JP6885356B2 - Recommended information identification device, recommended information identification system, recommended information identification method, and program - Google Patents

Description

The present invention relates to a recommended information specifying device for specifying recommended information suitable for the taste of a user who uses a store, a recommended information specifying system, a recommended information specifying method, and a program.

In electronic commerce and advertisements on the Internet, there is known a method of extracting so-called recommended information according to a user's taste from a huge amount of information and displaying it on a terminal device used by the user. The recommended information is also referred to as recommendation information, and corresponds to, for example, product information including recommended products that should be recommended that may match the taste of the user. The recommended information is specified based on behavior history such as history information of a user purchasing a product on a WEB site (product purchase history) and history information of browsing a product introduction page (product browsing history). .. The function of analyzing the behavior history, identifying and presenting the recommended information according to the user's preference based on a predetermined algorithm is called a recommendation function, and has been widely used in recent years.

Conventionally, a recommendation device having the recommendation function has been proposed (see Patent Document 1). The recommendation device displays recommended information according to the taste of a customer (visitor) on a tablet or the like used by an employee in a store such as a retail store.

Further, in recent years, a payment system using an information code such as a bar code or a QR code (registered trademark) indicating the content or price of a product has been proposed (see Patent Document 2). A mobile terminal (for example, a smartphone) of a customer who uses a store is equipped with a code reading function for reading an information code given to a product. According to the payment system, the customer uses a mobile terminal to read the information code attached to the product package or the like, puts the product in the shopping cart, and then charges the product amount included in the read information code. Make a payment. Payment of the product amount can be made, for example, by using a cash register that can communicate with the mobile terminal, payment by electronic money registered in the mobile terminal, online credit payment, or the like.

Japanese Unexamined Patent Publication No. 2017-215667 Japanese Unexamined Patent Publication No. 2015-185089

However, in stores such as retail stores, the product status differs depending on the multiple types of products handled, so the conventional recommendation function that presents recommended information determined only by the customer's taste is a store according to the product status. In some cases, the side (seller) cannot present the product that it wants to sell to the customer. For example, in a store that handles a certain product A and a similar product B, if the inventory of the product A is larger than that of the product B, the store wants to sell the product B preferentially. With the recommendation function, it is not possible to give priority to product B as recommended information and present it to the customer. In particular, in supermarkets that handle products with a short product life such as cooked foods (processed foods) such as fresh foods and prepared foods, the product status changes from moment to moment in a day, so recommended information that takes into account the product status Proposals are desired.

An object of the present invention is to specify recommended information according to the product situation, so that the user who receives the recommended information can quickly obtain the product information that matches his / her taste, and the store side can obtain the recommended information. The purpose is to provide a recommended information identification device, a recommended information identification system, a recommended information identification method, and a program that can increase the purchase rate of products that are preferentially sold.

The recommended information specifying device according to one aspect of the present invention is based on the user's behavior history with respect to the product provided in the store and a predetermined priority regarding the sale of a plurality of products provided in the store. It is provided with a recommended information specifying unit that specifies recommended information including recommended products according to the user from a plurality of product information related to a plurality of products.

The recommended information specifying device of the present invention further includes a ranking determination unit configured to determine the priority by machine learning. The ranking determination unit includes a fluctuation information acquisition unit that acquires fluctuation information that causes fluctuations in the priority, and a ranking learning unit that learns the priority based on the fluctuation information acquired by the fluctuation information acquisition unit. And have.

The recommended information identification system according to another aspect of the present invention includes a priority order acquisition unit, an action history acquisition unit, and a recommended information identification unit. The priority acquisition unit acquires the priority from the priority storage unit that stores the priority regarding the sale of a plurality of products provided in the store. The action history acquisition unit acquires the action history from the action history storage unit that stores the user's action history for the product provided in the store. The recommended information specifying unit is recommended according to the user from the plurality of product information based on the action history acquired by the action history acquisition unit and the priority acquired by the priority order acquisition unit. Identify recommended information, including products.

The method for specifying recommended information according to another aspect of the present invention includes an action history acquisition step of acquiring a user's action history for a product provided in a store, the acquired action history, and a plurality of products provided in the store. It is provided with a recommended information specifying step of identifying recommended information including recommended products according to the user from a plurality of product information related to the plurality of products based on a predetermined priority with respect to the sale of the product.

The present invention can also be regarded as a program for causing a computer to execute each step of the recommended information identification method, or as a computer-readable recording medium in which such a program is recorded non-temporarily.

According to the present invention, by specifying the recommended information according to the product situation, the user who is presented with the recommended information can quickly obtain the product information that matches his / her taste, and the store side ( The seller) can increase the purchase rate of products that he / she wants to sell preferentially.

Further, according to the present invention, by providing the ranking determination unit including the fluctuation information acquisition unit and the ranking learning unit, the priority regarding the sale of a plurality of products can be set based on the fluctuation information that causes fluctuations in the priority. It is learned by the ranking learning department. As a result, even if the product status changes from moment to moment in the store, the recommended information is specified using the priority determined according to the change, and as a result, the user who is presented with the recommended information. Can quickly obtain product information that matches their tastes, and the store side (seller) can increase the purchase rate of products that they want to sell preferentially in real time.

FIG. 1 is a diagram showing a configuration of an information processing system according to an embodiment of the present invention. FIG. 2 is a diagram showing an example of a screen displayed on a mobile terminal constituting an information processing system. FIG. 3 is a block diagram showing the configuration of each device constituting the information processing system. FIG. 4 is a block diagram showing a configuration of a ranking determination unit of a server device constituting an information processing system. FIG. 5 is a block diagram showing a configuration of a discount determination unit of a server device constituting an information processing system. FIG. 6 is a diagram showing the order of each product handled by the store. FIG. 7 is a flowchart showing an example of the procedure of the recommendation processing executed in the information processing system. FIG. 8 exemplifies each coefficient and the recommendation evaluation value when the extracted five candidate products are products having the first to fifth priority. FIG. 9 is a flowchart showing an example of a learning operation executed in the learning unit of the information processing system.

Hereinafter, embodiments of the present invention will be described with reference to the drawings as appropriate. It should be noted that the embodiments described below are merely examples that embody the present invention, and do not limit the technical scope of the present invention.

[Information processing system 100]
As shown in FIG. 1, the information processing system 100 (recommended information identification system of the present invention) according to the embodiment of the present invention includes a server device 11 (an example of the recommended information identification device of the present invention) and a gate device for entering a store. 12. The gate device 13 for leaving the store, the self-type register 14, the user terminal 15 (an example of the terminal device of the present invention), and the employee terminal 16 used by the employees of the store 200 are included. The gate device 12, the gate device 13, and the register 14 may all be one or may be plural. In the present embodiment, for example, an example in which the information processing system 100 is introduced into a store 200 such as a supermarket that handles a wide variety of products will be described. Of course, the store 200 in which the information processing system 100 is introduced is not limited to a supermarket, and may be an actual retail store that handles a plurality of products, or may be a virtual store on the Internet.

The gate device 12 is installed near the entrance of the store 200, and the gate device 13 is installed near the exit of the store 200. The user terminal 15 is a terminal device possessed and used by a user (user) who has entered the store 200 when purchasing a product displayed in the store 200 in the store 200, and is, for example, a smartphone or a smartphone. It is a mobile terminal such as a mobile phone or a tablet terminal. The employee terminal 16 is a terminal device used by employees of the store 200, and is, for example, a mobile terminal such as a laptop computer, a smartphone, or a tablet terminal.

Hereinafter, with reference to FIGS. 1 and 2, a series of flow from the user visiting the store 200 to purchasing the product and leaving the store will be illustrated.

