Program and data processing device

Abstract

This program provides an estimate of the likelihood of an advertiser purchasing a product or service that is being advertised. [Solution] The program is characterized in that it causes a computer to perform a process that acquires first pedestrian flow data representing the flow of people in a designated area who have not yet purchased the advertised target, and estimates the likelihood of purchase by advertisers for the advertised target based on the first pedestrian flow data and a trained model, and the trained model is generated by machine learning a plurality of training data that define the relationship between purchase data of actual purchasers for the advertised target and second pedestrian flow data representing the flow of people in the designated area of ​​those actual purchasers.

Description

Technical Field

[0001] This invention relates to programs and data processing devices.

Background Art

[0002] Manufacturers often want to know which customers are buying the products they manufacture. For example, if they can grasp the fact that their products are being purchased by many women in their 20s, they can utilize this fact in the development of new products.

[0003] For this reason, stores may collect ID-POS data that associates a customer ID (Identifier) that uniquely identifies a customer with POS (Point Of Sales) data indicating the sales performance of products, and provide it to the manufacturer. Also, an operator (so-called "data platform provider") that operates and provides a data platform may collect ID-POS data from stores and provide it to the manufacturer (see, for example, Patent Document 1).

[0004] Here, it is known that ID-POS data includes, for example, the date and time when a product was purchased, the customer ID of the customer who purchased the product, the purchased product, the unit price and the number of purchased products, and the total amount of the purchased products. Also, customer data including the customer ID, gender, postal code of the customer's residence, and birth month is known (see, for example, Patent Document 2).

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0006] As mentioned above, both ID-POS data and customer data include a customer ID. Therefore, if ID-POS data and customer data are provided from a store to a manufacturer, the manufacturer can estimate customer attributes from the ID-POS data based on the customer ID common to both the ID-POS data and the customer data. For example, the manufacturer can estimate customer attributes such as gender and birth month from the ID-POS data. Furthermore, based on the date and time of purchase and the birth month, the manufacturer can also estimate the age group in which the product was purchased as a customer attribute.

[0007] However, for customers for whom ID-POS data is unavailable, it is difficult for manufacturers to estimate the aforementioned customer attributes. In other words, for new products that have not yet entered the market, ID-POS data is not generated, making it difficult for manufacturers to estimate the likelihood of purchase by those who have not yet purchased the product.

[0008] Furthermore, even for older products that have already been released to the market, ID-POS data is generated at each store, meaning that each store can only collect ID-POS data for a fraction of the actual purchasers of older products. For this reason, it is difficult for individual stores and manufacturers to accurately estimate the overall purchase potential of older products based solely on ID-POS data.

[0009] Furthermore, advertisements for products are often delivered to target audiences based on such purchase likelihood. Therefore, if purchase likelihood cannot be estimated, there is a possibility that appropriate advertisements for products will not be delivered to the target audience. This possibility applies not only to products but also to services.

[0010] Therefore, one aspect of this is to provide a program and data processing device that estimates the likelihood of purchase by advertisers for advertised products and services. [Means for solving the problem]

[0011] In one embodiment, the program causes a computer to perform a process that acquires first pedestrian flow data representing the flow of people in a designated area who have not yet purchased the advertised target, and estimates the likelihood of purchase by advertisers of the advertised target based on the first pedestrian flow data and a trained model, wherein the trained model is generated by machine learning a plurality of training data that define the relationship between purchase data of actual purchasers of the advertised target and second pedestrian flow data representing the flow of people in the designated area of ​​those actual purchasers.

[0012] In the above configuration, the first human flow data, the second human flow data, and the purchase data all include unique advertising identifiers held by multiple mobile terminals, and the processing can be configured to identify a portion of the advertising identifiers based on the likelihood of purchase, and to use the identified portion of the advertising identifiers in the first human flow data.

[0013] In the above configuration, the purchase data includes the attributes of the actual purchaser, and the processing can be configured to estimate the purchase probability of the advertiser whose attributes are common to the actual purchaser.

[0014] In the above configuration, the purchase data may include either the number of purchases or the purchase frequency of the actual purchaser, and the processing may be configured to estimate the purchase tendencies of the advertisers based on either the number of purchases or the purchase frequency.

[0015] In one embodiment, the program causes the computer to perform the following processes: acquire first pedestrian flow data representing the flow of people in a designated area of ​​non-purchasers of an advertised target; generate visit history data of the non-purchaser to a specific facility based on the first pedestrian flow data and time data representing the time the non-purchaser stays at that specific facility included in the designated area; estimate the advertised target identifier of the advertised target to recommend to the non-purchaser based on the visit history data of the non-purchaser and a trained model; and estimate the likelihood of the advertiser purchasing the advertised target based on the advertised target tag representing the characteristics of the advertised target identified by the advertised target identifier and the attributes of the non-purchaser. The completed model is characterized by being generated by machine learning a plurality of training data that define the relationship between an advertising target tag representing the characteristics of the advertising target extracted based on purchase data of actual buyers for the advertising target, a facility tag representing the characteristics of the specific facility, which is generated based on a common attribute and is generated based on a second flow data representing the flow of people in the designated area of ​​the actual buyers and time data representing the time the actual buyers stay at the specific facility, and a combination data of the actual buyer's buyer identifier and the advertising target identifier of the advertising target.

[0016] In the above configuration, the first human flow data, the second human flow data, and the purchase data all include unique advertising identifiers held by multiple mobile terminals, and the processing can be configured to identify a portion of the advertising identifiers based on the likelihood of purchase, and to use the identified portion of the advertising identifiers in the first human flow data.

[0017] In the above configuration, the process may be configured such that, when it detects the input or selection of confirmation items related to the advertisement being advertised via a predetermined screen, it searches for the attributes of the non-purchaser corresponding to the confirmation items, and outputs the search results, which include the combination of the non-purchaser's attributes and the probability that the advertisement will be purchased by the non-purchaser, as the purchase possibility of the non-purchaser on the predetermined screen or a separate screen.

