Learning estimation system, learning estimation method, and learning estimation program

Abstract

We provide a learning estimation system that can estimate future sales even for stores with limited data. [Solution] In a communication system including a learning estimation system 100 and store terminals 200a and 200b, the learning estimation system comprises: an acquisition unit that acquires store location data indicating the location of a store, store business type data indicating the type of business of a store, and sales-related data related to the daily sales of a store; a conversion unit that converts the data acquired by the acquisition unit into training data; and a learning unit that trains a machine learning model that estimates sales-related data of a store based on weather or pedestrian flow data with the training data.

Description

Technical Field

[0001] This invention relates to a learning and estimation system, a learning and estimation method, and a learning and estimation program capable of estimating sales-related data of a store.

Background Art

[0002] It is known that sales in a store are affected by the weather and the flow of people on that day. Therefore, Patent Document 1 discloses a technique for predicting demand at a location based on the weather at each location. Further, Patent Document 2 discloses a technique for predicting the number of customers in a store such as a retail store or a restaurant using a Bayesian network.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Means for Solving the Problems

[0004] A learning and estimation system according to an aspect of the present invention includes an acquisition unit that acquires store location data indicating the location of a store, store format data indicating the format of the store, and sales-related data related to the daily sales of the store, a conversion unit that converts the data acquired by the acquisition unit into teacher data, and a learning unit that causes a machine learning model that estimates the sales-related data of the store based at least on the location of the store and the format of the store to learn the teacher data.

[0005] Further, in the learning and estimation system, the acquisition unit may further acquire weather data for each day, and the machine learning model may be a model that further estimates the sales-related data of the store based on the weather data.

[0006] Furthermore, in the above-described learning estimation system, the acquisition unit may further acquire pedestrian flow data for store locations on a given day, and the machine learning model may further estimate store sales-related data based on the pedestrian flow data.

[0007] Furthermore, in the above-described learning estimation system, the acquisition unit may further acquire calendar information indicating whether the date is a holiday or a weekday, and whether it is a day when a specific event occurred, and the machine learning model may further be a model that estimates store sales-related data based on the calendar information.

[0008] Furthermore, in the above-described learning estimation system, the acquisition unit may further acquire promotional data related to store sales promotions on a given day, and the machine learning model may further be a model that estimates store sales-related data based on the promotional data.

[0009] Furthermore, in the above-described learning estimation system, the transformation unit may logarithmically transform the sales-related data acquired by the acquisition unit into training data.

[0010] Furthermore, in the above-described learning estimation system, the transformation unit may logarithmically transform the sales-related data acquired by the acquisition unit, logarithmically transform the sales-related data from different days, and transform the difference between the two logarithmically transformed sales-related data into training data.

[0011] Furthermore, in the above-described learning estimation system, the conversion unit may, when the sales-related data to be estimated using the machine learning model is sales-related data from a predetermined date later, logarithmically transform the sales-related data acquired by the acquisition unit, logarithmically transform the sales-related data acquired by the acquisition unit from a predetermined date earlier, and convert the difference between the two sets of logarithmically transformed sales-related data into training data.

[0012] Furthermore, in the above-described learning estimation system, the transformation unit may convert the difference between the two sales-related data sets into training data by performing an inverse logarithmic transformation.

[0013] Furthermore, in the above-described learning estimation system, the conversion unit may, when the sales-related data to be estimated using the machine learning model is sales-related data from a predetermined date later, convert information based on the difference between the sales-related data acquired by the acquisition unit and the sales-related data acquired before the predetermined date of said sales-related data as training data.

[0014] Furthermore, the above-described learning estimation system may also include a reception unit that accepts input of date data relating to the location of the store where sales-related data is to be estimated, the type of business of the store, and the date; an estimation unit that inputs the date data received by the reception unit into a machine learning model to estimate sales-related data for the date indicated by the date data at the store; and an output unit that outputs the sales-related data estimated by the estimation unit.

[0015] Furthermore, the above-described learning estimation system may include a data acquisition unit that acquires weather data at the location of a store based on the store's location, a machine learning model that further estimates sales-related data for the store based on the weather data, and an estimation unit that further inputs the weather data acquired by the data acquisition unit into the machine learning model to estimate sales-related data.

[0016] Furthermore, the above-described learning estimation system may include a data acquisition unit that acquires weather data at the location of a store based on the store's location, a machine learning model that further estimates sales-related data for the store based on the weather data, and an estimation unit that further inputs the weather data acquired by the data acquisition unit into the machine learning model to estimate sales-related data.

[0017] Furthermore, the above-described learning estimation system includes a data acquisition unit that acquires calendar information indicating whether the date indicated by the date data is a holiday or a weekday, and whether it is a day when a specific event occurred. The machine learning model further estimates store sales-related data based on the calendar information, and the estimation unit may further input the calendar information acquired by the data acquisition unit into the machine learning model to estimate sales-related data.

[0018] Furthermore, the above-described learning estimation system may include a data acquisition unit that acquires promotional data related to store promotions on the date indicated by the date data, and the machine learning model further estimates store sales-related data based on the promotional data, and the estimation unit may further input the promotional data acquired by the data acquisition unit into the machine learning model to estimate sales-related data.

[0019] Furthermore, in the above-described learning estimation system, if the acquisition unit is unable to acquire sales-related data for a specific date, a supplementation unit may be provided to supplement the sales-related data for that specific date, and the estimation unit may estimate the sales-related data for the date indicated by the date data based on the sales-related data supplemented by the supplementation unit.

[0020] Furthermore, in the above-described learning estimation system, the supplementation unit may use past sales data of the store for the same day of the week as the specific date to supplement the sales-related data for that specific date.

