JP7589391B2 - Information processing device, determination method, and determination program - Google Patents

Description

The present disclosure relates to an information processing device, a decision method, and a decision program.

Measures are being taken to increase revenue. For example, coupons are provided to users. The users then visit a store and purchase many items using the coupons. This increases revenue. However, if a coupon is provided to a user who is planning to visit the store, it simply results in a price reduction, and the effect of the measure is low. Therefore, a technology has been proposed for granting incentives to users (see Patent Document 1). The information processing device of Patent Document 1 grants incentives to users based on the behavioral history and background information of multiple users.

Patent No. 6899350

However, inequality may arise depending on the policy. Inequality in policies is thought to lead to distrust in the services.

The purpose of this disclosure is to eliminate inequalities in policies.

An information processing device according to an aspect of the present disclosure is provided. The information processing device includes an acquisition unit that acquires a trained model, behavioral feature information indicating behavioral features of each user, attribute information indicating attributes of each user, policy candidate information indicating candidates for policies such as providing coupons or novelties to each user , a policy score that is a value for alleviating inequality in policies, and budget information indicating a budget, a generation unit that generates data based on the behavioral feature information, the attribute information, and the policy candidate information, a prediction unit that predicts sales or an increase in the number of store visits when a policy is implemented based on the generated data and the trained model, and a determination unit that performs optimization calculation within the budget using the prediction result that is a result of the prediction and the policy score to determine a policy for each user.

This disclosure will help eliminate inequalities in policies.

FIG. 1 illustrates a communication system. FIG. 2 is a diagram illustrating hardware included in an information processing device. FIG. 2 is a block diagram showing functions of the information processing device. FIG. 13 is a diagram illustrating an example of a behavior history table. FIG. 13 is a diagram illustrating an example of a behavioral characteristic table. FIG. 13 is a diagram illustrating an example of an attribute table. FIG. 13 is a diagram illustrating an example of a policy result table. FIG. 11 is a diagram illustrating an example of learning data. 13 is a flowchart illustrating an example of a process executed by a learning unit. FIG. 13 is a diagram illustrating an example of a policy candidate table. FIG. 11 is a diagram illustrating an example of generated data. FIG. 13 is a diagram illustrating an example of a prediction result. 13A and 13B are diagrams (part 1) showing an example of a method for calculating a policy score. FIG. 2 is a diagram (part 2) showing an example of a method for calculating a policy score. FIG. 11 is a diagram illustrating a specific example of a process executed by a determination unit. FIG. 13 is a diagram showing an example of a measure for each user. 11 is a flowchart illustrating an example of a process executed by an information processing device.

The following describes an embodiment with reference to the drawings. The following embodiment is merely an example, and various modifications are possible within the scope of this disclosure.

Embodiment
1 is a diagram showing a communication system. The communication system includes an information processing device 100 and a terminal device 200. The information processing device 100 and the terminal device 200 communicate with each other via a network.
The information processing device 100 is a device that executes the determination method. For example, the information processing device 100 is a server. The information processing device 100 may also be a personal computer, a smartphone, a tablet terminal, or the like.
The terminal device 200 is a device used by a user. For example, the terminal device 200 is a smartphone, a tablet terminal, etc. One terminal device is illustrated in Fig. 1. The number of terminal devices may be two or more.

Next, the hardware of the information processing device 100 will be described.
2 is a diagram showing hardware included in the information processing apparatus 100. The information processing apparatus 100 includes a processor 101, a volatile storage device 102, a non-volatile storage device 103, and a communication interface 104.

The processor 101 controls the entire information processing device 100. For example, the processor 101 is a CPU (Central Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), etc. The processor 101 may be a multiprocessor. The information processing device 100 may also have a processing circuit. Furthermore, the information processing device 100 may have a microcomputer or a SoC (System on Chip).

The volatile storage device 102 is a main storage device of the information processing device 100. For example, the volatile storage device 102 is a random access memory (RAM). The non-volatile storage device 103 is an auxiliary storage device of the information processing device 100. For example, the non-volatile storage device 103 is a read only memory (ROM), a hard disk drive (HDD), or a solid state drive (SSD).
The communication interface 104 communicates with the terminal device 200 .