[0018] In the above configuration, the process may be configured to acquire demographic data in the designated area and calculate the sales quantity of the advertised item in the designated area based on the demographic data and the purchase probability.

[0019] In the above configuration, the process may be configured to acquire unit price data representing the unit price of the advertised item, and to calculate the total purchase amount of the advertised item by the non-purchaser in the designated area based on the sales quantity and the unit price data.

[0020] In one embodiment, the data processing device includes an acquisition unit that acquires first pedestrian flow data representing the flow of people in a designated area of ​​non-purchasers of the advertised target, and an estimation unit that estimates the likelihood of purchase by advertisers of the advertised target based on the first pedestrian flow data and a trained model, wherein the trained model is generated by machine learning a plurality of training data that define the relationship between purchase data of actual purchasers of the advertised target and second pedestrian flow data representing the flow of people in the designated area of ​​the actual purchasers.

[0021] In one embodiment, a data processing apparatus includes an acquisition unit that acquires first pedestrian flow data representing the pedestrian flow in a designated area of non-purchasers for an advertising target, a generation unit that generates visit history data of the non-purchasers for a specific facility based on the first pedestrian flow data and time data representing the time the non-purchasers stay at the specific facility included in the designated area, a first estimation unit that estimates an advertising target identifier of the advertising target to be recommended to the non-purchasers based on the visit history data of the non-purchasers and a learned model, and a second estimation unit that estimates the purchase probability of an advertising delivery target for the advertising target based on an advertising target tag representing the characteristics of the advertising target identified by the advertising target identifier of the advertising target and the attributes of the non-purchasers. The learned model is generated by machine learning of a plurality of teacher data respectively defining the relationship between tag data obtained by combining, based on a common attribute, an advertising target tag representing the characteristics of the advertising target extracted based on the purchase data of actual purchasers for the advertising target, visit history data of the actual purchasers for the specific facility generated based on second pedestrian flow data representing the pedestrian flow of the actual purchasers in the designated area and time data representing the time the actual purchasers stay at the specific facility, and a facility tag representing the characteristics of the specific facility extracted based on the visit history data of the actual purchasers, and combined data of the purchaser identifier of the actual purchaser and the advertising target identifier of the advertising target.

Effect of the Invention

[0022] According to the present case, it is possible to estimate the purchase probability of an advertising delivery target for an advertising target such as a product or service.

Brief Description of the Drawings

[0023] [Figure 1] It is an example of a data processing system. [Figure 2] It is an example of the hardware configuration of a data processing server. [Figure 3] It is an example of the functional configuration of a data processing server. [Figure 4]This is an example of customer flow data from actual buyers. [Figure 5] This is an example of visit history data. [Figure 6] This is an example of buyer attribute data. [Figure 7] This is an example of product data. [Figure 8] This is an example of ID-POS data. [Figure 9] This is an example of POI data. [Figure 10] This is an example of demographic data. [Figure 11] This flowchart shows an example of how a data processing server operates. [Figure 12] This is an example of assigning purchasing tendencies to foot traffic data of non-purchasers. [Figure 13] This flowchart shows another example of how a data processing server operates. [Figure 14] (a) is an example of a first knowledge graph. (b) is an example of a second knowledge graph. (c) is an example of a knowledge graph collection. [Figure 15] This is an example of a screen for confirming purchase potential. [Modes for carrying out the invention]

[0024] The following describes the implementation of this invention with reference to the drawings. In the embodiments described later, a product is used as an example of the advertising target, but the advertising target is not limited to a product; it may also be a service such as a restaurant.

[0025] (First Embodiment) As shown in Figure 1, the data processing system ST is a computer system that includes a terminal device 10 and a data processing server 100. The terminal device 10 and the data processing server 100 are connected via a communication network NW. The communication network NW includes either a LAN (Local Area Network) or the Internet, or both.

[0026] In Figure 1, a PC (Personal Computer) is shown as an example of a terminal device 10, but the terminal device 10 is not limited to a PC. The terminal device 10 may also be a smart device such as a smartphone or tablet. Also, in Figure 1, a physical server device is shown as an example of a data processing server 100, but the data processing server 100 may also be a virtual server device. Furthermore, in Figure 1, one data processing server 100 is shown as an example, but multiple data processing servers 100 may be provided in the data processing system ST, and various data processing may be distributed among the multiple data processing servers 100.

[0027] The data processing system ST is used by users 11 belonging to business companies. Business companies may be manufacturers that produce goods or non-manufacturers that provide services. Manufacturers include, but are not limited to, food manufacturers, cosmetics manufacturers, and shoe manufacturers. Non-manufacturers include, but are not limited to, restaurants, retail stores, etc.

[0028] User 11 can use the data processing system ST by operating the input device 12 provided on the terminal device 10 and accessing the data processing server 100. For example, when user 11 performs a predetermined operation on the input device 12, the control device 13 of the terminal device 10 sends an instruction corresponding to the predetermined operation to the data processing server 100. Upon receiving the instruction, the data processing server 100 performs various data processing based on the received instruction and sends the processing results to the control device 13.

[0029] As will be explained in more detail later, for example, when the data processing server 100 receives instructions corresponding to a predetermined operation, it estimates the likelihood of purchase by non-buyers who have not yet purchased the new product (a product that has not yet been released to the market), and transmits the estimated likelihood of purchase as a processing result to the control device 13. When the control device 13 receives the processing result, it displays a predetermined screen including the processing result on the display device 14 of the terminal device 10. As a result, a predetermined screen appears on the display device 14. By viewing the predetermined screen, the user 11 can understand the likelihood of purchase by non-buyers who have not yet purchased the new product.

[0030] In this way, the data processing server 100 can estimate the purchase probability of non-purchasers, enabling it to deliver appropriate advertisements for new products to target audiences. Furthermore, even for older products that have already been released to the market, the data processing server 100 can accurately estimate the purchase probability of all older products based solely on ID-POS data from individual stores and manufacturers. Therefore, it can also deliver appropriate advertisements for older products to target audiences.