First, a user who visits the store 200 approaches the gate device 12 for entering the store from the entrance while possessing the user terminal 15. When the user approaches the gate device 12, the user terminal 15 communicates with the gate device 12 by short-range wireless communication represented by Bluetooth (registered trademark) or FeliCa (registered trademark). At this time, the user identification information (ID number, name, gender, age, etc.) stored in the user terminal 15 is transmitted to the gate device 12. The user identification information is information for identifying a user of the store 200. The user identification information is transferred from the gate device 12 to the server device 11 through a communication network N1 such as a wireless LAN (in-store LAN) installed in the store 200. As a result, the server device 11 recognizes that a user registered in advance has visited the store. If the user terminal 15 includes the purchase history information of the user in the store 200, the purchase history information is also transmitted to the gate device 12 and transferred to the server device 11 through the communication network N1. .. Here, the purchase history information is history information of products that the user has purchased in the past at the store 200.

A shopping application used in a system for shopping while scanning an information code (bar code, QR code, etc.) attached to a product is pre-installed on the user terminal 15. When the user terminal 15 communicates with the gate device 12, the user terminal 15 automatically activates the shopping application. At this time, for example, the screen shown in FIG. 2A is displayed on the operation display unit 22 (see FIG. 3) of the user terminal 15. As a result, the user can purchase the product in the store 200 by using the shopping application.

While buying around the store 200, the user picks up the product he / she wants to purchase (the product to be purchased) from the display shelf, and obtains the information code attached to the product from the imaging unit 24 of the user terminal 15. Image (scan) by (see FIG. 3). The shopping application of the user terminal 15 reads the price information (information including the product name, amount of money, and product ID) of the product included in the information code, and operates the operation display unit 22 of the user terminal 15 (see FIG. 3). ). For example, when the user scans three items a, b, c, the user terminal 15 includes the price information of the three items a, b, c selected as purchase targets in FIG. 2 ( The screen shown in B) is displayed. The read price information is stored in the user terminal 15, is transmitted from the user terminal 15 to the server device 11 through the communication network N1, and is also stored in the server device 11.

The server device 11 reads the purchase history information (an example of the action history of the present invention) corresponding to the user identification information transferred from the gate device 12 from the storage unit 32 (customer information storage unit 321), and will be described later. The recommendation process (see FIG. 7) is executed. In this recommendation process, recommendation information (an example of the recommended information of the present invention) according to the user is obtained from a plurality of product information related to all the products (products that can be sold at the store 200) managed in the server device 11. It is a process to identify. Specifically, product information that will match the taste of the user is extracted based on a predetermined algorithm, and the extracted product information is specified as the recommendation information. Then, the specified recommendation information is transmitted to the user terminal 15 through the communication network N1. In the user terminal 15, the recommendation information is displayed in a pop-up format or the like on the operation display unit 22 (see FIG. 3) of the user terminal 15 so that the user who is buying around can easily notice it.

When paying for the read product (settlement), the user goes to the register 14 and "pays" (Fig. 2 (B)) with the user terminal 15 held in the communication unit provided in the register 14. ) Refer to). As a result, the user terminal 15 communicates with the register 14 through the short-range radio, and the user identification information is transmitted from the user terminal 15 to the register 14. Further, the user identification information transmitted to the register 14 is transferred from the register 14 to the server device 11 through the communication network N1. After that, the server device 11 extracts the price information of the payment target corresponding to the user identification information and transfers it to the register 14. Then, in the register 14, the settlement process of the amount indicated by the price information is executed based on the settlement method designated by the user. When the payment process is completed, the user terminal 15 displays a screen indicating that the payment has been completed.

Note that the settlement process may be performed online between the user terminal 15 and the server device 11 without using the register 14. The settlement process is performed as follows. For example, when the user selects "payment" on the screen of FIG. 2B, the payment process is started. For example, the user terminal 15 transmits the price information of the products a, b, and c to be settled to the server device 11 together with the user identification information. The server device 11 transmits a payment request to a payment server (not shown) connected via the Internet based on the price information. The payment server executes payment processing by a payment system according to the payment business operator. After that, the execution of the payment process is transmitted to the user terminal 15 via the server device 11, and a message to that effect is displayed on the operation display unit 22 (see FIG. 3).

When the payment process is completed, the register 14 transmits information on the purchased product for which payment has been completed to the server device 11. The server device 11 updates the inventory quantity of the product managed by the server device 11 based on the information received from the register 14. Further, the server device 11 updates the purchase history information stored and managed for each user based on the actual purchase information received from the register 14.

When the user leaves the store 200, the user approaches the gate device 13 for leaving the store. When the user approaches the gate device 13, the user terminal 15 communicates with the gate device 13 by short-range wireless communication. At this time, the flag information indicating that the shopping is completed is transmitted to the gate device 13 together with the user identification information of the user, and the identification information and the flag information are transmitted from the gate device 13 to the server via the communication network N1. Transferred to device 11. As a result, the server device 11 recognizes that the shopping has been completed without fraud and that the user has left the store.

As described above, the user can perform shopping using the shopping application at the store 200.

As described above, in the information processing system 100, the recommendation process for specifying the recommendation information according to the user is executed, and the specified recommendation information is displayed on the user terminal 15.

Hereinafter, a specific configuration of the information processing system 100 for realizing the recommendation processing will be described with reference to FIG.

[User terminal 15]
As shown in FIG. 3, the user terminal 15 includes a control unit 21, an operation display unit 22, a storage unit 23, an imaging unit 24, a communication I / F 25, and the like. The user terminal 15 is, for example, a mobile terminal owned by a user of the store 200, for example, a smartphone, a mobile phone, or a tablet terminal. The user terminal 15 may be a mobile terminal lent from the store 200, or may be a cart terminal installed in a shopping cart.

The imaging unit 24 is a digital camera that captures an image of a subject and outputs it as digital image data. A user who makes a purchase in the store 200 takes an image of the information code attached to the product to be purchased by the image pickup unit 24. By reading the captured image of the information code by the reading processing unit 211 described later, price information (product name, amount of money, product ID, etc.) of the product to be purchased can be obtained.

The communication I / F 25 connects the user terminal 15 to the communication network N1 by wire or wirelessly, and executes data communication according to a predetermined communication protocol with an external device such as a server device 11 via the communication network N1. It is a communication interface for In addition, the communication I / F 25 enables the user terminal 15 to perform data communication with devices such as gate devices 12 and 13 and registers 14 via radio (for example, short-range radio such as Bluetooth).

The operation display unit 22 is a user interface including a display unit such as a liquid crystal display or an organic EL display that displays various information, and an operation unit such as a mouse, keyboard, or touch panel that accepts user operations. The recommendation information according to the user is displayed on the display unit of the operation display unit 22.

The storage unit 23 is a non-volatile storage medium such as a flash memory that stores various types of information. The storage unit 23 stores a control program for causing the control unit 21 to execute various processes.

In addition, the storage unit 23 stores user identification information for identifying the user. The user identification information is information for identifying a user, such as an ID number, a name, a gender, and an age. The user identification information is registered in advance in the server device 11 provided in the store 200, and is stored and managed in the storage unit 32 (customer information storage unit 321) in the server device 11 together with the purchase history information.

Further, the storage unit 23 stores the price information of the product to be purchased, which is read by the reading processing unit 211 described later.

The control unit 21 has control devices such as a CPU, a ROM, and a RAM. The CPU is a processor that executes various arithmetic processes. The ROM is a non-volatile storage unit in which control programs such as a BIOS and an OS for causing the CPU to execute various processes are stored in advance. The RAM is a volatile or non-volatile storage unit that stores various types of information, and is used as a temporary storage memory (work area) for various processes executed by the CPU. Then, the control unit 21 controls the user terminal 15 by executing various control programs stored in advance in the ROM or the storage unit 23 on the CPU.

