) may consist of at least one of a rule-based model and a trained machine learning model. A rule-based model is configured to derive an inference result (a predicted value of the likelihood of triggering a purchase behavior) from a given input according to rules. The rules may be set as appropriate. A machine learning model is configured to have one or more computational parameters that can be adjusted by machine learning. One or more computational parameters are used in the calculation of the desired inference (prediction of the likelihood of triggering a purchase behavior). A machine learning model may consist of, for example, a neural network, a regression model, a decision tree model, a support vector machine, or other functional equations (computational models). Performing machine learning (i.e., training a machine learning model) may involve adjusting (optimizing) the values ​​of the computational parameters using training data. The machine learning method may be appropriately selected depending on the machine learning model adopted. This is acceptable. The machine learning method may include, for example, known optimization techniques such as backpropagation, regression analysis, and random forests. The trained machine learning model may include large-scale generative models such as large-scale language models.

[0057] In one example, at least one of the immediate prediction model 5 and the long-term prediction model 6 may include a neural network. The structure of the neural network is not particularly limited and may be determined as appropriate depending on the embodiment. The structure of the neural network may be specified, for example, by the number of layers from the input layer to the output layer, the type of each layer, the number of nodes (neurons) included in each layer, and the connection relationships between the nodes in each layer. In one example, the neural network may include any mechanism such as a recursive structure, a self-attention mechanism, or an autoregressive model. The neural network may also include any layer such as a fully connected layer, a convolutional layer, a pooling layer, a deconvolutional layer, an unpooling layer, a normalization layer, a dropout layer, or an LSTM (Long short-term memory). The neural network may include any type of model such as a diffusion model, a Transformer model, or a generative model. The weights of the connections between each node included in the neural network and the thresholds of each node are examples of computational parameters.

[0058] (Input / Output Configuration) The input and output configurations of each model (5, 6) may be modified as appropriate depending on the embodiment. The inputs of each model (5, 6) are not particularly limited and may be determined as appropriate depending on the embodiment, as long as they include behavioral history (e.g., target sample 30). The inputs of each model (5, 6) may consist only of behavioral history, or they may further include other information besides behavioral history. The outputs of each model (5, 6) are not particularly limited and may be determined as appropriate depending on the embodiment, as long as they include predicted values ​​of the degree to which purchasing behavior is triggered (first degree 31, second degree 32). The outputs of each model (5, 6) may consist only of predicted values ​​of the degree to which purchasing behavior is triggered, or they may further include other information besides the predicted values ​​of the degree to which purchasing behavior is triggered.

[0059] As shown in Figure 1, in one example, at least one of the rapid prediction model 5 and the long-term prediction model 6 may be configured to accept input of the user's behavioral history and attribute information, and to predict the degree to which purchasing behavior will be triggered from the input behavioral history and attribute information. Accordingly, the information processing device 1 may further acquire attribute information 301 of the target user T1. The attribute information 301 may be acquired by any method. In one example, the attribute information 301 may be acquired by an input operation of the target user T1. In another example, the attribute information 301 may be pre-stored in any storage area. The information processing device 1 may acquire the attribute information 301 from any storage area. Any storage area may include the memory resources of the information processing device 1, the memory resources of another computer, an external storage device, a storage medium, or a combination thereof. The information processing device 1 may further provide the acquired attribute information 301, along with the target sample 30 (behavioral history), to at least one of the rapid prediction model 5 and the long-term prediction model 6. In other words, providing a target sample 30 to at least one of the rapid-response prediction model 5 and the long-term prediction model 6 may be comprised of providing the target sample 30 and attribute information 301.

[0060] Attribute information (attribute information 301) is an example of information other than behavioral history. Attribute information 301 used for prediction may also be called target attribute information. Attribute information (attribute information 301) may be appropriately configured to show one or more attributes that may be related to the purchasing behavior of a user (target user T1). Attributes may include, for example, age, gender, address, etc. A user's attributes may have a relationship with the products that are the target of their purchasing behavior (for example, a student may tend to purchase products used in student life). Therefore, according to this example, by further including attribute information 301 in the input, it is possible to expect an improvement in the accuracy of predicting the degree of purchase behavior by at least one of the rapid prediction model 5 and the long-term prediction model 6. When providing attribute information 301 to the rapid prediction model 5 and the long-term prediction model 6, the data provided to the rapid prediction model 5 and the long-term prediction model 6 as attribute information 301 is the same as the target sample 30. The elements may be exactly the same or at least partially different. Providing attribute information 301 to the immediate forecasting model 5 and the long-term forecasting model 6 may include at least partially different data elements of attribute information 301 provided between the immediate forecasting model 5 and the long-term forecasting model 6, insofar as it provides data belonging to attribute information 301.

[0061] (output format) In one example, the first degree 31 (predicted value) predicted by the rapid prediction model 5 may be configured to show the difference (relative value) between the case where an incentive is given and the case where no incentive is given. In another example, the first degree 31 (predicted value) predicted by the rapid prediction model 5 may be configured to show the absolute value when an incentive is given, rather than the difference with the case where no incentive is given. In this case, the information processing device 1 may use the first degree 31 predicted as an absolute value directly to determine whether or not to give an incentive. Alternatively, the information processing device 1 may predict the degree to which purchasing behavior is triggered when no incentive is given using another model, and calculate the difference between the predicted value obtained from the other model and the first degree 31 predicted as an absolute value by the rapid prediction model 5. The information processing device 1 may use the calculated difference value to determine whether or not to give an incentive. Another model may be constructed similarly to the rapid prediction model 5, except that the attributes of the dependent variable differ in some respects (it predicts the degree to which purchasing behavior is triggered when no incentive is provided).

[0062] The same may apply to the long-term prediction model 6. In one example, the second degree 32 (predicted value) predicted by the long-term prediction model 6 may be configured to show the difference (relative value) between the case where an incentive is given and the case where no incentive is given. In another example, the second degree 32 (predicted value) predicted by the long-term prediction model 6 may be configured to show the absolute value when an incentive is given. In this case, the information processing device 1 may use the second degree 32 predicted as an absolute value directly to determine whether or not to give an incentive. Alternatively, the information processing device 1 may predict the degree to which purchasing behavior is triggered when no incentive is given using another model, and calculate the difference between the predicted value obtained from the other model and the second degree 32 predicted as an absolute value by the long-term prediction model 6. The information processing device 1 may use the calculated difference value to determine whether or not to give an incentive. The other model may be configured similarly to the long-term prediction model 6, except that the attributes of the target variable are slightly different (it predicts the degree to which purchasing behavior is triggered when no incentive is given).

[0063] (Executor / Processing Order) At least one of the rapid-response forecasting model 5 and the long-term forecasting model 6 may be deployed on the information processing device 1, or on a computer other than the information processing device 1. The calculations for at least one of the rapid-response forecasting model 5 and the long-term forecasting model 6 may be executed on the information processing device 1, or on a computer other than the information processing device 1.

[0064] Predicting the first degree 31 using the rapid prediction model 5 may be comprised of executing the calculation process of the rapid prediction model 5 and obtaining the predicted first degree 31 from the rapid prediction model 5. Predicting the first degree 31 using the rapid prediction model 5 may also be comprised of having another computer execute the calculation process of the rapid prediction model 5 and obtaining the calculation result of the rapid prediction model 5 (the predicted first degree 31) directly or indirectly from that other computer.

[0065] The same may apply to the long-term prediction model 6. That is, predicting the second degree 32 using the long-term prediction model 6 may consist of performing the calculations of the long-term prediction model 6 and obtaining the predicted second degree 32 from the long-term prediction model 6. Predicting the second degree 32 using the long-term prediction model 6 may also consist of having another computer perform the calculations of the long-term prediction model 6 and obtaining the calculation results of the long-term prediction model 6 (the predicted second degree 32). It may be constructed by obtaining data directly or indirectly from other computers.

[0066] When both the rapid-response forecasting model 5 and the long-term forecasting model 6 are deployed on other computers, the computer on which the rapid-response forecasting model 5 is deployed may be the same as or different from the computer on which the long-term forecasting model 6 is deployed. Furthermore, the execution order of the process for predicting the first degree 31 and the process for predicting the second degree 32 is not particularly limited and may be changed as appropriate depending on the embodiment. The process for predicting the second degree 32 may be executed before the process for predicting the first degree 31. The process for predicting the first degree 31 and the process for predicting the second degree 32 may be executed in parallel, at least partially.

[0067] [Generating long-term prediction models] Figure 4 schematically shows an example of the generation of the long-term prediction model 6 according to this embodiment. In the example in Figure 4, it is assumed that the long-term prediction model 6 is constructed using a machine learning model and generated by machine learning. In one example, a model generation device 2 may be provided to generate the long-term prediction model 6. The model generation device 2 may be one or more computers configured to generate a trained machine learning model that can be used as the long-term prediction model 6 by controlling the execution of machine learning. An information processing system may be configured with the information processing device 1 and the model generation device 2.