[0031] The purchase possibility may include, for example, the number of people who can purchase in a designated area, the probability of purchase, the certainty of purchase, or the quantity or amount that can be purchased in a designated area. The data processing server 100 may also estimate the purchasing tendencies of non-purchasers, such as frugal or extravagant, and transmit the estimated purchasing tendencies to the control device 13 as a processing result. In this case, the user 11 can understand the purchasing tendencies of non-purchasers. In this way, the data processing server 100 can estimate the purchase possibility and purchasing tendencies of non-purchasers for new and old products.

[0032] Referring to Figure 2, the hardware configuration of the data processing server 100 will be described. Note that the terminal device 10 described above has essentially the same hardware configuration as the data processing server 100, so a detailed explanation will be omitted.

[0033] The data processing server 100 includes a CPU (Central Processing Unit) 100A as a processor, and RAM (Random Access Memory) 100B and ROM (Read Only Memory) 100C as memory. The data processing server 100 also includes a network interface 100D and an HDD (Hard Disk Drive) 100E. An SSD (Solid State Drive) may be used instead of the HDD (Hard Disk Drive) 100E.

[0034] The data processing server 100 may include, as necessary, at least one of the following: input I / F 100F, output I / F 100G, input / output I / F 100H, and drive device 100I. The CPU 100A to the drive device 100I are connected to each other by an internal bus 100J. In other words, the data processing server 100 can be implemented by a computer.

[0035] Input I / F 100F is connected to an input device 710. Examples of input devices 710 include keyboards, mice, and touch panels. Output I / F 100G is connected to a display device 720. Examples of display devices 720 include liquid crystal displays. Input / output I / F 100H is connected to a semiconductor memory 730. Examples of semiconductor memory 730 include USB (Universal Serial Bus) memory and flash memory. Input / output I / F 100H reads programs stored in the semiconductor memory 730. Input I / F 100F and Input / output I / F 100H are equipped with, for example, USB ports. Output I / F 100G is equipped with, for example, a DisplayPort.

[0036] A portable recording medium 740 is inserted into the drive unit 100I. The portable recording medium 740 can be a removable disk such as a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc). The drive unit 100I reads the program recorded on the portable recording medium 740. The network interface 100D includes, for example, a LAN port and a communication circuit. The communication circuit includes either a wired communication circuit or a wireless communication circuit, or both. The network interface 100D is connected to a communication network NW.

[0037] The CPU 100A temporarily stores programs stored in at least one of the ROM 100C, HDD 100E, and semiconductor memory 730 in RAM 100B. The CPU 100A also temporarily stores programs recorded on the portable recording medium 740 in RAM 100B. By executing the stored programs, the CPU 100A realizes various functions described later and also executes data processing methods including various processes described later. The programs should conform to the flowchart described later.

[0038] The functional configuration of the data processing server 100 will be described with reference to Figures 3 to 10. Figure 3 shows the main functions of the data processing server 100.

[0039] As shown in Figure 3, the data processing server 100 includes a storage unit 110, a processing unit 120, and a communication unit 130. The storage unit 110 can be implemented by either the RAM 100B or the HDD 100E, or both. The processing unit 120 can be implemented by the CPU 100A described above. The communication unit 130 can be implemented by the network interface 100D described above.

[0040] The memory unit 110, processing unit 120, and communication unit 130 are interconnected. The memory unit 110 includes a human flow memory unit 111, a visit history memory unit 112, a buyer attribute memory unit 113, and a product information memory unit 114. The memory unit 110 also includes a purchase history memory unit 115, a POI (Point of Interest) information memory unit 116, and a demographics memory unit 117. The memory unit 110 stores various data using the human flow memory unit 111, the visit history memory unit 112, the buyer attribute memory unit 113, the product information memory unit 114, the purchase history memory unit 115, the POI information memory unit 116, and the demographics memory unit 117.

[0041] The processing unit 120 includes a training data generation unit 121, a model generation unit 122, and a potential estimation unit 123. The processing unit 120 processes various data using the training data generation unit 121, the model generation unit 122, and the potential estimation unit 123.

[0042] The pedestrian flow memory unit 111 stores pedestrian flow data representing the flow of people in a designated area who are actual buyers of the old product. The old product is a product that was released to the market before the new product mentioned above. The pedestrian flow memory unit 111 also stores pedestrian flow data representing the flow of people in a designated area who are not actual buyers. The pedestrian flow data of non-buyers is an example of the first pedestrian flow data, and the pedestrian flow data of actual buyers is an example of the second pedestrian flow data.

[0043] For example, as shown in Figure 4, the foot traffic data of actual purchasers includes multiple items such as personal ID (Identifier), advertising ID, latitude, longitude, and measurement date and time. The personal ID field registers a unique identifier that identifies the individual purchaser. The advertising ID field registers a unique identifier that is held by the purchaser's mobile device and is used solely for delivering advertisements within the application software (hereinafter simply referred to as "app") on that mobile device. The mobile device may be a smartphone, tablet, smartwatch, or gaming device.

[0044] The latitude, longitude, and measurement date and time fields register the latitude and longitude of the mobile device, measured, for example, by the GPS (Global Positioning System) function of the mobile device, along with the measurement date and time. Alternatively, the latitude and longitude of the mobile device estimated based on the signal strength of Bluetooth® beacons transmitted wirelessly between the mobile device carried by the actual purchaser and beacon terminals installed in various facilities described later, and the installation location of the beacon terminals, may also be registered. In this way, the latitude and longitude that identify the location of the mobile device are measured periodically by the GPS function, etc. Therefore, if an actual purchaser moves while carrying a mobile device, the movement of the actual purchaser is represented as pedestrian flow. Note that the pedestrian flow data of non-purchasers is basically the same as the pedestrian flow data of actual purchasers, so a detailed explanation is omitted.

