JP7514123B2 - Information processing method, program, and information processing device - Google Patents

Description

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

There are technologies that provide beverage products according to consumer needs. For example, Patent Document 1 discloses a recommendation device that extracts specific keywords from language data entered by a user, selects sake products to recommend to the user based on the extracted keywords, and outputs information about the selected products.

JP 2019-139663 A

However, the invention described in Patent Document 1 selects recommended products from keywords according to rules created manually, and the accuracy of the recommendations is not necessarily high.

In one aspect, the objective is to provide an information processing method etc. that can suggest beer beverages that are suitable for customers.

In one aspect of the information processing method, a computer executes a process in which a customer's attributes are acquired, answers to a plurality of questions are acquired, and the acquired attributes and answers are input into a first model that has been trained to output a plurality of characteristic values indicating the characteristics of a beer beverage including beer, happoshu, new genre beer or a beer-flavored beverage when the attributes and answers are input, the acquired attributes and answers are input, the plurality of characteristic values are output, and a memory unit that stores the plurality of characteristic values and a serving start date in association with each beer beverage is referenced to calculate a similarity between the plurality of characteristic values of each beer beverage and the plurality of characteristic values output from the first model, and the beer beverage to be proposed to the customer is selected based on the calculated similarity and the serving start date .

One aspect is that it allows us to suggest beer drinks that are suitable for customers.

FIG. 1 is an explanatory diagram illustrating an example of the configuration of a beverage recommendation system. FIG. 2 is a block diagram showing an example of the configuration of a server. FIG. 2 is a block diagram showing a configuration example of a terminal. 3 is an explanatory diagram showing an example of a record layout of a beverage DB, a customer DB, a question table, and a history DB. FIG. 1 is an explanatory diagram showing an overview of a first embodiment. FIG. 11 is an explanatory diagram showing an example of an input screen. FIG. 13 is an explanatory diagram showing an example of a proposal screen. FIG. 13 is an explanatory diagram showing another example of the proposal screen. FIG. 13 is an explanatory diagram showing an example of an input screen for a second questionnaire. 13 is a flowchart showing a procedure for generating a first model. 13 is a flowchart showing a procedure for a beverage suggestion process. 13 is a flowchart showing a procedure for updating a first model. FIG. 11 is a block diagram showing a configuration example of a server according to the second embodiment. FIG. 11 is an explanatory diagram showing an overview of a second embodiment. 13 is a flowchart showing a procedure for generating a second model. 13 is a flowchart showing a procedure of a beverage suggestion process according to the second embodiment.

The present invention will now be described in detail with reference to the drawings showing embodiments thereof.
(Embodiment 1)
1 is an explanatory diagram showing an example of the configuration of a beverage recommendation system. In this embodiment, a beverage recommendation system that recommends beer beverages suited to a customer's preferences based on responses to a questionnaire entered by the customer will be described. The beverage recommendation system includes an information processing device 1 and terminals 2, 2, 2, . . . Each device is communicatively connected to a network N such as the Internet.

The information processing device 1 is an information processing device capable of various information processing and information transmission and reception, such as a server computer, a personal computer, etc. In this embodiment, the information processing device 1 is a server computer, and for simplicity, will be referred to as server 1 below. The server 1 functions as an estimation device that estimates beer information regarding a beer beverage to be proposed to a customer. Specifically, as described below, the server 1 estimates characteristic values of a beer beverage to be proposed to a customer from multiple attributes of the customer and the customer's answers to multiple questions (questionnaires) using a first model 51 described below generated by machine learning. The server 1 searches a database for beer beverages similar to the estimated characteristic values and proposes them to the customer.

In this specification, the beer beverage includes not only beer, but also happoshu, new genre beer (a beer-flavored sparkling alcoholic beverage made from ingredients or using a manufacturing method different from beer and happoshu), and beer-flavored beverages (beer-flavored sparkling carbonated beverages, including non-alcoholic beverages). Note that this system may also be applied to beverages other than beer beverages (alcoholic beverages other than beer beverages, or beverages other than alcoholic beverages).

The terminal 2 is a terminal device used by each customer, such as a smartphone, a personal computer, or a tablet terminal. The terminal 2 may be a terminal device owned by the customer, or may be a terminal device loaned to the customer by a beer beverage providing facility such as a restaurant or retail store. The server 1 acquires questionnaire responses from the terminal 2, and outputs and displays information on the beer beverage that is estimated to be recommended to the terminal 2.

2 is a block diagram showing an example of the configuration of the server 1. The server 1 includes a control unit 11, a main memory unit 12, a communication unit 13, and an auxiliary memory unit .
The control unit 11 has one or more arithmetic processing devices such as a central processing unit (CPU), a micro-processing unit (MPU), a graphics processing unit (GPU), etc., and performs various information processing, control processing, etc. by reading and executing a program P1 stored in the auxiliary storage unit 14. The main storage unit 12 is a temporary storage area such as a static random access memory (SRAM), a dynamic random access memory (DRAM), a flash memory, etc., and temporarily stores data necessary for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing processing related to communication, and transmits and receives information to and from the outside.

The auxiliary storage unit 14 is a non-volatile storage area such as a large-capacity memory or a hard disk, and stores the program P1 and other data necessary for the control unit 11 to execute processing. The auxiliary storage unit 14 also stores the first model 51, the beverage DB 141, the customer DB 142, the question table 143, and the history DB 144. The first model 51 is a machine learning model that has learned predetermined training data, and is a model that inputs multiple attributes of a customer and answers to multiple questions and outputs beer information about beer beverages to be proposed to the customer. The first model 51 is expected to be used as a program module that functions as part of artificial intelligence software. The beverage DB 141 is a database that stores information on each beer beverage. The customer DB 142 is a database that stores information on each customer. The question table 143 is a table that stores the content of questions when taking the questionnaire described below. The history DB 144 is a database that stores the history of beer beverage proposals to customers.