[0021] Furthermore, in the above-described learning estimation system, the learning unit may learn from the sales-related data supplemented by the supplementation unit.

[0022] Also, a learning estimation method according to one aspect of the present invention includes an acquisition step in which a computer acquires store location data indicating the location of a store, store format data indicating the format of the store, and sales-related data related to the daily sales of the store; a conversion step of converting the data acquired in the acquisition step into teacher data; and a learning step of causing a machine learning model that estimates the sales-related data of the store based on at least the location of the store and the format of the store to learn the teacher data.

[0023] Also, a learning estimation program according to one aspect of the present invention causes a computer to realize an acquisition function of acquiring store location data indicating the location of a store, store format data indicating the format of the store, and sales-related data related to the daily sales of the store; a conversion function of converting the data acquired by the acquisition function into teacher data; and a learning function of causing a machine learning model that estimates the sales-related data of the store based on at least the location of the store and the format of the store to learn the teacher data.

Brief Description of the Drawings

[0024] [Figure 1] It is a schematic diagram showing the outline of the invention. [Figure 2] It is a block diagram showing a configuration example of an information processing apparatus. [Figure 3] It is a block diagram showing a configuration example of a store terminal. [Figure 4] It is a flowchart showing an operation example of the information processing apparatus during learning. [Figure 5] It is a flowchart showing an operation example of the store terminal during learning. [Figure 6] It is a flowchart showing an operation example of the information processing apparatus during estimation. [Figure 7] It is a flowchart showing an operation example of the store terminal during estimation.

Modes for Carrying Out the Invention

[0025] Hereinafter, a learning estimation system according to the present invention will be described with reference to the drawings.

[0026] <Overview> Predicting sales-related data for stores is crucial. Sales-related data may include, but is not limited to, the number of customers visiting a store in a day or the total daily sales. Any other information related to store sales is acceptable, and the time period is not limited to a single day. If the number of customers can be predicted as sales-related data, stores such as restaurants can estimate the necessary inventory and reduce unnecessary purchases. Alternatively, if total sales can be predicted as sales-related data, it becomes easier to plan store renovations or the installation of new equipment. Such predictions can be performed using learning models.

[0027] Incidentally, such learning models require a large amount of training data for accurate predictions. Using actual historical sales data from a store as this training data allows for accurate predictions for that particular store. However, small stores often lack sufficient training data to make accurate predictions. This is also true for newly established stores. Furthermore, some stores may find it difficult to even prepare a learning model in the first place.

[0028] Therefore, the objective of this embodiment is to provide a learning estimation system, a learning estimation method, and a learning estimation program that can provide a machine learning model capable of making reasonably accurate predictions even if it is difficult for stores to prepare a learning model.

[0029] Figure 1 is a schematic diagram showing an overview of the processing of the learning estimation system 100 according to this embodiment. Figure 1 shows an example of the configuration of a communication system including the learning estimation system 100, and shows the learning estimation system 100 and store terminals 200a and 200b (hereinafter collectively referred to as store terminals 200; however, in Figure 1, stores are shown as a substitute for terminals). The learning estimation system 100 stores a machine learning model that estimates sales-related data for stores based on at least the store's location and business type. This machine learning model may estimate sales-related data based on one or more of the following in addition to the store's location and business type: weather at the store's location, foot traffic at the store's location, calendar information such as whether it is a holiday or weekday, or whether there is an event scheduled for that day, and promotional data for the store. In other words, the machine learning model is a model that has learned the correspondence between various store-related information such as the store's location and business type, and sales-related data. It may be a model that has learned the correspondence between sales-related data and one or more of the following: store location, store type, weather near the store, pedestrian traffic near the store, calendar information, and promotional data. The learning estimation system 100 can be said to have learned sales-related data from stores of various types in various locations. The store's business type may refer to, for example, the genre of service offered by a restaurant, such as Japanese food, soba noodles, ramen, Italian food, or French cuisine, or the operating style of the store, such as a food truck, a fixed store, a food court, a standing-only restaurant, or a reservation-only system. Furthermore, the store is not limited to a restaurant; it may be a clothing store or a furniture store, and the business type may change depending on the content of the store.

[0030] Here, by training the system with sales-related data from the store terminal 200, we can provide a learning estimation system 100 that can compensate for the lack of training data for the store terminal 200 and perform more accurate estimations of sales-related data from the store terminal 200. A detailed explanation follows below.

[0031] <Structure> <Configuration of Learning Estimation System 100> Figure 2 is a block diagram showing an example configuration of the learning estimation system 100. The learning estimation system 100 is a computer system (information processing device) that converts sales-related data from the store terminal 200 into training data for learning, and also estimates the sales-related data from the store terminal 200. The learning estimation system 100 may be implemented using a PC, server device, etc., but is not limited to these. Furthermore, the functions of the learning estimation system 100 may be implemented by parallel processing using multiple computers, or by a cloud server. The learning and estimation of sales-related data by the learning estimation system 100 may be provided as a web service.

[0032] As shown in Figure 2, the learning estimation system 100 includes a communication unit 110, a control unit 130, and a storage unit 140. The learning estimation system 100 may also include an input unit 120 and an output unit 150.

[0033] The communication unit 110 is a communication interface that has the function of communicating with an external device of the learning estimation system 100. The communication unit 110 has the function of communicating with the store terminal 200 as an external device. The communication unit 110 receives sales-related data from the store terminal 200 and transmits it to the control unit 130. The communication unit 110 also receives date data from the store terminal 200 indicating the date on which the sales-related data should be estimated and transmits it to the control unit 130. Here, the date indicated by the date data may be a specific day, or it may be a predetermined period including multiple days (for example, the weekdays of the next week, the next month, or the three months of one year from now, but it is not limited to these and may be at the discretion of the store staff). The communication unit 110 also transmits the estimated sales-related data to the store terminal 200 that requested the estimation, in accordance with instructions from the control unit 130.