[0045] Returning to Figure 3, the visit history storage unit 112 stores visit history data representing the history of visits made by actual purchasers to POIs. A POI is an example of a specific facility and includes commercial facilities such as shops, including restaurants and retail stores. A POI may also include public facilities such as parks, libraries, and train stations, sports facilities such as baseball fields and soccer stadiums, medical facilities such as hospitals and clinics, and roads such as sidewalks and roadways. Roads may be expressways (such as national expressways and urban expressways) including service areas (SAs) and parking areas (PAs), or general roads other than expressways including intersections and T-junctions. It should be noted that POIs are not limited to such man-made objects, and natural objects such as mountains, rivers, and lakes may also be included. In this way, a POI corresponds to a specific feature on map information, such as a man-made object or a natural object.

[0046] As shown in Figure 5, the visit history data includes multiple items such as personal ID, POI name, POI tag, visit date, start time of stay, end time of stay, and duration of stay. The personal ID field registers a unique identifier that identifies actual purchasers and non-purchasers. The POI name field registers the name of the POI visited by the actual purchaser or non-purchaser. The POI tag field registers the characteristics of the POI associated with it as POI tags. For example, if word-of-mouth information about the POI is posted on the internet, several words included in the word-of-mouth information will be registered as POI tags. The visit date field registers the date on which the actual purchaser or non-purchaser visited the POI.

[0047] The entries for start time of stay, end time of stay, and duration of stay record the time when the person began staying at the POI, the time when they ended their stay at the POI, and the duration of their stay at the POI. If actual buyers or non-buyers remain at a specific location for a certain period of time, it is estimated that they stayed at the POI located at that specific location for a certain period of time. The visit history data is generated based on the pedestrian flow data of actual buyers and non-buyers and the POI data described later, and is stored in the visit history storage unit 112.

[0048] Returning to Figure 3, the buyer attribute storage unit 113 stores buyer attribute data representing the attributes of actual buyers. As shown in Figure 6, the buyer attribute data includes multiple items such as buyer ID, gender, age group, and occupation. The buyer attribute data may also include items such as the postal code of the residential area, date of birth, and nationality. The buyer ID item registers a unique identifier that identifies actual buyers and non-buyers. The gender item registers the gender of actual buyers and non-buyers. The age group item registers the age group based on the date of birth of actual buyers and non-buyers. The occupation item registers the occupation of actual buyers and non-buyers. Note that the buyer attribute data is generated based on information entered when issuing a membership card usable at a store or information entered into a member app, and stored in the buyer attribute storage unit 113.

[0049] Returning to Figure 3, the product information storage unit 114 stores product data related to new and old products. As shown in Figure 7, the product data includes multiple items such as product ID, product name, product description, product tag, manufacturer, unit price, and price range. The product ID field registers a unique identifier that identifies the new or old product. For example, a JAN (Japanese Article Number) code may be used as the identifier registered in the product ID field. The product name field registers the name of the new or old product. The product description field registers a text that describes the characteristics of the new or old product.

[0050] The product tag field registers some of the words included in the description as product characteristics. The number of product characteristics registered in the product tag field may be one or more. The manufacturer field registers the name of the manufacturer that produces the new or old product. The unit price field registers the unit price of the new or old product. The price range field registers the price range of the new or old product, such as high price range or low price range.

[0051] Returning to Figure 3, the purchase history storage unit 115 stores ID-POS data that links the buyer ID with POS data showing the sales performance of the old product. ID-POS data is an example of purchase data of an actual buyer. As shown in Figure 8, ID-POS data includes multiple items such as purchase date and time, buyer ID, advertisement ID, product ID, unit price, quantity, and total amount. The purchase date and time item registers the date and time when the actual buyer, identified by the buyer ID, purchased the old product.

[0052] The Buyer ID field registers a unique identifier that identifies the actual purchaser. The Ad ID field registers a unique identifier that is held by the actual purchaser's mobile device and is used solely for delivering advertisements within the app on that mobile device. The Product ID field registers a unique identifier that identifies the old product. The Unit Price field registers the unit price of the old product. The Quantity field registers the quantity of the old product purchased by the actual purchaser. The Total Amount field registers the result of multiplying the unit price of the old product registered in the Unit Price field by the quantity of the old product registered in the Quantity field.

[0053] Returning to Figure 3, the POI information storage unit 116 stores the POI data related to the POI described above. As shown in Figure 9, the POI data includes multiple items such as POI-ID, POI name, POI tag, latitude range, longitude range, and price range. The POI-ID field stores a unique identifier that identifies the POI. The POI name field stores the name of the POI. The POI tag field stores the POI tag described above.

[0054] The latitude range field registers the latitude range in which the POI is located. The longitude range field registers the longitude range in which the POI is located. The latitude range and longitude range registered in each field uniquely identify the area occupied by the POI on the map information. Therefore, if the latitude and longitude included in the pedestrian flow data are within the area occupied by the POI, it is presumed that actual buyers or non-buyers located at those latitudes and longitudes visited the POI. The price range field registers the price ranges of old and new products handled at the POI.

[0055] Returning to Figure 3, the demographic data storage unit 117 stores demographic data representing the population statistics of the designated area. As shown in Figure 10, the demographic data includes multiple items such as postal code, area name, population by age group, and gender ratio. The postal code item registers the postal code that identifies the designated area. The area name item registers the name of the designated area (e.g., city or town name). The station name item may be used instead of the area name item. In this case, the station name item registers the name of the designated station. The population by age group item registers the population by age group in the designated area. In Figure 10, the population of people in their 30s is shown as an example, and other age groups are omitted. The gender ratio item registers the gender ratio in the designated area. Note that the population by age group and gender ratio in the designated area can be obtained using information provided by the local government that has jurisdiction over the designated area.