Specifically, as shown in FIG. 3, the control unit 21 includes various processing units such as a reading processing unit 211, a display processing unit 212, a notification processing unit 213, and an action result transmission unit 214. The control unit 21 functions as the various processing units by executing various processes according to the control program on the CPU. Further, a part or all of the processing units included in the control unit 21 may be composed of an electronic circuit. The control program may be a program for causing a plurality of processors to function as the various processing units.

The reading processing unit 211 reads the price information (product name, amount of money, product ID, etc.) of the product included in the information code based on the image data of the information code captured by the imaging unit 24. Hereinafter, for convenience of explanation, the price information read as a purchase target is referred to as purchase target information. The read purchase target information is stored in the storage unit 23 until it is canceled or the settlement of the purchase target product is completed.

The display processing unit 212 causes the display unit of the operation display unit 22 to display the recommendation information transmitted from the server device 11. For example, the display processing unit 212 displays the recommendation information on the display unit of the operation display unit 22 in a pop-up format. When the recommendation information includes a product image, a product name, or the like, the display processing unit 212 displays such information on the display unit of the operation display unit 22 in a pop-up format. The product image or product name is linked to, for example, an address in which more detailed information on the recommended product (for example, product description, product price, product display location in the store 200) is stored, and is used. When a person touches the product image or product name on the display screen, the display processing unit 212 reads out the detailed information stored in the address and displays it on the display unit. Further, when the recommendation information includes the URL of the WEB site including the photographic image and the description of the product, the display processing unit 212 displays the URL on the display unit of the operation display unit 22. In this case, when the user touches the URL, the display processing unit 212 accesses the URL and displays the WEB site on the display unit.

The notification processing unit 213 transmits various information to the server device 11, the gate devices 12, 13 and the register 14. For example, the notification processing unit 213 transmits the user identification information or the like to the gate device 12 when the user enters the store 200, and uses the gate device 13 when the user leaves the store 200. Send person identification information, etc. In addition, the notification processing unit 213 transmits the purchase target information read during the purchase to the server device 11.

The action result transmission unit 214 sets the action result indicating whether or not the user has acted on the recommendation information displayed on the display unit of the operation display unit 22 as the ranking determination unit 33 of the server device 11 (see FIG. 3). ). The actions include, for example, an action of browsing the recommendation information displayed on the display unit of the operation display unit 22 (browsing action), and an action of reading the product information of the product information code indicated by the recommendation information by the reading processing unit 211. (Reading behavior), or the behavior of purchasing the product indicated by the recommendation information displayed on the display unit (purchasing behavior), and the like can be mentioned. The action result includes a positive result indicating that each of the actions has occurred, or a negative result indicating that each of the actions has not occurred.

The control unit 21 performs a process of determining whether or not the user has taken an action with respect to the recommendation information displayed on the display unit of the operation display unit 22. The control unit 21 that performs such a determination process can be regarded as an action determination unit that determines whether or not the user has acted on the recommendation information.

For example, in the control unit 21, whether or not the product image or the product name is touch-operated between the time when the recommendation information is displayed on the display unit of the operation display unit 22 and the time when a certain time (time-up time) elapses. Or, a process of determining whether or not the URL has been touch-operated is performed. In this case, if it is determined that the touch operation is performed within the fixed time, the action result transmission unit 214 is the information of the affirmative result (action result) indicating that the user has browsed the recommendation information on the user terminal 15. Is transmitted to the ranking determination unit 33 of the server device 11.

In addition, whether or not the product information of the product information code indicated by the recommendation information has been read by the reading processing unit 211 until the fixed time elapses or the settlement processing is performed. Performs the process of determining whether or not. In this case, when it is determined that the product information of the product information code indicated by the recommendation information has been read, the action result transmission unit 214 is scheduled to purchase the product indicated by the recommendation information by the user of the user terminal 15. Information on the affirmative result (action result) indicating that there is is transmitted to the ranking determination unit 33 of the server device 11.

In addition, the control unit 21 performs a process of determining whether or not the product indicated by the recommendation information has been purchased. In this case, when the post-purchase product that has been settled includes the product indicated by the recommendation information, the action result transmission unit 214 shows an affirmative result indicating that the product indicated by the recommendation information has been purchased. The information (action result) is transmitted to the ranking determination unit 33 of the server device 11.

In addition, when the browsing action or the reading action is not performed in the fixed time, the action result transmitting unit 214 does not perform the reading action until the user performs the settlement process in the register 14. When the purchase action is not performed in the settlement process, information on a negative result (action result) indicating that each action has not occurred is transmitted to the ranking determination unit 33 of the server device 11.

[Server device 11]
The server device 11 includes a control unit 31, a storage unit 32, a ranking determination unit 33 (an example of a ranking determination unit of the present invention), a discount determination unit 34 (an example of a discount determination unit of the present invention), a communication I / F35, and the like. It is a server computer. The server device 11 is not limited to one computer, and may be a computer system in which a plurality of computers operate in cooperation with each other. Further, various processes executed by the server device 11 may be distributed and executed by one or a plurality of processors.

The communication I / F 35 connects the server device 11 to the communication network N1 by wire or wirelessly, and is predetermined between the user terminal 15, the gate devices 12, 13 and the register 14 via the communication network N1. It is a communication interface for executing data communication according to a communication protocol.

The storage unit 32 is a non-volatile storage unit including a semiconductor memory, an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like that stores various types of information. In the present embodiment, the configuration in which the storage unit 32 is provided in the server device 11 is illustrated. For example, a part or all of various information in the storage unit 32 is transmitted to the server device 11 through the communication network N1 or the Internet. It may be stored in an external device such as another server device or storage device capable of data communication with. In this case, the server device 11 reads necessary information from the external device or stores the information in the external device, if necessary.

The storage unit 32 stores a control program for causing the control unit 31 to execute various control processes and the recommendation process (recommendation extraction process, recommendation selection process) described later shown in FIG. 7. For example, the control program is non-temporarily recorded on a computer-readable recording medium such as a CD or DVD, and is read by a reading device (not shown) such as a CD drive or a DVD drive included in the server device 11. It is stored in the storage unit 32.

The storage unit 32 is divided into a plurality of storage areas. Specifically, the storage unit 32 includes a customer information storage unit 321 (an example of an action history storage unit of the present invention), a product information storage unit 322, a planned purchase product storage unit 323, and a priority storage unit 324 (of the present invention). An example of a priority storage unit) is provided.

The customer information storage unit 321 stores information registered in advance for each user (hereinafter, referred to as customer information). The customer information includes purchase history information regarding products that the user has purchased at the store 200 in the past, in addition to the user identification information including the user's ID number, name, gender, age, and the like. That is, the purchase history information is information indicating the user's behavior history with respect to the product sold at the store 200.

The customer information is stored (registered) as the user identification information in the customer information storage unit 321 of the storage unit 32 when the user registers as a member of the store 200, for example. After that, when the user purchases a product in the store 200, the customer is in a state where the purchase history information including the product information and the purchase date related to the purchased product is linked to the user identification information of the purchased user. It is overwritten and stored in the information storage unit 321. That is, the customer information is updated. The purchase history information is used for the recommendation extraction process by the recommendation extraction process unit 311 described later.

The product information storage unit 322 stores product information related to all products that can be sold at the store 200. The product information is information that can identify the product, and includes, for example, price information including a product name, a price of the product, and a product ID (JAN code, etc.). In addition, the product information includes the type of product, the life of the product (expiration date, expiration date, expiration date, etc.), the presence or absence of a discount, the discount rate, the display location of the product, the photograph and description of the product, and the like. Information such as a URL includes information that is convenient for the user when shopping at the store 200. If the product is a cooked food (processed food) such as prepared foods, the cooking time (including the cooking date and scheduled cooking time), which is the processing date and time of the cooked food, and the state after cooking (processed state). Information indicating the above may be included in the product information. Further, when the product is, for example, the target of the sales promotion campaign, the product information may include information for determining whether or not the product is the target of the sales promotion campaign. Furthermore, the product information may include information that should be grasped only by the seller of the store 200, for example, the profit margin of the product, the degree of disposal risk, the number of stocks, and the like.

