JP7757925B2 - Information processing device and method - Google Patents

Description

This disclosure relates to an information processing device and method for predicting products to recommend to customers.

A system has been disclosed that predicts the next product a customer is likely to purchase based on basic personal information about each customer, detailed information about the car-related products the customer owns, and the customer's purchase history of car-related products at each store (see, for example, Patent Document 1).

Japanese Patent Application Laid-Open No. 2005-346362

One aspect of the disclosure is to provide an information processing device and method that can improve the accuracy of predicting products to be recommended to customers. Machine learning models, for example, are used to predict products to be recommended to customers.

One aspect of the present disclosure is
estimating, for a first user, one or more candidates of a first type of product to be proposed to the first user based on first information about the user and second information about values;
extracting one or more of the first type of products from the one or more candidates based on third information related to a request for the first type of product based on a current state of the first user, and outputting one or more of the first type of products to be proposed to the first user;
a control unit that executes
The information processing device is provided with:
The control unit
The first information and the second information about the first user are input into a first learning model (e.g., a machine learning model) that has learned the relationship between the values of a second user and the first type of product that the second user has purchased or wanted to purchase, and the one or more candidates are obtained as output.

Another aspect of the present disclosure is
determining accuracy of first information for a first user, including information about the first user that may change over time, based on the time elapsed since the first information for the first user was last updated;
outputting a request for updating the first information about the first user if the accuracy of the first information is lower than a predetermined threshold;
a control unit that executes
The information processing device is provided with:
The control unit
inputting the first information about the first user and the elapsed time since the last update date and time of the first information about the first user into a second learning model (e.g., a machine learning model) that has learned changes in the first information about the second user over time;
The accuracy of the first information about the first user is obtained as an output.

Another aspect of the present disclosure is
The computer
estimating, for a first user, one or more candidates of a first type of product to be proposed to the first user based on first information about the user and second information about values;
extracting one or more of the first type of products from the one or more candidates based on third information related to a request for the first type of product based on a current state of the first user, and outputting one or more of the first type of products to be proposed to the first user;
This is a method for doing this.

One aspect of the present disclosure can improve the accuracy of predictions about products to be recommended to customers.

FIG. 1 is a diagram illustrating an example of a system configuration of a product recommendation system according to the first embodiment. FIG. 2 shows an example of the hardware configuration of the server. FIG. 3 is a diagram illustrating an example of a functional configuration of the server. FIG. 4 shows an example of information held in the customer information database. FIG. 5 shows an example of information stored in the value information database. FIG. 6 shows an example of information stored in the sales performance information database. FIG. 7 is a diagram illustrating an example of a customer information accuracy determination process performed by the information accuracy determination unit. FIG. 8 is a diagram illustrating an example of a process of acquiring candidates for a proposed product by the proposal candidate acquiring unit. FIG. 9 shows an example of extraction condition items for extracting suggested products from suggested product candidates. FIG. 10 is a diagram showing an example of a process for determining a proposed product based on an extraction condition related to "vehicle width 1." FIG. 11 is a diagram showing an example of a process for determining proposed products based on extraction conditions related to "delivery date." FIG. 12 is an example of a flowchart of the process of determining the accuracy of customer information by the server. FIG. 13 is an example of a flowchart of a suggested product determination process by the server. FIG. 14 is an example of a flowchart of the suggested product extraction process by the server. FIG. 15 is an example of a flowchart of a learning process of a learning model. FIG. 16 shows an example of a product suggestion screen of a store terminal.

For example, when predicting products to suggest based on information about a customer's preferences or values, the prediction results may include inappropriate products that do not meet the customer's requirements. Alternatively, for example, durable consumer goods such as automobiles have a long replacement cycle. Therefore, if the customer information used for the prediction remains outdated and has not been updated, predicting products to suggest using information that does not match the current situation may not lead to a purchase. One aspect of the present disclosure addresses these issues by improving the accuracy of predictions of products to suggest. The accuracy of predictions of products to suggest refers to whether the suggested products will be purchased or whether the customer will want to purchase them.

One aspect of the present disclosure is an information processing device that outputs one or more first types of products to be recommended to a first user. The information processing device is, for example, a computer such as a server, PC, smartphone, or tablet terminal. The information processing device includes a control unit. The control unit is, for example, a processor such as a CPU (Central Processing Unit) or an integrated circuit such as an FPGA (Field-Programmable Gate Array).

In one aspect of the present disclosure, the control unit estimates one or more candidates of a first type of product to propose to the first user based on first information about the first user and second information about the user's values. The control unit extracts one or more candidates and outputs one or more first type of product to propose to the first user based on third information about the first user's requirements for the first type of product based on the first user's current situation.

The first type of product is, for example, a durable consumer good such as an automobile, household appliance, or smartphone. The first information about the user includes, for example, the user's age, gender, marital status, whether or not they have children, the age of their eldest child, hobbies, annual income, and area of residence. However, the information included in the first information is not limited to these. The first information is an element that influences the user's values.

The second information regarding the user's values is, for example, information indicating a high level of safety awareness, information indicating a preference for price, and information indicating a preference for quality. The second information may be obtained, for example, as a response to a questionnaire exploring the user's values, or as a response to questions exploring the user's values in an interactive format. This first information and second information is information that changes over a relatively long period of time, in other words, information that changes little over time.

On the other hand, the third information is information regarding the first user's requirements for the first type of product based on the first user's current situation. Therefore, the third information is likely to change in response to changes in the first user's situation, and is likely to change in a shorter period of time than the first information and the second information. Examples of the third information include upper limit delivery times, upper limit size, required functions, and expected cost performance.

According to one aspect of the present disclosure, when proposing a first type of product, candidates to be suggested are estimated based on information that changes little over time, and from these candidates, products to be suggested are extracted based on third information that may change depending on the current situation. This makes it possible to suggest products that reflect the user's values, which do not change much, and that are suited to the user's current situation. In other words, it is possible to exclude products that are not suitable for the user's current situation from the products to be suggested to the user. Therefore, according to one aspect of the present disclosure, it is possible to improve the accuracy of predicting products to be suggested to the user.

In one aspect of the present disclosure, the control unit may input first information and second information about a first user to a first learning model and obtain one or more candidates as output. The first learning model is a model that has learned the relationship between the values of a second user and a first type of product that the second user has purchased or is interested in purchasing. The first learning model may be, for example, a machine learning model based on a predetermined algorithm, a neural network, a deep learning model, a convolutional neural network, a recurrent neural network, or the like. The second user is a user who is the source of learning data. By using the first learning model, it is possible to obtain a first type of product that reflects the values of the first user from the first information and second information about the first user.

The first learning model receives first information and second information about a second user,
The first learning model may have been trained using training data that outputs a first type of product that a user has purchased or wanted to purchase. The first learning model may output a degree of agreement between each of a plurality of first type products and the first user's values in response to input of first information and second information about the first user. In this case, the control unit may select, as one or more candidates to be proposed to the first user, products of the first type whose degree of agreement is equal to or greater than a predetermined threshold, or products of the first type that are among the top predetermined number in terms of degree of agreement. This makes it possible to propose to the first user products of the first type that have a higher degree of agreement with the first user's values.

