JP7229031B2 - NOTIFICATION METHOD SELECTION DEVICE, NOTIFICATION METHOD SELECTION METHOD, AND PROGRAM - Google Patents

Description

The present invention relates to a notification method selection device, a notification method selection method, and a program.

2. Description of the Related Art With the development of computer and communication network technologies such as the Internet, the method of giving business notices to customers has changed from mailing printed materials to electronic notices. Electronic notification is notification using e-mail, messaging service, etc. via a communication network. The messaging service is, for example, a mechanism for transmitting and receiving contents such as messages using functions such as RCS (Rich Communication Services), SMS (Short Message Service), and MMS (Multimedia Messaging Service). RCS is a standard for implementing an instant messenger for mobile phones at the terminal level, and uses telephone numbers instead of user names and e-mail addresses when communicating with other parties. In addition, in RCS, the upper limit of the number of characters that can be transmitted is set larger than in conventional SMS and MMS. Japanese Patent Laid-Open No. 2002-200000 discloses a technique for sending an e-mail to an address that is highly likely to be seen by the customer immediately when notifying the customer.

JP 2007-241732 A

However, many companies offer similar services with electronic notifications such as e-mail, and customers may receive numerous notifications. With a large number of electronic notices, it can happen that the notices are left unopened by the customer. In this case, the customer cannot be made aware of the contents of the notification. As a countermeasure, it is conceivable to return to the method of notifying the customer by mailing printed matter. This is because, in general, when a physical printed matter is delivered, it is believed that the will to check the contents of the matter is likely to work. However, since printing and shipping costs are required to send printed materials by mail, returning to the method of notifying by printed materials has the problem of increasing costs compared to electronic notification.

SUMMARY OF THE INVENTION The present invention has been made in view of such circumstances, and its object is to make it easier for customers to recognize the content of the notification when notifying the customer, and to prevent an increase in the cost required for notification. To provide a notification method selection device, a notification method selection method, and a program capable of

In order to solve the above-described problems, the present invention uses a notification method estimation model to select a notification method when notifying customers of information, such as a printed notification, which is a printed notification, or an electronic notification, which is an electronic notification. An analysis unit that selects at least one from notifications, and a processing unit that outputs the result of selection by the analysis unit, and the notification method estimation model is a customer that indicates the attributes of existing customers who have a track record of past notifications. A machine learning model created by executing machine learning using learning data in which attribute information is associated with the existing customer's reaction to the notification given to the existing customer , and when the print notification is performed and the response rate and opening rate in the case of performing the message notification as the electronic notification, and the analysis unit estimates the response rate of the message notification is smaller than the response rate of the print notification, In the notification method selection device, the message notification is set as the notification method if the opening rate of the notification is equal to or greater than the threshold, and the print notification is set as the notification method if the opening rate of the message notification is less than the threshold.

Further, in order to solve the above-described problems, the present invention uses a notification response estimation model to determine the notification method when the analysis unit notifies the customer of the information by printed notification or electronic notification. At least one is selected from electronic notifications that are notifications by method, the processing unit outputs the result of the selection by the analysis unit, and the notification response estimation model indicates attributes of existing customers who have a track record of past notifications. A machine learning model created by executing machine learning using learning data in which customer attribute information is associated with the existing customer's reaction to the notification given to the existing customer , and the printing notification is performed. and the response rate and the opening rate in the case of performing the message notification as the electronic notification, and the analysis unit estimates the response rate of the message notification when the response rate of the message notification is smaller than the response rate of the print notification. The notification method selection method selects the message notification as the notification method if the opening rate of the message notification is equal to or higher than the threshold, and sets the printing notification as the notification method if the opening rate of the message notification is less than the threshold.

Further, in order to solve the above-described problems, the present invention provides a program for operating a computer as the notification method selection device described above, wherein the computer functions as each unit included in the notification method selection device. It is a program for

As described above, according to the present invention, a notification method suitable for the customer can be selected according to the content estimated by the notification reaction estimation model, so that the content of the notification can be easily recognized by the customer. can be done. In addition, since each customer can choose to receive notification by printed matter or by electronic means, the costs required for notification can be reduced compared to returning to the method of notifying all customers by printed matter. Is possible. In other words, it is possible to make the contents of the notification easier for the customer to recognize, and to prevent the cost required for the notification from increasing.

1 is a block diagram showing the configuration of a notification method selection system 1 to which a notification method selection device 10 according to a first embodiment is applied; FIG. 1 is a block diagram showing the configuration of a notification method selection device 10 according to a first embodiment; FIG. 4 is a diagram showing a configuration example of information stored in a learning data storage unit 103 according to the first embodiment; FIG. 4 is a flowchart for explaining the operation of the notification method selection device 10 according to the first embodiment; FIG. 10 is a diagram showing a configuration example of information stored in a learning data storage unit 103A according to the second embodiment; FIG. FIG. 13 is a diagram showing a configuration example of information stored in a learning data storage unit 103B according to the third embodiment; FIG.