Next, functions of the information processing device 100 will be described.
3 is a block diagram showing the functions of the information processing device 100. The information processing device 100 includes a storage unit 110, a learning unit 120, an acquisition unit 130, a generation unit 140, a prediction unit 150, a determination unit 160, and an output unit 170.

The storage unit 110 may be realized as a storage area secured in the volatile storage device 102 or the non-volatile storage device 103 .
A part or all of the learning unit 120, the acquisition unit 130, the generation unit 140, the prediction unit 150, the determination unit 160, and the output unit 170 may be realized by a processing circuit. Also, a part or all of the learning unit 120, the acquisition unit 130, the generation unit 140, the prediction unit 150, the determination unit 160, and the output unit 170 may be realized as a module of a program executed by the processor 101. For example, the program executed by the processor 101 is also referred to as a determination program. For example, the determination program is recorded on a recording medium.

The memory unit 110 may store a behavior history table 111, a behavior feature table 112, an attribute table 113, a policy result table 114, a trained model 115, and a policy candidate table 116. The behavior history table 111, the behavior feature table 112, the attribute table 113, the policy result table 114, the trained model 115, and the policy candidate table 116 will be described later.

<Learning Phase>
The learning unit 120 generates the learned model 115. The function of the learning unit 120 will be described in detail.
The learning unit 120 acquires the behavior history table 111. The behavior history table 111 is shown below.

Figure 4 is a diagram showing an example of a behavior history table. The behavior history table 111 shows the behavior history of a user. The behavior history table 111 has items such as a user ID (identifier), date and time, and area of stay. The behavior history table 111 may also include GPS (Global Positioning System) data and ticket gate entry and exit history.

The learning unit 120 extracts behavioral characteristics of the user based on the behavior history table 111. The learning unit 120 registers the behavioral characteristics in the behavioral characteristics table 112. The behavioral characteristics table 112 is shown below.

5 is a diagram showing an example of a behavioral characteristic table 112. The behavioral characteristic table 112 shows behavioral characteristics for each user. The behavioral characteristic table 112 has fields for user ID, average store visit frequency, and average store visit time.
The learning unit 120 may extract, as the behavioral characteristics, facilities that the user frequently uses and times when the user frequently travels. The learning unit 120 may also extract, as the behavioral characteristics, behavioral patterns on holidays and weekdays.

The learning unit 120 acquires the attribute table 113. The attribute table 113 is shown.

Figure 6 is a diagram showing an example of an attribute table. The attribute table 113 shows attributes for each user. The attribute table 113 has items such as user ID, age, gender, and address.

The learning unit 120 acquires the policy result table 114. The policy result table 114 is shown below.

Figure 7 is a diagram showing an example of a policy result table. The policy result table 114 shows the results of policies implemented in the past. For example, the policy result table 114 shows that a user with user ID "00001" used a 100 yen coupon at Store A, and paid 1,500 yen to Store A. The 1,500 yen may also be expressed as sales for Store A. The policy result table 114 may also include an increase in the number of visits.

The learning unit 120 generates learning data based on the behavioral characteristics table 112, the attribute table 113, and the policy result table 114. The learning data is shown below.

8 is a diagram showing an example of training data 300. Training data 300 is data generated by the training unit 120.
The learning unit 120 generates the trained model 115 by using the training data 300. A multiple regression analysis may be used in generating the trained model 115. For example, the learning unit 120 performs the multiple regression analysis by using the formula (1).

Note that y is the objective variable. The values based on the behavioral characteristics table 112 and the attribute table 113 are x1 to xi. The values based on the policy result table 114 are z1 to zj. T is policy presence/absence information. α1 to αi and β1 to βj are correction coefficients. γ is a constant term. The policy effect may also be calculated using equation (2).

Note that a policy is in place when "1" is entered into T. A policy is not in place when "0" is entered into T.

A known method may be used to generate the trained model 115. For example, a meta-learner method such as a support vector machine (SVM), a gradient boosting decision tree (GBDT), S-Leaner, or T-Leaner, or a causal tree method such as a causal tree or causal forest may be used to generate the trained model 115.