[0034] The input unit 120 has the function of receiving input from an operator of the learning estimation system 100 and transmitting the input content to the control unit 130. The input unit 120 may be implemented by an input device such as a mouse, keyboard, or touch panel, and in the case of voice input, it may be implemented by a microphone. As an example, the input unit 120 receives input of sales-related data from an employee of a store terminal 200 by an operator of the learning estimation system 100 and transmits the information to the control unit 130.

[0035] The control unit 130 is a processor that has the function of controlling each part of the learning estimation system 100. The control unit 130 may be implemented as a single-core or multi-core processor. The control unit 130 executes various programs stored in the memory unit 140 and utilizes various data to realize the functions of the learning estimation system 100.

[0036] The control unit 130 converts the sales-related data received from the store terminal 200 into training data and uses it to train the machine learning model 141 in the storage unit 140. The control unit 130 then estimates the sales-related data for the date indicated by the date data received from the store terminal 200 and notifies the store terminal 200.

[0037] The control unit 130 includes, as functions implemented by the control unit 130, an acquisition unit 131, a conversion unit 132, a learning unit 133, a reception unit 134, a data acquisition unit 135, and an estimation unit 136.

[0038] The acquisition unit 131 acquires sales-related data from the store terminal 200. The acquisition unit 131 acquires sales-related data transmitted from the store terminal 200 and received by the communication unit 110. The acquisition unit 131 transmits the acquired sales-related data to the conversion unit 132. The sales-related data acquired by the acquisition unit 131 includes information on the number of customers or sales (which may be total sales or sales per SKU (Stock Keeping Unit)) for each day or a predetermined period at the store terminal 200, and may also include information on the weather at the store's location on that day, pedestrian traffic near the store, and whether or not sales promotions were conducted on that day. If the sales-related data does not include weather or pedestrian traffic information, the acquisition unit 131 may acquire this information from a weather information service that provides weather information at that time or a pedestrian traffic service that provides pedestrian traffic information, based on the store's location at the store terminal 200.

[0039] The conversion unit 132 converts the sales-related data transmitted from the acquisition unit 131 into training data. Here, converting to training data may mean extracting the necessary information as training data from the sales-related data transmitted from the acquisition unit 131 and the weather or pedestrian flow data transmitted from the data acquisition unit 135, and converting it into data in a format that can be used to train the machine learning model 141 in the storage unit 140. The conversion unit 132 may convert the sales-related data into training data by logarithmically transforming it. Since the absolute number of sales-related data can vary greatly depending on the location and size of the store, the learning estimation system 100 can normalize the numerical value by logarithmically transforming the sales-related data, enabling it to perform learning that corresponds to various stores, and as a result, it can provide more accurate estimated sales-related data to various stores. The conversion unit 132 transmits the generated training data to the learning unit 133.

[0040] The learning unit 133 trains the machine learning model 141 with the training data transmitted from the conversion unit 132. As a result, the machine learning model 141 can make estimations based on the actual sales at the store terminal 200, and thus can make more accurate estimations for that store than if no training were performed. The learning unit 133 may perform training using, for example, a gradient boosting decision tree or a deep neural network, but is not limited to these.

[0041] The reception unit 134 receives date data transmitted from the store terminal 200, which has been received by the communication unit 110. The date data received by the reception unit 134 is information indicating the date on which sales-related data in the store terminal 200 will be estimated (predicted). The reception unit 134 transmits the received date data to the estimation unit 136. The reception unit 134 also transmits information about the store of the store terminal 200 that sent the date data to the estimation unit 136. The information about the store may be read from the storage unit 140, or the information about the store may also be received from the store terminal 200 at the time of receiving the date data. The reception unit 134 also transmits information indicating the location of the store from the store information to the data acquisition unit 135.

[0042] The data acquisition unit 135 acquires weather or pedestrian flow data for the date indicated by the date data of the store terminal 200, based on the location of the store terminal 200. The data acquisition unit 135 may acquire either or both weather or pedestrian flow data. For example, the data acquisition unit 135 may access a server that stores and provides weather data for each day of each year to acquire the data. Alternatively, for example, the data acquisition unit 135 may access a server that stores and provides pedestrian flow data for each location each year to acquire the data. Here, when estimating sales-related data, the data acquisition unit 135 may acquire the average value of the weather or pedestrian flow data for that location for the past several years for the day to be estimated. The average value of weather data may be, for example, the rainfall rate based on whether it rained or was sunny on that day. The average value of pedestrian flow data may be the average number of people present on that day in the past in the regional mesh of the store's location. Furthermore, the weather or human flow data for the date indicated by the date data may be data estimated by a trained model that has learned the weather or human flow for each past day. The data acquisition unit 135 transmits the acquired weather or human flow data to the conversion unit 132 and the estimation unit 136.

[0043] The estimation unit 136 estimates sales-related data for the store terminal 200 based on information transmitted from the reception unit 134 and the data acquisition unit 135. The estimation unit 136 estimates sales-related data for the store terminal 200 on a (future) date transmitted from the reception unit 134. It inputs the date data received by the reception unit 134, information indicating the location of the store terminal 200, information indicating the type of business the store terminal 200 operates, and predicted weather or pedestrian flow data for the date indicated by the date data at the store terminal 200 into the machine learning model 141 to estimate the sales-related data for the store at the store terminal 200 on the transmitted date. The date for which this estimation is performed is specified by the date data and may be one day or multiple days. At this time, the machine learning model 141 may predict the face value of sales itself, the number of customers, or both as sales-related data. The estimation unit 136 transmits the estimated sales-related data to the store terminal 200 via the communication unit 110.