An embodiment of the present invention will be described below with reference to the drawings.

(First embodiment)
First, the first embodiment will be explained.
FIG. 1 is a block diagram showing the configuration of a notification method selection system 1 to which a notification method selection device 10 according to the first embodiment is applied. The notification method selection system 1 is a system for notifying specific customers of various types of information in response to requests from companies and the like. The notification method selection system 1 includes, for example, a notification method selection device 10, a printing device 20, a distribution server 30, a print layout generation device 40, and an electronic layout generation device 50.

The notification method selection device 10 acquires a notification request. The notification request here is a signal transmitted from a terminal device or the like provided in a company or the like to the notification method selection device 10, and is a signal requesting notification from the company to the customer. The notification request includes at least information indicating the destination of the notification, such as the customer's name, address, telephone number, e-mail address, and the like.
Also, the notification request may include the content of the notification. The notice of predetermined content here is a business notice that the company gives to the customer, for example, notifying the amount of the claim or notifying the customer that the insurance that the customer has subscribed to has reached maturity. It is something to do.
The notification method selection device 10 selects a notification method (hereinafter also referred to as a notification method) according to the customer to be notified based on the notification request, and outputs the selection result.

The notification methods in this embodiment are print notification and electronic notification. A print notification is a notification by printed matter, and is a method of notifying by, for example, physical printed matter being delivered to the customer's address. Electronic notification is notification by an electronic method, and is a method of notifying by sending an e-mail or a message using a messaging service to the terminal (desktop, tablet, smartphone, etc.) used by the customer. . The messaging service here is, for example, a mechanism for transmitting and receiving contents such as messages using functions such as RCS (Rich Communication Services), SMS (Short Message Service), and MMS (Multimedia Messaging Service). That is, the electronic notification includes notification by e-mail and notification using a messaging service (hereinafter referred to as message notification).

The notification method selection device 10 outputs print notification instruction data that instructs print notification or electronic notification instruction data that instructs electronic notification, according to the selection result of selecting the notification method.
The print notification instruction data is data obtained by adding information such as the name and address of the customer to whom the notification is to be sent to information to the effect that the print notification is instructed.
The electronic notification instruction data is data obtained by adding information such as the e-mail address of the customer to be notified, the telephone number and account for message notification, etc. to the information to instruct the electronic notification.
When the notification method is print notification, the notification method selection device 10 outputs print notification instruction data to the printing device 20 . When the notification method is electronic notification, the notification method selection device 10 outputs electronic notification instruction data to the distribution server 30 . When the notification method is electronic notification, notification method selection device 10 adds information designating notification by e-mail or notification using a messaging service (hereinafter referred to as message notification) to electronic notification instruction data. may

The printing device 20 uses the print notification instruction data acquired from the notification method selection device 10 to perform notification by printed matter.
When acquiring the print notification instruction data, the printing device 20 requests the layout data from the print layout generation device 40 and acquires the print layout data from the print layout generation device 40 . The print layout data is data indicating the content to be notified by the print notification, for example, print data for printing a paper surface as a printed matter. The print data includes, for example, words and images to be printed, and information indicating layout such as layout on paper.
The printing device 20 prints on paper using the print layout data to generate printed matter. The printing device 20 seals the generated printed matter, attaches the address indicated in the print notification instruction data to the sealed envelope, and performs processing for delivering the envelope to the customer. Send the printed matter to the customer. As a result, the customer is notified by printed matter.

The distribution server 30 performs electronic notification using the electronic notification instruction data acquired from the notification method selection device 10 .
When the distribution server 30 acquires the electronic notification instruction data, the distribution server 30 requests layout data from the electronic layout generation device 50 and acquires the electronic layout data from the electronic layout generation device 50 . The electronic layout data is data indicating the content to be notified by electronic notification, and is, for example, content data indicated in an e-mail or message. Content data includes titles, texts, images, layouts, and information indicating link destinations. Images may include moving images as well as still images.
The distribution server 30 generates the contents of an e-mail using the electronic layout data, and transmits the generated e-mail to the e-mail address used by the customer indicated in the electronic notification instruction data. The distribution server 30 generates message content using the acquired electronic layout data, and transmits the generated message to the message ID used by the customer indicated in the electronic notification instruction data. The message ID is, for example, the customer's telephone number, or the customer's identification number registered in the application that notifies the message.

The print layout generation device 40 transmits print layout data to the printing device 20 in response to a request from the printing device 20 .
The electronic layout generation device 50 transmits electronic layout data to the distribution server 30 in response to a request from the distribution server 30 .

In the above description, the print layout generation device 40 generates the print layout data and the electronic layout generation device 50 generates the electronic layout data. However, the present invention is not limited to this. For example, the notification method selection device 10 may generate either or both of the print layout data and the electronic layout data.

FIG. 2 is a block diagram showing the configuration of the notification method selection device 10 according to the first embodiment. The notification method selection device 10 includes, for example, a notification request acquisition unit 100, an analysis unit 101, a processing unit 102, a learning data storage unit 103, and a learned model information storage unit 104.