In one aspect of the present disclosure, the control unit may estimate one or more candidates to suggest to the first user based further on information regarding the first user's purchase history of the first type of product. Information regarding the first user's purchase history of the first type of product can be considered to be a type of information that reflects the first user's values regarding the first type of product. Therefore, by estimating one or more candidates to suggest to the first user based further on information regarding the first user's purchase history of the first type of product, the accuracy of predicting the products to be suggested can be further improved.

In one aspect of the present disclosure, the control unit may acquire the third information based on information transmitted by the first user other than items included in the first information and second information about the first user. The information transmitted by the first user may be, for example, a text recording of a conversation with the first user, the content of the first user's utterances, and information input by the first user. The third information is information regarding a request for a first type of product based on the first user's current situation, which is not included in the first information or the second information. By acquiring the third information based on information transmitted by the first user, the third information can be acquired as information reflecting the first user's current situation.

For example, if the information transmitted by the first user indicates that the first user wants a first type of product immediately, the control unit may acquire the third information as information including an upper limit on the delivery time for the first type of product. In this case, the control unit may determine, from among one or more candidates, candidates that can be delivered by the upper limit on the delivery time as the one or more first type of product to be proposed to the first user. In this case, first type products with long delivery times are excluded from the products proposed to the first user, even if they have a high degree of agreement with the first user's values. This can improve the accuracy of predictions of products to be proposed to users.

For example, if the information transmitted by the first user indicates that there is a limit to the storage space for the first type of product, the control unit may acquire the third information as information including an upper limit on the size of the first type of product. In this case, the control unit may determine, from among one or more candidates, candidates whose size is less than the upper limit as one or more first type of product to be suggested to the first user. In this case, first type of product that cannot be stored in the storage space is excluded from the products suggested to the first user, even if the product has a high degree of agreement with the first user's values. This can improve the accuracy of predicting products to be suggested to users.

The control unit may acquire third information further based on information regarding the first user's purchase history of the first type of product. For example, the control unit may acquire the upper limit of the delivery date for the first type of product and the upper limit of the size of the first type of product based on information regarding the first user's purchase history of the first type of product, i.e., information about the first type of product purchased in the past by the first user. This makes it possible to extract from the candidates a first type of product that is more suited to the first user's current situation.

In one aspect of the present disclosure, the first information may include information that may change over time. In this case, the control unit may determine the accuracy of the first information about the first user based on the elapsed time since the last update date and time of the first information, and if the accuracy of the first information is lower than a predetermined threshold, output a request to update the first information about the first user. The accuracy of the information refers to the degree of accuracy of the information.

If the first information includes information that may change over time, there is a possibility that the first information may differ from the current state of the first user if a long period of time has passed since the first information was last updated. Even if low-accuracy first information is used to estimate candidates for the first type of product to be suggested to the first user, there is a high possibility that the accuracy of the estimation will decrease. Therefore, according to one aspect of the present disclosure, the accuracy of the first information can be improved by determining the accuracy of the first information and having the first information updated if the accuracy of the first information is low, thereby improving the accuracy of the first information and the products suggested to the first user.

The control unit may also input the first information about the first user and the elapsed time since the last update date and time of the first information about the first user to the second learning model, and obtain the accuracy of the first information about the first user as output. The second learning model is a model that has learned changes in the first information about the second user over time. The second learning model may be, for example, a machine learning model according to a predetermined algorithm, a neural network, a deep learning model, a convolutional neural network, a recurrent neural network, or the like. By using the second learning model, the accuracy of the first information about the first user can be obtained from the first information about the first user and the elapsed time since the last update date and time. Note that the second user that is the source of learning data for the first learning model and the second user that is the source of learning data for the second learning model may or may not overlap.

Another aspect of the present disclosure can be identified as an information processing device having a control unit that performs the following operations: determines the accuracy of first information about a first user, including information about the first user that may change over time, based on the amount of time elapsed since the first information about the first user was last updated; and, if the accuracy of the first information is lower than a predetermined threshold, outputs a request to update the first information about the first user. The information processing device is, for example, a computer such as a server, PC, smartphone, or tablet terminal. The control unit is, for example, a processor such as a CPU.

The control unit may input the first information about the first user and the time elapsed since the first information about the first user was last updated to the second learning model, and obtain the accuracy of the first information about the first user as output. The second learning model is a model that has learned changes in the first information about the second user over time. The second learning model may be, for example, a machine learning model according to a predetermined algorithm, a neural network, a deep learning model, a convolutional neural network, a recurrent neural network, or the like.

The first information may include multiple items. The second learning model may have been trained using learning data that receives as input the first information about the second user before the update and the time elapsed between the first information before the update and the updated first information, and outputs the updated first information about the second user. The second learning model may output the accuracy of each of the multiple items of the first information in response to input of the first information about the first user and the time elapsed since the first information about the first user was last updated. In this case, the control unit may output an update request for items whose accuracy is below a predetermined threshold. This makes it possible to determine the accuracy for each item included in the first information.

The control unit may also determine the accuracy of the first information about the first user based on information about a first type of product currently owned by the first user. In this case, the inputs to the second learning model are the first information about the first user, the elapsed time since the first information was last updated, and information about the first type of product currently owned by the first user. The input data for the learning data of the second learning model may include the first information about the second user before the update, the elapsed time between the first information before the update and the first information after the update, and information about the first type of product owned by the second user at the time of the update.

The first type of product is a durable consumer good. The type of product owned is often influenced by the first information about the first user. For example, families with young children tend to own large vehicles such as station wagons, while single people are less likely to own large vehicles such as station wagons. Therefore, when determining the accuracy of the first information about the first user, the accuracy of the determination of the accuracy of the first information can be improved by further using information about the first type of product currently owned by the first user.

Another aspect of the present disclosure can be specified as a method by which a computer executes the processing of the information processing device. Furthermore, another aspect of the present disclosure can be specified as a program for causing a computer to execute the processing of the information processing device. Another aspect of the present disclosure can be specified as a computer-readable, non-transitory recording medium for the program.

Embodiments of the present disclosure will be described below with reference to the drawings. The configurations of the following embodiments are examples, and the present disclosure is not limited to the configurations of the embodiments.

First Embodiment
FIG. 1 is a diagram showing an example of the system configuration of a product proposal system 100 according to the first embodiment. The product proposal system 100 is a system that proposes products to customers. In the first embodiment, it is assumed that the product is an automobile. The product proposal system 100 includes a server 1 and a store terminal 2. The store terminal 2 is a terminal used by sales staff at an automobile sales store. The product proposal system 100 includes multiple stores and store terminals 2, but FIG. 1 shows only one store and one store terminal 2 within that store. Note that multiple store terminals 2 may be provided within one store.

Server 1 and store terminal 2 are connected to network N1 and can communicate through network N1. Network N1 is, for example, a public line network such as the Internet.