[0056] Returning to Figure 3, the training data generation unit 121 acquires pedestrian flow data of actual purchasers in the designated area (see Figure 4) from the pedestrian flow storage unit 111. The training data generation unit 121 also acquires ID-POS data (see Figure 8) from the purchase history storage unit 115. Once the training data generation unit 121 acquires the pedestrian flow data of actual purchasers and the ID-POS data, it generates multiple training data sets that define the relationship between the pedestrian flow data of actual purchasers and the ID-POS data, respectively.

[0057] For example, the training data generation unit 121 generates multiple training data based on the identifier of an actual purchaser that is included in both the actual purchaser's foot traffic data and the ID-POS data. The training data generation unit 121 may also generate multiple training data based on the identifier of an advertisement that is included in both the actual purchaser's foot traffic data and the ID-POS data. In this way, the training data generation unit 121 can generate multiple training data that define the relationship between the actual purchaser's foot traffic data and the ID-POS data, respectively.

[0058] The model generation unit 122 generates a trained model by machine learning through multiple training data generated by the training data generation unit 121. Here, there is a high probability that there is a correlation between the flow of actual buyers and the purchase of old products by actual buyers. Therefore, the model generation unit 122 adjusts and calculates coefficients that satisfy this correlation. In this way, the model generation unit 122 can estimate the correlation between the flow of actual buyers and the purchase of old products by actual buyers.

[0059] The potential estimation unit 123 acquires pedestrian flow data of non-purchasers in a designated area from the pedestrian flow memory unit 111. Once the potential estimation unit 123 acquires the pedestrian flow data of non-purchasers, it estimates the likelihood of non-purchasers purchasing the new product based on the pedestrian flow data of non-purchasers and the trained model generated by the model generation unit 122. Specifically, the potential estimation unit 123 acquires buyer attribute data (see Figure 6) from the buyer attribute memory unit 113 and estimates the likelihood of purchase by non-purchasers whose attributes are the same as those of actual buyers.

[0060] The potential estimation unit 123 may identify a portion of the advertisement identifier based on the likelihood of purchase, and may use the identified portion of the advertisement identifier in the foot traffic data of non-purchasers. This allows the potential estimation unit 123 to appropriately deliver advertisements for new products to non-purchasers with a high likelihood of purchase. As a result, the willingness of non-purchasers to purchase the new product is stimulated. The likelihood of purchase may be a degree of purchase probability, such as high or low purchase probability, or a numerical value such as purchase probability.

[0061] Furthermore, the potential estimation unit 123 can estimate the number of purchases and the purchase frequency of actual purchasers based on ID-POS data. Therefore, the potential estimation unit 123 can estimate the purchasing tendencies of non-purchasers, defined by various purchasing factors such as frugal or spendthrifty, based on either the estimated number of purchases or the purchase frequency of actual purchasers. Note that the purchasing factors are not particularly limited to frugal or spendthrifty. Various terms representing purchasing habits may be used as purchasing factors.

[0062] The operation of the data processing server 100 will be explained with reference to Figure 11.

[0063] First, the training data generation unit 121 generates training data (step S1). For example, when the training data generation unit 121 receives instructions corresponding to a predetermined operation, it acquires actual customer flow data and ID-POS data, and generates multiple training data sets that define the relationship between the actual customer flow data and ID-POS data.

[0064] When the training data generation unit 121 generates training data, the model generation unit 122 then generates a trained model using machine learning (step S2). As described above, the model generation unit 122 generates a trained model by machine learning multiple training data. When the model generation unit 122 generates a trained model, the potential estimation unit 123 then obtains pedestrian flow data of non-purchasers from the pedestrian flow memory unit 111 (step S3).

[0065] Once the potential estimation unit 123 acquires the flow data of non-purchasers, it estimates the likelihood of non-purchasers purchasing the new product based on the flow data of non-purchasers and the trained model (step S4). For example, as shown in Figure 12, the potential estimation unit 123 can assign purchasing tendencies of non-purchasers, such as frugal or extravagant, to the flow data of non-purchasers, who are identified by identifiers different from those of actual purchasers. Once the likelihood of non-purchasers purchasing has been estimated, the potential estimation unit 123 terminates its processing.

[0066] Thus, according to the first embodiment, the data processing server 100 can generate a trained model by machine learning multiple training data sets that define the relationship between actual purchaser foot traffic data and ID-POS data. Furthermore, once the trained model is generated, the data processing server 100 can estimate the likelihood of non-purchasers purchasing a new product based on the foot traffic data of non-purchasers and the trained model. By using the data processing server 100, users 11 can grasp the likelihood of purchasing a new product and develop effective new products for the market without waste.

[0067] (Second Embodiment) The second embodiment of this invention will be described with reference to Figures 13 to 15. In the first embodiment, it was explained that the likelihood of non-purchasers purchasing the product and the purchasing tendencies of non-purchasers, defined by various purchasing factors such as frugality and spending, are estimated. In the second embodiment, it will be explained that the number of non-purchasers who are likely to purchase the new product in a designated area, the purchase probability of non-purchasers purchasing the new product, and the planned purchase quantity of the new product in the designated area are estimated.

[0068] First, as shown in Figure 13, the training data generation unit 121 extracts product tags (step S11). More specifically, the training data generation unit 121 first obtains ID-POS data (see Figure 8) from the purchase history storage unit 115 and extracts the identifiers of old products registered in the ID-POS data along with the identifiers of the actual purchasers. After extracting the identifiers of old products, the training data generation unit 121 then obtains product data (see Figure 7) from the product information storage unit 114 and extracts the product characteristics of the old products associated with the identifiers of the old products as product tags.

[0069] Here, product tags may be defined by different words. However, even if defined by different words, product tags may share a common concept or meaning. For this reason, the training data generation unit 121 attempts to unify product tags that share a common concept, for example, based on LLM (Large Language Models). For example, the training data generation unit 121 identifies one of the product tags that share a common concept as a representative product tag. This reduces the number of product tags and improves the processing speed of subsequent processing.