[0044] The acquisition unit 131, conversion unit 132, learning unit 133, and data acquisition unit 135 are functional units that learn sales-related data from the store terminal 200. The reception unit 134, data acquisition unit 135, and estimation unit 136 are functional units that perform estimation processing using the machine learning model 141.

[0045] The memory unit 140 has the function of storing various programs and data necessary for the operation of the learning estimation system 100. The memory unit 140 can be implemented using, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), flash memory, etc., but is not limited to these. The memory unit 140 may also be cloud storage accessible to the learning estimation system 100. The memory unit 140 may store various programs and data necessary to realize the functions that the learning estimation system 100 should perform. The memory unit 140 may store a program that converts sales-related data from the store terminal 200 into training data to train the machine learning model 141, and a program that estimates and transmits sales-related data from the store terminal 200 at the date and time transmitted from the store terminal 200. The memory unit 140 also stores a machine learning model 141 that estimates store sales-related data based on weather or pedestrian flow. Machine learning model 141 is a trained model that uses the date and time of sales (total sales, number of customers, number of units sold per SKU, or a combination of these), the store's location, and the store's business type as explanatory variables, and sales-related data (total sales, number of customers, or both) as the dependent variable. Machine learning model 141 may also learn weather information, pedestrian flow information, calendar information, and store promotional information for the store's location as explanatory variables. The store's location may be managed in units of a predetermined range when the map is divided into units of 500m square, but is not limited to this, and may be 1km square, 5km square, etc. In this case, the weather data may be weather information corresponding to the day of the sales-related data within that predetermined range, and this weather information may be weather information for each hour (for example, time in 1-hour units, but is not limited to this, and may be in units of 30 minutes, 2 hours, AM, PM, or YON). Furthermore, the pedestrian flow data may be the number of people present at each time interval within a predetermined range (for example, it may be in units of one hour, but is not limited to this; it may also be in units of 10 minutes, 30 minutes, or even a full day).Weather information may be obtained from, for example, the Japan Meteorological Agency's servers, but is not limited to this. Human flow data may be the number of people estimated from the number of terminals connected to base stations of a telecommunications carrier, and may be obtained from the telecommunications carrier. Calendar information may indicate whether the date of the sales-related data to be learned is a weekday, a holiday, or a day with a specific event (for example, Valentine's Day or Christmas, but is not limited to these; it may also be an event set by the organizer, such as a fireworks display, a sports match, or a choral competition), and this information can be obtained from regular calendar data. Promotional information may be information on whether a store is conducting a promotion, and may also include information on the content of the promotion. Here, a promotion may be a measure taken by a store with the aim of increasing sales at the store terminal 200, and may include, for example, discounts during specific time periods, issuance of coupons or other benefits, or receiving a free gift with the purchase of a certain product, but is not limited to these. Furthermore, the memory unit 140 may also store information about the stores of the store terminals 200 that use the learning estimation system 100. This information may include the store's name, location, type of business, etc., and may also include other information such as the store's area and the number of customers it can attract.

[0046] The output unit 150 has the function of outputting specified information according to instructions from the control unit 130. For example, the output unit 150 may output character information or image information, in which case the output unit 150 is realized by a monitor provided with or connected to the learning estimation system 100. Alternatively, for example, the output unit 150 may output audio information, in which case the output unit 150 is realized by a speaker provided with or connected to the learning estimation system. The output unit 150 may output, for example, estimated sales-related data.

[0047] The above is an example of the configuration of the learning estimation system 100.

[0048] <Configuration of store terminal 200> Figure 3 is a block diagram showing an example configuration of a store terminal 200. The store terminal 200 is a terminal related to a store and may be a computer system that manages information such as sales at that store. The store terminal 200 may be implemented using a PC, tablet terminal, server device, smartphone, etc., but is not limited to these. The store terminal 200 may or may not be installed inside the store, as long as it is related to the store.

[0049] As shown in Figure 3, the store terminal 200 includes a communication unit 210, an input unit 220, a control unit 230, a storage unit 240, and an output unit 250.

[0050] The communication unit 210 is a communication interface that has the function of communicating with external devices of the store terminal 200. The communication unit 210 has the function of communicating with the learning estimation system 100 as an external device. The communication unit 210 transmits store sales-related data to the learning estimation system 100 according to instructions from the control unit 230.

[0051] The input unit 220 has the function of receiving input from the operator of the store terminal 200 and transmitting the input content to the control unit 230. The input unit 220 may be implemented by an input device such as a mouse, keyboard, or touch panel, and in the case of voice input, it may be implemented by a microphone. As an example, the input unit 220 receives date data from a store employee indicating the date on which they want to predict sales or the number of customers, and transmits it to the control unit 230. As mentioned above, the date indicated by the date data may be a specific day, or it may be a predetermined period including multiple days (for example, the weekdays of the next week, the next month, or the three months of one year from now, but it is not limited to these and may be at the discretion of the store employee). The input unit 220 may also receive information on sales amount and the number of customers from the store employee and transmit it to the control unit 230.

[0052] The control unit 230 is a processor that has the function of controlling various parts of the store terminal 200. The control unit 230 may be implemented as a single-core or multi-core processor. The control unit 230 executes various programs stored in the memory unit 240 and utilizes various data to realize the functions of the store terminal 200.