The auxiliary storage unit 14 may be an external storage device connected to the server 1. The server 1 may be a multi-computer consisting of multiple computers, or may be a virtual machine virtually constructed by software.

In the present embodiment, the server 1 is not limited to the above configuration, and may include, for example, an input unit that accepts operational input, a display unit that displays images, etc. The server 1 may also include a reading unit that reads portable storage medium 1a such as a CD (Compact Disk)-ROM or a DVD (Digital Versatile Disc)-ROM, and may read and execute the program P1 from the portable storage medium 1a. Alternatively, the server 1 may read the program P1 from the semiconductor memory 1b.

3 is a block diagram showing an example of the configuration of the terminal 2. The terminal 2 includes a control unit 21, a main memory unit 22, a communication unit 23, a display unit 24, an input unit 25, an imaging unit 26, a GPS (Global Positioning System) receiving unit 27, and an auxiliary memory unit 28.
The control unit 21 has one or more arithmetic processing devices such as a CPU, an MPU, etc., and performs various information processing, control processing, etc. by reading and executing the program P2 stored in the auxiliary storage unit 28. The main storage unit 22 is a temporary storage area such as a RAM, and temporarily stores data necessary for the control unit 21 to execute arithmetic processing. The communication unit 23 is a communication module for performing processing related to communication, and transmits and receives information to and from the outside. The display unit 24 is a display screen such as a liquid crystal display, and displays images. The input unit 25 is an operation interface such as a mechanical key, and accepts operation input from the user. The imaging unit 26 is a camera equipped with an imaging element such as a CMOS (Complementary MOS), and captures images. The GPS receiving unit 27 is a receiver that receives GPS signals, and obtains location information of the terminal 2. The auxiliary storage unit 28 is a non-volatile storage area such as a hard disk, a large-capacity memory, etc., and stores the program P2 and other data necessary for the control unit 21 to execute processing.

The terminal 2 may be provided with a reading unit that reads a portable storage medium 2a such as a CD-ROM, and may read and execute the program P2 from the portable storage medium 2a. Alternatively, the terminal 2 may read the program P2 from the semiconductor memory 2b.

FIG. 4 is an explanatory diagram showing an example of the record layout of the beverage DB 141, the customer DB 142, the question table 143, and the history DB 144.
The beverage DB 141 includes a beverage ID column, a product name column, a description column, a product image column, a manufacturer column, a production region column, a serving start date column, and a characteristic value column. The beverage ID column stores a beverage ID for identifying each beer beverage. The product name column, description column, product image column, manufacturer column, production region column, serving start date column, and characteristic value column store the product name, description, product image, manufacturer (e.g., brewery), production region (e.g., brewery location), serving start date (e.g., product release date), and characteristic value of the beer beverage, respectively, in correspondence with the beverage ID.

The characteristic values include, for example, component values such as original extract, apparent extract, apparent final fermentation degree, apparent fermentation degree, bitterness value, alcohol content, extract content, residual fermentation degree, total nitrogen, pH, color, ethyl acetate, n-propanol, isobutanol, isoamyl alcohol, isoamyl acetate, amyl alcohol, active amyl alcohol, acetaldehyde, diacetyl, amino acids, sulfate ions, phosphate ions, citric acid, succinic acid, malic acid, acetic acid, lactic acid, isobutyric acid, caprylic acid, capric acid, total polyphenols, β-glucan, turbidity, foam retention time, carbon dioxide, linalool, α acid, iso-α acid, S-fraction, etc. The characteristic values also include the amount of alcohol, calories, yeast used (top-fermenting yeast or bottom-fermenting yeast), whether or not specific raw materials (malt, hops, wheat, sugar, spices, fruits, vegetables, rice, miso) are used, and whether or not the product name includes the name of a region in Japan.

Customer DB142 includes a customer ID column, a customer name column, and an attribute column. The customer ID column stores a customer ID for identifying each customer. The customer name and attribute columns store the customer name and multiple attributes of the customer in association with the customer ID, respectively. Customer attributes include, for example, age, gender, and information about usual drinking habits (preferred brand of alcohol, usual amount of alcohol consumed, place of drinking, etc.).

The question table 143 includes a question ID column, a question attribute column, a question column, and an answer column. The question ID column stores a question ID for identifying each question to be presented to the customer. The question attribute column, question column, and answer column each store question attributes, question content (question text), and answer options in association with the question ID. The question attribute is metadata indicating whether the question is presented before a beer beverage is suggested or after a beer beverage is ordered, and indicates whether the question is from the first or second questionnaire described below.

History DB144 includes a date/time column, a customer column, a pre-proposal response column, a proposed beverage column, an order column, and a post-order response column. The date/time column stores the date/time when the beer beverage was proposed. The customer column, pre-proposal response column, proposed beverage column, order column, and post-order response column each store, in association with the proposal date/time, the customer ID of the customer who proposed the beer beverage, the response to the question before the proposal (first questionnaire), the beverage ID of the proposed beer beverage, whether or not the proposed beer beverage was ordered, and the response to the question after the order (second questionnaire).

Fig. 5 is an explanatory diagram showing an overview of the first embodiment. The overview of the present embodiment will be described with reference to Fig. 5.
In this embodiment, the server 1 provides a recommendation service that suggests beer beverages that match the customer's preferences based on the customer's attributes (age, sex, etc.) and answers to questions related to a questionnaire described below. The service is used, for example, in a restaurant that serves beer beverages, but may also be applied to other locations such as retail stores and EC (Electronic Commerce) sites. In this embodiment, the service is provided in a restaurant as an example.

For example, restaurants participating in this service are provided with a two-dimensional code (QR Code (registered trademark), etc.) containing a specific URL (Uniform Resource Locator). Terminal 2 reads the two-dimensional code from an image captured by imaging unit 26, launches a web browser, and accesses the homepage provided by server 1. Terminal 2 then displays the screens shown in Figure 6 and onwards.