In this manner, the trained model 115 is generated. The trained model 115 is a model that predicts an increase in sales or number of visits to the store. The learning unit 120 stores the trained model 115 in an external device that can be connected to the memory unit 110 or the information processing device 100. Note that a diagram of the external device is omitted.

Next, the process executed by the learning unit 120 will be described with reference to a flowchart.
FIG. 9 is a flowchart illustrating an example of a process executed by the learning unit.
(Step S<b>11 ) The learning unit 120 extracts behavioral features of the user based on the behavior history table 111 .
(Step S<b>12 ) The learning unit 120 acquires the behavioral features table 112 , the attribute table 113 , and the policy result table 114 from the storage unit 110 .
(Step S13) The learning unit 120 generates learning data based on the behavioral features table 112, the attribute table 113, and the policy result table 114.
(Step S14) The learning unit 120 generates a trained model 115 using the training data.
(Step S15) The learning unit 120 stores the learned model 115 in the memory unit 110.

In the above explanation, the case where the information processing device 100 generates the trained model 115 has been described. The trained model 115 may be generated by a learning device other than the information processing device 100.

<Utilization phase>
Returning to FIG. 3, the functions of the acquisition unit 130 and the like will be described.
The acquisition unit 130 acquires the trained model 115. For example, the acquisition unit 130 acquires the trained model 115 from the storage unit 110 or an external device.

The acquiring unit 130 acquires the behavior characteristic table 112 and the attribute table 113. For example, the acquiring unit 130 acquires the behavior characteristic table 112 and the attribute table 113 from the storage unit 110 or an external device. Here, the behavior characteristic table 112 is also referred to as behavior characteristic information, and the attribute table 113 is also referred to as attribute information.
Furthermore, the acquiring unit 130 acquires the policy candidate table 116. For example, the acquiring unit 130 acquires the policy candidate table 116 from the storage unit 110 or an external device. The policy candidate table 116 is shown below.

10 is a diagram showing an example of a policy candidate table. The policy candidate table 116 shows policy candidates. The policy candidate table 116 may be expressed as information showing policy candidates to be implemented on a plurality of users. The policy candidate table 116 is also referred to as policy candidate information.
10 shows nine policies. For example, policy No. 1 indicates that a 100 yen coupon for store A is provided. For example, policy No. 2 indicates that a 100 yen coupon for store B is provided.

The generation unit 140 generates data based on the behavioral characteristics table 112, the attribute table 113, and the policy candidate table 116. The data is input to the trained model 115. The data to be generated is shown.

Figure 11 is a diagram showing an example of generated data. The generation unit 140 generates nine pieces of data based on the behavioral characteristics, attributes, and nine measures of the user ID "00001." The generation unit 140 similarly generates data corresponding to all users such as the user ID "00002."

The prediction unit 150 predicts the increase in sales or the number of visits to the store when a measure is implemented based on the data generated by the generation unit 140 and the trained model 115. In detail, the prediction unit 150 inputs the data to the trained model 115, and the trained model 115 outputs the increase in sales or the number of visits to the store when a measure is implemented. The prediction result is shown.

Figure 12 is a diagram showing an example of a prediction result. The prediction result in Figure 12 indicates an increase in the number of visits to the store. For example, the prediction result "1.2" indicates that if a 100 yen coupon for Store A is provided to a user with user ID "00001", the number of visits by that user will increase to "1.2".

The acquisition unit 130 acquires a policy score, which is a value for mitigating inequality in a policy. For example, the acquisition unit 130 acquires the policy score from the storage unit 110 or an external device. Here, a method for calculating the policy score will be described.

13(A) and (B) are diagrams (part 1) showing an example of a method for calculating a policy score. FIG. 13(A) shows a policy score evaluating whether different users receive a policy on the previous day and today. "Coef ₁ " shown in the formula of FIG. 13(A) may be set by a user. Also, "Coef ₁ " may be set automatically. "0" or "1" is set to "i" in the formula of FIG. 13(A). "0" indicates that no policy is received. "1" indicates that a policy is received. When calculating a policy score evaluating whether different users receive a policy on the previous day and today, the previous day and plan 2 in the table of FIG. 13(A) are referenced. Then, the policy score is calculated using the inner product based on the previous day and plan 2 and "Coef ₁ ".