[0053] The control unit 230 transmits sales-related data to the learning estimation system 100 and outputs sales-related data for a date specified by a store employee or the like. Furthermore, if the control unit 230 receives date data for which sales-related data should be predicted from the input unit 220, it may transmit that date data to the learning estimation system 100 via the communication unit 210. In this case, if there are plans for promotional activities on the date indicated by the date data, the control unit 230 may also transmit the promotional information along with the data.

[0054] The control unit 230 includes a sales-related data transmission unit 231 and a display control unit 232, which are the functions that the control unit 230 is supposed to perform.

[0055] The sales-related data transmission unit 231 transmits sales-related data stored in the storage unit 240 of the store terminal 200 to the learning estimation system 100 via the communication unit 210. The sales-related data transmission unit 231 may transmit new sales-related data to the learning estimation system 100 each time new sales-related data is stored in the storage unit 240, or it may transmit the data in batches at predetermined intervals (for example, every week, every month, etc., but not limited to these). The sales-related data transmission unit 231 may also transmit sales-related data in response to instructions from store employees at the store terminal 200. The sales-related data transmission unit 231 may transmit sales-related data, including at least one day's sales and customer count information, along with identification information that can identify the store, to the learning estimation system 100. In addition, it may also transmit other store-related information, such as information indicating the store's location (e.g., latitude and longitude information and address), information indicating the store's business type, store area, number of parking spaces, number of cash registers, number of seats, opening time, closing time, business hours, and promotional data if any promotions were conducted. If the store information is registered in the learning estimation system 100, this information does not need to be transmitted. The sales-related data transmission unit 231 may also transmit sales-related data to the learning estimation system 100 upon instruction from a store employee at the store terminal 200.

[0056] The display control unit 232 causes the output unit 250 to output information based on the sales forecast data transmitted from the learning estimation system 100.

[0057] The storage unit 240 has the function of storing various programs and data necessary for the operation of the store terminal 200. The storage unit 240 can be implemented using, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), or flash memory, but is not limited to these. The storage unit 240 may store various programs and data necessary to realize the functions that the store terminal 200 should perform. The storage unit 240 may also store sales-related data 241. The sales-related data 241 is a database containing information related to sales at the store that the store terminal 200 corresponds to. Sales-related information includes information such as daily sales amount and number of customers. The storage unit 240 may also store information indicating the store's location (such as latitude and longitude information and address), information indicating the store's business type, as well as information such as store area, number of parking spaces, number of cash registers, number of seats, opening time, closing time, business hours, and information related to sales promotions.

[0058] The output unit 250 has the function of outputting specified information according to instructions from the control unit 230. For example, the output unit 250 may output text information or image information, in which case it is implemented by a monitor provided on or connected to the store terminal 200. Alternatively, for example, the output unit 250 may output audio information, in which case it is implemented by a speaker provided on or connected to the learning estimation system. The output unit 250 displays information indicating estimated sales-related data, that is, information on the store's sales amount or the number of customers on a future date.

[0059] The above is an example of the configuration of the store terminal 200.

[0060] <Operation> Next, we will explain the operation of the learning estimation system 100 and the store terminal 200.

[0061] Figure 4 is a flowchart illustrating an example of how the learning estimation system 100 operates when learning sales-related data from a store terminal 200.

[0062] As shown in Figure 4, the acquisition unit 131 of the learning estimation system 100 acquires sales-related data transmitted from the store terminal 200 via the communication unit 110. The sales-related data includes information on the number of customers and sales at the store, and may also include promotional information if promotions are being conducted. The acquisition unit 131 also acquires store location data indicating the location of the store and store business type data indicating the type of business the store operates (step S401). The acquisition unit 131 may acquire the store location data indicating the location of the store and the store business type data indicating the type of business the store operates from the store terminal 200 via the communication unit 110, or it may acquire the data from the storage unit 140 if the store information is already stored in the storage unit 140. The acquisition unit 131 transmits the acquired sales-related data to the conversion unit 132.

[0063] The conversion unit 132 converts each acquired data into training data (step S402). The conversion unit 132 transmits the generated training data to the learning unit 133. The training data includes information indicating the location of the store, information indicating the type of business the store operates, and sales-related data for the store. It may also include weather information for the store's location, pedestrian traffic information in the vicinity of the store, calendar information indicating whether the date of the sales-related data is a weekday, holiday, or a day with a specific event, and promotional information indicating whether promotions were conducted on the date of the sales-related data.

[0064] The learning unit 133 trains the machine learning model 141 with the training data transmitted from the conversion unit 132 (step S403). As a result, the machine learning model 141 has also learned the sales-related data of the store terminal 200, enabling it to make more accurate predictions when forecasting the sales-related data of the store terminal 200.

[0065] Figure 5 is a flowchart illustrating an example of the operation when uploading sales-related data from the store terminal 200 to the learning estimation system 100.

[0066] As shown in Figure 5, the sales-related data transmission unit 231 of the store terminal 200 determines whether the timing for transmitting sales-related data has arrived (step S501). The transmission timing may be set in advance by the store employee of the store terminal 200, and may be, for example, the arrival of a specific time such as midnight on that day, but is not limited to this. The unit waits until the timing for transmitting sales-related data arrives (NO in step S501).

[0067] When the timing for transmitting sales-related data arrives (YES in step S501), the sales-related data transmission unit 231 retrieves untransmitted sales-related data from the storage unit 240 (step S502). Then, the sales-related data transmission unit 231 transmits the retrieved sales-related data to the learning estimation system 100 via the communication unit 210 (step S503). At this time, the sales-related data transmission unit 231 may also send identification information to identify the store, information indicating the store's location, and information indicating the store's business type, but this information does not need to be sent if the learning estimation system 100 has already stored it. In addition, if any sales promotions were conducted on that day, the transmitted sales-related data may include information about those promotions.