In this embodiment, a series of screen displays and operations are performed on a web page, but it is also possible to install a dedicated application on terminal 2 and perform a series of processes on that application, for example.

FIG. 6 is an explanatory diagram showing an example of an input screen. When an access is received from the terminal 2, the server 1 transitions to the input screen shown in FIG. 6A. The screen in FIG. 6A is an input screen for inputting customer attributes, and is a screen for inputting basic information required for using this service. In addition to the customer's age, sex, etc., the terminal 2 accepts input of the brand of alcoholic beverages usually consumed, the amount of alcohol consumed, and the place of drinking (whether the customer drinks at home or when eating out) on the screen in FIG. 6A. Note that these are examples of customer attributes, and the customer's location, occupation, and other information may also be input. The server 1 stores the input customer attributes in the customer DB 142 in association with the customer ID. In addition, the terminal 2 stores the input customer attributes in the auxiliary storage unit 28, and may skip attribute input on the screen in FIG. 6A when using this service for the second or subsequent time.

Next, terminal 2 transitions to the input screen of FIG. 6B. The screen of FIG. 6B is an input screen that accepts input of responses from the customer to a specific questionnaire that asks about the customer's preferences. As shown in FIG. 6B, terminal 2 displays a specific character 61 (object) that provides guidance on this service in the center of the screen, and displays a question asking about the customer's preferences as the character's dialogue. Terminal 2 also displays answer options for the question and accepts answer selection input from the customer. Terminal 2 switches questions in sequence, accepts answer input, and acquires answers to each of the multiple questions. Server 1 stores the acquired answers in history DB 144 in association with the customer ID.

In this embodiment, interactions with customers are basically carried out through text display, but voice input and output may also be carried out using, for example, a chatbot engine.

As described below, in this embodiment, a survey is also conducted after a beer beverage is ordered (see FIG. 9). Therefore, for the sake of convenience in the following description, the survey in FIG. 6B conducted before a beer beverage is suggested is referred to as the "first survey," and the survey conducted after the order is placed is referred to as the "second survey."

Returning to FIG. 5 for further explanation, the server 1 estimates beer information regarding the beer beverage to be recommended to the customer based on the customer's attributes and answers entered on each input screen. Specifically, the server 1 inputs the customer's attributes and answers into a first model 51 generated by machine learning, and estimates (outputs) the beer information.

The first model 51 is a machine learning model generated by learning predetermined training data, and is a decision tree model that branches according to each attribute and answer. Specifically, the first model 51 is a decision tree model (e.g., LightGBM) generated using a gradient boosting technique, which is a type of ensemble learning, and is generated by sequentially generating multiple weak classifiers.

For example, the server 1 uses the attributes of a customer who actually ordered a beer beverage and the customer's response to the first questionnaire as training input data, and reads out beer information about the beer beverage ordered by the customer from the beverage DB 141 and uses this as the correct value of the output data. Note that the training data may not be data of actual customers, but may be fictitious data created manually. Alternatively, the training data may be data that has been inflated by a data generation means such as a Generative Adversarial Network (GAN). In other words, the training data may be data in which correct beer information is associated with a person's attributes and responses to questions, and the data is not limited to data of actual people (customers).

The server 1 generates a first model 51 based on the above training data. That is, the server 1 generates one weak classifier (decision tree) using the training data, and generates a next weak classifier based on the error between the predicted value of the beer information by the generated weak classifier and the correct value of the beer information indicated by the training data. The server 1 sequentially generates the next weak classifier by referring to the gradient of the loss function defined by the error between the predicted value and the correct value so as to take into account the learning result of the previous weak classifier, and generates the final classifier, i.e., the first model 51.

In this embodiment, gradient boosting is used as the learning method, but other ensemble learning methods, such as bagging, which generates multiple weak classifiers in parallel, may also be used. Furthermore, the first model 51 may be generated by a learning method other than ensemble learning.

Furthermore, the first model 51 is not limited to a decision tree model, but may be other machine learning models such as a neural network, a reinforcement learning model, or a support vector machine (SVM).

The server 1 inputs the customer's attributes and answers acquired from the terminal 2 into the above-mentioned first model 51, and estimates beer information regarding the beer beverage to be recommended to the customer. The beer information is, for example, characteristic values of the beer beverage, specifically, component values of various components such as the apparent extract and the apparent final fermentation degree. The server 1 uses the first model 51 to estimate the component values of the beer beverage to be recommended to the customer. The server 1 may also estimate the alcohol content and calories of the beer beverage as characteristic values other than the component values.

For example, the server 1 may estimate a value (such as an ID) indicating the yeast used as a characteristic value. Specifically, the server 1 estimates whether the yeast used is top-fermenting yeast or bottom-fermenting yeast. Note that naturally fermenting yeast may also be included in the yeast to be estimated. This allows estimation to take into account the difference between so-called ales and lagers.

For example, the server 1 may estimate a value indicating whether or not specific ingredients are used as a characteristic value. Specific ingredients include, for example, malt, hops, wheat, sugar, spices, fruits, vegetables, rice, miso, etc. By taking into account the use of these ingredients, it is possible to appropriately estimate a beer beverage that suits the customer's preferences, and this is also useful for, for example, allergy countermeasures.

For example, the server 1 may estimate a value indicating the presence or absence of a specific region name in the product name as a characteristic value. Specifically, the server 1 estimates the presence or absence of a region name in Japan in the product name. In this way, in the case of craft beer, for example, it is possible to suggest a beer beverage taking into consideration whether the craft beer is domestically produced or from which region in Japan. Note that the region names to be estimated are not limited to region names within Japan, and may be other region names (such as country names or region names of countries other than Japan).

For example, the server 1 may estimate a value (such as an ID) indicating the manufacturer, such as the manufacturer or region of the beer beverage to be proposed, as a characteristic value. This makes it possible to appropriately estimate which brewery and region of craft beer should be proposed, for example, in the case of craft beer.