In the first embodiment, based on a product proposal request from the store terminal 2, the server 1 estimates candidate products to propose to the target customer from the customer information, value information, and sales performance information of the target customer. In the first embodiment, the estimation of candidate products to propose to the target customer is performed using a learning model. The server 1 extracts products that meet the target customer's requirements for a car from the candidate products estimated by the learning model, and determines the products to propose. The target customer's requirements for a car include, for example, requirements regarding delivery time, vehicle width, functions, and design. The server 1 transmits information regarding the proposed products to the store terminal 2.

In the first embodiment, the server 1 determines the accuracy of the customer information of the target customer used to estimate product candidates to be suggested. In the first embodiment, a learning model is used to estimate the customer information. The customer information includes, for example, age, gender, marital status, whether or not the customer is married, whether or not the customer has children, the age of the eldest child, and annual income. Some of the items included in the customer information change over time. Furthermore, automobiles are a type of durable consumer good, and replacement intervals are often measured in years. Therefore, years often pass since the last update date and time of the customer information, and it is highly likely that the customer information held by the store has changed. If estimation is performed using customer information that does not reflect the customer's current situation, it is highly likely that products that match the customer's values will not be suggested. Therefore, by determining the accuracy of the customer information and having the customer information updated if the accuracy of the customer information is low, the accuracy of the estimation of the products to be suggested can be improved. The accuracy of the customer information refers to the degree to which the customer information matches the customer's current situation. The customer is an example of a "first user."

Figure 2 shows an example of the hardware configuration of server 1. Server 1 includes a CPU 101, memory 102, auxiliary storage device 103, and communication unit 104 as its hardware configuration. Memory 102 and auxiliary storage device 103 are each examples of computer-readable recording media.

The auxiliary storage device 103 stores various programs and data used by the CPU 101 when executing each program. The auxiliary storage device 103 is, for example, a hard disk drive (HDD) or a solid state drive (SSD). Programs stored in the auxiliary storage device 103 include, for example, an operating system (OS) and a control program for the product recommendation system 100.

The memory 102 is a storage device that provides the CPU 101 with a storage area for loading programs stored in the auxiliary storage device 103, a work area, and is used as a buffer. The memory 102 may be, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), or the like.
This includes semiconductor memory such as IEEE 802.11a/b/g (IEEE 802.11b/g) and IEEE 802.11b/g (IEEE 802.11b/g).

The CPU 101 executes various processes by loading the OS and various other programs stored in the auxiliary storage device 103 into the memory 102 and executing them.
The number of CPUs 101 is not limited to one, and may be multiple. The CPU 101 is an example of a "control unit."

The communication unit 104 is a module that connects, for example, a LAN (Local Area Network) card and a network cable such as an optical module, and that includes a signal processing circuit. The communication unit 104 is not limited to a circuit that can connect to a wired network, but may also be a wireless signal processing circuit that can process wireless signals from a wireless communication network such as Wi-Fi. Note that the hardware configuration of the server 1 is not limited to that shown in FIG. 2.

The store terminal 2 is, for example, a tablet terminal, a smartphone, or a PC. The store terminal 2 includes a hardware configuration such as a CPU, memory, an auxiliary storage device, a wireless communication unit, a touch panel display, a speaker, and a microphone. An application program for clients of the product recommendation system 100 is installed in the auxiliary storage device of the store terminal 2, and by executing this program, a request for suggested products can be sent from the store terminal 2 to the server 1.

Figure 3 is a diagram showing an example of the functional configuration of server 1. The functional configuration of server 1 includes an information accuracy determination unit 11, a proposal candidate acquisition unit 12, a proposed product determination unit 13, a customer information DB 14, a value information DB 15, a sales performance information DB 16, and a product information DB 17. Processing by these functional components is achieved when the CPU 101 of server 1 executes the control program of the product proposal system 100 stored in the auxiliary storage device 103.

When the information accuracy determination unit 11 receives a product suggestion request from the store terminal 2, it reads customer information about the target customer from the customer information DB 14 and determines the accuracy of the customer information. For example, identification information of the target customer is also received along with the product suggestion request. Details of the accuracy determination process will be described later. If the accuracy of the customer information about the target customer is below a predetermined threshold, the information accuracy determination unit 11 transmits a request to update the customer information to the store terminal 2. For example, a message requesting the update of the customer information and the names of the items to be updated are also transmitted together with the request to update the customer information.

When the proposal candidate acquisition unit 12 receives a product proposal request from the store terminal 2, it reads customer information from the customer information DB 14, value information from the value information DB 15, and sales performance information from the sales performance information DB 16 for the target customer. The proposal candidate acquisition unit 12 acquires proposed product candidates based on the customer information, value information, and sales performance information for the target customer. Details of the proposed product candidate acquisition process by the proposal candidate acquisition unit 12 will be described later. The proposal candidate acquisition unit 12 outputs information regarding the proposed product candidates to the proposed product determination unit 13.

When information about the suggested product candidates is input from the suggested candidate acquisition unit 12, the suggested product determination unit 13 extracts candidates that satisfy predetermined conditions and determines the suggested products. The predetermined conditions are the conditions that the target customer requires for the product. The process of extraction from the suggested product candidates by the suggested product determination unit 13 will be described in detail below. The suggested product determination unit 13 reads information about the suggested products from the product information DB 17 and transmits it to the store terminal 2. The store terminal 2 outputs the received information about the suggested products on its display.

The customer information DB 14, value information DB 15, sales performance information DB 16, and product information DB 17 are created in a storage area within the auxiliary storage device 103 of the server 1. The customer information DB 14 stores information about customers. The value information DB 15 stores information about customer values. The sales performance information DB 16 stores information about products that have been sold to customers. Details of the information stored in the customer information DB 14, value information DB 15, and sales performance information DB 16 will be described later.

Product information DB 17 stores information about products. Product information includes, for example, product name, model, color, size, and information about running costs. In the first embodiment, the product is an automobile. Therefore, product information includes, for example, information such as the car name, model, model, body color, body length, body width, and fuel efficiency. Running costs include, for example, fuel efficiency in the case of an automobile, power consumption in the case of a household appliance, and time until a smartphone is fully charged or continuous usable time in the case of a smartphone. Note that the information included in product information is merely an example and is not limited to these.

Figure 4 is an example of information stored in customer information DB 14. Customer information DB 14 stores customer information related to customers. The customer information shown in Figure 4 includes fields such as customer identification information (customer ID), date of birth, age, gender, generation, marriage flag, hobbies, life stage, annual household income, and last updated date and time. Customer information is registered by store staff via the store terminal 2, for example, based on information filled out in a questionnaire when the customer visits the store for the first time. Furthermore, when updating, for example, store staff asks the customer for information and updates the information via the store terminal 2. However, this is not limited to this, and the customer themselves may operate the store terminal 2 to enter customer information. The last updated date and time field stores the date and time the customer information was last updated.

Among the customer information items shown in Figure 4, gender, age group, marriage flag, hobbies, life stage, and annual household income may be selected from multiple options and stored. For example, the marriage flag item has two options: married and not married. If the customer is married, "married" is selected and "yes" is stored in the marriage flag item. If the customer is single, "not married" is selected and the marriage flag item is left blank.