[0068] The store terminal 200 periodically transmits sales-related data to the learning estimation system 100, allowing the learning estimation system 100 to learn from this information and achieve accurate estimations when estimating sales-related data.

[0069] Figure 6 is a flowchart showing an example of how the learning estimation system 100 operates when estimating sales-related data for stores.

[0070] The reception unit 134 of the learning estimation system 100 receives store information and date data for which sales, customer numbers, or both are to be predicted (step S601). The reception unit 134 receives the date data from the store terminal 200 via the communication unit 110. The reception unit 134 may also receive store information from the store terminal 200 or acquire it from the storage unit 140. Store information includes at least the store's location and the type of business the store operates. The reception unit 134 transmits the acquired store location information and date data to the data acquisition unit 135. The reception unit 134 also transmits the acquired store information and date data to the estimation unit 136. If there are plans for a sales promotion at the store on the date indicated by the date data, the reception unit 134 may also receive that sales promotion information from the store terminal 200 and transmit it to the estimation unit 136.

[0071] The data acquisition unit 135 may acquire weather information, pedestrian flow data, or calendar information predicted for the date indicated by the transmitted date data at the location of the store, based on the transmitted store location (step S602). The data acquisition unit 135 transmits the acquired weather information or pedestrian flow data to the estimation unit 136. If weather information or pedestrian flow data is not used in estimating sales-related data, step S602 can be omitted. The estimation unit 136 inputs the date and time indicated by the transmitted date and time information, the store information including the store's location and business type, and, if necessary, weather information, pedestrian flow data, calendar information, and promotional information into the machine learning model 141 to estimate sales-related data for the date and time indicated by the date and time information (step S603).

[0072] The estimation unit 136 then outputs the estimated sales-related data (step S604). The estimation unit 136 outputs the estimated sales-related data by transmitting it to the store terminal 200 via the communication unit 110, and then terminates the process.

[0073] Figure 7 is a flowchart illustrating an example of how the store terminal 200 presents predicted sales-related data to the store clerk.

[0074] As shown in Figure 7, the store terminal 200 receives date data from the store clerk to the input unit 220 for which sales forecasts are to be made (step S701). The input unit 220 transmits the received date data to the control unit 230.

[0075] When the control unit 230 receives date data, it transmits the received date data and information indicating the store to the learning estimation system 100 via the communication unit 210 (step S702). The information indicating the store may include store identification information and information indicating the store's location and type of business. If the learning estimation system 100 has stored information indicating the store's location and type of business, only the store identification information is required.

[0076] The communication unit 210 of the store terminal 200 receives sales-related data estimated by the learning estimation system 100 (step S703). The communication unit 210 transmits the received sales-related data to the control unit 230.

[0077] When the display control unit 232 of the control unit 230 receives sales-related data, it outputs that information to the output unit 250 (monitor) (step S704) and terminates the process.

[0078] This allows store employees using the store terminal 200 to recognize the store's sales and customer numbers for a specified date, which can be used for purposes such as purchasing inventory.

[0079] The above is an example of the operation of the learning estimation system 100 and the store terminal 200 according to the embodiment.

[0080] <Summary> As described above, the learning estimation system 100 stores a machine learning model 141 that was originally trained to associate various sales-related data from different stores with the date of sale, the store's location, the store's business type, and the weather or foot traffic at that time. By preparing a machine learning model that has been trained in advance with sales-related data from stores in various locations and business types, it is possible to estimate sales-related data with a certain degree of accuracy even for small stores or new stores where it is difficult to prepare a large sample size of training data. The learning estimation system 100 then learns the sales-related data sent from the store terminal 200 by associating it with the store's location, business type, weather or foot traffic at that time, calendar information, promotional information, etc., and is therefore able to estimate the sales-related data of that store. The more store sales-related data the learning estimation system 100 learns, the better the estimation accuracy can be. Furthermore, the store terminal 200 simply sends sales-related data to the learning estimation system 100, which automatically converts it into training data in a format suitable for the machine learning model 141 and trains it. As a result, store employees and others using the store terminal 200 can easily obtain sales-related data for a specified date and time from the learning estimation system 100, which can then be used to improve store operations.

[0081] <Variation> It goes without saying that the learning estimation system 100 and store terminal 200 according to the above embodiment are not limited to the above embodiment and may be implemented by other methods. Various modifications will be described below.

[0082] (1) In the above embodiment, the learning and estimation system 100 was described as a computer system that performs both learning and estimation, but learning and estimation may be configured to be performed by separate devices, respectively.

[0083] (2) In the above embodiment, the learning estimation system 100 was described as estimating sales-related data for a specified date based on the store's location, business type, weather, or foot traffic. However, the learning estimation system 100 may also use other information about the store from the store terminal 200, namely store area, number of parking spaces, number of cash registers, number of seats, opening time, closing time, and business hours, to estimate sales-related data. In this case, the machine learning model 141 needs to be pre-trained with information used to estimate store area, number of parking spaces, number of cash registers, number of seats, opening time, closing time, and business hours. By adding this additional information, more accurate estimation can be expected.

[0084] (3) In the above embodiment, the learning estimation system 100 may include a supplement unit that receives sales-related data from the store terminal 200 for a certain period and, if there is a day within that period for which there is no sales-related data despite it being a business day, supplements the sales-related data for that day.

[0085] As an example, the supplementation unit may calculate the average value of sales-related data for the same day in the past for a number of years for which sales-related data is available, or for a predetermined number of years, and supplement the data with the calculated average value. The supplementation unit may also supplement the data with sales-related data estimated by the estimation unit 136, or, if it receives consecutive sales-related data for multiple days from the store terminal 200 and there is a day in between for which there is no sales-related data, it may supplement the data with sales-related data for that day.