In addition, the server 1 may estimate the price, quantity, etc. of the beer beverage to be proposed. In this way, the characteristic values output from the first model 51 are not limited to the above parameters, and may be other parameters.

In this embodiment, the customer's attributes and the answers to the first questionnaire are used as inputs to the first model 51, but other information may be used as inputs to the first model 51. For example, the server 1 may use meteorological information (weather, temperature, etc.) around the customer's current location as input to the first model 51. In this case, for example, the server 1 obtains location information (GPS signal, etc.) from the terminal 2, obtains meteorological information for the customer's current location indicated by the obtained location information from an external API (Application Programmable Interface), and inputs it to the first model 51. For example, if the terminal 2 is a store-installed terminal, the location of the terminal 2 is fixed, so it is sufficient to obtain meteorological information for the fixed location without obtaining location information from the terminal 2. This makes it possible to estimate the characteristic values of the beer beverage to be proposed to the customer while taking into account the external environment.

For example, the server 1 may also use data indicating the type of food served at the restaurant (Japanese cuisine, Chinese cuisine, izakaya, Italian cuisine, etc.) as other information used for input to the first model 51. For example, the server 1 refers to a website that lists reviews of restaurants to obtain data indicating the food type of the restaurant used by the customer, and inputs this data into the first model 51. This makes it possible to suggest beer beverages that are suitable for meals served at the restaurant.

Although not specifically mentioned above, the number of dimensions of the input data (attributes, responses, etc.) may be reduced by using techniques such as principal component analysis.

The server 1 acquires characteristic values of the beer beverage from the first model 51, and acquires a confidence factor indicating the accuracy of the characteristic values. The confidence factor is a value indicating the accuracy of the estimation result by the first model 51, and is expressed, for example, as a probability value. Note that the confidence factor referred to here is different from the confidence factor used for pruning the branches (roof) of the decision tree during learning.

In this embodiment, the first model 51 estimates the characteristic values of the beer beverage, but the present embodiment is not limited to this. For example, the first model 51 may estimate the brand (beverage ID) of the beer beverage to be proposed to the customer as the beer information. In this case, the server 1 may regard the processing in the first model 51 as a classification problem, and have the beverage ID of the beer beverage to be proposed to the customer learn as the correct value of the training data. In this way, the beer beverage to be proposed to the customer may be directly estimated in the first model 51.

Based on the characteristic values estimated by the first model 51, the server 1 selects a beer beverage to be proposed to the customer from the beverage DB 141 and proposes it to the customer. Specifically, the server 1 calculates the similarity (e.g., cosine similarity) between the characteristic values estimated by the first model 51 and the characteristic values of each beer beverage stored in the beverage DB 141, selects a beer beverage with a high similarity, and proposes it to the customer. Note that the similarity may be an index value other than cosine similarity, and may be any parameter that can appropriately evaluate the degree of similarity of the characteristic values of each beer beverage. For example, the server 1 compares the calculated similarity with a predetermined threshold, and selects a beer beverage with a similarity equal to or greater than the threshold as a beer beverage to be proposed to the customer. Note that the server 1 may select, for example, a beer beverage with the highest similarity, regardless of the threshold.

In addition to or instead of the similarity, the server 1 may select the beer beverage to be proposed based on the start date of the provision of each beer beverage, the proposal history of beer beverages proposed to customers in the past, the confidence level obtained from the first model 51, etc.

For example, the server 1 selects beer beverages to be recommended to customers, even if the similarity of the beer beverages is less than a threshold value, if the beer beverages have a service start date within a predetermined period from the current time. This makes it possible to recommend new beer beverage products to customers and promote sales.

For example, when the degree of certainty obtained from the first model 51 is less than a predetermined threshold, the server 1 may preferentially select beer beverages whose service start date is within a predetermined period, regardless of the degree of similarity. This makes it possible to suitably suggest beer beverages when there is a high possibility that the estimation result by the first model 51 is uncertain.

For example, the server 1 may include a question in the first questionnaire asking whether or not the customer has a preference for new products (such as "Would you like to try new products?"), and when a response indicating a desire for new products is received, the server 1 may preferentially select beer beverages whose release date is within a specified period of time. This allows the server 1 to suggest new products that take into account the customer's requests.

For example, the server 1 may extract from the beverage DB 141 beer beverages whose similarity is equal to or greater than a threshold value and beer beverages within a specified period from the start date of service, and select beer beverages to be proposed to the customer from each of the extracted beer beverage groups (beer beverage groups whose similarity is equal to or greater than a threshold value and beer beverages within a specified period from the start date of service) in a fixed ratio (for example, 1:1). This makes it possible to propose beer beverages that are thought to match the customer's preferences together with new beer beverage products for which sales promotion is desired.

In addition, for example, if the beer beverage selected based on the similarity (e.g., the beer beverage with the highest similarity) has been previously proposed to the customer, the server 1 may change it to another beer beverage (e.g., the beer beverage with the next highest similarity). This prevents the same product from being proposed multiple times, which is expected to increase the number of repeat customers.

For example, the server 1 may change to another beer beverage if the customer has previously given a low rating to the selected beer beverage in the second questionnaire described below. The server 1 displays a question item in the second questionnaire described below asking about the rating of the beer beverage, and accepts input of the rating value, for example, on a five-point scale. The rating value may be expressed as a score or as an option such as "like" or "dislike." The server 1 changes to another beer beverage if the customer has previously received a response below a certain value (for example, "2" or less on a five-point scale from "1" to "5") about the selected beer beverage. This makes it possible to suggest a beer beverage taking into consideration the customer's past ratings.

The server 1 selects beer beverages to be proposed to the customer using one or more of the multiple selection methods exemplified above. Note that all of the above selection methods are merely examples, and other selection methods may be used. In addition, the number of beer beverages selected may not be one, but multiple. The server 1 outputs information about the selected beer beverage to the terminal 2 and proposes it to the customer.