[0070] After extracting product tags, the training data generation unit 121 then extracts POI tags (step S12). More specifically, the training data generation unit 121 first obtains visit history data (see Figure 5) generated based on actual purchaser foot traffic data from the visit history storage unit 112, and extracts the POI tags registered in the visit history data along with the individual identifiers of the actual purchasers.

[0071] Even if POI tags are defined with different words, they may share a common concept or meaning. Therefore, the training data generation unit 121 unifies POI tags with a common concept based on LLM and identifies representative POI tags. This reduces the number of POI tags and improves the processing speed of subsequent processing. Note that POI tags are an example of facility tags.

[0072] After extracting POI tags, the training data generation unit 121 then combines the product tags and POI tags (step S13). For example, the training data generation unit 121 combines the product tags and POI tags based on predetermined rules, such as CDR (Cross Domain Recommendation). For example, the training data generation unit 121 can combine the product tags and POI tags based on the degree of overlap of words contained in each of the product tags and POI tags. The training data generation unit 121 may also discover word overlaps based on LLM and combine the product tags and POI tags.

[0073] When product tags and POI tags are combined, the training data generation unit 121 generates training data (step S14). More specifically, the training data generation unit 121 generates multiple training data sets that define the relationships between tag data formed by combining product tags and POI tags, and combination data that combines the identifier of an individual actual purchaser with the identifier of a previous product. This associates individual actual purchasers who have both a product domain representing the area of ​​the previous product and a POI domain representing the area of ​​the POI, and identifies the purchasing preferences of individual actual purchasers.

[0074] Once training data is generated, the model generation unit 122 generates a knowledge graph of POIs by machine learning the training data (step S15). By using this knowledge graph as a trained model, the product domain is inferred from the POI domain.

[0075] For example, based on ID-POS data (see Figure 8) and product data (see Figure 7), actual buyer P1, identified by buyer ID "P#001", purchased the old product "Chicken Breast Bar". Also, based on visit history data (see Figure 5), actual buyer P1, identified by personal ID "P#001", visited the POI "Fitness Gym". Here, the old product "Chicken Breast Bar" is associated with the product tag "Health-Oriented". Also, the POI "Fitness Gym" is associated with the POI tag "Health-Oriented". As a result, when the model generation unit 122 generates a knowledge graph, a first knowledge graph KG1 related to actual buyer P1 is generated, as shown in Figure 14(a).

[0076] Similarly, based on ID-POS data (see Figure 8) and product data (see Figure 7), actual buyer P2, identified by buyer ID "P#002", purchased the old product "cosmetics". Also, based on visit history data (see Figure 5), actual buyer P2, identified by personal ID "P#002", visited the POI "department store". Here, the old product "cosmetics" is associated with the product tag "beautiful skin". Also, the POI "department store" is associated with the POI tag "skirt". Therefore, when the model generation unit 122 generates a knowledge graph, a second knowledge graph KG2 concerning actual buyer P2 is generated, which is different from the first knowledge graph KG1, as shown in Figure 14(b).

[0077] Furthermore, based on visit history data, if non-purchaser P51, identified by personal ID "P#501," visits the POI "Fitness Gym," then, as shown in Figure 14(a), non-purchaser P51's POI "Fitness Gym" is associated with the first knowledge graph KG1. The same applies to non-purchaser P61. Also, based on visit history data, if non-purchaser P52, identified by personal ID "P#502," visits the POI "Department Store," then, as shown in Figure 14(b), non-purchaser P52's POI "Department Store" is associated with the second knowledge graph KG2.

[0078] The model generation unit 122 generates various knowledge graphs, such as the first knowledge graph KG1 and the second knowledge graph KG2, and then embeds the nodes and edges in the knowledge graphs into a three-dimensional vector space. TransE is one known embedding method. In the knowledge graph, knowledge is represented in a form called a triple, such as "for s (subject), the value (object) of r (predicate) is o." The subject (s) and object (o) are called entities, and the predicate (r) is called a relation. Entities correspond to nodes, and relations correspond to edges. For example, in this embodiment, actual buyers P1 and P2, POI "Fitness Gym," and POI "Department Store" correspond to entities (or nodes). The relationships between entities, such as purchases and visits, correspond to relations.

[0079] A triple is symbolically represented as [s,r,o], and the three elements of a triple are represented by three vectors in the embedding space. Embedding involves representing knowledge as triple data and entities and relations as vectors. By embedding the nodes and edges in the knowledge graph into a three-dimensional vector space, a knowledge graph collection KB is generated, which stores the ternary relationships (triplets) in the knowledge graph, as shown in Figure 14(c). Such embedding makes it possible to predict unknown triples.

[0080] Returning to Figure 13, once the model generation unit 122 generates a knowledge graph, the potential estimation unit 123 then acquires pedestrian flow data of non-purchasers (step S16) and generates visit history data of non-purchasers (step S17). More specifically, the potential estimation unit 123 acquires pedestrian flow data of non-purchasers from the pedestrian flow storage unit 111 and POI data from the POI information storage unit 116. Once the pedestrian flow data and POI data of non-purchasers are acquired, the potential estimation unit 123 generates visit history data of non-purchasers for the POI based on time data representing the time non-purchasers stay at the POI, the latitude and longitude registered in the pedestrian flow data, and the latitude and longitude range registered in the POI data.

[0081] Once visit history data for non-purchasers is generated, the potential estimation unit 123 estimates the product IDs of new products to recommend (step S18). More specifically, the potential estimation unit 123 estimates the product IDs of new products to recommend to non-purchasers P51, P52, and P61 based on the visit history data of non-purchasers, the knowledge graph collection KB, and a known KGAT (Knowledge Graph Attention Network) model. KGAT can output the purchase probability of the new products for non-purchasers P51, P52, and P61. The KGAT model can be referenced from Non-Patent Literature 1 below. <Non-Patent Document 1> Fumiyo Ito, et al., "A Study on a Purchase Behavior Analysis Model Based on Knowledge Graph Attention Networks", Transactions of the Information Processing Society of Japan, Vol. 63, No. 1, pp. 205-217, 2022-01

[0082] Once the product ID is estimated, the potential estimation unit 123 extracts the product tag for that product ID (step S19). For example, the potential estimation unit 123 accesses the product information storage unit 114 and extracts the product tag corresponding to the estimated product ID for each product ID. The potential estimation unit 123 may extract a single product tag or multiple product tags for each product ID. Once the product tags are extracted, the potential estimation unit 123 complements the knowledge graph collection KB based on the extracted product tags and the KGAT model.