The "life stage" field offers multiple options that combine information on marital status, the customer's generation, the age of the eldest child, and whether the child is independent. The age of the eldest child can be divided into four categories: under 7, 7 to 17, 17 to 23, and 23 or older. This is because statistically, there are trends in the type of car purchased depending on the age of the eldest child. This is an example of a category based on whether the eldest child is attending school, can drive, and is likely to be independent. For example, if a customer is in their 50s, married, has an eldest child 23 or older, and lives with their child, the "life stage" field would store information such as "family household, 50s, eldest child 23 or older." For example, if a customer is in their 30s and single, the "life stage" field would store information such as "single, 30s." Note that in the example shown in Figure 4, some of the information stored in the "life stage" field has been omitted for convenience.

For items for which no response is obtained from the customer, information indicating "unknown" may be stored. The items included in the customer information shown in FIG. 4 are only examples, and the items included in the customer information are not limited to those shown in FIG. 4. The items included in the customer information can be set arbitrarily by the administrator of the product recommendation system 100 depending on the embodiment. Customer information is an example of "first information." Age, generation, marriage status, hobbies, life stage, and annual household income are examples of "information that may change over time."

Figure 5 is an example of information stored in the value information DB 15. The value information DB 15 stores value information related to customer values. For example, when a customer visits the store for the first time, the customer answers a questionnaire asking about their values, and the value information is registered by store staff via the store terminal 2 based on the answers. However, this is not limited to this, and customers themselves may operate the store terminal 2 to enter their answers to the questionnaire. For each question in the questionnaire, options such as "applies," "somewhat applies," "can't say," "somewhat doesn't apply," and "doesn't apply" are provided, and customers respond by selecting one of the options. Note that the options for each question are not limited to these and are set according to the content of the question.

In the example shown in Figure 5, the value information includes items such as "actively take in information," "try to be a smart consumer," "cars are a means of transportation," and "expectations from cars: support for safe driving," and each item stores the options selected by the customer. The content of the questionnaire questions for obtaining value information is not limited to those shown in Figure 5. The content of the questionnaire questions for obtaining value information can be set arbitrarily by the administrator of the product proposal system 100 depending on the embodiment.

In the example shown in FIG. 5, the value information DB 15 also includes staff memos. Staff memos are records of information that store staff feel reflects the customer's values regarding car sales during conversations with the customer. Staff memos are freely entered by store staff via the store terminal 2. However, this is not limited to this, and for example, staff memos may be recorded by inputting conversations between customers and store staff into a learning model and recording keywords obtained as output. However, this is not limited to this, and the staff memo field may be information entered by the customer in the free-form entry field of the questionnaire, entered by store staff via the store terminal 2, or entered by the customer themselves using the store terminal 2. Although FIG. 5 shows the staff memo field recording the information entered by the store staff as is, keywords may be extracted from the record and tagged.

The staff memo records information that is not included in the customer information and the value information, such as, for example, "The parking lot is small, so we can't fit a large car in" or "I'm tired of red, so I want a different color." For items for which no answer is obtained from the customer, information indicating "unknown" may be stored. An answer to a question about values is an example of "second information." The information recorded in the staff memo is an example of "information sent by the first user other than items included in the first information and the second information about the first user."

Figure 6 is an example of information stored in the sales performance information DB 16. The sales performance information DB 16 stores information about products that customers have purchased in the past. In the example shown in Figure 6, the sales performance information includes the following items: customer ID, currently owned vehicle, previously owned vehicle, etc. Each vehicle item further includes the vehicle name, grade, options, price, and payment method. The sales performance information is registered, for example, by store staff via the store terminal 2 after a contract for the purchase of a product is concluded. Note that the sales performance information shown in Figure 6 is an example, and the information included in the sales performance information can be arbitrarily set by the administrator of the product recommendation system 100 depending on the embodiment. The sales performance information is an example of "information about the purchase history of the first type of product for the first user."

Figure 7 is a diagram showing an example of the customer information accuracy determination process performed by the information accuracy determination unit 11. The information accuracy determination unit 11 determines the accuracy of the customer information using a learning model for accuracy determination. The learning model for accuracy determination may be, for example, a machine learning model following a predetermined algorithm, a neural network, a deep learning model, a convolutional neural network, a recurrent neural network, etc.

For example, when the learning model for accuracy assessment receives input of information for each item of customer information, the last update date and time, information about the currently owned vehicle, and the current date and time for one customer, it outputs the accuracy (%) for each item of customer information as the accuracy. The information about the currently owned vehicle input into the learning model for accuracy assessment may be, for example, vehicle identification information, or multiple pieces of information such as the vehicle name, model, and body color. Note that for each item of customer information and information about the currently owned vehicle that has not been obtained, a value indicating "unknown" is input into the learning model for accuracy assessment. Furthermore, for example, if a customer does not currently own a vehicle, a value indicating "not owned" may be input into the learning model for accuracy assessment for information about the currently owned vehicle.

The information accuracy determination unit 11 identifies items in the output of the learning model for accuracy determination whose accuracy is below a predetermined threshold as having low accuracy, i.e., items that may have changed. The information accuracy determination unit 11 sends a message to the store terminal 2 requesting an update for items with low accuracy. The learning model for accuracy determination may also output the possibility (%) of change for each item of customer information. In this case, the information accuracy determination unit 11 identifies items whose possibility of change is equal to or greater than a predetermined threshold as having low accuracy, i.e., items that may have changed.

The learning model for accuracy assessment uses learning data about training users to learn the relationship between pre-update customer information and updated customer information. More specifically, the input data for the learning data of the learning model for accuracy assessment is, for example, the pre-update customer information, the last update date and time of the pre-update customer information, information about the vehicle owned at the time of learning, and the update date and time of the updated customer information. The output data for the learning model for accuracy assessment is, for example, information indicating whether or not there has been a change in each item of the updated customer information.

Each vehicle has a tendency toward a certain buyer demographic depending on its characteristics. For example, the buyers of sports cars tend to be mostly single people and few families with children. For example, the buyers of station wagons tend to be mostly families with children and few single people. Therefore, in the first embodiment, the learning model for accuracy learning is trained so that if the customer characteristics indicated by the input customer information do not match the tendency of the buyer demographic of the currently owned vehicle, the accuracy of the customer information is output with a low value.

More specifically, for example, if the customer information indicates that the customer is a single woman in her 30s and the vehicle currently owned is an SUV (Sport Utility Vehicle), the characteristics of the customer are SUV.
It is suspected that the life stage of the customer in question may have changed, as they are a minority among the buyers of V. In this case, the learning model for accuracy assessment outputs, in response to input containing customer information about the customer in question, that there is a high possibility that the life stage items have changed.

However, if the customer information includes information that the customer has outdoor hobbies such as camping, the customer's characteristics will match the trends of the majority of SUV buyers. In this case, the learning model used to assess accuracy will not output a high probability of change in the life stage category for input containing customer information about the customer.

The learning model for accuracy assessment may also use a portion of the value information as input. For example, if customer information indicates that the customer is single in their twenties and currently owns a minivan, this customer's characteristics are in the minority among minivan buyers, raising the possibility that the customer's life stage may have changed. Furthermore, if the value information item "Car is a means of transportation" is "applicable," this indicates that the minivan was chosen due to the customer's environment rather than their preference, and there is a higher possibility that the customer's life stage has changed from single to family household. In this case, the learning model for accuracy assessment will output, in response to input including customer information about the customer and a portion of the value information, that there is a high possibility that the life stage item has changed.