Figure 7 is an explanatory diagram showing an example of a proposal screen. Figure 7 illustrates an example of a screen when proposing a beer beverage to a customer. The proposal screen includes a character 61, a beer display column 71, an order button 72, and a cancel button 73. Similar to Figure 6B, the character 61 is a character that provides guidance to the customer. The terminal 2 displays the character 61 in the display area (second area) at the top of the screen, and also displays a comment (text) from the character 61 recommending the beer beverage displayed in the beer display column 71.

The beer display field 71 is a display field (first area) that displays information on the beer beverage proposed to the customer, i.e., the beer beverage selected by the server 1. For example, the terminal 2 displays the product name, description, and product image of the proposed beer beverage in the beer display field 71. The terminal 2 also displays a detailed profile of the proposed beer beverage, such as the beer style (beer classification according to ingredients, production method, alcohol content, etc.), alcohol content (concentration), calories, manufacturer (brewery), and production region. In addition, the terminal 2 displays a radar chart that shows the flavor of the beer beverage on multiple scales. The radar chart may be automatically created by the server 1 by processing the component values of the beer beverage stored in the beverage DB 141, for example, or may be created manually by an administrator of the system.

The order button 72 and the cancel button 73 are operation objects for selecting whether or not to order (drink) the proposed beer beverage when linkage with the store's ordering system is possible. When the order button 72 is operated, the proposed beer beverage is ordered. In this case, for example, the server 1 notifies the store staff of the order for the beer beverage. When the cancel button 73 is operated, the proposed beer beverage is not ordered and the process ends. In this case, for example, the server 1 may select another beer beverage (such as the beer beverage with the next highest similarity) and display it on the proposal screen, or may end the series of proposal processes.

Figure 8 is an explanatory diagram showing another example of the proposal screen. For example, the server 1 may change the display mode of the proposal screen depending on the confidence level of the estimation result in the first model 51. Figure 8 illustrates an example of a screen when the confidence level is low.

For example, the server 1 changes the display mode of the character 61 depending on whether the confidence level is equal to or higher than a predetermined threshold. Specifically, the server 1 changes the facial expression of the character 61 and the comment of the character 61. For example, as shown in FIG. 8, when the confidence level is less than the threshold, the server 1 changes the facial expression of the character 61 from the screen of FIG. 7 to a negative expression and displays a comment that the server is not confident that the proposed beer beverage matches the customer's preferences and requests the customer to cooperate in the second questionnaire described below. This allows the customer to easily grasp the likelihood that the proposed beer beverage matches their preferences, and makes it easier to obtain cooperation in the second questionnaire.

For example, when the degree of certainty is less than a threshold, the server 1 may display other beer beverages than the beer beverage displayed in the beer display field 71 as candidates and accept the order. For example, the terminal 2 displays one or more other beer beverages as candidates in the candidate field 81 at the bottom of the proposal screen, with product images, product names, etc. The server 1 displays in the candidate field 81 beer beverages to be proposed next to the beer beverage displayed in the beer display field 71. Note that the next to the beer beverage is, for example, a beer beverage with the next highest similarity to the beer beverage displayed in the beer display field 71, but is not limited to this. For example, when the start date of service is used as a selection criterion for the beer beverage to be proposed, the beer beverage with the next most recent start date of service may be a candidate, and the candidate beer beverage depends on the selection method. By also proposing other beer beverages when the degree of certainty is low, the beer beverage can be suitably proposed in consideration of the estimation result in the first model 51.

In the above, other beer beverages are displayed as candidates only when the degree of certainty is below the threshold value, but the candidate column 81 may also be displayed and other beer beverages may be suggested when the degree of certainty is equal to or above the threshold value.

In addition, in the above, the display mode of the character 61 is changed according to the confidence level in the first model 51, but this embodiment is not limited to this. For example, the server 1 may use the service start date as a selection criterion for beer beverages, and when a beer beverage within a predetermined period from the service start date is selected, the facial expression of the character 61 may be changed and a comment such as "How about our new product XX?" may be displayed. In addition, for example, the server 1 may use a proposal history or past evaluations to a customer as a selection criterion for beer beverages, and when a beer beverage that has been previously proposed or a beer beverage that has been previously rated low is changed and another beer beverage is proposed, the facial expression of the character 61 may be changed and a comment such as "XX is the most recommended, but how about this one this time?" may be displayed. In this way, the information that is the basis for changing the display of the character 61 is not limited to the confidence level, and may be other information.

9 is an explanatory diagram showing an example of the input screen of the second questionnaire. When the order button 72 is operated and a beer beverage is ordered, the terminal 2 transitions to the screen of FIG. 9. The screen is a display screen of the second questionnaire about the beer beverage ordered (drank) by the customer, and displays questions asking about impressions of the beer beverage, etc., and options for answers to the questions. For example, the terminal 2 displays a question item asking about the evaluation value of the beer beverage as shown in FIG. 9, and accepts input of an answer on a five-point scale. In addition, the terminal 2 sequentially displays each question item asking about meals eaten together, how many glasses it is, etc., and accepts input of answer options. Note that the content of the questions in the second questionnaire is not limited to the above, and may be a question asking about the beer beverage desired by the customer, such as a more preferred flavor (e.g., "a higher alcohol content would be better"), and the content of the questions is not limited. The server 1 associates the answers to each question item with the answers to the first questionnaire obtained before the proposal, the proposed beer beverage, etc., and stores them in the history DB 144 as a proposal history.

In this embodiment, the second survey is conducted only if the proposed beer beverage is ordered, but the second survey may also be conducted if the beverage is not ordered.

For example, the server 1 uses the proposal history stored in the history DB 144 to re-learn the first model 51. Specifically, the server 1 uses the attributes of the customer who proposed the beer beverage and the answers to the first questionnaire entered by the customer as input data for re-learning (training), performs learning using the proposed beer beverage as the correct answer value for re-learning, and updates the first model 51.