[0068] In one example of this embodiment, the model generation device 2 may control the machine learning of the long-term prediction model 6. The model generation device 2 may output the results of the machine learning. The long-term prediction model 6 may be appropriately configured to predict the degree to which purchasing behavior is triggered over a longer period than the immediate period (second degree) by providing incentives based on the user's behavioral history related to purchasing behavior. The machine learning may include training the long-term prediction model 6 by providing training samples 70 to the long-term prediction model 6 so that the predicted value of the degree (second degree) derived by the long-term prediction model 6 approaches the true value 75 corresponding to the training samples 70. The true value 75 may be called a label, a teacher signal, etc.

[0069] The combination of training sample 70 and true value 75 may be collected as training dataset D1. The model generator 2 may acquire multiple training datasets D1, each composed of a combination of training sample 70 and true value 75. The model generator 2 may control the execution of machine learning for each acquired training dataset D1 so that the long-term prediction model 6 is trained so that the second degree prediction value derived by providing the training sample 70 fits the corresponding true value 75. As a result of performing this machine learning, a long-term prediction model 6 (trained machine learning model) can be generated that has acquired the ability to predict the degree to which purchasing behavior will be triggered over a long period of time from the user's behavior history. The generated long-term prediction model 6 may be provided to the information processing device 1 as appropriate.

[0070] Each training dataset D1 may be collected from sample user S1 as appropriate. Training samples 70 may be constructed in the same way as target samples 30. Training samples 70 may be generated as appropriate from behavioral history obtained up to the time of collecting training dataset D1. Training samples 70 may be generated as appropriate from, for example, purchase data (POS data), web browser browsing history, etc. The true value 75 may be generated as appropriate from the degree to which purchase behavior of sample user S1 was triggered (number of purchases, purchase quantity, purchase amount, etc.) after incentives were given in the past. The true value 75 may be generated as appropriate from, for example, purchase data (POS data). In one example, the true value 75 may be constructed to show the difference (relative value) between the case where an incentive was given and the case where an incentive was not given. In another example, the true value 75 may be constructed to show the absolute value when an incentive was given. Also, in one example, the true value 75 may be constructed from the measured value of the degree to which purchase behavior was triggered measured over a long period (second period) after an incentive was given. In other words, the true value of 75 provided an incentive. The true value may be obtained by measuring the degree to which purchasing behavior is triggered over a long period in the past. For example, the training sample 70 may be generated from the behavioral history before the Kth incentive was given, and the true value 75 may be generated from the result of measuring the degree to which purchasing behavior is triggered after the Kth incentive was given (K is any natural number). The number of sample users S1 is not particularly limited and may be determined as appropriate depending on the embodiment. Sample users S1 may or may not include target users T1. In one example, if sample users S1 include target users T1, at least a portion of the target samples 30 acquired in the past may be used as the training sample 70.

[0071] Each training dataset D1 may be collected by one or more computers. The one or more computers involved in collecting training dataset D1 may or may not include the model generator 2. Each training dataset D1 may be automatically generated by computer processing, or it may be manually generated with at least partial intervention from an operator. The computer storing the training dataset D1 may or may be the same as the computer generating the training dataset D1. At least a portion of each training dataset D1 may be generated on the model generator 2, or it may be generated on a computer other than the model generator 2. The model generator 2 may obtain at least a portion of each training dataset D1 from an external computer. The external computer may include an external storage device such as a NAS (Network Attached Storage).

[0072] If the long-term prediction model 6 is configured to accept further input of user attribute information, each training dataset D1 collected may further include attribute information 701 of a sample user S1. The attribute information 701 may be configured in the same way as the attribute information 301 above. The attribute information 701 may be obtained from the sample user S1 as appropriate. The long-term prediction model 6 may be trained so that the second degree prediction value derived by providing the training sample 70 and the attribute information 701 fits the corresponding true value 75. The attribute information 701 used for training may be called training attribute information.

[0073] According to one example of this embodiment, it is possible to generate a long-term prediction model 6 (trained machine learning model) that has acquired the ability to predict the degree to which incentives induce long-term purchasing behavior (second degree) from the user's behavioral history. By using the generated long-term prediction model 6 to identify users with a high degree to which long-term purchasing behavior is induced, it is possible to expect effective incentive provision as described above.

[0074] (Substitution with predicted values) For example, the true value 75 for all training datasets D1 may be obtained by measuring the degree to which purchasing behavior is triggered over a long period. However, if the measured value of the degree to which purchasing behavior is triggered over a long period is used as the true value 75 for all training datasets D1, collecting the true value 75 will be time-consuming and costly. Therefore, in another example, for the training sample 70 of the behavioral history of sample user S1, the true value 75 of the degree to which purchasing behavior is triggered over a long period may be derived from the measured value 45 of the degree to which purchasing behavior is triggered by providing incentives, using the transformation model 4. In other words, by providing the measured value 45 of the degree to which purchasing behavior is triggered over a short period, the predicted value of the degree to which purchasing behavior is triggered over a long period may be derived from the measured value 45. The derived predicted value may be used as the true value 75 instead of the measured value.

[0075] The short-term period may be any period shorter than the long-term period (second period). The short-term period may be equivalent to or longer than the immediate period (first period). For example, the short-term period may be from the time the incentive is granted until the third predetermined time in the future. The period up to the point in time when the interval has elapsed (the third predetermined period) may be defined. The third predetermined period may be defined to be shorter than the second predetermined period. For example, the short period may be defined as a period of several weeks. For example, the short period may be defined as between 3 weeks and 16 weeks. The short period may also be defined in relation to the long period. In one example, the short period may be defined as less than half of the long period. In another example, the short period may be defined as less than 3 / 4 of the long period. The short period may be called the third period. The degree to which purchasing behavior is triggered during the short period may be called the third degree.

[0076] Except for the shorter measurement period, the measured value 45 may be obtained in the same way as the true value 75 when using the measured value. The measured value 45 may be generated appropriately from, for example, purchase data (POS data). Furthermore, if the true value 75 is derived from a portion of the multiple training datasets D1 used for machine learning using the transformation model 4, a reduction in the effort required to collect the true value 75 can be expected. Therefore, the range from which the true value 75 is derived by the transformation model 4 may be at least a portion of the multiple training datasets D1 used for machine learning. In one example, the true value 75 of all training datasets D1 may be derived from the measured value 45. In another example, the true value 75 of a portion of the training dataset D1 may be derived from the measured value 45. The fact that the true value 75 for the training sample 70 of sample user S1 is derived from the measured value 45 may be achieved by deriving the true value 75 from the measured value 45 for at least a portion of the multiple training datasets D1.

[0077] The transformation model 4 may be configured as appropriate to predict a degree of purchase behavior ranging from a third degree, which induces purchase behavior in the short term, to a second degree, which induces purchase behavior in the long term, by providing incentives. That is, the transformation model 4 may be configured to convert the third degree value into a second degree prediction. Similar to the immediate prediction model 5, the transformation model 4 may consist of at least one of a rule-based model and a trained machine learning model. In one example, the transformation model 4 may include a neural network.

[0078] The input and output configuration of the transformation model 4 may be modified as appropriate depending on the embodiment. The inputs of the transformation model 4 are not particularly limited and may be determined as appropriate depending on the embodiment, as long as they include the third degree value. The inputs of the transformation model 4 may consist only of the third degree value, or they may further include other information in addition to the third degree value. The output of the transformation model 4 is not particularly limited and may be determined as appropriate depending on the embodiment, as long as it includes the second degree prediction value. The output of the transformation model 4 may consist only of the second degree prediction value, or it may further include other information in addition to the second degree prediction value.

[0079] As shown in Figure 4, in one example, the conversion model 4 may be configured to accept input from at least one of the user's behavior history and attribute information, and to derive a predicted value for the second degree from at least one of the user's behavior history and attribute information, along with the value for the third degree. Accordingly, the conversion model 4 may be further given at least one of the training sample 70 and attribute information 701 in addition to the measured value 45 for the third degree. That is, in deriving the true value 75 (predicting the second degree), at least one of the training sample 70 and attribute information 701 may be further used in addition to the measured value 45 for the third degree. The user's behavior history (training sample 70) and attribute information (attribute information 701) are examples of information other than the value for the third degree. According to one example of this embodiment, by further using at least one of the training sample 70 and attribute information 701, it is possible to expect an improvement in the accuracy of deriving the predicted value for the second degree.

[0080] Furthermore, when providing behavioral history (training sample 70) to the transformation model 4 and the long-term prediction model 6, the data elements provided as behavioral history to the transformation model 4 and the long-term prediction model 6 may be completely identical or at least partially different, similar to the target sample 30 mentioned above. As long as data belonging to behavioral history is provided, the behavioral history can be used with the transformation model 4 and the long-term prediction model 6. Providing data to the long-term prediction model 6 may include at least partially different data elements of behavioral history provided between the transformation model 4 and the long-term prediction model 6. The same may apply to attribute information (attribute information 701). That is, when providing attribute information (attribute information 701) to the transformation model 4 and the long-term prediction model 6, the data elements provided to the transformation model 4 and the long-term prediction model 6 as attribute information may be exactly the same or at least partially different. Providing attribute information to the transformation model 4 and the long-term prediction model 6 may include at least partially different data elements of attribute information provided between the transformation model 4 and the long-term prediction model 6, as long as the data belonging to attribute information is provided.