[0083] Here, the potential estimation unit 123 generates matching rules (step S20) when it detects the input or selection of confirmation items related to product advertisements based on the user 11's operation. For example, as shown in Figure 15, when any product is selected on the purchase potential confirmation screen displayed on the display device 14 and a search is instructed by the pointer Pt, the terminal device 10 transmits the confirmation items related to the advertisement of the selected product to the potential estimation unit 123. Note that the purchase potential confirmation screen is just one example of a predetermined screen. Furthermore, the confirmation items are not limited to product selection; for example, they may be the selection or input of the product's brand name or JAN code, or the selection or input of the event name or venue for various events. The confirmation items may also be the selection or input of the age and gender of the target audience for advertising, the selection or input of the facility name including the store that sells the product, or the selection or input of the name of a region or area.

[0084] When the potential estimation unit 123 receives confirmation items, it detects the selection of confirmation items and extracts the attributes of non-purchasers from the buyer attribute storage unit 113 based on the identifier of the individual non-purchaser registered in the human flow data of non-purchasers not associated with ID-POS data. After extracting the attributes of non-purchasers, the potential estimation unit 123 associates the attributes of non-purchasers with the extracted product tags, the product names corresponding to the confirmation items, and the purchase probability of new products output by KGAT. It then narrows down the edge weights in the knowledge graph by a threshold and adds edges to the original knowledge graph that have been narrowed down to all users and purchasable products. After adding edges to the original knowledge graph, the potential estimation unit 123 generates matching rules based on the Apriori algorithm and a Bayesian network and outputs them to the purchase potential confirmation screen. The edge weights represent the importance of the relationship between the user and the product (see, for example, Figure 3 in Non-Patent Document 1), and the thresholds are set in advance. Furthermore, all users consist of those who have visit history data but no ID-POS data, and those who have both visit history data and ID-POS data.

[0085] As a result, as shown in Figure 15, matching rules appear on the purchase potential confirmation screen as matching results. For example, for a combination of attributes of non-purchasers for the product "shoes" and product tags representing the product's characteristics, user 11 can understand the specific product name and the probability of non-purchasers purchasing that product. As a confirmation item, if, for example, an event name is entered, a combination of multiple items related to the event name will appear on the purchase potential confirmation screen as matching rules. The potential estimation unit 123 may output one optimal matching rule as a matching result, or it may output multiple matching rules that have high advertising effectiveness. User 11 can understand various matching rules and use the matching rules themselves to consider how to deliver advertisements.

[0086] Once matching rules are generated, the potential estimation unit 123 then estimates the purchase feasibility (step S21). For example, as shown in Figure 15, when any matching result is selected by the pointer Pt on the purchase potential confirmation screen displayed on the display device 14, the terminal device 10 transmits a display instruction for area designation to the potential estimation unit 123.

[0087] When the potential estimation unit 123 receives a display instruction, it outputs an area selection field to the purchase potential confirmation screen, as shown in Figure 15, in which the user can specify one of several areas. When an area is selected by the pointer Pt on the purchase potential confirmation screen and a search is instructed by the pointer Pt, the terminal device 10 sends an instruction to the potential estimation unit 123 to calculate the number of people who can purchase in the selected area.

[0088] When the potential estimation unit 123 receives a calculation instruction, it accesses the demographics memory unit 117 and calculates the number of potential buyers by multiplying the population of the age group corresponding to the specified area by the male-female ratio, based on the gender, age group, and specified area included in the selected matching result. Once the number of potential buyers is calculated, the potential estimation unit 123 multiplies the number of potential buyers by the purchase probability included in the matching result to calculate the number of people who can purchase in the specified area. Once the number of people who can purchase is calculated, the potential estimation unit 123 outputs the number of people who can purchase to the purchase potential confirmation screen.

[0089] As a result, as shown in Figure 15, the number of people who can purchase in the designated area will appear on the purchase potential confirmation screen as a purchase possibility. The potential estimation unit 123 may also calculate and output the purchase quantity, purchase price, etc., by multiplying the number of people who can purchase by a predetermined coefficient. This allows the user 11 to understand the number of people who can purchase, the purchase quantity, the purchase price, etc., in the designated area.

[0090] Furthermore, by preparing in advance the past sales history of products belonging to the same category as the selected product, the potential estimation unit 123 may calculate a future sales forecast for the selected product during a specified period based on the number of potential buyers, the sales history, and known methods. By outputting the future sales forecast to the purchase potential confirmation screen, the potential estimation unit 123 allows the user 11 to understand the future sales performance of the selected product. As for known methods, for example, Non-Patent Literature 2 below can be consulted. <Non-Patent Document 2> Kenji Tanaka, "A sales forecasting model for new-released and nonlinear sales trend products", Expert Systems with Applications, Vol. 37, Issue. 11, pp. 7387-7393, Nov 2010

[0091] In addition, the potential estimation unit 123 may output the processing results to a screen separate from the purchase potential confirmation screen. Furthermore, if the potential estimation unit 123 detects a press of the ad ID download button Bt provided on the purchase potential confirmation screen by a pointer Pt, it may generate a list of ad IDs of target advertisers corresponding to the number of people who can make a purchase and download them to the terminal device 10. This makes it possible to deliver targeted advertisements to mobile devices that possess ad IDs.