In this case, it is preferable that the server 1 performs re-learning according to whether or not the proposed beer beverage is ordered, the answer to the second questionnaire entered after the order, etc. For example, if the customer does not order the proposed beer beverage, or if the customer orders a beer beverage displayed in the candidate column 81 instead of the beer display column 71, the server 1 modifies the characteristic value of the proposed beer beverage (for example, changes it to the characteristic value of another beer beverage ordered by the customer) and updates the first model 51 with the correct characteristic value for re-learning. Also, for example, if the beer beverage drunk by the customer is low-rated (evaluation value is equal to or less than a certain value) in the second questionnaire, the server 1 modifies the characteristic value of the proposed beer beverage (for example, changes it to the characteristic value of the beer beverage with the next highest similarity), gives the corrected characteristic value to the first model 51 as the correct answer value, and updates the first model 51. Alternatively, if the beer beverage drunk by the customer is highly rated (evaluation value is equal to or more than a certain value), the server 1 may give the characteristic value of the beer beverage as it is as the correct answer value to update the first model 51.

The correct answer values (characteristic values) for re-learning may be correct labels created manually based on the responses to the second questionnaire, etc.

Note that the above re-learning method is just one example, and the present embodiment is not limited to this. For example, when the second questionnaire contains a question about the beer beverage desired by the customer, the server 1 may change the characteristic value (e.g., change the alcohol content) of the proposed beer beverage in accordance with the answer to the question (e.g., "A higher alcohol content would be better"), and use the changed characteristic value as the correct value for re-learning to update the first model 51. In other words, the correct value of the beer information (characteristic value) used for re-learning may be changed in accordance with the answer to the second questionnaire.

In this way, the accuracy of suggestions can be improved through the operation of this system based on responses to the second questionnaire, etc.

Fig. 10 is a flowchart showing the procedure of the generation process of the first model 51. The process content when the first model 51 is generated by machine learning will be described with reference to Fig. 10.
The control unit 11 of the server 1 acquires training data for generating the first model 51 (step S11). The training data includes a plurality of attributes of a person, answers to questions related to the first questionnaire, and correct values of the beer information. The beer information is, for example, characteristic values of a beer beverage, and includes values indicating, in addition to component values of the beer beverage, yeast used, whether or not a specific raw material is used, whether or not a specific region name is included in the product name, and the like. For example, the control unit 11 uses the attributes of an actual customer and the answers to the first questionnaire as training data, and reads out characteristic values of the beer beverage actually drunk by the customer from the beverage DB 141 and uses them as correct values.

Based on the training data, the control unit 11 generates a first model 51 that outputs beer information when multiple responses to the first questionnaire and multiple customer attributes are input (step S12). Specifically, as described above, the control unit 11 generates a decision tree model using gradient boosting. The control unit 11 generates one weak classifier (decision tree) using the training data, and generates a next weak classifier based on the error between the predicted value by the generated weak classifier and the correct value. The control unit 11 sequentially generates weak classifiers by referring to the gradient of a loss function defined by the error between the predicted value and the correct value, and generates a final classifier, i.e., the first model 51. The control unit 11 ends the series of processes.

11 is a flowchart showing the procedure of the beverage suggestion process. The process of suggesting beer to a customer will be described with reference to FIG.
The control unit 11 of the server 1 accepts access from the terminal 2, for example, by reading a two-dimensional code (step S31). The control unit 11 outputs an input screen for inputting information necessary for proposing beer beverages to the terminal 2 and causes it to be displayed (step S32). The control unit 11 accepts input of multiple attributes of the customer on the input screen (step S33). If the customer is already registered in the customer DB 142, the control unit 11 skips step S33. The control unit 11 sequentially displays multiple questions related to the first questionnaire on the input screen and accepts input of answers to each question from the customer (step S34).

The control unit 11 inputs the multiple attributes and answers input by the customer into the first model 51, and estimates the characteristic values (beer information) of the beer beverage to be proposed to the customer (step S35). The control unit 11 calculates the similarity between the estimated characteristic values and the characteristic values of each beer beverage stored in the beverage DB 141 (step S36). The control unit 11 selects a beer beverage from the beverage DB 141 according to the similarity calculated in step S36 as well as the start date of service, etc., and outputs and displays a proposal screen displaying information on the beer beverage to the terminal 2 (step S37). The proposal screen includes a beer display section 71 displaying the selected beer beverage, as well as a character 61 introducing the beer beverage. For example, the control unit 11 changes the display mode (facial expression, comment, etc.) of the character 61 according to the degree of certainty indicating the accuracy of the characteristic values output from the first model.

The control unit 11 determines whether or not an order for the proposed beer beverage has been accepted (step S38). If it is determined that the order has not been accepted (S38: NO), the control unit 11 ends the series of processes. Note that if the answer is NO in step S38, the control unit 11 may return the process to step S37 and re-suggest another beer beverage.

If it is determined that the order has been accepted (S38: YES), the control unit 11 causes the terminal 2 to display each question related to the second questionnaire and accepts input of answers from the customer (step S39). The control unit 11 then ends the series of processes.

12 is a flowchart showing the procedure of the process of updating the first model 51. The process of updating the first model 51 by performing re-learning will be described with reference to FIG.
The control unit 11 of the server 1 acquires training data for re-learning from the history DB 144 etc. (step S51). The training data for re-learning includes the answer to the first questionnaire entered by the customer, the attributes of the customer, the characteristic values of the beer beverage proposed to the customer, etc., as well as whether or not the proposed beer beverage was ordered, the answer to the second questionnaire entered after the order, etc.

The control unit 11 updates the first model 51 based on the training data for re-learning (step S52). For example, if the customer does not order the proposed beer beverage, the control unit 11 updates the parameters of the first model 51 by taking the characteristic value of the beer beverage ordered by the customer as the correct answer value. Also, for example, if the evaluation value of the proposed beer beverage is equal to or lower than a certain value, the control unit 11 updates the parameters of the first model 51 by taking the characteristic value of the other beer beverage as the correct answer value. The control unit 11 ends the series of processes.