[0081] The conversion model 4 may be deployed on the model generation device 2, or on a computer other than the model generation device 2. The calculation process for deriving the true value 75 from the measured value 45 may be executed on the model generation device 2, or on a computer other than the model generation device 2. In one example, the model generation device 2 may obtain the true value 75 corresponding to the training sample 70 by providing the conversion model 4 with the measured value 45 of the third degree corresponding to the training sample 70 and executing the calculation process of the conversion model 4. In this case, at the time of acquisition, the training dataset D1 may be provisionally composed of a combination of the training sample 70 and the measured value 45 of the third degree. In another example, the calculation process for deriving the true value 75 from the measured value 45 of the third degree using the conversion model 4 may be executed in advance on one or more arbitrary computers. Thus, the true value 75 may be derived in advance. The model generation device 2 may obtain the true value 75 that has been derived in advance from the measured value 45 of the third degree.

[0082] In one example of this embodiment, by using the transformation model 4 for at least a portion of the multiple training datasets D1, predicted values ​​are derived from measured values ​​45 measured over shorter periods rather than longer periods, and these derived predicted values ​​are used as substitutes for true values ​​75. This shortens the period for measuring the degree of purchase behavior in order to obtain true values ​​75, thereby reducing the effort required to collect true values ​​75 for machine learning of the long-term prediction model 6. Furthermore, according to this example of the embodiment, a reduction in the time required to obtain the long-term prediction model 6 can be expected.

[0083] [Generating a transformation model] Figure 5 schematically shows an example of the generation scene of the transformation model 4 according to this embodiment. In the example in Figure 5, it is assumed that the transformation model 4 is constructed using a machine learning model and generated by machine learning. In one example, the model generation device 2 may be further configured to generate a trained machine learning model that can be used as the transformation model 4 by controlling the execution of machine learning.

[0084] In one example of this embodiment, the model generation device 2 may control the machine learning of the transformation model 4. The model generation device 2 may output the results of the machine learning of the transformation model 4. The transformation model 4 may be appropriately configured to predict the degree of inducing purchasing behavior over a long period (second degree) from the degree of inducing purchasing behavior over a short period (third degree) by providing incentives. The machine learning may include training the transformation model 4 so that the predicted value of the second degree derived by the transformation model 4 approaches the true value of the second degree corresponding to the training value 71, by providing the transformation model 4 with a training value 71 of the third degree.

[0085] The combination of training values ​​71 and true values ​​76 may be collected as training dataset D2. The model generator 2 may acquire multiple training datasets D2, each composed of a combination of training values ​​71 and true values ​​76. For each acquired training dataset D2, the model generator 2 may control the execution of machine learning so that the transformation model 4 is trained so that the second degree predicted value derived by providing a third degree training value 71 fits the corresponding true value 76. As a result of performing this machine learning, the third degree to the second degree A transformation model 4 (a trained machine learning model) that has acquired the ability to predict the result can be generated. The generated transformation model 4 may be used as appropriate in deriving the true value 75. If the calculation to derive the true value 75 is performed on another computer (an external computer), the generated transformation model 4 may be provided as appropriate to the other computer that derives the true value 75.

[0086] Each training dataset D2 may be collected as appropriate from sample user S2. The training values ​​71 and true values ​​76 may be generated as appropriate from the degree to which purchasing behavior (number of purchases, purchase quantity, purchase amount, etc.) of sample user S2 has been given incentives in the past. The training values ​​71 and true values ​​76 may be generated as appropriate from purchase data (POS data), for example. The true value 76 may be constructed in the same way as the true value 75 above. The true value 76 may be obtained in the same way as the true value 75 above. In one example, the true value 76 may be constructed from measured values ​​of the degree to which purchasing behavior has been measured over a long period (second period) after incentives have been given. That is, the true value 76 may be obtained by measuring the degree to which purchasing behavior has been triggered over a long period in the past after incentives have been given. In one example, the training value 71 may be obtained in the same way as the true value 76, except that the measurement period is different (it is a short period). In other words, the training value 71 may be obtained by measuring the degree to which purchasing behavior is elicited in the short period in the past after an incentive has been given. The number of sample users S2 is not particularly limited and may be determined as appropriate depending on the embodiment. Sample users S2 may or may not overlap with sample users S1 at least partially. If sample users S2 overlap with sample users S1 at least partially, at least a portion of the collected true values ​​76 may be used as true values ​​75.

[0087] Each training dataset D2 may be collected by one or more computers. The one or more computers involved in collecting each training dataset D2 may or may not include the model generator 2. Each training dataset D2 may be automatically generated by computer processing, or it may be manually generated with at least partial intervention from an operator. The computer storing the training dataset D2 may be the same as or different from the computer generating the training dataset D2. At least a portion of each training dataset D2 may be generated on the model generator 2, or it may be generated on a computer other than the model generator 2. The model generator 2 may obtain at least a portion of each training dataset D2 from an external computer. The external computer may include an external storage device such as a NAS.

[0088] If the transformation model 4 is configured to accept at least one of the user's behavior history and attribute information as input, each collected training dataset D2 may further include at least one of the training samples 72 and attribute information 721 of the sample user S2's behavior history. The training samples 72 may be generated in the same way as the training samples 70 above. At least some combinations of training samples 72 and true values ​​76 may be used as training dataset D1. The attribute information 721 may be configured in the same way as the attribute information 701 above. The attribute information 721 may be obtained from the sample user S2 as appropriate. At least some of the attribute information 721 may be used as attribute information 701 above. The attribute information 721 may be referred to as training attribute information. The transformation model 4 may be trained such that the second degree prediction derived by providing at least one of the training samples 72 and attribute information 721 together with the training values ​​71 fits the corresponding true values ​​76.

[0089] According to one example of this embodiment, a transformation model 4 (a trained machine learning model) can be generated that has acquired the ability to predict the degree of purchase behavior over a long period from the degree of purchase behavior in a short period. By obtaining at least a portion of the true values ​​75 used for machine learning of the long-term prediction model 6 using the transformation model 4, the effort required to collect the true values ​​75 can be reduced as described above.

[0090] [Generating a rapid-response predictive model] Figure 6 schematically shows an example of the generation scene of the rapid prediction model 5 according to this embodiment. In the example in Figure 6, it is assumed that the rapid prediction model 5 is constructed using a machine learning model and generated by machine learning. In one example, the model generation device 2 may be further configured to generate a trained machine learning model that can be used as the rapid prediction model 5 by controlling the execution of machine learning.

[0091] In one example of this embodiment, the model generation device 2 may control the machine learning of the rapid prediction model 5. The model generation device 2 may output the results of the machine learning of the rapid prediction model 5. The rapid prediction model 5 may be appropriately configured to predict the degree to which purchasing behavior is triggered (first degree) in a rapid period by the provision of incentives, based on the user's behavioral history related to purchasing behavior. The machine learning may include training the rapid prediction model 5 by providing the training sample 73 to the rapid prediction model 5 so that the predicted value of the degree (first degree) derived by the rapid prediction model 5 approaches the true value 77 corresponding to the training sample 73.

[0092] The combination of training sample 73 and true value 77 may be collected as training dataset D3. The model generator 2 may acquire multiple training datasets D3, each composed of a combination of training sample 73 and true value 77. The model generator 2 may control the execution of machine learning for each acquired training dataset D3 so that the responsive prediction model 5 is trained so that the first degree of predicted value derived by providing the training sample 73 fits the corresponding true value 77. As a result of performing this machine learning, a responsive prediction model 5 (trained machine learning model) can be generated that has acquired the ability to predict the degree to which purchasing behavior will be triggered in an immediate period from the user's behavior history. The generated responsive prediction model 5 may be provided to the information processing device 1 as appropriate.

[0093] Each training dataset D3 may be collected as appropriate from sample users S3. Training samples 73 may be generated in the same way as training samples 70. The true value 77 may be obtained in the same way as true value 75, etc., except that the measurement period is an immediate period. For example, true value 77 may be obtained by measuring the degree to which purchasing behavior was elicited in the immediate period in the past after the incentive was given. The number of sample users S3 is not particularly limited and may be determined as appropriate depending on the embodiment. Sample users S3 may or may not overlap with sample users S1 at least partially. If sample users S3 overlap with sample users S1 at least partially, at least a portion of the collected training samples 73 may be used as training samples 70 of training dataset D1. That is, the training samples (70, 73) of each training dataset (D1, D3) may be collected in common at least partially.

[0094] Each training dataset D3 may be collected by one or more computers. The one or more computers involved in collecting each training dataset D3 may or may not include the model generator 2. Each training dataset D3 may be automatically generated by computer processing, or it may be generated manually with at least partial intervention from an operator. The computer storing the training dataset D3 may or may not be the same as the computer generating the training dataset D3. At least a portion of each training dataset D3 may be generated on the model generator 2, or it may be generated on a computer other than the model generator 2. The model generator 2 may obtain at least a portion of each training dataset D3 from an external computer. The external computer may include an external storage device such as a NAS.

[0095] If the rapid prediction model 5 is configured to accept further input of user attribute information: Each training dataset D3 collected may further include attribute information 731 of a sample user S3. Attribute information 731 may be structured similarly to attribute information 701. Attribute information 731 may be obtained from sample user S3 as appropriate. Attribute information 731 may be used as attribute information 701. Attribute information 731 may be referred to as training attribute information. The rapid prediction model 5 may be trained so that the first degree prediction value derived by providing the training sample 73 and attribute information 731 fits the corresponding true value 77.