Alternatively, the learning model for accuracy assessment can also use staff memos as input. For example, if customer information indicates that the customer is single in their 30s and the current vehicle is an SUV, and the staff memo indicates that the parking lot is small, the width of the current vehicle does not match the narrowness of the parking lot, and the user of the vehicle may not be the customer. In this case, users of the customer's vehicle other than the customer may be family members or employees of a company run by the customer, raising the possibility that the customer's life stage may have changed. In this case, the learning model for accuracy assessment will output, in response to input including the customer information and staff memos about the customer, that there is a high possibility that the life stage item has changed.

Note that in Figure 7, a car is assumed as the product, and therefore information about currently owned vehicles is used as input to the learning model for accuracy assessment; however, any information about the product targeted by the product recommendation system 100 may be input. Also, in Figure 7, customer information and product information are used as input to the learning model for accuracy assessment; however, only customer information, the last update date and time, and the current date and time (elapsed time) may be used as input to the learning model for accuracy assessment. In this case, the learning model for accuracy assessment learns changes in customer information due solely to the passage of time, and estimates the probability that each item of customer information will not change over time (or the probability that it has changed due solely to the passage of time). The learning model for accuracy assessment is an example of a "second learning model." The learning data for the learning model for accuracy assessment is an example of a "second user."

8 is a diagram showing an example of a process for acquiring suggested product candidates by the suggestion candidate acquisition unit 12. The suggestion candidate acquisition unit 12 estimates suggested product candidates using a learning model for candidate acquisition. The learning model for candidate acquisition may be, for example, any of a machine learning model according to a predetermined algorithm, a neural network, a deep learning model, a convolutional neural network, a recurrent neural network, etc.

When information about each item of customer information, each item of value information, and information about the currently owned vehicle for one customer is input, the learning model for candidate acquisition outputs the degree of match (%) with the customer's values for each vehicle handled by the product recommendation system 100. The information about the currently owned vehicle input into the learning model for candidate acquisition may be, for example, the vehicle's identification information or the vehicle name. Note that for each item of customer information and value information that has not been obtained, a value indicating "unknown" is input into the learning model for candidate acquisition. Also, for example, if the customer does not currently own a vehicle, a value indicating "not owned" may be input into the learning model for candidate acquisition for information about the currently owned vehicle.

The proposal candidate acquisition unit 12 acquires, from the output of the learning model for acquiring accuracy, for example, vehicles whose degree of match with the customer's values is equal to or greater than a predetermined threshold, as candidates for proposed products. However, this is not limited to this, and the proposal candidate acquisition unit 12 may acquire, as candidates for proposed products, a predetermined number of vehicles that have the highest degree of match with the customer's values. Which of these is adopted is up to the design of the administrator of the product proposal system 100.

The candidate acquisition learning model uses learning data about a learning user to learn the relationship between the learning user's values and the vehicles that the learning user has purchased or is considering purchasing. More specifically, the input data for the candidate acquisition learning model is, for example, customer information, value information, and information about the vehicle currently owned. The output data for the candidate acquisition learning model is, for example, whether the learning user has purchased or intends to purchase each vehicle handled by the product recommendation system 100.

Each vehicle has a tendency towards a certain demographic, depending on its characteristics. By learning using the learning data described above, the candidate acquisition learning model reflects the tendency of each vehicle's buyer demographic. For example, in the case of a single male customer whose hobby is driving, the candidate acquisition learning model will output vehicles classified as sports cars that match the customer's values. For example, in the case of a customer who is a family with children, the candidate acquisition learning model will output vehicles classified as minivans or wagons that match the customer's values.

Note that in Figure 8, a car is assumed as the product, and therefore information about currently owned vehicles is used as input to the learning model for candidate acquisition; however, any information about the products targeted by the product recommendation system 100 may be input. Also, in Figure 8, customer information, value information, and information about the currently owned vehicle are used as input to the learning model for candidate acquisition; however, customer information and only customer information may be used as input to the learning model for candidate acquisition. In this case, the learning model for candidate acquisition estimates the degree of match between each vehicle and the customer's value, without taking into account the vehicle currently owned by the customer. Note that the input to the learning model for candidate acquisition is not limited to the information shown in Figure 8; any information that is thought to affect the customer's value toward the product can be used. The learning model for candidate acquisition is an example of a "first learning model." The learning user of the learning model for candidate acquisition is an example of a "second user."

The learning data for the learning model for accuracy assessment and the learning model for candidate acquisition may be obtained from the sales history of each store, or may be obtained by preparing a separate learning user. When the learning data is obtained from the sales history of each store, the customer of the target data is an example of a "second user." Note that the learning users of the learning model for accuracy assessment and the learning model for candidate acquisition do not need to be the same. Furthermore, the learning of the learning model for accuracy assessment and the learning model for candidate acquisition may be performed, for example, at a predetermined interval, or may be performed in real time each time new learning data is acquired.

In either case, the learning model is trained by adjusting the parameters of the learning model so that, for example, for one piece of learning data, the difference between the output value obtained by inputting input data and the value of the output data for the learning data is reduced. Learning for one piece of learning data is completed when this process is repeated a predetermined number of times, or when the difference between the output value of the learning model and the value of the output data for the learning data becomes smaller than a predetermined value. By performing this process for each piece of untrained learning data, the learning model is trained.

Figure 9 shows an example of extraction condition items for extracting suggested products from suggested product candidates. The suggested product determination unit 13 extracts candidates that match the extraction conditions from the suggested product candidates determined by the suggested candidate acquisition unit 12 and determines the suggested products. The extraction condition items are set in advance. In Figure 9, the extraction condition items are shown as delivery date, vehicle width 1, vehicle width 2, functions, and fuel efficiency. Keywords are associated with each extraction condition item, and if a corresponding keyword is found in the customer's staff memo, extraction conditions for the corresponding item are set. Extraction conditions vary depending on the item, but for items that can be expressed numerically, an upper limit value is set. The upper limit value is set by the suggested product determination unit 13.

For example, if the staff memo for the target customer in the value information DB 16 contains keywords such as "moving," "changing jobs," or "want it as soon as possible," the suggested product determination unit 13 sets an extraction condition for "delivery date." In this case, the suggested product determination unit 13 obtains the delivery dates of vehicles previously purchased from the customer's sales performance information, and sets an upper limit for the delivery date based on the delivery date of the vehicle. This upper limit for the delivery date becomes the extraction condition for "delivery date."

For example, if the keyword "narrow parking lot" is found in the staff memo for the target customer in the value information DB 16, the suggested product determination unit 13 sets an extraction condition for "vehicle width 1." In this case, the suggested product determination unit 13 obtains the vehicle widths of one or more vehicles previously purchased from the customer's sales performance information, and sets an upper limit for the vehicle width based on the vehicle widths of the one or more vehicles. This upper limit for the vehicle width becomes the extraction condition for "vehicle width 1."