In addition, the product information storage unit 322 stores the past sales record and the past disposal history for each product. The past sales record includes the date and time when the product was sold in the past, the price, the number of sales, the presence or absence of a discount, the discount rate, and the like. In addition, the past disposal history includes the presence or absence of disposal of the unsold product, the date and time of disposal performed in the past, the amount of disposal, and the like.

The product information stored in the product information storage unit 322 is updated every time a change occurs. For example, when the user pays for the product (settlement), the inventory quantity of the product is immediately updated by the control unit 31 of the server device 11. In addition, the price of the product, the number of stocks, the presence or absence of a discount, the discount rate, the cooking time, the presence or absence of a sales promotion campaign, the exhaust risk, the increase in the number of stocks due to arrival, etc. are based on the information input by the employee from the employee terminal 16. Is updated by the control unit 31.

The purchase target information sequentially transmitted from the user terminal 15 is temporarily stored in the product storage unit 323 to be purchased each time the reading process is performed by the reading processing unit 211. The planned purchase product storage unit 323 functions as a virtual shopping cart. The purchase target information is stored in the planned purchase product storage unit 323 until it is canceled or the settlement of the purchase target product is completed.

The priority storage unit 324 stores a plurality of product information as well as priorities relating to the sale of the plurality of products. FIG. 6 is a ranking list TB1 showing the priority of each of the plurality of products. In the ranking list TB1, among all the products stored in the product information storage unit 322, a plurality of products (hereinafter referred to as priority products) that the store 200 side (seller) desires to sell with priority are included. It is included and prioritizes sales for each priority product. The priority order indicates the order of the products for which sales are desired to be promoted, and the product having the highest order is the product for which sales are most desired to be promoted in the store 200. The priority may be set for each of the products stored in the product information storage unit 322.

The priority of each of the priority products in the ranking list TB1 is determined based on the fluctuation information regarding the fluctuation factors that cause fluctuations in the priority. The fluctuation information includes, for example, the price, type, life span, presence / absence of discount, discount rate of the priority product, cooking time of the cooked food such as prepared food, and whether or not the product is subject to the sales promotion campaign. Information such as profit margin, degree of disposal risk, inventory quantity, past sales performance, and past disposal history. Further, when the priority product is a cooked food such as "yakitori" represented by prepared foods, the fluctuation information includes a cooking date, a cooking time, and a state after cooking. In addition, the fluctuation information includes temperature, humidity, atmospheric pressure, and weather at the location of the store 200. In addition, the fluctuation information includes the current time, the number of visitors to the store 200 on the day, and the visitor history. The store visitor history is statistical information indicating the number of visitors by month or day of the week in the past. In addition, the number of visitors includes the expected number of visitors predicted from the visitor history. It can be said that such fluctuation information is information that causes fluctuations in the situation (product status) surrounding the product, including the state of the product and the purchase motivation of the user. In determining the priority order of the priority product, the variation information used in the determination process may include any one or more of the above-mentioned specific plurality of variation elements.

The priority order varies depending on the product status and various factors on the store 200 side (seller). For example, a product with a large inventory is given a higher priority, and a product with a small inventory is given a lower priority. This is because the larger the number of stocks, the more likely it is that unsold products will occur. In addition, products with a short product life are set at the top, and products with a long product life are set at the bottom. Products with a high rate of return are ranked high, and products with a low profit margin are ranked low. Products with a high disposal risk are ranked high, and products with a low risk of disposal are ranked low. In addition, the priority order varies depending on the type of product. For example, ingredients / foodstuffs and foods (including cooked foods) and beverages that are ingredients for cooking are set at the top, and detergents and stationery are placed at the bottom. It is decided. In addition, even for cooked foods such as prepared foods, the priority order varies depending on what time the cooking time is or the state after cooking, and the elapsed time after cooking is long or until the cooking time. If the waiting time is long, if the state after cooking is bad, it is ranked high, and if the elapsed time is short, the waiting time is short, or the state after cooking is good, it is ranked low.

In addition, the price of the product, the presence or absence of a discount, the discount rate, the past sales record, the past disposal history, the temperature, humidity, atmospheric pressure, and weather at the location of the store 200, the number of visitors, the visitor history, etc. are also priorities. Is a factor that fluctuates. Further, when a plurality of fluctuation information exists, the plurality of fluctuation information described above may affect each other and the priority may change. For example, a large number of visitors are expected on a sunny day, so in this case, the ranking of each priority product is determined by focusing on the effect of the profit margin rather than the effect of the number of inventories. In addition, when the weather is good but the temperature is high, the ranking is determined by emphasizing the type of product. For example, if it is a cooked food, it is presumed that it is difficult to sell from the past sales results. Cooked foods that are kept warm are ranked high, and cold dishes that are presumed to sell well are ranked low.

In the present embodiment, the priority order is determined or updated by the order determination unit 33, which will be described later, based on the plurality of fluctuation information described above. The ranking determination unit 33 will be described later.

The control unit 31 has control devices such as a CPU, a ROM, and a RAM. The CPU is a processor that executes various arithmetic processes. The ROM is a non-volatile storage unit in which control programs such as a BIOS and an OS for causing the CPU to execute various arithmetic processes are stored in advance. The RAM is a volatile or non-volatile storage unit that stores various types of information, and is used as a temporary storage memory (work area) for various processes executed by the CPU. Then, the control unit 31 controls the server device 11 by executing various control programs stored in advance in the ROM or the storage unit 32 on the CPU.

Specifically, as shown in FIG. 3, the control unit 31 includes a recommendation extraction processing unit 311, a recommendation selection processing unit 312, a recommendation information transmission unit 313 (an example of the recommended information output unit of the present invention), and an information acquisition unit 314 ( It includes various processing units such as a priority acquisition unit and an action history acquisition unit) of the present invention. Here, the recommendation extraction processing unit 311 and the recommendation selection processing unit 312 are examples of the recommended information specifying unit of the present invention. The control unit 31 functions as the various processing units by executing various processes according to the control program on the CPU. Further, a part or all of the processing units included in the control unit 31 may be composed of an electronic circuit. The control program may be a program for causing a plurality of processors to function as the various processing units.

The recommendation extraction processing unit 311 executes a recommendation extraction process for extracting recommendation information suitable for the user's taste from a plurality of product information based on the purchase history information of the user stored in the customer information storage unit 321. .. As a control program for executing the recommendation extraction process, the storage unit 32 stores a recommendation engine (recommendation program) that extracts the recommendation information by using a conventionally known predetermined recommendation algorithm that uses the purchase history information. ing. The recommendation extraction processing unit 311 uses the recommendation engine to extract a preset candidate number of two or more recommended products (recommended product candidates, corresponding to the candidate information of the present invention) from a plurality of product information. That is, the recommendation extraction processing unit 311 executes the recommendation extraction processing based on a predetermined algorithm by using the recommendation engine stored in the storage unit 32.

The recommendation selection processing unit 312 executes a recommendation selection process for further selecting a specific recommended product candidate from the plurality of recommended product candidates extracted by the recommendation extraction processing unit 311. A control program for executing the recommendation selection process is stored in the storage unit 32. In the present embodiment, the recommendation selection processing unit 312 is specified from the plurality of recommended product candidates extracted by the recommendation extraction processing unit 311 based on the priority of the priority product stored in the priority storage unit 324. Select the specific recommended product candidate.