[0096] According to one example of this embodiment, it is possible to generate an immediate prediction model 5 (a trained machine learning model) that has acquired the ability to predict the degree to which immediate purchasing behavior is triggered by the provision of incentives (degree 1) from the user's behavior history. By using the long-term prediction model 6 to identify users who are highly likely to be triggered by long-term purchasing behavior, and by using the generated immediate prediction model 5 to identify users who are highly likely to be triggered by immediate purchasing behavior, it is possible to expect effective incentive provision as described above.

[0097] [Machine Learning Control] The machine learning computations may be performed on the model generator 2, or on another computer (external computer) other than the model generator 2. Controlling machine learning may include at least one of the following: performing machine learning within the model generator 2, or giving instructions to an external computer to perform machine learning. That is, in one example, the model generator 2 may perform machine learning for each model (4, 5, 6). In another example, the model generator 2 may give instructions to an external computer to perform machine learning for each model (4, 5, 6). The external computer may perform machine learning for each model (4, 5, 6) in response to instructions from the model generator 2. In yet another example, the model generator 2 may perform some of the processing for machine learning for each model (4, 5, 6), and the external computer may perform the remaining processing for machine learning for each model (4, 5, 6). The external computer may consist of one or more computers. The external computer may be connected to the model generator 2 via a network, or it may be connected directly to the model generator 2. The type of network is not particularly limited and may be appropriately selected depending on the embodiment.

[0098] [Output machine learning results] For each model (4, 5, 6), the machine learning results may be output in any way. The content and destination of the output information may be appropriately selected depending on the embodiment. In one example, outputting the machine learning results may include generating training result data that shows the machine learning results (trained machine learning model) and saving the generated training result data to any memory area. Any memory area may include, for example, the memory resources of the model generation device 2, the memory resources of another computer, an external storage device, a storage medium, or a combination thereof. In another example, outputting the machine learning results may include outputting the convergence results of training in machine learning. The convergence results may include, for example, loss, number of time steps, learning curve, etc.

[0099] [System Configuration] In one example, as shown in Figures 4 to 6, the information processing device 1 and the model generation device 2 may be connected to each other via a network. The type of network may be appropriately selected from, for example, the Internet, a wireless communication network, a mobile communication network, a telephone network, a dedicated network, etc. However, the method of data exchange between each device (1, 2) is not limited to this example and may be appropriately selected depending on the embodiment. In another example, data exchange between each device (1, 2) may be performed via a storage medium.

[0100] Furthermore, in the examples shown in Figures 4 to 6, the information processing device 1 and the model generation device 2 are separate. It is a computer. However, the configuration of the system according to this embodiment is not limited to this example and may be modified as appropriate depending on the embodiment. In another example, the information processing device 1 and the model generation device 2 may be configured as a single computer. That is, the information processing device 1 may also operate as the model generation device 2. At least one of the information processing device 1 and the model generation device 2 may be configured as multiple computers.

[0101] §2 Example Configuration [Hardware configuration] (Information processing device) Figure 7 schematically shows an example of the hardware configuration of the information processing device 1 according to this embodiment. In one example, the information processing device 1 may be configured as a computer in which a control unit 11, a storage unit 12, a communication module 13, an input device 14, and an output device 15 are electrically connected.

[0102] The control unit 11 may include a hardware processor such as a CPU, RAM (Random Access Memory), and ROM (Read Only Memory), and is configured to perform information processing based on a program and various data. The control unit 11 (CPU) is an example of a processor resource.

[0103] The storage unit 12 may include, for example, a hard disk drive, a solid-state drive, or a semiconductor memory, and is configured to hold arbitrary data. The storage unit 12, RAM, and ROM are examples of memory resources of the information processing device 1. In one example of this embodiment, the storage unit 12 may store various information such as a program 81, immediate prediction model data 500, and long-term prediction model data 600.

[0104] Program 81 is a program that causes the information processing device 1 to perform information processing related to the granting of incentives (Figure 14, described later). Program 81 includes a series of instructions for said information processing. The immediate prediction model data 500 is configured to represent the immediate prediction model 5. The long-term prediction model data 600 is configured to represent the long-term prediction model 6. The configuration of the immediate prediction model data 500 is not particularly limited and may be determined appropriately depending on the embodiment, as long as it can hold information for performing calculation processing of the immediate prediction model 5. For example, if the immediate prediction model 5 is generated by machine learning, the immediate prediction model data 500 may be configured to include information indicating the values ​​of calculation parameters adjusted by machine learning. The immediate prediction model data 500 may also be configured to include information indicating the configuration of the immediate prediction model 5 (e.g., the structure of the neural network). Similarly, the configuration of the long-term prediction model data 600 is not particularly limited and may be determined appropriately depending on the embodiment, as long as it can hold information for performing calculation processing of the long-term prediction model 6. At least one of the immediate forecast model data 500 and the long-term forecast model data 600 may be managed separately from program 81 or incorporated into program 81. If the calculation processing of immediate forecast model 5 is performed on an external computer, the immediate forecast model data 500 may be omitted from the storage unit 12. If the calculation processing of long-term forecast model 6 is performed on an external computer, the long-term forecast model data 600 may be omitted from the storage unit 12.

[0105] In one example, at least one of the program 81, the immediate prediction model data 500, and the long-term prediction model data 600 may be stored in the storage medium 91 instead of or together with the storage unit 12. The storage medium 91 is configured to store various types of information (stored programs, etc.) by electrical, magnetic, optical, mechanical, or chemical means so that a machine such as a computer can read the information. The storage unit 12 and the storage medium 91 are examples of non-temporary storage media. The information processing device 1 may acquire at least one of the program 81, the immediate prediction model data 500, and the long-term prediction model data 600 from the storage medium 91. The storage medium 91 may be a disk-type storage medium (CD, DVD, etc.). The storage medium may be a non-disk type semiconductor memory (such as flash memory). Any drive device may be used to read the information stored in the storage medium 91. The type of drive device may be selected according to the storage medium 91. The drive device may be connected to the information processing device 1 in any way. The storage medium 91 may include an external storage device.

[0106] The communication module 13 is configured to perform wired or wireless communication over a network. The communication module 13 may consist of, for example, a wired LAN (Local Area Network) module, a wireless LAN module, etc. The network standard is not particularly limited and may be appropriately selected depending on the embodiment. For example, the type of network may be appropriately selected from the Internet, wireless communication network, mobile communication network, telephone network, dedicated network, etc. The information processing device 1 may perform data communication with other computers (for example, a model generation device 2, an external computer, etc.) via the communication module 13.

[0107] The input device 14 is configured to accept information input. The input device 14 may consist of, for example, a mouse, keyboard, control, microphone, etc. The output device 15 is configured to output information. The output device 15 may consist of, for example, a display, speaker, etc. The information processing device 1 may be operated via the input device 14 and the output device 15. The input device 14 and the output device 15 may be directly connected to the information processing device 1, or they may be indirectly connected via at least one of the communication module 13 and the external interface. The external interface may be, for example, USB (Universal Serial The devices may be configured to connect to external devices via wired or wireless connections using a Bus port, dedicated port, etc. The input device 14 and the output device 15 may be integrated in at least part of their configuration using a touch panel display or the like.

[0108] Regarding the specific hardware configuration of the information processing device 1, components can be omitted, replaced, and added as appropriate depending on the embodiment. For example, the control unit 11 may include multiple hardware processors. Hardware processors include microprocessors, FPGAs (field-programmable gate arrays), DSPs (digital signal processors), and GPs. It may be composed of a U (Graphics Processing Unit), an ASIC (application-specific integrated circuit), etc. A small number of communication modules 13, input devices 14 and output devices 15 At least one of the following may be omitted. Program 81, immediate prediction model data 500, and long-term prediction model data 600 may be stored in an external storage device such as a NAS. An external storage device is also an example of a non-temporary storage medium. The information processing device 1 may consist of multiple computers. In this case, the hardware configuration of each computer may or may not be the same. The information processing device 1 may consist of a computer designed specifically for the services provided, as well as a general-purpose server device, a general-purpose PC (Personal Computers, laptops, terminal devices, etc. Terminal devices include smartphones, etc. This may include user terminals such as tablet devices.

[0109] (Model generation device) Figure 8 schematically shows an example of the hardware configuration of the model generation device 2 according to this embodiment. In one example, the model generation device 2 may be configured as a computer in which a control unit 21, a storage unit 22, a communication module 23, an input device 24, and an output device 25 are electrically connected.

[0110] The control units 21 to output devices 25 and the storage medium 92 of the model generation device 2 may be configured in the same way as the control units 11 to output devices 15 and the storage medium 91 of the information processing device 1. The control unit 21 (CPU) is an example of the processor resources of the model generation device 2. The storage unit 22, RAM, and ROM are examples of the memory resources of the model generation device 2. In one example of this embodiment, the storage unit 22 stores various information such as the model generation program 82, the converted model data 400, the immediate prediction model data 500, and the long-term prediction model data 600. That's fine.