[0086] The learning unit 133 may or may not use the supplemented data as training data for the store terminal 200.

[0087] (4) In the above embodiment, the learning estimation system 100 improves the accuracy of estimating store sales-related data by acquiring and learning store sales-related data. However, the learning estimation system 100 may also estimate sales-related data without store sales-related data, i.e., without learning store sales. In this case, only the store's location and business type are input to the learning estimation system 100 to estimate sales-related data. As an example of such use, if you want to open a new store in a specific location, you can acquire sales-related data for that location, so you can get information to judge whether the location is suitable for the new store.

[0088] (5) In the above embodiment, the learning estimation system 100 was shown to collect, learn, and estimate sales-related data on a daily basis, but this is not limited to that. The learning estimation system 100 may perform learning and estimation of sales-related data in units in which learning and estimation are desired. For example, sales-related data may be collected, learned, and estimated in time units according to the estimation unit required by the learning estimation system 100, such as hourly units, predetermined time units such as AM / PM, or monthly units. This will enable the learning estimation system 100 to perform estimations in a range of time units.

[0089] (6) In the above embodiment, the learning estimation system 100 was described in which the sales-related data acquired by the acquisition unit 131 is logarithmically transformed into training data and then trained by the learning unit 133. However, this is not limited to this. The transformation unit 132 may logarithmically transform the sales-related data, then calculate the difference between the logarithmically transformed value and the logarithmically transformed value of sales-related data from a different day, and use the calculated difference as training data. Here, the sales-related data from a different day may refer to sales-related data from a predetermined date prior to the acquired sales-related data, when the sales-related data estimated by the machine learning model 141 is sales-related data from a predetermined date later. Furthermore, the training data may be the value obtained by further inversely transforming the calculated difference. In this case, the output of the machine learning model 141 may be the value obtained by inversely transforming the difference between the logarithmically transformed value of the acquired sales-related data and the logarithmically transformed value of the estimated sales-related data from a predetermined date later.

[0090] Furthermore, the learning unit 133 may learn the difference between sales-related data that has not undergone logarithmic transformation as training data. That is, when the machine learning model 141 estimates sales-related data after a predetermined date, the learning unit 133 may learn the difference between the sales-related data acquired by the acquisition unit 131 and the sales-related data acquired before the predetermined date. This difference may also be a second-order difference. In the case of a first-order difference, the amount of change in sales-related data is learned, and in the case of a second-order difference, the change in the amount of change in sales-related data is learned. Furthermore, the learning unit 133 may learn the difference between both.

[0091] Furthermore, the learning unit 133 may learn training data that shows this difference. By learning the difference, the learning unit 133 can make the time series data (sales-related data along the time series) stationary. For time series data to be stationary means that the statistical properties of the time series data become less likely to change. Therefore, the machine learning model 141 that the learning unit 133 has learned and generated can perform estimations that correspond to various conditions on the day the sales-related data is estimated (following the trend).

[0092] (7) In the above embodiment, the estimation unit 136 of the learning estimation system 100 was described as estimating sales-related data for a date indicated by date data received by the reception unit 134, but this is not limited to that. The estimation unit 136 may estimate sales-related data for predetermined dates and transmit it to the store terminal 200. For example, if the store terminal 200 transmits sales-related data for a certain day, the estimation unit 136 may estimate sales-related data for a predetermined period from that day (for example, two weeks, but not limited to this).

[0093] (8) The program for the learning estimation system 100 of this disclosure to learn by converting store sales-related data into training data and to predict store sales-related data may be provided stored on a computer-readable storage medium. The storage medium is a “non-temporary tangible medium” capable of storing the program. The storage medium may include any suitable storage medium such as an HDD or SSD, or two or more suitable combinations thereof. The storage medium may be volatile, non-volatile, or a combination of volatile and non-volatile. However, the storage medium is not limited to these examples and may be any device or medium capable of storing the program.

[0094] The learning estimation system 100 can realize the functions of the multiple functional units shown in each embodiment by, for example, reading a program stored on a storage medium and executing the read program. Furthermore, the program may be provided to the learning estimation system 100 via any transmission medium (such as a communication network or broadcast waves). The learning estimation system 100 realizes the functions of the multiple functional units shown in each embodiment by, for example, executing a program downloaded via the Internet or the like. This program may be executed by the learning estimation system 100 or the like.

[0095] This program can be implemented using, but is not limited to, scripting languages ​​such as ActionScript and JavaScript®, object-oriented programming languages ​​such as Objective-C, Java®, and Python®, and markup languages ​​such as HTML5.

[0096] At least a portion of the processing in the learning estimation system 100 may be implemented by cloud computing, which consists of one or more computers. Furthermore, each functional unit of the learning estimation system 100 may be implemented by one or more circuits that implement the functions shown in the above embodiment, or the functions of multiple functional units may be implemented by one circuit.

[0097] (9) The above embodiments and their modifications may be combined as appropriate to the extent that they are in line with the objective of generating training data of sales-related data of a store and learning from it, and predicting sales-related data of the store.