As described above, according to this embodiment 1, by using the first model 51 that has already learned beer information (characteristic values), it is possible to appropriately select and propose to the customer a beer beverage that matches the customer's preferences.

Furthermore, according to the present embodiment 1, a beer beverage can be suitably proposed via the character 61. In particular, in this embodiment, the display mode of the character 61 is changed according to the confidence level obtained from the first model 51, so that the reliability of the proposed beer beverage can be easily grasped.

In addition, according to the first embodiment, the beer beverage to be proposed can be suitably selected by calculating the similarity between the characteristic values output from the first model 51 and the characteristic values of each beer beverage stored in the beverage DB 141. In particular, in this embodiment, the beer beverage can be more suitably proposed by selecting based on the start date of service, confidence level, etc. in addition to the similarity.

(Embodiment 2)
In the first embodiment, a beer beverage to be proposed to a customer is selected by calculating the similarity with the characteristic value output from the first model 51. In the present embodiment, a beer beverage to be proposed to a customer is selected by using a second model 52 constructed by unsupervised learning so as to classify the beer beverage according to the characteristic value. Note that the same reference numerals are used for the contents overlapping with the first embodiment, and the description thereof is omitted.

FIG. 13 is a block diagram showing an example of the configuration of the server 1 according to the second embodiment. The auxiliary storage unit 14 of the server 1 according to the present embodiment stores the second model 52. The second model 52 is a machine learning model that has learned training data, and is a clustering model that inputs multiple characteristic values (beer information) of beer beverages and classifies the beer beverages corresponding to the characteristic values into one of multiple clusters. The second model 52 is expected to be used as a program module that functions as part of artificial intelligence software.

Figure 14 is an explanatory diagram showing an overview of the second embodiment. The overview of this embodiment will be explained based on Figure 14.

In this embodiment, the server 1 inputs the characteristic values output from the first model 51 to the second model 52, and classifies the beer beverages into one of a number of clusters that are collections of the beer beverages stored in the beverage DB 141. The second model 52 is a clustering model that classifies the beer beverages according to the characteristic values, and is generated by unsupervised learning using, for example, the t-SNE method (t-distribution type stochastic neighbor embedding method).

In this specification, "unsupervised learning" refers to generating a model by learning from training data where the correct answer is unknown.

For example, the server 1 learns using the characteristic values of each beer beverage stored in the beverage DB 141 as training data. The training data may be characteristic values related to beer beverages other than the beer beverages provided (proposed) by this system, or may be data inflated by a known data generation means. The training data of the second model 52 may be common to the training data of the first model 51. The server 1 regards multiple characteristic values of each beer beverage as features, compresses them from high dimensions to low dimensions (e.g., two or three dimensions), and maps them into a feature space so that beer beverages with similar characteristic values are close to each other. The server 1 then aggregates groups of beer beverages that are close in distance (e.g., Euclidean distance) in the feature space to generate multiple clusters.

In the above, the t-SEN method is mentioned as a clustering method, but other methods such as the k-means method may be used. In addition, in this embodiment, a clustering model generated by unsupervised learning is used as the second model 52, but the classification problem in this embodiment may be regarded as class classification, and the second model 52 may be generated by supervised learning from training data in which a set (class) of beer beverages is known.

The server 1 inputs the characteristic values output from the first model 51 into the second model 52 and classifies them into one of a number of clusters. The server 1 then selects one of the beer beverages that belongs to the one or more classified clusters as the beer beverage to be proposed to the customer.

In this case, for example, the server 1 may select a beer beverage to be proposed based on the similarity, serving start date, confidence level, proposal history, etc. described in the first embodiment in addition to the classification result into clusters. For example, the server 1 may select a beer beverage that has the highest similarity to the characteristic value output from the first model 51 from among the beer beverages belonging to the classified clusters. Also, for example, the server 1 may select a beer beverage whose serving start date is within a predetermined period from among the beer beverages belonging to the classified clusters. Also, for example, the server 1 may select a beer beverage that belongs to a cluster other than the classified cluster when the confidence level is less than a threshold value. Also, for example, the server 1 may select a beer beverage that has not been proposed to the customer from among the classified clusters. In this way, many methods for selecting a beer beverage are possible in this embodiment as well.

The server 1 outputs information about the beer beverage selected above to the terminal 2 and proposes it to the customer. The subsequent processing is the same as in embodiment 1, so a description thereof will be omitted.

15 is a flowchart showing the procedure of the generation process of the second model 52. The process contents when the second model 52 is generated will be described with reference to FIG.
The control unit 11 of the server 1 acquires a plurality of characteristic values corresponding to a plurality of beer beverages as training data from the beverage DB 141 (step S201). The control unit 11 performs unsupervised learning to classify each beer beverage into a plurality of clusters according to the characteristic values, thereby generating a second model 52 that classifies the plurality of characteristic values as input into one of a plurality of clusters that are a set of beer beverages (step S202). The control unit 11 ends the series of processes.

16 is a flowchart showing the procedure of the beverage suggestion process according to the embodiment 2. After estimating the characteristic value of the beer beverage using the first model 51 (step S35), the server 1 executes the following process.
The control unit 11 of the server 1 inputs the characteristic values output from the first model 51 to the second model 52 and classifies the beer beverage into one of a plurality of clusters that are a collection of beer beverages (step S221). The control unit 11 selects a beer beverage from the beverage DB 141 according to the classified cluster, and outputs a proposal screen proposing the selected beer beverage to the terminal 2 (step S222). The control unit 11 shifts the process to step S38.

As described above, according to the second embodiment, by using the second model 52 that classifies beer beverages based on characteristic values, it is possible to appropriately select beer beverages to be recommended to customers. In particular, in this embodiment, the second model 52 can be constructed without the need to manually label the correct answer values.