[0111] The model generation program 82 is a program that causes the model generation device 2 to execute information processing related to the generation of the transformation model 4 (Figure 11 described later), information processing related to the generation of the long-term forecast model 6 (Figure 12 described later), and information processing related to the generation of the immediate forecast model 5 (Figure 13 described later). The model generation program 82 includes a series of instructions for said information processing. The instructions for information processing related to the generation of each model (4, 5, 6) may be separated into a separate program. If information processing related to the generation of the transformation model 4 is omitted, the instructions for information processing related to the generation of the transformation model 4 may be omitted from the model generation program 82. If information processing related to the generation of the immediate forecast model 5 is omitted, the instructions for information processing related to the generation of the immediate forecast model 5 may be omitted from the model generation program 82. If information processing related to the generation of the long-term forecast model 6 is omitted, the instructions for information processing related to the generation of the long-term forecast model 6 may be omitted from the model generation program 82.

[0112] The transformed model data 400, the immediate prediction model data 500, and the long-term prediction model data 600 may be generated as a result of executing the model generation program 82. The transformed model data 400 is configured to represent the transformed model 4. Similar to the immediate prediction model data 500, the configuration of the transformed model data 400 is not particularly limited and may be determined as appropriate depending on the embodiment, as long as it can hold information for executing the calculation processing of the transformed model 4. When the model generation device 2 performs calculations to derive the true value 75, the transformed model data 400 may be managed separately from the model generation program 82 (instructions for information processing related to the generation of the long-term prediction model 6), or it may be incorporated into the model generation program 82.

[0113] At least one of the model generation program 82, the converted model data 400, the immediate prediction model data 500, and the long-term prediction model data 600 may be stored in the storage medium 92 instead of or together with the storage unit 22. The model generation device 2 may obtain the model generation program 82 from the storage medium 92. In one example, when the machine learning calculation processing of the long-term prediction model 6 is performed on the model generation device 2, at least a portion of the multiple training datasets D1 may be stored in at least one of the storage unit 22 and the storage medium 92. The model generation device 2 may obtain at least a portion of the multiple training datasets D1 from at least one of the storage unit 22 and the storage medium 92. At least a portion of the multiple training datasets D1 may be stored in an external storage device. In one example, when the machine learning calculation processing of the converted model 4 is performed on the model generation device 2, at least a portion of the multiple training datasets D2 may be stored in at least one of the storage unit 22 and the storage medium 92. The model generation device 2 may obtain at least a portion of multiple training datasets D2 from at least one of the storage unit 22 and the storage medium 92. At least a portion of the multiple training datasets D2 may be stored in an external storage device. Also, in one example, when the machine learning calculation processing of the rapid prediction model 5 is performed on the model generation device 2, at least a portion of the multiple training datasets D3 may be stored in at least one of the storage unit 22 and the storage medium 92. The model generation device 2 may obtain at least a portion of multiple training datasets D3 from at least one of the storage unit 22 and the storage medium 92. At least a portion of the multiple training datasets D3 may be stored in an external storage device.

[0114] The model generation device 2 may perform data communication with other computers (information processing device 1, external computer, external storage device, etc.) via the communication module 23. The model generation device 2 may be operated via the input device 24 and the output device 25.

[0115] Furthermore, regarding the specific hardware configuration of the model generation device 2, depending on the embodiment, components can be omitted, replaced, and added as appropriate. For example, the control unit 21 may include multiple hardware processors. The hardware processors may be microprocessors, F It may consist of PGA, DSP, GPU, ASIC, etc. At least one of the communication module 23, input device 24, and output device 25 may be omitted. The model generation device 2 may consist of multiple computers. In this case, the hardware configuration of each computer may or may not be the same. The model generation device 2 may be an information processing device designed specifically for the service provided, or it may be a general-purpose server device, a general-purpose PC, a tablet PC, a terminal device, etc.

[0116] [Software Configuration] (Information processing device) Figure 9 schematically shows an example of the software configuration of the information processing device 1 according to this embodiment. The control unit 11 of the information processing device 1 executes instructions contained in the program 81 stored in the storage unit 12 using the CPU. As a result, the information processing device 1 operates as a computer equipped with a sample acquisition unit 111, a first prediction unit 112, a second prediction unit 113, an assignment determination unit 114, a quantity determination unit 115, and an output processing unit 116 as software modules. In other words, in this example of the embodiment, each software module of the information processing device 1 is realized by the control unit 11 (CPU).

[0117] The sample acquisition unit 111 is configured to acquire target samples 30 of the behavioral history related to the purchasing behavior of target user T1. The first prediction unit 112 is configured to predict a first degree 31 that will induce purchasing behavior in an immediate period by providing incentives, by providing the acquired target samples 30 to the immediate prediction model 5. When the calculation processing of the immediate prediction model 5 is performed on the information processing device 1, the first prediction unit 112 may include the immediate prediction model 5 by holding immediate prediction model data 500. The second prediction unit 113 is configured to predict a second degree 32 that will induce purchasing behavior in a period longer than the immediate period by providing incentives, by providing the acquired target samples 30 to the long-term prediction model 6. When the calculation processing of the long-term prediction model 6 is performed on the information processing device 1, the second prediction unit 113 may include the long-term prediction model 6 by holding long-term prediction model data 600. The grant determination unit 114 is configured to determine whether or not to grant an incentive to the target user T1 based on the predicted first degree 31 and second degree 32.

[0118] The quantity determination unit 115 is configured to determine the amount of incentive 33 according to the predicted first degree 31 and second degree 32. The output processing unit 116 is configured to output incentive information 35 regarding the granting of incentives to target user T1 if it is determined to grant an incentive to target user T1. If the output processing unit 116 is determined not to grant an incentive to target user T1, the information processing device 1 is configured to omit the output of the incentive information 35. If the quantity determination unit 115 determines the amount of incentive 33, the incentive information 35 may be configured to relate to the granting of the determined amount 33 of incentives.

[0119] (Model generation device) Figure 10 schematically shows an example of the software configuration of the model generation device 2 according to this embodiment. The control unit 21 of the model generation device 2 executes instructions contained in the model generation program 82 stored in the storage unit 22 using the CPU. As a result, the model generation device 2 operates as a computer equipped with a data acquisition unit 211, a learning processing unit 212, and an output processing unit 213 as software modules. In other words, in one example, each software module of the model generation device 2 may also be implemented by the control unit 21 (CPU).

[0120] When the transformation model 4 is generated by the model generation device 2, the data acquisition unit 211 may be configured to acquire multiple training datasets D2. The learning processing unit 212 generates the transformation model The output processing unit 213 may be configured to control the machine learning of model 4. The output processing unit 213 may be configured to output the results of the machine learning of the transformation model 4.

[0121] When the long-term prediction model 6 is generated by the model generation device 2, the data acquisition unit 211 may be configured to acquire multiple training datasets D1. In one example, the true values ​​75 of at least some of the multiple training datasets D1 may be derived from measured values ​​45 of the degree to which purchasing behavior is triggered in the short term by the provision of incentives, using a transformation model 4. When the calculation process for deriving the true values ​​75 is performed on the model generation device 2, the data acquisition unit 211 may include a transformation model 4 by holding transformation model data 400. The learning processing unit 212 may be configured to control the machine learning of the long-term prediction model 6. The output processing unit 213 may be configured to output the results of the machine learning of the long-term prediction model 6.

[0122] When the rapid prediction model 5 is generated by the model generation device 2, the data acquisition unit 211 may be configured to acquire multiple training datasets D3. The learning processing unit 212 may be configured to control the machine learning of the rapid prediction model 5. The output processing unit 213 may be configured to output the results of the machine learning of the rapid prediction model 5.

[0123] (others) In this example of the embodiment described above, each software module of the information processing device 1 and the model generation device 2 are implemented by a general-purpose CPU. However, the method of implementing each of the above modules is not limited to this example and may be appropriately modified depending on the embodiment. Some or all of the above software modules may be implemented by one or more dedicated processors or chipsets. Each of the above modules may be implemented as a hardware module. Regarding the software configuration of the information processing device 1 and the model generation device 2, modules may be omitted, replaced, and added as appropriate depending on the embodiment.

[0124] §3 Example of Operation [Generating a transformation model] Figure 11 is a flowchart showing an example of the processing procedure for generating the conversion model 4 by the model generation device 2 according to this embodiment. The following processing procedure is an example of an information processing method (model generation method for the conversion model 4) executed by a computer. The following processing procedure is merely an example, and each step may be modified as much as possible. Furthermore, depending on the embodiment, steps in the following processing procedure can be omitted, replaced, and added as appropriate.

[0125] (Step S501) In step S501, the control unit 21 operates as a data acquisition unit 211 and acquires multiple training datasets D2. Each training dataset D2 may consist of a combination of a third-degree training value 71 and a second-degree true value 76 of a sample user S2. In one example, each training dataset D2 may further include at least one of a training sample 72 and attribute information 721. Once multiple training datasets D2 have been acquired, the control unit 21 proceeds to the next step S502.