The notification request acquisition unit 100 acquires a notification request and outputs the acquired notification request to the analysis unit 101 . The notification request includes information about the notification destination and information about the contents of the notification. The information about the notification destination is information indicating the notification destination, and includes, for example, the name, address, telephone number, e-mail address, message ID, etc. of the customer to be notified. The information about the content of the notification is information indicating the content to be notified and the conditions of the notification. The notification conditions are information indicating conditions for notification, such as a period for notification. For example, in the case of a notice informing the customer of the billed amount, it is necessary to notify the customer before the billed amount is debited, so the notification period is set to one week before the debit date. .

The analysis unit 101 creates a notification method estimation model in advance. A notification method reaction model is a trained model that estimates a customer's reaction to an electronic and/or printed notification given to the customer. The analysis unit 101 stores the created notification reaction estimation model in the learned model information storage unit 104 . The notification method reaction model is described in detail later.

A response to a notification is a response indicating that the customer has recognized the content of the notification, and there are, for example, two responses, response and opening. The reaction here may be both response and unsealing, or may be either response or unsealing.

A response is a response made by a customer according to the content of the notification. For example, when the customer is notified of the billed amount, it is possible to determine that the customer has responded to the notification, such as when the customer applies for a point redemption guide corresponding to the amount. In addition, if you notify that the insurance has reached maturity, if the insurance company has contacted you to continue enrolling in the insurance after maturity, or if you notify that the insurance has reached maturity, , it can be determined that the customer has responded to the notification, such as when an offer to join the student endowment insurance is made.

An open is a customer-initiated opening of a notification. Whether or not the notice has been opened is determined for the electronic notice. For example, in the case of message notification, it is determined whether or not the message has been opened (read), that is, whether or not the electronic notification has been opened by the application that performs the message notification. In the application, for example, when a message is displayed on the application of the terminal used by the customer of the notification destination, it is determined that the message has been opened by transmitting the message to the management server or the like even though the message has been opened. . In the case of notification by e-mail, an e-mail containing an image for confirming opening (hereinafter referred to as an opening confirmation image) and information indicating a link destination is sent. The tamper-evident image, for example, is pre-programmed to access a specific URL (Uniform Resource Locator) when the image is displayed. In this case, it is considered that the e-mail has been opened when the e-mail is opened and the confirmation image is displayed or when a specific URL is accessed by clicking a link destination.

Based on the notification request from the notification request acquisition unit 100, the analysis unit 101 selects a notification method from print notification and electronic notification. The analysis unit 101 estimates the customer's response to the electronic notification and/or print notification by inputting the customer attribute information of the customer to be notified into the notification response estimation model, and selects a notification method according to the estimation result. do. The customer attribute information is information indicating the attributes of the customer, such as age, sex, address, occupation, family structure, etc. of the customer.

Based on the notification request, the analysis unit 101 acquires the customer attribute information of the customer of the notification destination. The customer attribute information is included in the notification request, for example. Alternatively, the customer information may be stored in a customer attribute information storage server (not shown) communicably connected to the notification method selection device 10 via a communication network or the like. In this case, the customer attribute information storage server is a server device that stores customer attribute information of various customers. For example, the analysis unit 101 accesses the customer attribute information storage server based on the customer's name or the like indicated in the notification request, and acquires the customer information of the customer to be notified. Customer attribute information of various customers may be stored in a customer attribute information storage unit (not shown) of the notification method selection device 10 .

The analysis unit 101 inputs the acquired customer attribute information of the customer to be notified to the notification method estimation model. The notification reaction estimation model estimates the customer's reaction to the electronic notification and/or the printed notification based on the input customer attribute information. The notification reaction estimation model estimates, for example, the response rate and the opening rate when electronic notifications are sent to customers. In addition, the notification response estimation model estimates the response rate when printing notification is sent to the customer. The response rate here is the estimated probability that there will be a response to the notification. Also, the open rate is the probability that the notification will be opened.

The analysis unit 101 causes the notification reaction estimation model to estimate the response rate for each of print notification and electronic notification (e-mail notification and message notification). The analysis unit 101 selects the notification method estimated to have the highest response rate as the notification method for the customer. Specifically, when the customer is estimated to have the highest response rate to notification by e-mail, the analysis unit 101 selects notification by e-mail as the notification method for the customer. When it is estimated that the customer response rate for message notification is higher (larger) than the response rate for print notification, analysis unit 101 selects message notification as a notification method for the customer.

Here, when a printed matter is delivered as an actual object, it is common for people to have the will to pick up the printed matter and check the contents. On the other hand, message notifications do not have a physical form like printed matter, so even if a message is notified to a terminal device or the like, people do not have the will to check the content. . Although the open rate of printed notices cannot be measured like electronic notices, it is considered to be higher than the open rate of electronic notices. In addition, it is thought that most customers respond to print notifications after confirming the contents of such printed materials that they have opened voluntarily. good.
For this reason, if the response rate for message notification is not estimated to be higher (larger) than the response rate for print notification, the print notification will not be sent to the majority of customers if the user selects to send the print notification to all customers. becomes. In this case, there is a concern that the cost (printing fee, delivery fee, etc.) for delivering the printed matter will be excessive.