In the present embodiment, the configuration in which the recommended information specifying unit of the present invention is realized by the recommendation extraction processing unit 311 and the recommendation selection processing unit 312 is illustrated, but the present invention is not limited to such a configuration. For example, when the recommendation extraction process is executed, both the purchase history information of the user stored in the customer information storage unit 321 and the priority of the priority product stored in the priority storage unit 324 are displayed. It is also applicable to a configuration in which the specific recommendation information is extracted based on the above. In this case, a program that realizes the recommendation selection process is incorporated in the recommendation engine that uses the above-mentioned conventionally well-known predetermined recommendation algorithm.

The recommendation information transmission unit 313 outputs the recommendation information including the product information of the specific recommended product candidate selected by the recommendation selection processing unit 312 to the user terminal 15 through the communication network N1. When the user terminal 15 receives the transmitted recommendation information, the user terminal 15 displays the recommendation information on the operation display unit 22 of the user terminal 15. In addition, the recommendation information transmission unit 313 also transmits the recommendation information to the employee terminal 16.

The information acquisition unit 314 acquires the priority regarding the sale of a plurality of products provided at the store 200 from the priority storage unit 324. Specifically, the information acquisition unit 314 acquires the priority of the necessary products from the priority storage unit 324. More specifically, when the product information of a plurality of recommended product candidates presumed to match the user's taste is selected by the recommendation extraction process by the recommendation extraction processing unit 311, the information acquisition unit 314 is selected. The priority of a plurality of product information is acquired from the priority storage unit 324. The acquired priority order is used by the recommendation selection processing unit 312.

In addition, the information acquisition unit 314 acquires the user's purchase history information for the product provided in the store 200 from the customer information storage unit 321. Specifically, when the user identification information of the user is sent from the gate device 12 when the user visits the store, the information acquisition unit 314 transfers the customer information corresponding to the user identification information to the customer information. The purchase history information extracted from the storage unit 321 and included in the customer information is extracted. The acquired purchase history information is transferred to the recommendation extraction processing unit 311.

[Ranking unit 33]
Hereinafter, the ranking determination unit 33 will be described with reference to FIG. The ranking determination unit 33 is a learning device configured to determine the priority by machine learning, and has a learning model for determining the priority. Here, FIG. 4 is a block diagram showing the configuration of the ranking determination unit 33, and shows an example in which reinforcement learning is applied as a machine learning algorithm.

As shown in FIG. 4, the ranking determination unit 33 learns from the fluctuation information observation unit 331 (an example of the fluctuation information acquisition unit of the present invention) and the action result acquisition unit 332 (an example of the action result acquisition unit of the present invention). A unit 40 (an example of a ranking learning unit of the present invention) and a data update processing unit 333 are provided. The ranking determination unit 33 can use a general-purpose CPU, but uses, for example, GPGPU (General-Purpose computing on Graphics Processing Units), a large-scale PC cluster, or the like in order to enable higher-speed arithmetic processing. Is desirable.

The fluctuation information observation unit 331 acquires the fluctuation information used for determining the priority of each of the priority products shown in the ranking list TB1 of FIG. The fluctuation information is the price of the product, the type of the product, the life of the product (expiration date, expiration date, expiration date, etc.), whether or not there is a discount, the discount rate, the cooking time or scheduled cooking time of the cooked food, and the target of the sales promotion campaign. In the case of the determination information of whether or not the product, the profit margin of the product, the degree of disposal risk, the number of stocks, the past sales record, and the past disposal history, the fluctuation information observation unit 331 performs the product information in the product information storage unit 322. Is monitored, and real-time information is constantly acquired from the product information storage unit 322.

When the fluctuation information is the temperature, humidity, and atmospheric pressure at the location of the store 200, the fluctuation information observation unit 331 is output from the temperature sensor, humidity sensor, and atmospheric pressure sensor provided indoors or outdoors of the store 200. Receives the sensor output value and acquires the real-time temperature, humidity, and atmospheric pressure indicated by the sensor output value. When the fluctuation information is the weather around the store 200, the control unit 31 downloads the weather information from the weather information database when the server device 11 is connected to the Internet, and the fluctuation information observation unit 331 is connected to the control unit 31. Receive the weather information.

The acquired fluctuation information is information used for learning of the learning unit 40, and is stored in a storage unit (not shown) in the ranking determination unit 33. The fluctuation information is input from the storage unit to the learning unit 40.

The action result acquisition unit 332 acquires the action result information transmitted from the action result transmission unit 214 of the user terminal 15. That is, the action result acquisition unit 332 acquires the action result (affirmative result, negative result) indicating whether or not the user has browsed or purchased the recommended product indicated by the recommendation information. The acquired action result information is information used for learning of the learning unit 40, and is stored in a storage unit (not shown) in the ranking determination unit 33. The action result is input from the storage unit to the learning unit 40.

The learning unit 40 learns the priority order based on the fluctuation information acquired by the fluctuation information observation unit 331. In the present embodiment, the learning unit 40 uses not only the acquired fluctuation information but also the action result acquired by the action result acquisition unit 332, and the store 200 side preferentially sells the product in the priority order. Machine learning. The learning unit 40 will be described later.

The data update processing unit 333 refers to the priority order learned by the learning unit 40, and refers to the priority order stored in the priority order storage unit 324, and in detail, each of the priority products shown in the order list TB1 of FIG. Update the priority of to the order after learning.

The learning unit 40 extracts useful rules, rules, knowledge representations, judgment criteria, etc. in the set of the input fluctuation information by analysis, outputs the judgment result, and learns the knowledge. It has a function. Machine learning includes algorithms such as supervised learning, unsupervised learning, and reinforcement learning, and in order to realize these methods, the feature quantity itself is extracted. A technique called "deep learning" is used for learning.

Here, supervised learning is an algorithm that learns the "relationship between input and output" from data input in advance. The input data is given the correct answer of the data together with the input value, and by giving a large amount of such data to the learning unit 40, the learning unit 40 learns the features in those data. , Estimate the output (result) from the input. If the relationship between given input / output data can be learned, the output can be predicted by applying the relationship to unknown data.

Unsupervised learning is an algorithm that learns features such as the structure, characteristics, and new knowledge of data from a large amount of input data without being given output data that is the correct answer. It differs from supervised learning in that the correct answer is not given to the data that is the source of learning.

In addition, intensive learning is not learning based on fixed and clear data such as supervised learning and unsupervised learning, but the program itself observes a given environment (current state) and is a continuous series. It is an algorithm that evaluates the behavior of the above and learns the appropriate behavior based on the interaction that the behavior gives to the environment, that is, the behavior to maximize the reward obtained in the future. TD learning and Q-learning are known as typical methods.

In reinforcement learning, the consequences of behavior are unknown. Therefore, when reinforcement learning is adopted as machine learning of the learning unit 40, learning starts from a state in which the priority of each product to be learned is not known at all, and in the initial stage, the priority obtained by learning is obtained. The ranking may not be the ranking desired by the store 200. Therefore, it is desirable to perform pre-learning by supervised learning, set the pre-learned priority as the initial state, and then perform reinforcement learning.

For example, when the reinforcement learning in the learning unit 40 is Q-learning, the general update formula (behavior value table) of the action value function Q (s, a) is expressed by the following formula (1).

In the equation (1), _{s t} represents the environment at time t, _{a t} represents the action at time t. By the action _{a t,} the environment is changed to _{s t + 1.} rt _{+ 1} is a reward that can be obtained by changing the environment. The term with max is the Q value multiplied by γ when the action a with the highest Q value known at that time is selected under _{the environment st + 1.} Further, γ is called a discount rate, and is a parameter of 0 <γ ≦ 1. Further, α is a learning coefficient in the range of 0 <α ≦ 1. When Q-learning is applied, the behavior of browsing the recommendation information displayed on the user terminal 15, reading the product information code indicated by the recommendation information, or the user purchasing the product (browsing behavior, reading behavior, purchase behavior) becomes the action _{a t.}

As shown in FIG. 4, the learning unit 40 includes a reward calculation unit 401 (an example of a reward output unit of the present invention) and a value function update unit 402 (an example of a learning model update unit of the present invention). The reward calculation unit 401 calculates and outputs a reward based on the fluctuation information acquired by the fluctuation information observation unit 331. Further, the value function update unit 402 updates the value function related to the priority based on the fluctuation information and the reward calculated by the reward calculation unit 401. This value function corresponds to the learning model.