[0126] (Step S502) In step S502, the control unit 21 operates as a learning processing unit 212 and controls the machine learning of the transformation model 4. The machine learning may include training the transformation model 4 so that the predicted value of the second degree derived by the transformation model 4 approaches the true value of the second degree corresponding to the training value 71, by providing the transformation model 4 with a training value of the third degree 71.

[0127] In one example, the control unit 21 processes each acquired training dataset D2 and determines the third degree of training. The machine learning process may be controlled so that the transformation model 4 is trained to approach the corresponding true value 76 when the training value 71 is provided to the transformation model 4. When the machine learning calculations for the transformation model 4 are performed on the model generation device 2, each acquired training dataset D2 may be used as appropriate for the machine learning of the transformation model 4. When the machine learning calculations for the transformation model 4 are performed on an external computer, the control unit 21 may provide each acquired training dataset D2 to the external computer and have the external computer perform the machine learning of the transformation model 4. If each training dataset D2 further includes at least one of the training samples 72 and attribute information 721, the transformation model 4 may be trained so that the predicted value of the second degree derived by providing at least one of the training samples 72 and attribute information 721 together with the training value 71 fits the corresponding true value 76. As a result of performing the machine learning in step S502, a transformation model 4 (trained machine learning model) that has acquired the ability to predict the second degree from the third degree can be generated. Once machine learning is performed on the conversion model 4, the control unit 21 proceeds to the next step S503.

[0128] (Step S503) In step S503, the control unit 21 operates as an output processing unit 213 and outputs the results of the machine learning of the conversion model 4.

[0129] In one example, the control unit 21 may generate transformed model data 400 as training result data showing the results of machine learning (trained machine learning model). The control unit 21 may store the generated transformed model data 400 in a predetermined memory area. The predetermined memory area may be, for example, RAM within the control unit 21, memory unit 22, external storage device, storage media, or a combination thereof. The storage media may be, for example, a CD, DVD, semiconductor memory, etc. The external storage device may be, for example, a data server such as a NAS. The external storage device may be, for example, an external storage device. If machine learning of the transformed model 4 is performed on an external computer, the transformed model data 400 may be generated on the external computer. In another example, the control unit 21 may output the convergence results obtained during the machine learning calculation process. The output destination may be, for example, RAM within the control unit 21, memory unit 22, output device 25, external computer, external storage device, storage media, or a combination thereof.

[0130] Once the output of the machine learning results is complete, the control unit 21 terminates the processing procedure for generating the conversion model 4 related to this example of operation.

[0131] Furthermore, if the calculation to derive the true value 75 using the conversion model 4 is performed on an external computer, the generated conversion model data 400 may be provided to the external computer at any timing and in any way. For example, the control unit 21 may send the conversion model data 400 to the external computer as an output process of step S503 or separately from the output process of step S503. The external computer may acquire the conversion model data 400 (conversion model 4) by receiving it. For example, the external computer may acquire the conversion model data 400 by accessing the model generation device 2 or an external storage device via a network. For example, the external computer may acquire the conversion model data 400 via a storage medium. Alternatively, for example, the conversion model data 400 may be pre-loaded into the external computer.

[0132] Furthermore, the control unit 21 may update or generate a new transformation model 4 (transformation model data 400) by repeatedly executing the series of processes in steps S501 to S503 periodically or irregularly. During this repetition, at least a portion of the training dataset D2 used for machine learning may be modified, corrected, added, deleted, etc., as appropriate. As a result, the control unit 21 may update the transformation model 4 used to derive the true value 75.

[0133] [Generating long-term prediction models] Figure 12 is a flowchart showing an example of the processing procedure for generating a long-term forecast model 6 by the model generation device 2 according to this embodiment. The following processing procedure is an example of an information processing method (model generation method for the long-term forecast model 6) executed by a computer. The following processing procedure is merely an example, and each step may be modified as much as possible. Furthermore, depending on the embodiment, steps in the following processing procedure can be omitted, replaced, and added as appropriate.

[0134] (Step S521) In step S521, the control unit 21 operates as a data acquisition unit 211 and acquires multiple training datasets D1. Each training dataset D1 may consist of a combination of the training sample 70 of the sample user S1 and the true value 75 of the second degree.

[0135] In one example, for at least a portion of multiple training datasets D1, the true value 75 may be derived from measured values ​​45 of the third degree measured over a shorter period rather than a longer period, by using a transformation model 4. For example, the control unit 21 may appropriately acquire measured values ​​45 of the third degree corresponding to the training sample 70. The control unit 21 may provide the acquired measured values ​​45 to the transformation model 4 and execute the calculation process of the transformation model 4. As a result of this calculation process, the control unit 21 may obtain the true value 75. Alternatively, for example, the calculation process of the transformation model 4 may be executed on an external computer. The control unit 21 may obtain the derived true value 75 directly or indirectly from the external computer. When deriving the true value 75, the transformation model 4 may be further provided with at least one of the training sample 70 and attribute information 701 in addition to the measured values ​​45 of the third degree. Also, in one example, each training dataset D1 may further include attribute information 701. Once multiple training datasets D1 are acquired, the control unit 21 proceeds to the next step S522.

[0136] (Step S522) In step S522, the control unit 21 operates as a learning processing unit 212 and controls the machine learning of the long-term prediction model 6. The machine learning may include training the long-term prediction model 6 by providing the training samples 70 to the long-term prediction model 6 so that the predicted value of the degree (second degree) derived by the long-term prediction model 6 approaches the true value 75 corresponding to the training samples 70.

[0137] In one example, the control unit 21 may control the machine learning process so that the long-term prediction model 6 is trained so that the second degree prediction value derived by the long-term prediction model 6 approaches the corresponding true value 75 when the training sample 70 of each acquired training dataset D1 is provided to the long-term prediction model 6. When the machine learning calculations for the long-term prediction model 6 are performed on the model generation device 2, each acquired training dataset D1 may be used as appropriate for the machine learning of the long-term prediction model 6. When the machine learning calculations for the long-term prediction model 6 are performed on an external computer, the control unit 21 may provide each acquired training dataset D1 to the external computer and have the external computer perform the machine learning of the long-term prediction model 6. If each training dataset D1 further includes attribute information 701, the long-term prediction model 6 may be trained so that the second degree prediction value derived by providing the training sample 70 and attribute information 701 fits the corresponding true value 75. As a result of performing the machine learning in step S522, a long-term prediction model 6 (trained machine learning model) can be generated that has acquired the ability to predict the degree to which purchasing behavior will be triggered over a long period of time from the user's behavior history. Once the machine learning of the long-term prediction model 6 is performed, the control unit 21 proceeds to the next step S523.

[0138] (Step S523) In step S523, the control unit 21 operates as an output processing unit 213 and outputs the machine learning results of the long-term prediction model 6.

[0139] In one example, the control unit 21 may generate long-term prediction model data 600 as training result data showing the results of machine learning (trained machine learning model). The control unit 21 may store the generated long-term prediction model data 600 in a predetermined memory area. The predetermined memory area may be, for example, RAM within the control unit 21, memory unit 22, an external storage device, a storage medium, or a combination thereof. If the machine learning of the long-term prediction model 6 is performed on an external computer, the long-term prediction model data 600 may be generated on the external computer. In another example, the control unit 21 may output the convergence results obtained during the calculation process of the machine learning of the long-term prediction model 6. The output destination may be, for example, RAM within the control unit 21, memory unit 22, output device 25, an external computer, an external storage device, a storage medium, or a combination thereof.

[0140] Once the output of the machine learning results is complete, the control unit 21 terminates the processing procedure for generating the long-term prediction model 6 related to this example of operation.

[0141] Furthermore, when the calculation process for predicting the second degree 32 is performed on the information processing device 1, the generated long-term prediction model data 600 may be provided to the information processing device 1 at any timing and in any way. For example, the control unit 21 may transmit the long-term prediction model data 600 to the information processing device 1 as an output process of step S523 or separately from the output process of step S523. The information processing device 1 may acquire the long-term prediction model data 600 (long-term prediction model 6) by receiving it. For example, the information processing device 1 may acquire the long-term prediction model data 600 by accessing the model generation device 2 or an external storage device via a network. For example, the information processing device 1 may acquire the long-term prediction model data 600 via the storage medium 91. Alternatively, for example, the long-term prediction model data 600 may be pre-loaded into the information processing device 1. Similarly, when the calculation process for predicting the second degree 32 is performed on an external computer, the generated long-term prediction model data 600 may be provided to the external computer at any timing and in any way.

[0142] Furthermore, the control unit 21 may update or generate a new long-term prediction model 6 (long-term prediction model data 600) by repeatedly executing the series of processes in steps S521 to S523 periodically or irregularly. During this repetition, at least a portion of the training dataset D1 used for machine learning may be changed, modified, added, deleted, etc., as appropriate. The updated or newly generated long-term prediction model 6 (long-term prediction model data 600) may be provided as appropriate to a computer (information processing device 1, external computer) that performs calculation processing to predict the second degree 32. As a result, the control unit 21 may update the long-term prediction model 6 used to predict the second degree 32.

[0143] [Generating a rapid-response predictive model] Figure 13 is a flowchart showing an example of the processing procedure for generating the rapid prediction model 5 by the model generation device 2 according to this embodiment. The following processing procedure is an example of an information processing method (model generation method for the rapid prediction model 5) executed by a computer. The following processing procedure is merely an example, and each step may be modified as much as possible. Furthermore, depending on the embodiment, steps in the following processing procedure can be omitted, replaced, and added as appropriate.