As a countermeasure, in the present embodiment, the analysis unit 101, even if the response rate for message notifications is not estimated to be higher (larger) than the response rate for print notifications, does not allow customers who are presumed to open message notifications to Choose to be notified. Then, a selection of sending printed notices only to customers whose response rate to message notifications is not estimated to be higher (larger) than the response rate to printed notifications, and who are estimated to rarely open message notifications. do. By doing so, the cost for delivering the printed matter is suppressed as much as possible.
Specifically, when the message notification response rate is estimated to be lower (smaller) than the print notification response rate, the analysis unit 101 estimates that the message notification opening rate is equal to or greater than a predetermined threshold, Select message notification as the notification method for the customer.

The processing unit 102 creates either print notification instruction data or electronic notification instruction data based on the selection result (selection result) of the notification method for the customer by the analysis unit 101 . When the print notification instruction data is created, the processing unit 102 outputs the created data to the printing device 20 . When the electronic notification instruction data is created, the processing unit 102 outputs the created data to the distribution server 30 . Here, the print notification instruction data is an example of the “selection result” by the analysis unit 101 . Also, the electronic notification instruction data is an example of the “selection result” by the analysis unit 101 .

The learning data storage unit 103 stores learning data. The learning data is data that the model learns when the analysis unit 101 creates the notification reaction estimation model. The learning data is input from another device (such as a server device that aggregates responses to notifications) via an input unit (not shown) of the notification method selection device 10, for example. The learned model information storage unit 104 stores a notification reaction estimation model.

FIG. 3 is a diagram showing a configuration example of information (learning data) stored in the learning data storage unit 103 according to the first embodiment. The learning data is created based on the results of past notifications.

As shown in FIG. 3, the learning data consists of a customer ID, an input and an output. A customer ID is identification information that uniquely identifies a customer. The input indicates customer attribute information of the customer corresponding to the customer ID, and in this example includes age, sex, address, occupation, family structure, and the like. The output includes items such as customer notification source, notification method, response, and opening. The notification source indicates the name of the company that is the notification source. The notification means indicates the notification method of the notification sent from the notification source to the customer during a predetermined period in the past, that is, print notification, e-mail, or message. The response indicates the number of notifications for which the notification method indicated in the notification means has been responded to. Opened indicates the number of notifications that have been opened for the notification method indicated in the notification means. In this example, a male customer in his 40s who lives in Tokyo, is engaged in the advertising business, and has a preschool child receives five message notifications from Company A, one of which is It is indicated that a response was received and that all five were opened. In addition, it is shown that this man received 20 message notifications from Company B, none of which received a response, and 18 of which were opened. In addition, for a customer who is a woman in her 60s who is a housewife living in Tokyo and whose child is already independent, Company A has not sent a print notification, and Company B has sent 30 messages. 5 were answered and 28 were opened.

The notification method estimation model will now be described.
The analysis unit 101 uses the learning data stored in the learning data storage unit 103 to make the machine learning model learn the correspondence relationship between the input (customer attribute information) and the output (reaction to the notification), thereby estimating the notification reaction. Create a model. General machine learning techniques such as hidden Markov models (HMM), deep learning, recurrent neural networks, convolutional neural networks, support vector machines (SVM), decision tree learning, genetic programming, etc. may be used.

The analysis unit 101 inputs unlearned input information (customer attribute information of a target customer for selecting a notification method, that is, a new customer) to the notification reaction estimation model.
When unlearned input information is input from the analysis unit 101, the notification reaction estimation model is a combination of learned input information (customer attribute information of existing customers) in learned learning data and unlearned input information. Calculate the degree of similarity. Here, the degree of similarity is, for example, the distance in vector space between a value obtained by quantifying the feature amount of unlearned input information by vector expression and a value obtained by digitizing the feature amount of learned input information by vector expression. calculated as The distance in the vector space is proportional to the amount of correlation between the values quantified by mutual vector representation, and is calculated, for example, by the inner product of the vectors.

The notification response estimation model extracts learned input information whose distance to unlearned input information in vector space is less than a predetermined threshold, that is, similarity. The trained model converts the learned output information (reactions to notifications of existing customers) associated with the extracted learned input information to the estimated output information (new customer response to notifications).

In the above description, the case where analysis unit 101 creates a notification reaction estimation model has been described. You may make it estimate the reaction to the notification in a customer.

FIG. 4 is a flowchart for explaining the operation of the notification method selection device 10 according to the first embodiment.
First, the notification method selection device 10 creates a notification reaction estimation model using the analysis unit 101 (step S10). The analysis unit 101 stores the created notification reaction estimation model in the learned model information storage unit 104 .
Next, the notification method selection device 10 acquires a notification request from the notification request acquisition unit 100 (step S11). The notification request acquisition unit 100 outputs the content of the acquired request notification to the analysis unit 101 .