As described above, the fluctuation information changes in the priority order, but the fluctuation information itself changes from moment to moment. Therefore, when the fluctuation information changes, the priority order is also updated. Whether or not the priority after the change is appropriate can be evaluated by whether the action result of each action of browsing action, reading action, and purchasing action is a positive result or a negative result. Thus, for example, the case of applying the Q-learning, action result (positive result, a negative result) is action a _t in substantially formula (1), action value represented by the formula (1) function Q (s the a), is used as a value function for changing the action a _t.

In the present embodiment, the reward calculation unit 401 outputs a positive reward when the action result is a positive result indicating that the recommended product indicated by the recommendation information has been viewed or purchased, and the action result is the recommended product. Outputs a negative reward if the result is negative, indicating that the item was not viewed or purchased. The absolute value of the output reward may differ depending on the content of the action result. For example, the reward for the behavioral result of the browsing behavior may be set to the lowest, and the reward corresponding to the behavioral result of the purchasing behavior may be set to the highest. Further, a weighting coefficient for the reward may be set for each user. For example, when the reward based on the action result of the user who frequently uses the store 200 is emphasized, the weighting coefficient multiplied by the reward is used as the usage frequency. It may be set to a proportional value.

[Discount determination unit 34]
Hereinafter, the discount determination unit 34 will be described with reference to FIG. The discount determination unit 34 determines whether or not to carry out a discount on a low-ranked product whose priority is lower than a predetermined reference rank (10th in the present embodiment) among a plurality of products of the same type. It is a learning device configured to judge by learning. Here, FIG. 5 is a block diagram showing the configuration of the discount determination unit 34, and shows an example in which reinforcement learning is applied as a machine learning algorithm. Since the configuration of the discount determination unit 34 is the same as the configuration of the ranking determination unit 33 except for the determination result notification unit 341, only the different configurations will be described in detail below, and the similar configuration will be described in detail in the ranking determination unit 33. The detailed description thereof will be omitted by adding the same reference numerals as the above.

In the present embodiment, the discount not only further reduces the selling price of the target product, but also includes an act corresponding to a substantial discount of the target product. For example, giving a coupon (gold ticket, discount ticket) when the target product is purchased, giving points that can be exchanged for the product, and the like correspond to the above discount.

As shown in FIG. 5, the discount determination unit 34 learns from the fluctuation information observation unit 331 (an example of the deterioration information acquisition unit of the present invention) and the action result acquisition unit 332 (an example of the action result acquisition unit of the present invention). A unit 40 (an example of a determination learning unit of the present invention) and a determination result notification unit 341 are provided. That is, the discount determination unit 34 is different from the ranking determination unit 33 in that the determination result notification unit 341 is provided instead of the data update processing unit 333.

For example, in the ranking list TB1 of FIG. 6, we consider a case where two products of the same type "yakitori" are ranked within the 10th place (standard ranking) and also exist in the 19th place, which is lower than the 10th place. To do. For example, for the product "Yakitori" ranked within the 10th place, since the priority is high, there is a high possibility that the user who is buying around the store 20 will be notified as the recommendation information, and the product will be used. It is more likely to be purchased by a person than the other-ranked product "Yakitori". On the other hand, the 19th-ranked product "Yakitori" (corresponding to the low-ranking product), which is lower than the standard ranking, has a lower ranking, and therefore is unlikely to be notified to the user as the recommendation information. In this case, at present, even if the lower-ranked product (low-ranked product) is not a product that the store 200 side (seller) wants to sell preferentially, the ranking determination unit 33 depends on the fluctuation content of the fluctuation information. It may be necessary to improve the sales situation without waiting for the rank up or down (rank change) after the ranking is decided.

Therefore, in the discount determination unit 34, the fluctuation information observation unit 331 acquires the fluctuation information as described above, but acquires the fluctuation information as the deterioration information that causes the deterioration of the product. Further, the learning unit 40 refers to a product (low-ranked product) whose priority is lower than the reference rank among a plurality of products of the same type based on the fluctuation information acquired by the fluctuation information observation unit 331. Machine learning whether or not to implement a discount.

By the way, the products handled in the store 200 include short-lived products that cannot be stored for a long period of time, such as fresh foods and prepared foods. In addition, daily necessities such as detergents and stationery, processed foods such as canned foods and instant noodles that can be stored for a long time, confectionery and luxury items, seasonings, alcoholic beverages such as liqueurs, and other products with long expiration dates and expiration dates. Is also included. Therefore, even if the low-ranked products are long-life products such as daily necessities, there is little need to discount them even if the number of stocks is large. On the other hand, for short-lived products such as fresh foods and prepared foods, if the number of stocks is short but the product is near the end of its life, it is highly necessary to sell out even if it is discounted in order to avoid the risk of unsold waste. Therefore, the discount determination unit 34 extracts a low-life product having a life shorter than a predetermined threshold value from the low-order / non-order products, and machine-learns whether or not to carry out a discount for the short-life product. It is preferable to do so. That is, in the present embodiment, the discount determination unit 34 machine-learns whether or not to carry out the discount by excluding the products having a life longer than the predetermined threshold value from the low-ranked products. The predetermined threshold value is arbitrarily set according to the target product, for example, one day for prepared foods and several days depending on the type of food for fresh foods. To.

The determination result notification unit 341 performs a process of transmitting the determination result to the employee terminal 16 with reference to the determination result learned by the learning unit 40. On the employee terminal 16 that has received this determination result, the determination result is displayed on the display unit of the terminal 16. Therefore, the employee of the store 200 recognizes which product should be discounted by checking the displayed determination result, and promptly discounts the product, that is, updates (replaces) the information code. It can be carried out.

In the above example, the reference ranking is set to 10th, but for example, does the discount determination unit 34 give a discount to each of the products other than the highest ranking product among the plurality of products of the same type? It may be something to learn whether or not. In this case, the ranking of the product having the highest ranking among the plurality of products of the same type corresponds to the reference ranking of the present invention.

Further, in the above-mentioned example, the discount for the product (product) whose priority is lower than the reference order among the plurality of products of the same type is illustrated. For example, the discount determination unit 34 is used for all the products. Among them, it may be learned whether or not to discount each of the products having a lower rank than the reference rank (lower rank products).

Further, not only the determination of whether or not to carry out the discount of the product, but also, for example, when it is determined that the discount is to be carried out, the learning unit 40 of the discount judgment unit 34 is made to learn the discount rate and the execution timing of the discount. You may do it.

Hereinafter, the recommended information identification method of the present invention will be described together with the recommendation process executed by the control unit 31 of the server device 11 using the flowchart of FIG. 7. In the following description, it is assumed that the recommendation process is performed from the time when the user passes through the gate device 12 of the store 200 to the time when the user leaves the store after passing through the gate device 13 of the store 200. .. That is, the recommendation process is performed when the user is buying around in the store 200. In FIG. 7, S11, S12, ... Indicates a processing procedure number (step number).

First, in step S11, the control unit 31 acquires the purchase history information of the user who entered the store from the customer information storage unit 321. Such processing is performed by the information acquisition unit 314 of the control unit 31. Step S11 corresponds to the action history acquisition step of the present invention. Further, in the next step S12, the control unit 31 extracts the action content of the user from the acquired purchase history information.