[0098] (10) According to each aspect of the disclosure described above, it is possible to estimate sales-related data of stores and optimize store operations, thereby contributing to the achievement of Sustainable Development Goal (SDG) 9, "Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation." [Explanation of symbols]

[0099] 100 Learning Estimation Systems 110 Communications Department 120 Input section 130 Control Unit 131 Acquisition Department 132 Conversion section 133 Learning Department 134 Reception Department 135 Data Acquisition Unit 136 Estimation Department 140 Storage section 150 Output section 200 store terminals 210 Communications Department 220 Input section 230 Control Unit 231 Sales-related data transmission unit 232 Display Control Unit 240 Storage section 250 Output section

Claims

1. An acquisition unit that acquires store location data indicating the location of the store, store business type data indicating the type of business of the store, and sales-related data related to the daily sales of the store. A conversion unit that converts the data acquired by the acquisition unit into training data, A learning unit that trains a machine learning model, which estimates store sales-related data based at least on the store's location and business type, with the aforementioned training data. A learning estimation system equipped with the following features.

2. The acquisition unit further acquires weather data for the aforementioned date, The aforementioned machine learning model is further a model that estimates store sales-related data based on weather data. The learning estimation system described in command 1.

3. The acquisition unit further acquires pedestrian flow data for the store location on the aforementioned date, The aforementioned machine learning model further estimates store sales-related data based on pedestrian flow data. The learning estimation system described in command 1.

4. The acquisition unit further acquires calendar information indicating whether the date is a holiday or a weekday, and whether it is a day when a specific event occurred. The aforementioned machine learning model further estimates store sales-related data based on calendar information. The learning estimation system described in command 1.

5. The acquisition unit further acquires promotional data relating to the store's promotions on the aforementioned date, The aforementioned machine learning model is further a model that estimates store sales-related data based on promotional data. The learning estimation system described in command 1.

6. The conversion unit performs a logarithmic transformation on the sales-related data acquired by the acquisition unit and converts it into the training data. A learning estimation system according to any one of claims 1 to 5, characterized by the features described above.

7. The conversion unit performs a logarithmic transformation on the sales-related data acquired by the acquisition unit, performs a logarithmic transformation on the sales-related data from different days, and converts the difference between the two logarithmic sales-related data sets into the training data. The learning estimation system described in Item 6.

8. The conversion unit, when the sales-related data to be estimated using the machine learning model is sales-related data from a predetermined date later, performs a logarithmic transformation on the sales-related data acquired by the acquisition unit, and also performs a logarithmic transformation on the sales-related data from a predetermined date earlier than the sales-related data acquired by the acquisition unit, and converts the difference between the two logarithmic sales-related data sets into the training data. The learning estimation system described in Item 6.

9. The conversion unit performs an inverse logarithmic transformation on the difference between the two sales-related data to convert it into the training data. The learning estimation system described in statement 8.

10. The conversion unit, when the sales-related data to be estimated using the machine learning model is sales-related data from a predetermined date later, converts the information based on the difference between the sales-related data acquired by the acquisition unit and the sales-related data acquired before the predetermined date of said sales-related data into the training data. A learning estimation system according to any one of claims 1 to 5.

11. A reception desk that accepts input of store location, store type, and date data related to sales-related data, The reception unit inputs the date data received by the reception unit into the machine learning model and estimates sales-related data for the date indicated by the date data at the store; The output unit outputs the sales-related data estimated by the estimation unit, The learning estimation system according to claim 1, comprising:

12. The system includes a data acquisition unit that acquires weather data at the location of the store based on the location of the store. The aforementioned machine learning model is further a model that estimates store sales-related data based on weather data. The estimation unit further inputs the weather data acquired by the data acquisition unit into the machine learning model to estimate sales-related data. The learning estimation system described in command 11.

13. The system includes a data acquisition unit that acquires weather data at the location of the store based on the location of the store. The aforementioned machine learning model is further a model that estimates store sales-related data based on weather data. The estimation unit further inputs the weather data acquired by the data acquisition unit into the machine learning model to estimate sales-related data. The learning estimation system described in command 11.

14. The system includes a data acquisition unit that acquires calendar information indicating whether the date indicated by the aforementioned date data is a holiday or a weekday, and whether it is a day on which a specific event occurred. The aforementioned machine learning model is further a model that estimates store sales-related data based on calendar information. The estimation unit further inputs the calendar information acquired by the data acquisition unit into the machine learning model to estimate sales-related data. The learning estimation system described in command 11.

15. The system includes a data acquisition unit that acquires promotional data relating to the sales promotion of the store on the date indicated in the aforementioned date data. The aforementioned machine learning model is further a model that estimates store sales-related data based on promotional data. The estimation unit further inputs the promotional data acquired by the data acquisition unit into the machine learning model to estimate sales-related data. The learning estimation system described in command 11.

16. If the acquisition unit is unable to acquire sales-related data for a specific date, the acquisition unit includes a compensation unit that compensates for the sales-related data for that specific date. The estimation unit estimates the sales-related data for the date indicated by the date data, based on the sales-related data supplemented by the supplementation unit. The learning estimation system described in command 11.

17. The supplementation unit uses past sales data of the store, specifically sales-related data for the same day of the week as the specific date, to supplement the sales-related data for that specific date. The learning estimation system described in command 16.

18. The learning unit learns the sales-related data that the supplementation unit has supplemented. The learning estimation system according to feature 17.

19. Computers Acquisition step: Acquisition step of acquiring store location data indicating the location of the store, store business type data indicating the type of business of the store, and sales-related data related to the daily sales of the store. The acquisition step includes a conversion step that converts the acquired data into training data, A learning step in which a machine learning model that estimates store sales-related data based on at least the store's location and business type is trained with the aforementioned training data, A learning estimation method that performs this task.

20. On the computer, A data acquisition function that acquires store location data indicating the store's location, store business type data indicating the store's business type, and sales-related data relating to the store's daily sales. The aforementioned acquisition function includes a conversion function that converts the acquired data into training data, A machine learning model that estimates sales-related data for stores based on at least the store's location and business type is trained on the aforementioned training data, and A learning estimation program that achieves this.

Images