Next, the notification method selection device 10 uses the notification response estimation model to estimate the response rate and the response rate of the new customer to be notified to the notification by the analysis unit 101 (step S12). The analysis unit 101 acquires the request notification from the notification request acquisition unit 100, refers to the customer attribute information storage server based on the information such as the name of the customer indicated as the notification target in the acquired notification request, and determines the notification target. Get customer attribute information for new customers.
Next, the analysis unit 101 determines whether or not the estimated value of the e-mail response rate is the highest (step S13).
Then, when the analysis unit 101 determines that the estimated value of the e-mail response rate is the highest, the notification method selection device 10 uses the result of the selection by the analysis unit 101 for e-mail notification by the processing unit 102. Electronic notification instruction data for e-mail is output (step S13).

On the other hand, if the analysis unit 101 determines in step S13 that the estimated value of the e-mail response rate is not the highest, the analysis unit 101 determines that the estimated value of the message response rate is the estimated value of the print notification response rate. It is determined whether or not the above is satisfied (step S15).
If the analysis unit 101 determines that the estimated value of the message response rate is greater than or equal to the estimated value of the response rate of the print notification, the processing unit 102 selects a message for message notification as a result of selection by the analysis unit 101. is output (step S16).

On the other hand, in step S15, when the analysis unit 101 determines that the estimated value of the message response rate is not equal to or greater than the estimated value of the response rate of the print notification, the analysis unit 101 determines that the estimated value of the message opening rate is a predetermined value. It is determined whether or not it is equal to or greater than the threshold (step S17).
When the analysis unit 101 determines that the estimated message open rate is equal to or greater than the predetermined threshold value, the processing unit 102 outputs print notification instruction data as a result of the selection by the analysis unit 101 (step S18). .

As explained above, the notification method selection device 10 of the first embodiment includes the analysis unit 101 and the processing unit 102 . The analysis unit 101 uses the notification reaction estimation model to select at least one of print notification and electronic notification as a notification method when notifying the customer of information. The processing unit 102 outputs the result of the selection by the analysis unit 101 (either print notification instruction data indicating a print notification instruction or electronic notification instruction data indicating an electronic notification instruction). The notification reaction model uses machine learning to perform machine learning using learning data that associates customer attribute information of existing customers who have a track record of notifying them in the past with the existing customer's reaction to the notification given to the existing customer. Machine learning model created.
As a result, the notification method selection device 10 of the first embodiment can estimate the reaction of the customer to the notification by the notification reaction estimation model, and select the notification method according to the tendency of the reaction estimated by the customer. can be easily confirmed by the customer. Also, for customers who tend to respond to both printed and electronic notifications because they can choose between printed or electronic notifications, electronic notifications can be selected. While making it easier for customers to confirm the notification, it is possible to reduce the cost required for the notification as compared with the case where all the customers are notified by printed matter.

In the above-described embodiment, if the estimated value of the message notification response rate is lower (smaller) than the estimated value of the print notification response rate, if the estimated value of the message notification opening rate is equal to or greater than a predetermined threshold, , the case where message notification is selected as the notification method for the customer. The predetermined threshold value in this case need not be a fixed value and may be changed according to circumstances.
For example, if the allotted budget is large, more expenses can be spent on printed notices. In this case, the predetermined threshold is set high (large) so that print notification is likely to be selected as the notification method for the customer. On the other hand, when the allocated budget is small, the predetermined threshold is set low (small) to make it difficult for the customer to select print notification as the notification method. As a result, within the allocated budget, it is possible to make a notification that is most easily confirmed by the customer.

(Second embodiment)
Next, a second embodiment will be described. In the present embodiment, the analysis unit 101 uses the notification date and time estimation model to estimate the date and time of notification (notification date and time) at which the content of the notification is likely to be confirmed, and uses the estimated result to determine the date and time of notification, and determine the interval of each notification when performing multiple notifications.

Here, the date and time of notification are the date and time when the notification is made in the print notification and the date and time when the notification is made in the electronic notification. The notification interval is the interval between sending dates when multiple print notifications are sent, or the interval between delivery times when multiple electronic notifications are sent.
In the following description, matters different from the above-described embodiment will be explained, and the same reference numerals will be given to the same matters as those of the above-described embodiment, and the explanation thereof will be omitted.

In the present embodiment, for example, when performing a notification on a date or time period when it is estimated that the content of the notification is likely to be confirmed, and performing multiple notifications using the same notification method for the same customer, Specify the notification order and notification interval.

For example, for customers who are presumed to be more likely to open (read) messages if they send message notifications in the afternoon on weekdays, send message notifications during that time period. will allow 10 minutes between sending each message notification.
Alternatively, if it is presumed that a customer who is scheduled to send three print notifications in a month tends to be more likely to respond to the print notification sent at the beginning of each month, , print notifications are sequentially sent at intervals of 3 days. As a result, notifications are sent on dates and times that are presumed to be easy for customers to confirm, and when multiple notifications are sent during those times, notifications are not too concentrated and all notifications are sent to customers. It is possible to make it easier to confirm

In this embodiment, the notification method selection device 10A includes an analysis unit 101A, a processing unit 102A, a learning data storage unit 103A, and a learned model information storage unit 104A.