FIG. 13B shows a measure score that evaluates whether or not a specific measure is biased among multiple measures. The "Coef ₂ " shown in the formula in FIG. 13B may be set by the user. Also, "Coef ₂ " may be set automatically. A case in which there is no bias in a specific measure among multiple measures is the case of plan 2 in the table in FIG. 13B. For example, plan 2 indicates that coupon A is provided to 21 people, coupon B is provided to 20 people, and coupon C is provided to 19 people. In this way, plan 2 indicates a case in which coupons A to C are provided to many people. In other words, plan 2 indicates a case in which one coupon (e.g., coupon A) is not provided to many people. For example, this case is the case of plan 1. The measure score is calculated using the standard deviation based on plan 2 and "Coef ₂ ".

FIG. 14 is a diagram (part 2) showing an example of a method for calculating a policy score. FIG. 14 shows a policy score for evaluating implementation of a policy on different users. "Coef ₃ " shown in the formula in FIG. 14 may be set by the user. Also, "Coef ₃ " may be set automatically. The case where a policy is implemented on different users is the case of plan 2 in the table in FIG. 14. The policy score is calculated using the standard deviation based on plan 2 and "Coef ₃ ".
13 and 14 are merely examples. Therefore, the policy score may be calculated by a method other than the above.

In addition, the acquisition unit 130 acquires budget information indicating the budget. For example, the acquisition unit 130 acquires the budget information from the memory unit 110 or an external device.

The decision unit 160 uses the prediction results and the policy scores to perform optimization calculations within the budget and determine policies for each user. In detail, the decision unit 160 performs optimization calculations using formula (3). In addition, formula (4) is used as a constraint condition.

Here, formula (3) is also referred to as the objective function. The determination unit 160 performs optimization calculations using formula (3) in order to maximize the objective function. In addition, the optimization calculations may use optimization methods such as the greedy method, gradient descent, and parameter search methods such as Bayesian optimization.

The processing performed by the determination unit 160 is explained using a concrete example.

Figure 15 is a diagram showing a specific example of processing executed by the decision unit. First, the budget is set to 600 yen. The decision unit 160 considers policy A within the budget. Policy A indicates that a 300 yen coupon for store C is provided to a user with user ID "00001", a 100 yen coupon for store A is provided to a user with user ID "00002", and a 200 yen coupon for store A is provided to a user with user ID "00004". The decision unit 160 calculates the total value of the prediction results corresponding to policy A. The decision unit 160 calculates an evaluation value for policy A using the total value of the prediction results and the policy score.

The decision unit 160 considers measure B within the budget. Measure B indicates that a 200 yen coupon for store B will be provided to user with user ID "00001", a 100 yen coupon for store C will be provided to user with user ID "00002", and a 300 yen coupon for store C will be provided to user with user ID "00003". The decision unit 160 calculates the total value of the prediction results corresponding to measure B. The decision unit 160 calculates an evaluation value for measure B using the total value of the prediction results and the measure score.

The decision unit 160 compares the evaluation value for measure A with the evaluation value for measure B. Based on the comparison result, the decision unit 160 determines that measure A is optimal. The decision unit 160 repeats the same process to consider the optimal measure. Then, the decision unit 160 decides on the optimal measure as the measure for each user. Here, an example of a measure for each user is shown.

Fig. 16 is a diagram showing an example of a campaign for each user. The table in Fig. 16 shows the determined campaign for each user. For example, a 100 yen coupon for Store A is provided to a user with a user ID "00001" as a campaign.
Moreover, the table in FIG. 16 shows the predicted results and costs corresponding to the determined measures.

In the above, a case has been described in which the policy score acquired by the acquisition unit 130 is used. The decision unit 160 may calculate a policy score based on the policy selected within the budget, and use the calculated policy score. For example, the decision unit 160 may calculate a policy score based on policy A, and use the calculated policy score. Furthermore, if the calculated policy score is greater than the policy score acquired by the acquisition unit 130, the decision unit 160 may redo the policy decision.

The determination unit 160 may also perform optimization calculations using equation (5). Furthermore, equation (4) is used as a constraint.