Next, in step S13, the control unit 31 uses the recommendation engine to perform a process of extracting a plurality of recommended recommended product candidates that will match the user's taste based on the action content. .. Such processing is performed by the recommendation extraction processing unit 311 of the control unit 31. In the present embodiment, the recommendation extraction processing unit 311 determines, for example, five candidates.

In the next step S14, the control unit 31 determines whether or not the high-ranking products belonging to the ranking list TB1 of FIG. 6 are included in the plurality of candidates extracted in step S13. For example, the higher-ranked product may be a product whose priority is within a predetermined reference order (10th in the present embodiment).

In step S14, when it is determined that the higher-ranking product is included in the plurality of candidates (Yes side of S14), in the next step S15, the highest-ranking product is selected from the plurality of candidates. Then, the selected product is specified as a recommended product. Such processing is performed by the recommendation selection processing unit 312 of the control unit 31. Step S15 corresponds to the recommended information specifying step of the present invention. After that, the recommendation information including the recommended product is transmitted to the user terminal 15 of the user by the recommendation information transmission unit 313. On the user terminal 15, the received recommendation information is displayed on the display unit of the operation display unit 22. The recommendation information is also transmitted to the employee terminal 16.

On the other hand, in step S14, when it is determined that the higher-ranking product is not included in the plurality of candidates (No side of S14), the control unit 31 again uses the recommendation engine to perform the action content. Based on this, a process of extracting a plurality of candidates different from the candidates extracted in step S13 is performed (step S17). After that, the process of step S14 is executed, and the processes of steps S14 and S17 are repeated until it is determined that the higher-ranking product is included in the plurality of candidates.

As described above, in the present embodiment, the recommendation information that matches the user's taste is displayed on the user terminal 15, so that the user who is presented with the recommendation information can use the product that matches his / her taste. Information (information on recommended products) can be obtained quickly. In addition, the recommendation information output to the user terminal 15 includes information on recommended products in which the priority of products that the store 200 side (seller) wants to sell preferentially is taken into consideration. Therefore, the possibility that the recommended product is viewed or purchased is increased, and as a result, the store side (seller) can improve the purchase rate of the product that he / she wants to sell preferentially.

Further, since the recommendation information is also transmitted to the employee terminal 16, the employee can confirm the received recommendation information of each user.

In the above-described embodiment, an example of specifying a recommended product by performing the processes of steps S14 and S15 has been described, but the present invention is not limited to such an example. For example, from the plurality of candidates extracted in step S13, the product having the highest recommendation evaluation value, which will be described later, may be specified as the recommended product.

As the recommendation evaluation value, for example, as shown in FIG. 8, numerical values Z1 and Z2 obtained by multiplying a predetermined compatibility coefficient Z1 and a ranking coefficient Z2 for each of the five extracted candidate products A1 to A5 are used. Can be done. Here, the compatibility coefficient Z1 is a numerical value of compatibility (preference) with the user for each of the candidate products A1 to A5, and is based on, for example, data such as purchase history information and product information of the user. It is calculated by the Probabilistic Latent Semantic Analysis (PLSA).

Further, the rank coefficient Z2 is an arbitrary coefficient determined based on a predetermined rule. In the present embodiment, an arbitrary numerical value "0.50" is set for the first-ranked product in the ranking list TB1 (see FIG. 6), and half the numerical value "0" of the first-ranked coefficient is set for the second-ranked product. .25 ”, 1/3 of the coefficient of the 1st place“ 0.16 ”for the 3rd place product, 1/4 of the coefficient of the 1st place“ 0.13 ”, A numerical value "0.10", which is one-fifth of the coefficient of the first place, is set for the product of the fifth place. The rule for setting the rank coefficient Z2 is not limited to the above rule, as long as the rank coefficient Z2 is set to a smaller value as the rank goes from higher to lower. It should be noted that each of the above-mentioned numerical values is displayed with the third decimal place truncated. Here, FIG. 8 illustrates each coefficient and the recommendation evaluation value when the five candidate products extracted in step S13 are products having the first to fifth priority.

In the recommendation process, the control unit 31 specifies the product having the largest numerical value among the recommendation evaluation values defined as described above as the recommended product, instead of the processes in steps S14 and S15. By specifying the recommended product in this way, it is possible to prevent the same product from being presented as a recommended product to the same user when there is no change in the ranking of the higher-ranked products.

The product with the highest recommendation evaluation value is not specified as the recommended product. For example, for each of the five extracted candidate products A1 to A5, the higher the value of the recommendation evaluation value, the higher the value. The selection rate may be set, and the recommended product may be stochastically specified based on each selection rate.

The recommendation evaluation value is obtained by multiplying the numerical values Z1 and Z2 obtained by multiplying the compatibility coefficient Z1 and the rank coefficient Z2 by the noise coefficient Z3 randomly determined for each of the five candidate products A1 to A5. You may use it. The noise coefficient Z3 is a coefficient for varying the recommendation evaluation value. By multiplying the numerical values Z1 and Z2 by the noise figure Z3, the recommended product specified is also affected. That is, the specified recommended product may change before and after the noise figure Z3 is multiplied. The timing for multiplying the noise figure Z3 is, for example, that the cumulative number of times the user has entered the store has reached a predetermined number of times, and that the number of times the user has entered the store has reached a predetermined number of times in the latest set period. , The user returned to the store after a predetermined period of time has passed, and the recommended product indicated by the presented recommendation information has a predetermined number of consecutive positive action results (browsing behavior, reading behavior, etc.) (Purchase behavior) can be the timing when any one or more of the fact that the user did not occur is satisfied. By adding such a noise coefficient Z3 to the recommendation evaluation value, it is possible to give diversity to the recommendation information including the specified recommended product.

The recommendation information having such diversity does not necessarily include the recommended product reflecting the purchase history of the user. However, when a user who browses the recommendation information that does not include the recommended product purchases the product indicated by the recommendation information using the recommendation information as an opportunity, the purchase record of such a product is overwritten in the user's purchase history. To. As a result, at the next timing, the recommendation information including the recommended product is specified by using the purchase history information after overwriting, and the consistency between the recommended product indicated by the recommendation information and the product preferred by the user is improved.

As the recommendation evaluation value, the numerical values Z2 and Z3 obtained by multiplying the rank coefficient Z2 by the noise figure Z3 may be used instead of using the compatibility coefficient Z1.

Hereinafter, an example of the learning operation flow of the learning unit 40 of the ranking determination unit 33 using reinforcement learning will be described with reference to the flowchart of FIG. In FIG. 9, S21, S22, ... Indicates a processing procedure number (step number). Since the learning operation of the learning unit 40 in the discount determination unit 34 is the same as the learning operation flow shown in FIG. 9, the description thereof will be omitted.

First, in step S21, the action result acquisition unit 332 of the ranking determination unit 33 acquires the action result (action result of browsing action, reading action, purchasing action) transmitted from the user terminal 15. In this step S21, the action results of all the users who use the store 200 are transmitted from each of the user terminals 15.

In step S22, the reward calculation unit 401 of the learning unit 40 calculates a determination evaluation value for determining whether or not the current priority is appropriate based on the action result acquired in step S21. The judgment evaluation value is calculated by adding evaluation points determined for each content of the action result. For example, positive evaluation points are set for positive results of browsing behavior, reading behavior, and purchasing behavior, and negative evaluation points are set for negative results. The judgment evaluation value is calculated by adding the evaluation points for all the action results. When a weighting coefficient is set for the evaluation point for each user, for example, a weighting coefficient of a numerical value proportional to the usage frequency is set for the evaluation point for the action result of the user who frequently uses the store 200. If so, the determination evaluation value may be calculated in consideration of the weighting coefficient.

In the next step S23, it is determined whether or not the determination evaluation value is within a predetermined allowable range. If it is determined that the determination evaluation value is within the permissible range, the process proceeds to step S24, assuming that the current priority is appropriate, and a positive reward is set. On the other hand, if it is determined that the determination evaluation value exceeds the permissible range, the process proceeds to step S25 and a negative reward is set, assuming that the current priority is inappropriate.