The analysis unit 101A creates a notification date/time estimation model. The notification date and time model is a trained model that estimates the notification date and time at which the notification is more likely to be recognized by the customer when the customer is notified. Analysis unit 101A creates a notification date/time estimation model using learning data stored in learning data storage unit 103A.

FIG. 5 is a diagram showing a configuration example of information (learning data) stored in the learning data storage unit 103A according to the second embodiment.
As shown in FIG. 5, the response history information learning data consists of a customer ID, an input and an output. A customer ID is identification information that uniquely identifies a customer. The input indicates customer attribute information of the customer corresponding to the customer ID, and in this example includes age, sex, address, occupation, family structure, and the like. The output includes items such as notification date and time. The date and time of notification indicates the date and time of notification determined based on reactions to past notifications. For example, for a man in his 40s working in the advertising industry, 6:00 to 7:00 in the morning on weekdays and during the daytime on weekdays are suitable (described as OK), while nighttime on weekdays and holidays are inappropriate (described as NG). Shown as date and time.

The analysis unit 101A uses the learning data stored in the learning data storage unit 103A to cause the machine learning model to learn the correspondence relationship between the input (customer attribute information) and the output (notification date and time), thereby estimating the notification date and time. Create a model. The analysis unit 101A stores the created notification date/time estimation model in the learned model information storage unit 104A.

The analysis unit 101A inputs unlearned input information (customer attribute information of a new customer, a customer whose notification method is to be selected) to the notification date/time estimation model.
The notification date and time estimation model estimates the probability (opening rate) that a notice delivered (delivered) to that category will be opened for each category of notification date and time, in response to input of unlearned input information from the analysis unit 101A. . The division of notification date and time may be any division combining date and time, for example, division such as weekday morning, weekday afternoon, weekday evening, and holiday, or division by day of the week. Alternatively, it may be divided into predetermined time intervals (for example, one hour).

Then, the analysis unit 101A determines the notification date and time based on the opening rate estimated for each category. For example, the analysis unit 101A selects a category randomly selected from among the estimated values of the opening rate for each category, the highest estimated value of the opening rate, or the category whose estimated rate is equal to or greater than a predetermined threshold, according to the notification date and time. to decide.

The analysis unit 101A outputs the determined notification date and time to the processing unit 102A together with the result of selection of the notification method. At this time, if there are a plurality of notifications for which electronic notification has been selected for the same customer, the analysis unit 101A outputs the interval between each notification to the processing unit 102A.

When the analysis unit 101A selects the print notification, the processing unit 102A receives information specifying the delivery date as the notification date and time, or when there are multiple print notifications addressed to the same customer, the delivery date of other print notifications. to the printing device 20, to the printing device 20. Here, the information specifying the delivery date and the information specifying the interval from the delivery dates of other print notifications are examples of the "determination result" by the analysis unit 101A.
When electronic notification is selected by the analysis unit 101A, the processing unit 102A receives information specifying the transmission time as the notification date and time, and when there are multiple print notifications addressed to the same customer, the transmission time of other electronic notifications. to the delivery server 30, to the distribution server 30. Here, the information specifying the delivery time and the information specifying the interval from the delivery time of another electronic notification are examples of the "determination result" by the analysis unit 101A.

The learning data storage unit 103A stores, as learning data, information obtained by inputting customer attribute information of an existing customer and outputting the notification date and time of the existing customer. The learned model information storage unit 104A stores a notification date/time estimation model.

When the printer 20 acquires the print notification instruction data to which the information specifying the delivery date is added, the printer 20 instructs the delivery company on the delivery date so that the printed matter is delivered on the specified delivery date.
When the distribution server 30 acquires the electronic notification instruction data to which the information designating the transmission time is added, the distribution server 30 prepares an e-mail or message designating the distribution date and time so as to be electronically notified at the designated transmission time.

As described above, in the notification method selection device 10A according to the second embodiment, the analysis unit 101A determines the notification date and time for notifying the customer of information using the notification date and time estimation model. The processing unit 102 outputs the result of determination by the analysis unit 101 .
As a result, in the notification method selection device 10A of the second embodiment, even if the date and time at which a new customer is likely to recognize the notification is unknown, the date and time at which the notification is likely to be recognized is determined based on the attributes of the customer. It is possible to make a notification at the date and time when it is estimated that it is likely to be recognized.

In addition, in the above-described second embodiment, the order of notifications may be determined when a plurality of notifications are made. By appropriately determining the order of notifications, any notification can be easily recognized.
The order of notifications is determined, for example, based on responses in past notifications to customers. Specifically, the analysis unit 101A derives the response rate and the opening rate for each notification source and notification content. The contents of the notification are, for example, notification of the amount to be claimed, notification of expiration of insurance, and the like. The contents of the notification may include the presence or absence of additional information such as information on campaigns and information on educational endowment insurance.