[0144] (Step S541) In step S541, the control unit 21 operates as a data acquisition unit 211 and acquires multiple training datasets D3. Each training dataset D3 may consist of a combination of a training sample 73 of a sample user S3 and a true value 77 of the first degree. In one example, each training dataset D3 may further include attribute information 731. Once multiple training datasets D3 have been acquired, the control unit 21 proceeds to the next step S542.

[0145] (Step S542) In step S542, the control unit 21 operates as a learning processing unit 212 and controls the machine learning of the rapid prediction model 5. The machine learning may include training the rapid prediction model 5 by providing the training sample 73 to the rapid prediction model 5 so that the predicted value of the degree (first degree) derived by the rapid prediction model 5 approaches the true value 77 corresponding to the training sample 73.

[0146] In one example, the control unit 21 may control the machine learning process so that the rapid prediction model 5 is trained to approach the corresponding true value 77 when the first degree of prediction derived by the rapid prediction model 5 is provided with the training sample 73 for each acquired training dataset D3. When the machine learning calculations for the rapid prediction model 5 are performed on the model generation device 2, each acquired training dataset D3 may be used as appropriate for the machine learning of the rapid prediction model 5. When the machine learning calculations for the rapid prediction model 5 are performed on an external computer, the control unit 21 may provide each acquired training dataset D3 to the external computer and have the external computer perform the machine learning of the rapid prediction model 5. If each training dataset D3 further includes attribute information 731, the rapid prediction model 5 may be trained so that the first degree of prediction derived by providing the training sample 73 and attribute information 731 fits the corresponding true value 77. As a result of performing the machine learning in step S542, it is possible to generate an immediate prediction model 5 (trained machine learning model) that has acquired the ability to predict the degree to which purchasing behavior will be triggered in an immediate period based on the user's behavior history. Once the machine learning of the immediate prediction model 5 is performed, the control unit 21 proceeds to the next step S543.

[0147] (Step S543) In step S543, the control unit 21 operates as an output processing unit 213 and outputs the machine learning results of the rapid prediction model 5.

[0148] In one example, the control unit 21 may generate responsive prediction model data 500 as learning result data showing the results of machine learning (trained machine learning model). The control unit 21 may store the generated responsive prediction model data 500 in a predetermined memory area. The predetermined memory area may be, for example, RAM within the control unit 21, memory unit 22, an external storage device, a storage medium, or a combination thereof. If the machine learning of the responsive prediction model 5 is performed on an external computer, the responsive prediction model data 500 may be generated on the external computer. In another example, the control unit 21 may output the convergence results obtained during the calculation process of the machine learning of the responsive prediction model 5. The output destination may be, for example, RAM within the control unit 21, memory unit 22, output device 25, an external computer, an external storage device, a storage medium, or a combination thereof.

[0149] Once the output of the machine learning results is complete, the control unit 21 terminates the processing procedure for generating the rapid prediction model 5 related to this example of operation.

[0150] Furthermore, when the calculation process for predicting the first degree 31 is performed on the information processing device 1, the generated immediate prediction model data 500 may be provided to the information processing device 1 at any timing and in any way. For example, the control unit 21 may transmit the immediate prediction model data 500 to the information processing device 1 as an output process of step S543 or separately from the output process of step S543. The information processing device 1 may acquire the immediate prediction model data 500 (immediate prediction model 5) by receiving it. For example, the information processing device 1 may acquire the immediate prediction model data 500 by accessing the model generation device 2 or an external storage device via a network. For example, the information processing device 1 may acquire the immediate prediction model data 500 via the storage medium 91. Also, for example, the immediate prediction model data 500 may be pre-loaded into the information processing device 1. Similarly, when the calculation process for predicting the first degree 31 is performed on an external computer, the generated immediate prediction model data 500 may be provided to the external computer at any timing and in any way.

[0151] Furthermore, the control unit 21 may update or generate a new rapid prediction model 5 (rapid prediction model data 500) by repeatedly executing the series of processes in steps S541 to S543 periodically or irregularly. During this repetition, at least a portion of the training dataset D3 used for machine learning may be changed, modified, added, deleted, etc., as appropriate. The updated or newly generated rapid prediction model 5 (rapid prediction model data 500) may be provided as appropriate to a computer (information processing device 1, external computer) that performs calculation processing to predict the first degree 31. As a result, the control unit 21 may update the rapid prediction model 5 used to predict the first degree 31.

[0152] [Granting of incentives] Figure 14 is a flowchart showing an example of a processing procedure for providing incentives by the information processing device 1 according to this embodiment. The following processing procedure is an example of an information processing method executed by a computer. The following processing procedure is merely an example, and each step may be modified as much as possible. Furthermore, depending on the embodiment, steps in the following processing procedure can be omitted, replaced, and added as appropriate.

[0153] (Step S101) In step S101, the control unit 11 operates as a sample acquisition unit 111 and acquires target samples 30 of the behavioral history related to the purchasing behavior of the target user T1. In one example, the control unit 11 may also acquire attribute information 301 of the target user T1. Once the target samples 30 are acquired, the control unit 11 proceeds to the next step S102.

[0154] (Step S102) In step S102, the control unit 11 operates as a first prediction unit 112 and provides the acquired target sample 30 to the immediate prediction model 5 to predict a first degree 31 that will induce purchasing behavior in an immediate period by providing an incentive.

[0155] In one example, the control unit 11 may provide the acquired target sample 30 to the rapid prediction model 5 and execute the calculation process of the rapid prediction model 5. As a result of this calculation process, the control unit 11 may obtain the result of predicting the first degree 31 from the rapid prediction model 5. In another example, the calculation process of the rapid prediction model 5 may be performed by an external computer. The control unit 11 may provide the target sample 30 to the external computer and have the calculation process of the rapid prediction model 5 executed by the external computer. The control unit 11 may obtain the calculation result of the rapid prediction model 5 (the result of predicting the first degree 31) directly or indirectly from the external computer. Also, in one example, if attribute information 301 has been acquired, the control unit 11 may use the rapid prediction model 5 to predict the first degree 31 by providing the target sample 30 and attribute information 301 to the rapid prediction model 5. Once the first degree 31 is predicted, the control unit 11 proceeds to the next step S103.

[0156] (Step S103) In step S103, the control unit 11 operates as a second prediction unit 113 and provides the acquired target sample 30 to the long-term prediction model 6 to predict a second degree 32 that will induce purchasing behavior over a long period of time by providing incentives.

[0157] In one example, the control unit 11 may provide the acquired target sample 30 to the long-term prediction model 6 and execute the calculation process of the long-term prediction model 6. As a result of this calculation process, the control unit 11 may obtain the result of predicting the second degree 32 from the long-term prediction model 6. In another example, the calculation process of the long-term prediction model 6 may be performed by an external computer. The control unit 11 may provide the target sample 30 to the external computer and have the external computer execute the calculation process of the long-term prediction model 6. The control unit 11 performs the calculation of the long-term prediction model 6. The result (the result of predicting the second degree 32) may be obtained directly or indirectly from an external computer. Also, in one example, if attribute information 301 has been obtained, the control unit 11 may use the long-term prediction model 6 to predict the first degree 31 by providing the target sample 30 and attribute information 301 to the long-term prediction model 6. Once the second degree 32 is predicted, the control unit 11 proceeds to the next step S104.

[0158] Note that the processing order of step S103 is not limited to this example and may be changed as appropriate depending on the embodiment. In another example, the processing of step S103 may be performed before the processing of step S102. The processing of step S103 may be performed at least partially in parallel with the processing of step S102.

[0159] (Step S104) In step S104, the control unit 11 decides whether or not to grant an incentive to the target user T1 based on the predicted first degree 31 and second degree 32. In one example, the control unit 11 may compare each predicted degree (31, 32) with a threshold and decide whether or not to grant an incentive depending on the result of the comparison. Once a decision has been made on whether or not to grant an incentive, the control unit 11 proceeds to the next step S105.

[0160] (Step S105) In step S105, the control unit 11 operates as an output processing unit 116 and determines the branch destination of the process according to the decision result in step S104. If it is decided to grant an incentive to target user T1, the control unit 11 proceeds to the next step S106. On the other hand, if it is decided not to grant an incentive to target user T1, the control unit 11 omits the processing in the next steps S106 and S107 and terminates the processing procedure related to granting incentives in this example.

[0161] (Step S106) In step S106, the control unit 11 operates as a quantity determination unit 115 and determines the amount of incentive 33 according to the predicted first degree 31 and second degree 32.

[0162] In one example, the control unit 11 may determine the amount of incentive 33 such that the higher the predicted second degree 32, the greater the amount of incentive 33. Alternatively, in another example, the control unit 11 may determine the amount of incentive 33 such that the lower the second degree 32 and the higher the first degree 31, the smaller the amount of incentive 33. Once the amount of incentive 33 is determined, the control unit 11 proceeds to the next step S107.

[0163] Note that the processing order in step S106 is not limited to this example and may be changed as appropriate depending on the embodiment. In another example, the processing in step S106 may be executed at least partially in parallel with the processing in step S104.