After that, in step S26, the value function update unit 402 of the learning unit 40 updates the value function related to the priority, that is, the action value function Q (s, a), returns to step S21 again, and repeats the same process. Will be executed.

In the above-described embodiment, the information processing system 100 applied to the actual store 200 is illustrated, but the present invention is for a virtual store existing on the Internet where the user makes a purchase via the user terminal 15. Is also applicable.

11: Server device 12: Gate device 13: Gate device 14: Register 15: User terminal 16: Employee terminal 21: Control unit 22: Operation display unit 23: Storage unit 24: Imaging unit 25: Communication I / F
31: Control unit 32: Storage unit 33: Ranking determination unit 34: Discount determination unit 35: Communication I / F
40: Rank learning unit 100: Information processing system 200: Store 211: Reading processing unit 212: Display processing unit 213: Notification processing unit 214: Action result transmission unit 311: Recommendation extraction processing unit 312: Recommendation selection processing unit 313: Recommendation information Transmission unit 314: Information acquisition unit 321: Customer information storage unit 322: Product information storage unit 323: Purchased product storage unit 324: Priority storage unit 331: Fluctuation information observation unit 332: Action result acquisition unit 333: Rank update processing unit 401: Reward calculation unit 402: Value function update unit TB1: Ranking list

Claims (16)

Based on the user's behavior history for the products provided in the store and the predetermined priorities regarding the sale of the plurality of products provided in the store, the user is responded to from the plurality of product information related to the plurality of products. Recommended information identification department that identifies recommended information including recommended products,
It is provided with a discount determination unit configured to determine by machine learning whether or not to implement a discount on a low-ranked product whose priority is lower than a predetermined reference rank.
The discount determination unit
A deterioration information acquisition unit that acquires deterioration information including the life of the product, which is information that causes fluctuations in the deterioration of the product.
Based on the lifespan acquired by the deterioration information acquisition unit, a short-life product having a life shorter than a predetermined threshold value is extracted from the low-ranked products, and the short-life product is based on the deterioration information. A recommended information identification device having a judgment learning unit for learning the judgment of whether or not to carry out a discount. Further, a ranking determination unit configured to determine the priority by machine learning is provided.
The ranking determination unit
A fluctuation information acquisition unit that acquires fluctuation information that causes fluctuations in the priority order,
The recommended information specifying device according to claim 1, further comprising a ranking learning unit that learns the priority based on the fluctuation information acquired by the fluctuation information acquisition unit. The fluctuation information is
Price, type, life of the product, presence / absence of discount, discount rate, profit margin, disposal risk, inventory quantity, past sales record, and past disposal history, processing date and time and processing state when the product is processed food , The recommended information identifying device according to claim 2, which includes at least one of the current time, the temperature, humidity, atmospheric pressure, and weather at the location of the store, and the number of visitors to the store. The ranking determination unit
Further having an action result acquisition unit for acquiring an action result indicating whether or not the user has browsed or purchased the recommended product indicated by the recommended information.
The second or third claim, wherein the rank learning unit learns the priority based on the fluctuation information acquired by the fluctuation information acquisition unit and the action result acquired by the action result acquisition unit. Recommended information identification device. The rank learning unit has a learning model for learning the priority, and has a learning model.
The ranking learning department
If the action result indicates that the recommended product has been viewed or purchased, a positive reward is output, and if the action result indicates that the recommended product has not been viewed or purchased, a negative reward is output. Reward output unit that outputs the reward of
The recommended information specifying device according to claim 4, further comprising a learning model updating unit that updates the learning model based on the reward output by the reward output unit. The recommended information specifying unit specifies, as the recommended information, the recommended product having the highest priority from a plurality of candidate information extracted from the plurality of product information based on the action history. The recommended information identification device according to any one of 5 to 5. The recommended information specifying unit has the highest evaluation value obtained by multiplying the noise coefficient randomly determined for each of the plurality of candidate information from the plurality of candidate information by the priority of each of the plurality of candidate information. The recommended information specifying device according to claim 6, wherein a product including a recommended product is specified as the recommended information. According to any one of claims 1 to 7, the terminal device used by the user when purchasing a product at the store is further provided with a recommended information output unit that outputs the recommended information specified by the recommended information specifying unit. Recommended information identification device described. The deterioration information is
At the price, type, profit margin, disposal risk, inventory quantity, past sales record, and past disposal history of the product, if the product is a processed food, the processing date and time, the current time, and the location of the store. The recommended information identifying device according to claim 1, further comprising temperature, humidity, barometric pressure, and weather, and at least one or more of the number of visitors to the store. A priority acquisition unit that acquires the priority from a priority storage unit that stores priorities related to the sale of a plurality of products provided at the store, and a priority acquisition unit.
The behavior history acquisition unit that acquires the behavior history from the behavior history storage unit that stores the user's behavior history for the products provided in the store, and the behavior history acquisition unit.
Based on the action history acquired by the action history acquisition unit and the priority acquired by the priority acquisition unit, a recommended product corresponding to the user is included from a plurality of product information related to the plurality of products. The recommended information identification department that identifies recommended information, and
It is provided with a discount determination unit configured to determine by machine learning whether or not to implement a discount on a low-ranked product whose priority is lower than a predetermined reference rank.
The discount determination unit
A deterioration information acquisition unit that acquires deterioration information including the life of the product, which is information that causes fluctuations in the deterioration of the product.
Based on the lifespan acquired by the deterioration information acquisition unit, a short-life product having a life shorter than a predetermined threshold value is extracted from the low-ranked products, and the short-life product is based on the deterioration information. A recommended information identification system having a judgment learning unit that learns the judgment of whether or not to carry out a discount. Further, a ranking determination unit configured to determine the priority by machine learning is provided.
The ranking determination unit
A fluctuation information acquisition unit that acquires fluctuation information that causes fluctuations in the priority order,
The recommended information identification system according to claim 10, further comprising a ranking learning unit that learns the priority based on the fluctuation information acquired by the fluctuation information acquisition unit. The behavior history acquisition step to acquire the user's behavior history for the products provided in the store,
Based on the acquired behavior history and predetermined priorities regarding the sale of a plurality of products provided at the store, a recommendation including a recommended product according to the user from a plurality of product information related to the plurality of products is recommended. Recommended information identification steps to identify information and
One or more processors execute a discount determination step configured by machine learning to determine whether or not to implement a discount on a low-ranked product whose priority is lower than a predetermined reference order. ,
The discount determination step is
Information that causes fluctuations in the deterioration of the product, such as deterioration information including the life of the product, is acquired, and based on the acquired life, the short life of the low-ranked product is shorter than a predetermined threshold value. A recommended information identification method for extracting a product and learning a determination as to whether or not to carry out a discount based on the deterioration information for the short-life product. A program for causing one or more processors to execute each step of the recommended information identification method according to claim 12. A terminal device used by users when purchasing products in stores.
A display processing unit that receives the recommended information specified by the recommended information specifying unit of the recommended information specifying device according to claim 4 or 5 and displays the recommended information on a predetermined display unit.
A terminal device including an action result transmitting unit that transmits an action result indicating whether or not the user has taken an action with respect to the recommended information displayed on the display unit to the recommended information specifying device. It is a method applied to the terminal device used by the user when purchasing products in the store.
A display step of receiving the recommended information specified by the recommended information specifying unit of the recommended information specifying device according to claim 4 or 5 and displaying the recommended information on a predetermined display unit.
One or a plurality of processors execute an action result transmission step of transmitting an action result indicating whether or not the user has taken an action with respect to the recommended information displayed on the display unit to the recommended information specifying device. Method. A program for causing one or more processors to perform each step of the method according to claim 15.

Images