For example, if the keyword "narrow road width" is found in the staff memo for the target customer in the value information DB 16, the suggested product determination unit 13 sets an extraction condition for "vehicle width 2." In the case of the "vehicle width 2" item, if the customer information includes items such as address, the suggested product determination unit 13 identifies the customer's residential area and obtains the width of narrow roads in that residential area (e.g., average value, etc.). Information on the width of narrow roads in each area may be stored in a database on the server 1, for example, or may be obtained from an external organization via a network. The suggested product determination unit 13 sets an upper limit for vehicle width based on the width of narrow roads in the customer's residential area. This upper limit for vehicle width becomes the extraction condition for "vehicle width 2."

For example, if the staff memo for the target customer in the value information DB 16 contains keywords such as "I want feature A" or "the same feature as my current car," the suggested product determination unit 13 sets an extraction condition for "feature." Since features cannot be quantified, extraction conditions are set as features that should be included. For example, the suggested product determination unit 13 obtains the features included in the customer's previously owned vehicle from the customer's sales performance information, and sets the inclusion of those features as an extraction condition for "feature."

For example, if the keyword "good fuel economy is better" is found in the staff memo about the target customer in the value information DB 16, the suggested product determination unit 13 sets an extraction condition for "fuel economy." In this case, the suggested product determination unit 13 acquires the fuel economy of one or more vehicles previously purchased from the customer's sales performance information, and sets an upper limit value for fuel economy based on the fuel economy of the one or more vehicles. This upper limit value for fuel economy becomes the extraction condition for "fuel economy."

The suggested product determination unit 13 collects information about candidate vehicles for suggested products via the product information DB 17 and the network, and extracts vehicles that match the extraction conditions as suggested products. The extraction conditions may be set for items that match keywords in the staff memo, and therefore multiple extraction conditions may be set. If there are multiple extraction conditions, the suggested product determination unit 13 may extract vehicles that satisfy all of the multiple extraction conditions. The extraction condition items shown in FIG. 9 are an example, and the extraction condition items can be set arbitrarily by the administrator of the product proposal system 100 depending on the embodiment. Information included in the staff memo items in the value information DB 16 is an example of "information sent by the first user." The extraction conditions for each extraction item are an example of "third information."

Figure 10 shows an example of a suggested product determination process based on extraction conditions related to "Vehicle Width 1." The suggested product determination unit 13 determines to set extraction conditions for the "Vehicle Width 1" item because the keyword "parking lot is narrow" is included in the staff memo for the customer. The suggested product determination unit 13 obtains the widths of the currently owned vehicle and the vehicle owned immediately before from the sales performance information of the target customer and sets an upper limit for the vehicle width. For example, in Figure 10, the suggested product determination unit 13 adds an allowable value to the larger of the widths of the currently owned vehicle and the vehicle owned immediately before, thereby setting the upper limit for the vehicle width to 1800 mm. The suggested product determination unit 13 obtains the widths of each suggested product candidate selected by the suggested candidate acquisition unit 12 from the product information DB 17 and determines CAR #4, whose width is less than the upper limit of 1800 mm, as the suggested product. Note that if there are multiple candidates whose vehicle width is less than the upper limit, all of those candidates are determined as suggested products. Note that the method for setting the upper limit of vehicle width is not limited to that described in Figure 10.

Figure 11 is a diagram showing an example of a process for determining suggested products based on extraction conditions related to "delivery date." Because the keyword "job change" is found in the staff memo about the customer, the suggested product determination unit 13 determines to set an extraction condition for the "delivery date" item. The suggested product determination unit 13 obtains the delivery date of the currently owned vehicle from the target customer's sales performance information and sets an upper limit for the delivery date. For example, in Figure 11, the suggested product determination unit 13 sets the upper limit for the delivery date to 3 months because the delivery date of the currently owned vehicle was 2-3 months at the time. The suggested product determination unit 13 obtains, for example, via the Internet, the delivery dates of each of the suggested product candidates selected by the suggested candidate acquisition unit 12, and determines CAR#1 and CAR#4, which have delivery dates less than the upper limit of 3 months, as suggested products. Note that the method for setting the upper limit for the delivery date is not limited to that described in Figure 11.

Candidate products for suggestion are obtained using customer information and value information. Customer information and value information are information that changes little over short periods. On the other hand, customer requirements for products, such as delivery time, vehicle width, functions, and fuel economy, are likely to change depending on the situation at the time. When using a learning model to predict suggested products using information with different change periods like this, the accuracy of the information is likely to decrease, and the learning model needs to be trained more frequently, making it difficult to maintain the accuracy of the prediction. However, the accuracy of the prediction can be improved by performing a two-stage process: using information with a long change period to predict candidate products for suggestion using a learning model, and then narrowing down the candidates using information with a short change period.

<Processing flow>
FIG. 12 is an example of a flowchart of the customer information accuracy determination process of the server 1.
The process shown in FIG. 12 is started, for example, when a product suggestion request is received from the store terminal 2. The identification information of the target customer is also received together with the product suggestion request. The process shown in FIG. 12 is executed by the CPU 101 of the server 1, but for convenience, the process will be described mainly with reference to the functional components. The same applies to the flowcharts in FIG. 13 and subsequent figures.

In OP101, the information accuracy determination unit 11 acquires customer information about the target customer from the customer information DB 14, and acquires information about the vehicle currently owned by the target customer from the sales performance information DB 16. In OP102, the information accuracy determination unit 11 inputs the customer information about the target customer, the last update date and time, the current date and time, and information about the vehicle currently owned by the target customer into a learning model for accuracy determination, and acquires the accuracy of each item of the customer information (see Figure 7).

In OP103, the information accuracy determination unit 11 determines whether there is any customer information item whose accuracy is below the threshold. If there is any customer information item whose accuracy is below the threshold (OP103: YES), the processing proceeds to OP104. If there is no customer information item whose accuracy is below the threshold (OP103: NO), the processing shown in FIG. 12 ends.

In OP104, the information accuracy determination unit 11 sends an instruction to the store terminal 2 to output a message requesting that the item whose accuracy is below the threshold be updated. Upon receiving this instruction from the server 1, the store terminal 2 outputs a message such as "Please update this item" on its screen. For example, the message continues to be displayed on the screen of the store terminal 2 until the customer information is updated.

In OP105, the information accuracy determination unit 11 determines whether the customer information has been updated by the store terminal 2. If the customer information of the target customer has been updated (OP105: YES), the processing proceeds to OP106. If the customer information of the target customer has not been updated (OP105: NO), the processing enters a standby state until the customer information is updated.

In OP106, the information accuracy determination unit 11 sends an instruction to the store terminal 2 to stop outputting the message requesting an update of customer information. Upon receiving this instruction from the server 1, the store terminal 2 stops displaying the message that was being displayed on the screen, and the message disappears from the screen. Thereafter, the processing shown in FIG. 12 ends.

Figure 13 is an example of a flowchart of the suggested product determination process of the server 1. The process shown in Figure 13 is initiated, for example, when a product suggestion request is received from the store terminal 2. Identification information of the target customer is also received along with the product suggestion request.