The analysis unit 101A compares the derived notification source and the response rate for each notification content, and determines the notification source whose response rate tends to be high or the notification content whose response rate tends to be high. Further, the analysis unit 101A compares the derived notification source and the opening rate for each notification content, and determines the notification source whose opening rate tends to be high or the content of the notification whose opening rate tends to be high.

Then, the analysis unit 101A determines the notification order of the plurality of notifications using the determined result. For example, when each notification source in a plurality of notifications includes a notification source that tends to have a high response rate, the analysis unit 101A determines that the notification should be made first (or last). Alternatively, if the content of each notification in a plurality of notifications includes content that tends to have a high response rate, it is determined that the notification is given first (or last). Whether the notification is first or last may be arbitrarily determined according to the customer's tendency.

(Third Embodiment)
Next, a third embodiment will be described. In this embodiment, a notification expression estimation model is used to estimate a notification expression (notification expression) that makes it easier to confirm the content of the notification, and the estimated result is used to select the notification expression.

The expression of the notification here is the expression made in the notification, for example, the font, font size, color, presence or absence of bold or underline, amount of information, presence or absence of still images or moving images, etc. is.

In this embodiment, the notification is made using an expression that is presumed to make it easier for the customer to confirm the content of the notification.
For example, it is determined that a customer who is presumed to tend to respond easily to notations using large characters will be notified using notations using large characters. Alternatively, it is determined that a customer who is presumed to tend to respond easily to a notification using an image is to be notified using an expression using an image. By providing a notification that is presumed to be easy for the customer to confirm, it is possible to make it easier for the customer to confirm the content of the notification.

In this embodiment, the notification method selection device 10B includes an analysis unit 101B, a processing unit 102B, a learning data storage unit 103B, and a learned model information storage unit 104B.

The analysis unit 101B creates a notification expression estimation model. The notification date and time model is a trained model that estimates a notification expression that makes the notification easier for the customer to recognize when the customer is notified. Analysis unit 101B creates a notification expression estimation model using learning data stored in learning data storage unit 103B.

FIG. 6 is a diagram showing a configuration example of information (learning data) stored in the learning data storage unit 103B according to the third embodiment.
As shown in FIG. 6, the learning data is composed of, for example, a customer ID, input and output. A customer ID is identification information that uniquely identifies a customer. The input indicates customer attribute information of the customer corresponding to the customer ID, and in this example includes age, sex, address, occupation, family structure, and the like. The output comprises items such as notification expressions. The notification expression indicates a notification expression that is suitable as an expression of notification to the customer indicated by the customer ID. For example, the size and thickness of characters described in the notification, the amount of information, the presence or absence of still images and moving images, etc. are indicated as notification expressions. This example shows that notifications using images and notifications with a reduced amount of information are suitable for men in their 40s who work in the advertising industry. It has been shown that women in their 60s prefer notifications with a larger font size.

The analysis unit 101B uses the learning data stored in the learning data storage unit 103B to cause the machine learning model to learn the correspondence relationship between the input (customer attribute information) and the output (notification expression), thereby estimating the notification expression. Create a model. The analysis unit 101B stores the created notification expression estimation model in the learned model information storage unit 104B.

The analysis unit 101B inputs unlearned input information (customer attribute information of a target customer for selecting a notification method, that is, a new customer) to the notification expression estimation model.
The notification expression estimation model estimates, for each category of notification expression, the probability that a notification with that expression is considered suitable for input of unlearned input information from the analysis unit 101B. The classification of notification expression is the classification of characters and images that can be expressed in the notification, for example, the size and thickness of characters, the amount of information, and the presence or absence of still images and moving images.

Then, the analysis unit 101B determines a notification expression based on the probability estimated for each classification of the notification expression. The analysis unit 101A, for example, determines the expression with the highest probability among the estimated values of the probability for each category as the notification expression. The analysis unit 101B outputs the determined notification expression to the processing unit 102B together with the selection result of selecting the notification method.

When print notification is selected by the analysis unit 101B, the processing unit 102B sends print notification instruction data to the printing device, to which information indicating an expression corresponding to the content determined by the analysis unit 101B (hereinafter, also referred to as expression information) is added. 20.
When electronic notification is selected by analysis unit 101B, processing unit 102B outputs electronic notification instruction data to distribution server 30 to which expression information according to the content determined by analysis unit 101B is added.
Here, the expression information according to the contents is an example of the "determination result" by the analysis unit 101B. Note that the representation information in the print notification and the representation information in the electronic notification may have the same content or different contents.

The learning data storage unit 103B stores, as learning data, information obtained by inputting the customer attribute information of the existing customer and outputting the notification expression of the existing customer. The learned model information storage unit 104B stores notification expression estimation models.

When the printing apparatus 20 acquires the print notification instruction data to which the expression information is attached, the printing apparatus 20 requests the print layout generation apparatus 40 for layout data corresponding to the contents of the expression information.
When the distribution server 30 acquires the electronic notification instruction data to which the expression information is added, the distribution server 30 requests the electronic layout generation device 50 for layout data corresponding to the contents of the expression information.