The output unit 170 outputs measures for each user as policy information.

Next, the process executed by the information processing device 100 will be described with reference to a flowchart.
FIG. 17 is a flowchart illustrating an example of processing executed by the information processing device.
(Step S21) The acquisition unit 130 acquires the trained model 115.
(Step S22) The acquiring unit 130 acquires the behavioral characteristics table 112, the attribute table 113, and the policy candidate table 116.
(Step S23) The generation unit 140 generates data based on the behavioral characteristics table 112, the attribute table 113, and the policy candidate table 116.

(Step S24) The prediction unit 150 predicts the increase in sales or the number of visits to the store when the measure is implemented, based on the generated data and the trained model 115.
(Step S25) The acquisition unit 130 acquires the policy score.
(Step S26) The acquiring unit 130 acquires budget information.
(Step S27) The determination unit 160 performs optimization calculations within the budget using the prediction results and the policy scores, and determines a policy for each user.
(Step S28) The output unit 170 outputs the policy for each user as policy information.

Here, if the budget is infinite, the best coupon can be provided to all users. However, the budget is limited. Therefore, the information processing device 100 performs optimization calculations to maximize the effect of the measures within the budget. Furthermore, when the optimization calculations are performed normally, there is a possibility that the calculation result that the best coupon is provided to a specific user is output every time. For example, there is a possibility that the calculation result that a 300 yen coupon from Store A is provided to a user with user ID "00001" is output every time. In order to prevent such biased calculation results from being output, the information processing device 100 uses a policy score in the optimization calculation. And preventing biased calculation results from being output eliminates inequality in the policies. Therefore, according to the embodiment, the information processing device 100 can eliminate inequality in the policies.

In the above, a coupon is provided as a campaign. For example, a novelty item may be provided as a campaign.

100 Information processing device, 101 Processor, 102 Volatile storage device, 103 Non-volatile storage device, 104 Communication interface, 110 Storage unit, 111 Action history table, 112 Action feature table, 113 Attribute table, 114 Policy result table, 115 Learned model, 116 Policy candidate table, 120 Learning unit, 130 Acquisition unit, 140 Generation unit, 150 Prediction unit, 160 Decision unit, 170 Output unit, 200 Terminal device, 300 Learning data.

Claims (3)

An acquisition unit that acquires a trained model, behavioral feature information indicating behavioral features of each user, attribute information indicating attributes of each user, policy candidate information indicating candidates for policies such as providing coupons or novelties to each user, a policy score that is a value for mitigating inequality in policies, and budget information indicating a budget;
A generation unit that generates data based on the behavioral characteristic information, the attribute information, and the policy candidate information;
A prediction unit that predicts sales or an increase in the number of visits to the store when a measure is implemented based on the generated data and the trained model;
a decision unit that performs optimization calculation within the budget using the prediction result, which is a result of the prediction, and the policy score, and decides a policy for each user;
An information processing device having the above configuration. An information processing device,
Acquire a trained model, behavioral characteristic information indicating behavioral characteristics of each user, attribute information indicating attributes of each user, policy candidate information indicating candidates for policies such as providing coupons or novelties to each user, a policy score which is a value for mitigating inequality in policies, and budget information indicating a budget, generate data based on the behavioral characteristic information, the attribute information, and the policy candidate information, and predict sales or an increase in the number of store visits when a policy is implemented based on the generated data and the trained model,
performing an optimization calculation within the budget using the prediction result, which is a result of the prediction, and the action score, and determining an action for each user;
How to decide. In the information processing device,
Acquire a trained model, behavioral characteristic information indicating behavioral characteristics of each user, attribute information indicating attributes of each user, policy candidate information indicating candidates for policies such as providing coupons or novelties to each user, a policy score which is a value for mitigating inequality in policies, and budget information indicating a budget, generate data based on the behavioral characteristic information, the attribute information, and the policy candidate information, and predict sales or an increase in the number of store visits when a policy is implemented based on the generated data and the trained model,
performing an optimization calculation within the budget using the prediction result, which is a result of the prediction, and the action score, and determining an action for each user;
A decision program that causes processing to occur.

Images