In OP201, the proposal candidate acquisition unit 12 acquires customer information of the target customer from the customer information DB 14, acquires value information of the target customer from the value information DB 15, and acquires information about the vehicle currently owned by the target customer from the sales performance information DB 16. In OP202, the proposal candidate acquisition unit 12 inputs the customer information, value information, and information about the currently owned vehicle of the target customer into a learning model for candidate acquisition, and acquires the degree of match between each vehicle handled by the product proposal system 100 and the customer's value (see FIG. 8).

In OP203, the proposal candidate acquisition unit 12 selects proposed product candidates based on the degree of match and outputs them to the proposed product determination unit 13. For example, the proposal candidate acquisition unit 12 selects vehicles whose degree of match is equal to or greater than a threshold, or a predetermined number of vehicles with the highest degree of match. The processing in OP202 and OP203 corresponds to the processing for acquiring proposed product candidates.

In OP204, the suggested product determination unit 13 executes a suggested product extraction process to extract products that satisfy the extraction conditions from among the suggested product candidates. The suggested product extraction process will be described in detail later. The suggested products are determined by the suggested product extraction process. In OP205, the suggested product determination unit 13 transmits information about the suggested products to the store terminal 2. The store terminal 2 displays the information about the suggested products. Thereafter, the process shown in FIG. 13 ends.

Figure 14 is an example flowchart of the suggested product extraction process on server 1. The process shown in Figure 14 corresponds to the process executed in OP204 of Figure 13.

In OP301, the suggested product determination unit 13 determines whether there is an item of extraction conditions that matches the contents of the customer's staff memo. If there is an item of extraction conditions that matches the contents of the customer's staff memo (OP301: YES), the process proceeds to OP302. If there is no item of extraction conditions that matches the contents of the customer's staff memo (OP301: NO), the process shown in FIG. 14 ends, and the process proceeds to OP205 in FIG. 13.

The processing of OP302 and OP303 is repeated the number of times corresponding to the number of extraction condition items that match the contents of the customer's staff memo. In OP302, the suggested product determination unit 13 sets extraction conditions according to the extraction condition items. In OP303, the suggested product determination unit 13 extracts products that satisfy the extraction conditions set in OP302 from the candidates. The subsequent processing of OP303 is executed for the extracted candidates.

When the processing in OP302 and OP303 has been completed for all extraction condition items that match the contents of the customer's staff memo, the processing shown in FIG. 14 ends, and the processing proceeds to OP205 in FIG. 13.

Figure 15 is an example of a flowchart of the learning process for a learning model. The process shown in Figure 15 is executed for each of the learning model for accuracy assessment and the learning model for candidate acquisition. Furthermore, for example, control of learning of the learning model for accuracy assessment is executed by the information accuracy determination unit 11. For example, control of learning of the learning model for candidate acquisition is executed by the proposal candidate acquisition unit 12. However, this is not limited to this, and if the server 1 is provided with a functional component that controls learning of the learning model, that functional component may control learning of both learning models. In the explanation of Figure 15, it is assumed that the proposal candidate acquisition unit 12 controls learning of the learning model for candidate acquisition.

The process shown in FIG. 15 is executed at a predetermined timing, for example, once a day, once a week, once a month, etc. However, this is not limited to this, and the process shown in FIG. 15 may be started in response to an instruction to start learning from an administrator of the product recommendation system 100. Alternatively, it may be executed when learning data is added.

In OP401, the proposal candidate acquisition unit 12 acquires learning data. The learning data may be acquired from a database separate from the server 1, or the server 1 may maintain a database of learning data and acquire the data from that database.

The processes from OP402 to OP405 are repeatedly executed for each piece of learning data. In OP402, the proposal candidate acquisition unit 12 inputs input data from the learning data into a learning model for candidate acquisition and obtains an output. In OP403, the proposal candidate acquisition unit 12 compares the output value of the learning model for candidate acquisition with the output data of the learning data. In OP404, the proposal candidate acquisition unit 12 adjusts the parameters of the learning model based on the comparison result of OP403.

In OP405, the proposal candidate acquisition unit 12 determines whether or not a termination condition is satisfied. The termination condition is, for example, that the difference value between the output value of the learning model for candidate acquisition and the output data of the learning data is less than a predetermined threshold, or that the processes from OP402 to OP405 have been repeatedly executed a predetermined number of times. If the termination condition is not satisfied (OP405: NO), the process proceeds to OP402, and the processes from OP402 to OP405 are repeatedly executed for the target learning data. If the termination condition is satisfied (OP405: YES), the processes from OP402 to OP405 are executed for the next learning data. When the processes from OP402 to OP405 have been completed for all learning data, the process illustrated in FIG. 15 ends.

Figure 16 is an example of a product proposal screen on the store terminal 2. The product proposal screen is a screen that is displayed when information on suggested products is received from the server 1 in response to a product proposal request to the server 1. On the product proposal screen shown in Figure 16, the "Customer Information" field displays customer information and information about the currently owned vehicle. The "Predicted Vehicle" field displays the names of three suggested products. The product proposal screen includes a "Get Suggested Vehicle" button. When a user operation is input to select the "Get Suggested Vehicle" button, the store terminal 2 sends a product proposal request to the server 1. The product proposal screen includes an "Update Information" button. When a user operation is input to select the "Update Information" button, the store terminal 2 sends an information update request and the updated information to the server 1.

In the example shown in FIG. 16, a message requesting an update, "Please update," is displayed in the "Life Stage" field in the "Customer Information" column. This indicates that the server 1 has determined that the accuracy of the "Life Stage" field for the customer in question is low. In the example shown in FIG. 16, the suggested products displayed in the "Predicted Vehicle" field are vehicles estimated using the customer information before the update. After this, when a store staff member inquires about the customer's current life stage, enters the information, and selects the "Update Information" button, the customer's life stage is updated via the store terminal 2. The product proposal screen then changes to show the updated information. When the store staff member then selects the "Get Suggested Vehicle" button, a product proposal request is sent to the server 1 via the store terminal 2. The server 1 estimates the products to suggest using the updated customer information and sends information about the suggested products to the store terminal 2. As a result, the suggested products displayed in the "Predicted Vehicle" field on the product proposal screen of the store terminal 2 are based on the updated customer information.

Store staff can view the suggested products displayed in the "Predicted Vehicle" column and make suggestions to the customer. If a purchase contract is subsequently concluded with the customer, the store staff may register the purchased vehicle as sales performance information via the store terminal 2. The newly registered sales performance information may also be used as learning data for a learning model used to acquire candidates.

<Operation and effect of the first embodiment>
In the first embodiment, the server 1 estimates candidate products to be suggested from customer information and value information, and further extracts candidates that correspond to product requirements based on the customer's current situation, and determines the suggested products. This prevents the suggestion of products that do not satisfy the product requirements based on the customer's current situation, even if they match the customer's value system, thereby improving the accuracy of estimating suggested products.

In addition, in the first embodiment, the server 1 determines the accuracy of the customer information and requests an update if the accuracy is low. This makes it possible to improve the accuracy of the suggested products estimated using the customer information.

<Other Modifications>
The above-described embodiment is merely an example, and the present disclosure can be modified and implemented as appropriate within the scope that does not deviate from the gist of the disclosure.