[0164] (Step S107) In step S107, the control unit 11 operates as an output processing unit 116 and outputs incentive information 35 regarding the provision of incentives to the target user T1.

[0165] In one example, the control unit 11 may output incentive information 35 regarding the granting of an incentive of a determined amount 33. In another example, the control unit 11 may send the incentive information 35 indicating the incentive to the destination of the target user T1 (the target user T1's terminal, account, etc.). In yet another example, the control unit 11 may provide the incentive information 35, which includes a command instructing the granting of the incentive, to another computer. This allows the control unit 11 to have the other computer issue the incentive to the target user T1.

[0166] Once the output of incentive information 35 is complete, the control unit 11 terminates the processing procedure for granting incentives related to this example of operation. The control unit 11 may execute the series of processes from steps S101 to S107 at any time. The control unit 11 may also repeatedly execute the series of processes from steps S101 to S107 periodically or irregularly.

[0167] (Features) In this embodiment, the processing in steps S102 and S103 predicts not only the degree to which immediate purchasing behavior is triggered (first degree 31) but also the degree to which long-term purchasing behavior is triggered (second degree 32). In the processing in step S104, not only the predicted first degree 31 but also the predicted second degree 32 is used as an indicator to decide whether or not to provide an incentive. This makes it easier to provide incentives to users who have a relatively low degree to which immediate purchasing behavior is triggered (first degree 31) but a high degree to which long-term purchasing behavior is triggered (second degree 32). In other words, it becomes easier to appeal not only to type A users in Figure 2 but also to type B users. Therefore, according to this embodiment, it is possible to expect the provision of effective incentives.

[0168] §4 Variant While embodiments of this disclosure have been described in detail above, the above description is merely illustrative in all respects. The processes and means described in this disclosure can be freely combined and implemented as long as no technical inconsistencies arise. Furthermore, various improvements or modifications may be made to the above embodiments as appropriate. For example, the following modifications are possible. In the following, the same reference numerals are used for components similar to those in the above embodiments, and explanations of points similar to those in the above embodiments have been omitted as appropriate. The following modifications can be combined as appropriate.

[0169] <4.1> For example, in the examples shown in Figures 4 and 12 above, the model generation device 2 generates the long-term forecast model 6. However, the computer that generates the long-term forecast model 6 is not limited to the model generation device 2. The long-term forecast model 6 may be generated by a computer other than the model generation device 2. If the long-term forecast model 6 is not generated by the model generation device 2, the components related to the generation of the long-term forecast model 6 may be omitted from the model generation device 2.

[0170] In the examples shown in Figures 4, 5, and 11 above, the model generation device 2 generates the transformation model 4 and the long-term forecast model 6. However, the computer that generates the transformation model 4 is not limited to the model generation device 2. The transformation model 4 may be generated by a different computer than the one that generates the long-term forecast model 6. The transformation model 4 may be generated by a computer other than the model generation device 2. If the transformation model 4 is not generated by the model generation device 2, the components related to the generation of the transformation model 4 may be omitted from the model generation device 2.

[0171] In the example shown in Figures 4-6 and Figure 13 above, the model generation device 2 generates the transformation model 4, the immediate prediction model 5, and the long-term prediction model 6. However, the computer that generates the immediate prediction model 5 is not limited to the model generation device 2. The immediate prediction model 5 may be generated by a different computer than the one that generates the transformation model 4. The immediate prediction model 5 may be generated by a different computer than the one that generates the long-term prediction model 6. The immediate prediction model 5 may be generated by a computer other than the model generation device 2. If the immediate prediction model 5 is not generated by the model generation device 2, the components related to the generation of the immediate prediction model 5 may be omitted from the model generation device 2.

[0172] <4.2> With respect to each processing step of the model generation device 2 according to the above embodiment, at least one of the following may be omitted, replaced, or added. For example, if the model generation device 2 does not need to perform the process of acquiring multiple training datasets D2, such as having a computer that performs machine learning calculations acquire the training dataset D2, the processing in step S501 may be omitted from the processing steps related to the generation of the transformed model 4. Similarly, if the model generation device 2 does not need to perform the process of acquiring multiple training datasets D1, the processing in step S521 may be omitted from the processing steps related to the generation of the long-term prediction model 6. If the model generation device 2 does not need to perform the process of acquiring multiple training datasets D3, the processing in step S541 may be omitted from the processing steps related to the generation of the immediate prediction model 5. If the processing of acquiring each training dataset (D1, D2, D3) in each step (S501, S521, S541) is omitted, the data acquisition unit 211 may be omitted from the software configuration of the model generation device 2.

[0173] If the model generation device 2 does not generate the transformation model 4, the processes in steps S501 to S503 may be omitted from the processing procedures executed by the model generation device 2. If the model generation device 2 does not generate the long-term forecast model 6, the processes in steps S521 to S523 may be omitted from the processing procedures executed by the model generation device 2. If the model generation device 2 does not generate the immediate forecast model 5, the processes in steps S541 to S543 may be omitted from the processing procedures executed by the model generation device 2.

[0174] Furthermore, the processing procedure of the information processing device 1 according to the above embodiment may be modified by omitting, substituting, or adding steps. For example, the processing in step S106 may be omitted from the processing procedure of the information processing device 1. If the processing to determine the amount of incentive 33 in step S106 is omitted, the amount determination unit 115 may be omitted from the software configuration of the information processing device 1.

[0175] §5 Experimental Examples To verify that the degree of purchasing behavior over a long period can be predicted from the degree of purchasing behavior over a short period, the following experiment was conducted. However, this disclosure is not limited to the following experimental example.

[0176] As an indicator of the degree to which purchasing behavior is stimulated over a short period, the purchase amount over a 4-week period was adopted. As an indicator of the degree to which purchasing behavior is stimulated over a long period, the purchase amount over a 24-week period was adopted. Coupon distribution was adopted as an incentive. For retail store customers (4380 people), the first purchase data for 24 weeks before coupon distribution and the second purchase data for 24 weeks after coupon distribution were extracted from ID-POS data. A trained machine learning model (regression model) was generated using the first purchase data. In machine learning, the measured purchase amount for the first 4 weeks of the first purchase data was adopted as the training value, and the measured purchase amount for 24 weeks of the first purchase data was adopted as the true value. LightGBM (Light Gradient Boosting Machine) was used for machine learning. In the experimental example, for each customer, the generated trained model Using a machine learning model, we predicted the purchase amount over 24 weeks from the actual purchase amount during the first four weeks of the second purchase data. The difference (absolute value) between the actual purchase amount (true value) and the predicted value by the trained machine learning model was evaluated as the error. In contrast, in the comparative example, the actual purchase amount during the first four weeks of the second purchase data was directly used as the predicted purchase amount over 24 weeks. The difference (absolute value) between the actual purchase amount (true value) and the predicted value for the 24-week purchase data was evaluated as the error. The results showed that the experimental example reduced the error by approximately 31.7% compared to the comparative example. From these results, we can verify that it is possible to predict the degree of purchase behavior in a long-term period from the degree of purchase behavior in a short-term period, that is, that the above transformation model can be generated. And it was done. [Explanation of Symbols]

[0177] 1… Information processing equipment, 11...Control unit, 12...Storage unit, 81...Program, 2...Model generation device, 21...Control unit, 22...Storage unit, 82...Model generation program, 4...Transformation model, 5...Rapid prediction model, 6...Long-term prediction model, T1...Target user, 30...Target sample, 31...1st degree, 32...2nd degree, 33...Quantity, 35...Incentive information, S1, S2... Sample users, 70...Training sample, 75...True value, 45...Measured value 71...Training value, 76...True value

Claims

1. A model generation method performed by a computer, Controlling machine learning for long-term prediction models, and Outputting the results of the aforementioned machine learning Includes, The aforementioned long-term prediction model is configured to predict the degree to which purchasing behavior will be triggered over a longer period than the immediate period by providing incentives based on the user's behavioral history related to purchasing behavior. With respect to the training sample of the sample user's behavioral history, the true value of the degree to which purchasing behavior is induced over the long period is derived from the measured value of the degree to which purchasing behavior is induced by providing the incentive, using a transformation model, and is obtained from measured values ​​measured over a shorter period than the long period. The machine learning includes training the long-term prediction model by providing the training samples to the long-term prediction model so that the predicted value of the degree derived by the long-term prediction model approaches the true value corresponding to the training samples. Model generation method.

2. Controlling machine learning for long-term prediction models, and Outputting the results of the aforementioned machine learning A control unit configured to perform the following: The aforementioned long-term prediction model is configured to predict the degree to which purchasing behavior will be triggered over a longer period than the immediate period by providing incentives based on the user's behavioral history related to purchasing behavior. With respect to the training sample of the sample user's behavioral history, the true value of the degree to which purchasing behavior is induced over the long period is derived from the measured value of the degree to which purchasing behavior is induced by providing the incentive, using a transformation model, and is obtained from measured values ​​measured over a shorter period than the long period. The machine learning includes training the long-term prediction model by providing the training samples to the long-term prediction model so that the predicted value of the degree derived by the long-term prediction model approaches the true value corresponding to the training samples. Model generation device.

Images