As described above, in the notification method selection device 10B according to the third embodiment, the analysis unit 101B uses the notification expression estimation model to determine whether or not the customer has responded to the notification and the expressions made in past notifications. determines the representation of the notification that informs the customer of the information based on the relationship between The processing unit 102 outputs the result of determination by the analysis unit 101 .
As a result, in the notification method selection device 10B of the third embodiment, even if the date and time at which a new customer can easily recognize the notification is unknown, an expression that the customer can easily confirm is selected based on the customer's attributes. It is possible to presume and provide notice in a specified language that is easy for the customer to ascertain.

In at least one of the above-described embodiments, the case where either print notification or electronic notification is selected has been described as an example, but both print notification and electronic notification may be selected. For example, the analysis unit 101 knows that the customer for whom the print notification is selected is planning to move recently, and the address to which the print notification is to be sent cannot be determined. Select both printed and electronic notifications, for example, if you are unsure whether the As a result, even if the address of the customer who has selected the print notification cannot be determined, the notification can be made by both the print notification and the electronic notification, allowing the customer to more reliably confirm the content of the notification. .

All or part of the notification method selection device 10 (10A, 10B) in the above-described embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded in a computer-readable recording medium, and the program recorded in this recording medium may be read into a computer system and executed. It should be noted that the "computer system" referred to here includes hardware such as an OS and peripheral devices. The term "computer-readable recording medium" refers to portable media such as flexible discs, magneto-optical discs, ROMs and CD-ROMs, and storage devices such as hard discs incorporated in computer systems. Furthermore, "computer-readable recording medium" means a medium that dynamically retains a program for a short period of time, like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. It may also include something that holds the program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or client in that case. Further, the program may be for realizing a part of the functions described above, or may be capable of realizing the functions described above in combination with a program already recorded in the computer system. It may be implemented using a programmable logic device such as an FPGA (Field Programmable Gate Array).

Although the embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and design and the like are included within the scope of the gist of the present invention.

DESCRIPTION OF SYMBOLS 1... Notification method selection system 10... Notification method selection apparatus 100... Notification request acquisition part 101... Analysis part 102... Processing part 103... Learning data storage part 104... Learned model information storage part 20... Printing Device, 30... Distribution server

Claims (5)

an analysis unit that selects at least one notification method when notifying a customer of information from printed notification, which is notification by printed matter, or electronic notification, which is notification by electronic method, using the notification reaction estimation model;
a processing unit that outputs a result of selection by the analysis unit;
with
The notification reaction estimation model is machine-learned using learning data that associates customer attribute information indicating the attributes of an existing customer with a track record of past notification and the reaction of the existing customer to the notification given to the existing customer. is a machine learning model created by executing, estimating the response rate in the case of performing the print notification, and the response rate and opening rate in the case of performing the message notification as the electronic notification,
When the response rate of the message notification is smaller than the response rate of the print notification, the analysis unit sets the notification method to the message notification if the opening rate of the message notification is equal to or greater than a threshold, and sets the notification method to the message notification if the opening rate of the message notification is less than the threshold. the print notification is used as the notification method;
Notification method selection device. The analysis unit uses a notification date and time estimation model to determine the notification date and time when notifying the customer of the information,
the processing unit outputs a result of the determination by the analysis unit;
The notification date and time estimation model is created by executing machine learning using learning data that associates customer attribute information indicating the attributes of the existing customer with the results of the notification date and time selected by the existing customer. is a machine learning model,
The notification method selection device according to claim 1. The analysis unit uses a notification expression estimation model to determine a notification expression when notifying the customer of information,
the processing unit outputs a result of the determination by the analysis unit;
The notification expression estimation model is created by executing machine learning using learning data that associates customer attribute information indicating the attributes of the existing customer with the results of notification expressions selected by the existing customer. is a machine learning model,
The notification method selection device according to claim 1. The analysis unit uses the notification response estimation model to select at least one notification method when notifying the customer of information from print notification, which is notification by printed matter, or electronic notification, which is notification by electronic method,
a processing unit that outputs a result of the selection by the analysis unit;
The notification reaction estimation model is machine-learned using learning data that associates customer attribute information indicating the attributes of an existing customer with a track record of past notification and the reaction of the existing customer to the notification given to the existing customer. is a machine learning model created by executing, estimating the response rate in the case of performing the print notification, and the response rate and opening rate in the case of performing the message notification as the electronic notification,
When the response rate of the message notification is smaller than the response rate of the print notification, the analysis unit sets the notification method to the message notification if the opening rate of the message notification is equal to or greater than a threshold, and sets the notification method to the message notification if the opening rate of the message notification is less than the threshold. the print notification is used as the notification method;
Notification method selection method. A program for causing a computer to operate as the notification method selection device according to any one of claims 1 to 3, the program for causing the computer to function as each unit included in the notification method selection device.

Images