The embodiments disclosed herein are illustrative in all respects and should not be considered limiting. The scope of the present invention is indicated by the claims, not by the meaning described above, and is intended to include all modifications within the scope and meaning equivalent to the claims.

1 Server (information processing device)
REFERENCE SIGNS LIST 11 Control unit 12 Main memory unit 13 Communication unit 14 Auxiliary memory unit P1 Program 51 First model 52 Second model 141 Beverage DB
142 Customer DB
143 Question table 144 History DB
2 Terminal 21 Control unit 22 Main memory unit 23 Communication unit 24 Display unit 25 Input unit 26 Imaging unit 27 GPS receiving unit 28 Auxiliary memory unit P2 Program

Claims (13)

Get customer attributes,
Get answers to multiple questions,
inputting the acquired attributes and answers into a first model that has been trained to output a plurality of characteristic values indicative of characteristics of a beer beverage, including beer, happoshu, new genre beer, or a beer-flavored beverage, when the attributes and answers are input, and outputting the plurality of characteristic values ;
A storage unit that stores a plurality of characteristic values and a serving start date in association with each beer beverage is referred to, and a similarity between the plurality of characteristic values of each beer beverage and the plurality of characteristic values output from the first model is calculated;
The beer beverage to be proposed to the customer is selected according to the calculated similarity and the service start date.
An information processing method in which processing is performed by a computer. The information processing method according to claim 1 , further comprising the step of selecting the beer beverage whose similarity is less than a predetermined threshold and whose serving date is within a predetermined period of time from the serving start date. obtaining a confidence level indicating the likelihood of the plurality of characteristic values output from the first model;
The information processing method according to claim 1 or 2 , further comprising the steps of: selecting a beer beverage within a predetermined period from the start date of service, regardless of the degree of similarity, when the degree of certainty is less than a predetermined threshold value. The plurality of questions includes a question asking whether the person likes a new beer beverage;
The information processing method according to any one of claims 1 to 3, further comprising the step of selecting a beer beverage within a predetermined period from the start date of the service, regardless of the degree of similarity, when the answer indicating a preference for a new beer beverage is obtained. The information processing method according to any one of claims 1 to 4, further comprising the step of selecting, at a certain ratio, the beer beverages whose similarity is equal to or greater than a predetermined threshold value and the beer beverages within a predetermined period from the start date of service. inputting the plurality of characteristic values output from the first model into a second model that has been trained by unsupervised learning so as to classify the plurality of characteristic values into one of a plurality of clusters that are a set of beer beverages when the plurality of characteristic values are input, and classifying the plurality of characteristic values into one of the plurality of clusters;
The information processing method according to claim 1 , further comprising the step of selecting the beer beverage to be proposed to the customer according to the classified cluster. obtaining an answer to a second question about the selected beer beverage;
The information processing method according to claim 1 , further comprising updating the first model based on an acquired answer to the second question. Acquire location information of the customer;
Obtaining meteorological information corresponding to the customer's current location based on the location information;
The information processing method according to claim 1 , further comprising inputting the attributes, the responses and the meteorological information into the first model, and outputting the plurality of characteristic values. The information processing method according to claim 1 , wherein the characteristic values include values indicating the yeast used in the beer beverage, whether a specific raw material is used, or whether a specific region name is included in the product name. Get customer attributes,
Get answers to multiple questions,
inputting the acquired attributes and answers into a first model that has been trained to output a plurality of characteristic values indicative of characteristics of beer, happoshu, or beer beverages including beer-flavored beverages when the attributes and answers are input, and outputting the plurality of characteristic values ;
A storage unit that stores a plurality of characteristic values and a serving start date in association with each beer beverage is referred to, and a similarity between the plurality of characteristic values of each beer beverage and the plurality of characteristic values output from the first model is calculated;
The beer beverage to be proposed to the customer is selected according to the calculated similarity and the service start date.
A program that causes a computer to carry out processing. A first acquisition unit that acquires attributes of a customer;
a second acquisition unit that acquires answers to the plurality of questions;
an output unit that inputs the acquired attributes and answers into a first model that has been trained to output a plurality of characteristic values that indicate characteristics of beer beverages including beer, happoshu, new genre beer, or beer-taste beverages when the attributes and answers are input, and outputs the plurality of characteristic values ;
A calculation unit that refers to a storage unit that stores a plurality of characteristic values and a serving start date in association with each beer beverage, and calculates a similarity between the plurality of characteristic values of each beer beverage and the plurality of characteristic values output from the first model;
a selection unit that selects the beer beverage to be proposed to the customer according to the calculated similarity and the serving start date;
An information processing device comprising: Get customer attributes,
Get answers to multiple questions,
inputting the acquired attributes and answers into a first model that has been trained to output a plurality of characteristic values indicative of characteristics of a beer beverage, including beer, happoshu, new genre beer, or a beer-flavored beverage, when the attributes and answers are input, and outputting the plurality of characteristic values;
inputting the plurality of characteristic values output from the first model into a second model that has been trained by unsupervised learning so as to classify the plurality of characteristic values into one of a plurality of clusters that are a set of beer beverages when the plurality of characteristic values are input, and classifying the plurality of characteristic values into one of the plurality of clusters;
Selecting the beer beverage to be proposed to the customer according to the classified clusters
An information processing method in which processing is performed by a computer. Get customer attributes,
Get answers to multiple questions,
inputting the acquired attributes and answers into a first model that has been trained to output a plurality of characteristic values indicative of characteristics of beer beverages including beer, happoshu, new genre beer, or beer-taste beverages when the attributes and answers are input, and outputting the plurality of characteristic values;
With reference to a storage unit that stores a plurality of characteristic values in association with each beer beverage, a similarity between the plurality of characteristic values of each beer beverage and the plurality of characteristic values output from the first model is calculated;
Selecting the beer beverage to be recommended to the customer according to the calculated similarity;
obtaining an answer to a second question about the selected beer beverage;
updating the first model based on the obtained answers to the second questions;
An information processing method in which processing is performed by a computer.

Images