The processes and means described in this disclosure can be freely combined and implemented as long as no technical contradictions arise.

Furthermore, processing described as being performed by one device may be shared and executed by multiple devices. Alternatively, processing described as being performed by different devices may be executed by a single device. In a computer system, the hardware configuration (server configuration) used to realize each function can be flexibly changed.

The present disclosure can also be realized by supplying a computer program that implements the functions described in the above embodiments to a computer, and having one or more processors in the computer read and execute the program. Such a computer program may be provided to the computer via a non-transitory computer-readable storage medium connectable to the computer's system bus, or via a network. Non-transitory computer-readable storage media include, for example, any type of disk, such as a magnetic disk (e.g., a floppy disk, a hard disk drive (HDD), etc.), an optical disk (e.g., a CD-ROM, a DVD disk, a Blu-ray disk), a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic card, a flash memory, an optical card, or any type of medium suitable for storing electronic instructions.

1. Server 2. Store terminal 11. Information accuracy determination unit 12. Proposal candidate acquisition unit 13. Proposal product determination unit 14. Customer information DB
15. Values information DB
16. Sales performance information DB
17.Product information DB
100: Product recommendation system 101: CPU
102: Memory 103: Auxiliary storage device 104: Communication unit

Claims (19)

estimating, for a first user, one or more candidates of a first type of product to be proposed to the first user based on first information about the user and second information about values;
extracting one or more of the first type of products from the one or more candidates based on third information related to a request for the first type of product based on a current state of the first user, and outputting one or more of the first type of products to be proposed to the first user;
a control unit that executes
Equipped with
the first information includes information that may change over time;
The control unit
determining accuracy of the first information about the first user based on an elapsed time since the first information was last updated;
outputting a request for updating the first information about the first user if the accuracy of the first information is lower than a predetermined threshold;
An information processing device that executes the above . The control unit
inputting the first information and the second information about the first user into a first learning model that has learned the relationship between the values of a second user and the first type of product that the second user has purchased or is interested in purchasing, and obtaining the one or more candidates as an output;
The information processing device according to claim 1 . The first learning model is
The system has been trained using training data in which the first information and the second information about the second user are input and the first type of product that the second user has purchased or is interested in purchasing is output;
outputting a degree of agreement between each of the plurality of first type products and the first user's values in response to input of the first information and the second information about the first user;
the control unit selects, as the one or more candidates, products of the first type whose degree of match is equal to or greater than a predetermined threshold, or products that are among a top predetermined number of products of the first type having the degree of match;
The information processing device according to claim 2 . the control unit estimates the one or more candidates further based on information regarding a purchase history of the first type of product for the first user;
The information processing device according to claim 1 . The control unit
The third information is acquired based on information transmitted from the first user other than items included in the first information and the second information about the first user.
The information processing device according to claim 1 . The control unit
The third information is acquired as information including an upper limit of a delivery time of the first type of product;
determining, from among the one or more candidates, the candidates that can be delivered by the upper limit of the delivery time as the one or more first type products to be proposed to the first user;
The information processing device according to claim 5 . The control unit
The third information is acquired as information including an upper limit value of the size of the first type of product;
determining, from among the one or more candidates, the candidates whose size is less than the upper limit value as the one or more first type products to be proposed to the first user;
The information processing device according to claim 5 . The control unit
acquiring the third information further based on information regarding the first user's purchase history of the first type of product;
8. The information processing device according to claim 6 or 7. The control unit
inputting the first information about the first user and the time elapsed since the last update date and time of the first information about the first user into a second learning model that has learned changes in the first information about the second user over time, and obtaining the accuracy of the first information about the first user as an output;
The information processing device according to claim 1 . determining accuracy of first information for a first user, including information about the first user that may change over time, based on the time elapsed since the first information for the first user was last updated;
outputting a request for updating the first information about the first user if the accuracy of the first information is lower than a predetermined threshold;
a control unit that executes
An information processing device comprising: The control unit
inputting the first information about the first user and the time elapsed since the last update date and time of the first information about the first user into a second learning model that has learned changes in the first information about the second user over time, and obtaining the accuracy of the first information about the first user as an output;
The information processing device according to claim 10 . the first information includes a plurality of items;
The second learning model is
learning has been completed using learning data that receives as input the first information before the update regarding the second user and an elapsed time from the first information before the update to the first information after the update, and outputs the updated first information regarding the second user;
outputting accuracy for each of the plurality of items of the first information in response to input of the first information about the first user and an elapsed time since the last update date and time of the first information about the first user;
the control unit outputs an update request for an item whose accuracy is less than a predetermined threshold.
The information processing device according to claim 11 . The control unit
determining accuracy of the first information about the first user further based on information about a first type of product currently owned by the first user;
The information processing device according to claim 10 . The control unit
inputting the first information about the first user, the time elapsed since the last update date and time of the first information, and information about the first type of product currently owned by the first user into a second learning model that has already learned about the relationship between changes in the first information over time and the first type of product currently owned by the second user, and obtaining the accuracy of the first information about the first user as an output;
The information processing device according to claim 13 . the first information includes a plurality of items;
The second learning model is
learning has been completed using learning data in which the first information about the second user before the update, the elapsed time between the first information about the second user before the update and the first information after the update, and information about the first type of product held by the second user at the time of the update are input, and the updated first information about the second user is output;
outputting accuracy for each of the plurality of items of the first information in response to input of the first information about the first user, the time elapsed since the last update date and time of the first information about the first user, and information about the first type of product currently owned by the first user;
the control unit outputs an update request for an item whose accuracy is less than a predetermined threshold.
The information processing device according to claim 14 . The control unit
estimating one or more candidates of a first type of product to be proposed to the first user based on the first information and second information related to values about the first user;
extracting one or more of the first type of products from the one or more candidates based on third information related to a request for the first type of product based on a current state of the first user, and outputting one or more of the first type of products to be proposed to the first user;
To execute
The information processing device according to claim 10 . The control unit
inputting the first information and the second information about the first user into a first learning model that has learned the relationship between the values of a second user and the first type of product that the second user has purchased or is interested in purchasing, and obtaining the one or more candidates as an output;
The information processing device according to claim 16 . The first learning model is
The system has been trained using training data in which the first information and the second information about the second user are input and the first type of product that the second user has purchased or is interested in purchasing is output;
outputting a degree of agreement between each of the plurality of first type products and the first user's values in response to input of the first information and the second information about the first user;
the control unit selects, as the one or more candidates, products of the first type whose degree of match is equal to or greater than a predetermined threshold, or products that are among a top predetermined number of products of the first type having the degree of match;
The information processing device according to claim 17 . The computer
estimating, for a first user, one or more candidates of a first type of product to be proposed to the first user based on first information about the user and second information about values;
extracting one or more of the first type of products from the one or more candidates based on third information related to a request for the first type of product based on a current state of the first user, and outputting one or more of the first type of products to be proposed to the first user;
Run
the first information includes information that may change over time;
The computer
determining accuracy of the first information about the first user based on an elapsed time since the first information was last updated;
outputting a request for updating the first information about the first user if the accuracy of the first information is lower than a predetermined threshold;
How to perform .