[0092] As described above, according to the second embodiment, the data processing server 100 combines product tags of old products extracted based on ID-POS data and POI tags extracted based on actual purchaser foot traffic data and visit history data, based on common attributes. The data processing server 100 also generates combination data of the actual purchaser's purchaser ID and the old product's product ID. The data processing server 100 can then generate a knowledge graph collection KB by machine learning multiple training data sets that define the relationship between the tag data, which is a combination of product tags and POI tags, and the combination data.

[0093] Once the Knowledge Graph KB is generated, the data processing server 100 generates visit history data of non-purchasers to POIs based on pedestrian flow data in the designated area of ​​non-purchasers and time data of non-purchasers staying at POIs included in the designated area. Then, based on the visit history data and the Knowledge Graph KB, the data processing server 100 estimates the product ID of a new product to recommend to non-purchasers, and estimates the purchase probability of the target audience based on the product tag of the new product identified by the product ID and the attributes of the non-purchaser. By utilizing the purchase probability of target audiences through the data processing server 100, users 11 can grasp the number of non-purchasers, the quantity purchased, and the purchase amount for new products, enabling them to develop effective new products for the market without waste. Furthermore, by utilizing the purchase probability of target audiences, users 11 can leverage this purchase probability in promotional activities for all products, including old and new products.

[0094] Although preferred embodiments of the present invention have been described in detail above, the present invention is not limited to specific embodiments, and various modifications and changes are possible within the scope of the gist of the invention as described in the claims. For example, the present invention may be used when a restaurant already has a service offering Japanese food in the market as an existing service, and it is looking to introduce a service offering Western food as a new service. [Explanation of Symbols]

[0095] ST Data Processing System 100 Data Processing Servers 110 Storage section 111 Human flow memory department 112 Visit history storage unit 113 Purchaser attribute storage unit 114 Product information storage section 115 Purchase history storage unit 116 POI information storage section 117 Demographics Memory Department 120 Processing Unit 121 Training Data Generation Unit 122 Model Generation Unit 123 Potential Estimation Unit

Claims

1. First, we obtain pedestrian flow data representing the flow of people in a designated area who have not yet purchased the advertised product. Based on the first pedestrian flow data and the trained model, the likelihood of an advertiser purchasing the advertised product is estimated. Let the computer perform the process, The trained model is generated by machine learning a plurality of training data sets that define the relationship between the purchase data of actual buyers for the advertised target and second pedestrian flow data representing the pedestrian flow of the actual buyers in the designated area. A program characterized by the following features.

2. The first pedestrian flow data, the second pedestrian flow data, and the purchase data all include unique advertising identifiers held by multiple mobile devices. The process involves identifying a portion of the advertising identifier based on the likelihood of purchase, and using the identified portion of the advertising identifier in the first human flow data. The program according to feature 1.

3. The aforementioned purchase data includes the attributes of the actual purchaser, The process described above estimates the likelihood of purchase by the advertisers whose attributes are the same as those of the actual purchaser. The program according to feature 1 or 2.

4. The purchase data includes either the number of purchases or the purchase frequency of the actual purchaser. The process estimates the purchasing tendencies of the advertisers based on either the number of purchases or the frequency of purchases. The program according to feature 1 or 2.

5. First, we obtain pedestrian flow data representing the flow of people in a designated area who have not yet purchased the advertised product. Based on the first pedestrian flow data and time data representing the time the non-purchaser stays at a specific facility included in the designated area, visit history data of the non-purchaser to the specific facility is generated. Based on the visit history data of the non-purchaser and the trained model, the advertiser identifier of the advertiser to recommend to the non-purchaser is estimated. Based on the ad target tag representing the characteristics of the ad target identified by the ad target identifier of the ad target, and the attributes of the non-purchaser, the likelihood of the ad targeter making a purchase of the ad target is estimated. Let the computer perform the process, The trained model is generated by machine learning a plurality of training data that define the relationship between tag data, which is a combination of A program characterized by the following features.

6. The first pedestrian flow data, the second pedestrian flow data, and the purchase data all include unique advertising identifiers held by multiple mobile devices. The process involves identifying a portion of the advertising identifier based on the likelihood of purchase, and using the identified portion of the advertising identifier in the first human flow data. L according to feature 5.

7. The process, when it detects input or selection of confirmation items related to the advertised item via a predetermined screen, searches for the attributes of the non-purchaser corresponding to the confirmation items, and outputs the search results, which include the combination of the non-purchaser's attributes and the probability that the advertised item will be purchased by the non-purchaser, as the purchase possibility of the non-purchaser on the predetermined screen or a separate screen. L according to feature 5.

8. The process involves obtaining demographic data in the designated area and calculating the sales quantity of the advertised item in the designated area based on the demographic data and the purchase probability. The program according to feature 7.

9. The process involves obtaining unit price data representing the unit price of the advertised item, and calculating the total purchase amount of the advertised item by the non-purchaser in the designated area based on the sales quantity and the unit price data. The program according to feature 8.

10. An acquisition unit that acquires first pedestrian flow data representing the flow of people in a designated area who have not yet purchased the advertisement, An estimation unit that estimates the likelihood of an advertiser purchasing the advertised target based on the first pedestrian flow data and a trained model, A data processing device having, The trained model is generated by machine learning a plurality of training data sets that define the relationship between the purchase data of actual buyers for the advertised target and second pedestrian flow data representing the pedestrian flow of the actual buyers in the designated area. A data processing device characterized by the following features.

11. An acquisition unit that acquires first pedestrian flow data representing the flow of people in a designated area who have not yet purchased the advertisement, A generation unit generates visit history data of the non-purchaser to the specific facility based on the first pedestrian flow data and time data representing the time the non-purchaser stays at the specific facility included in the designated area. A first estimation unit estimates the advertising target identifier of the advertising target to be recommended to the non-purchaser based on the visit history data of the non-purchaser and a trained model, A second estimation unit estimates the likelihood of an advertiser purchasing the advertised target based on an advertised target tag representing the characteristics of the advertised target identified by the advertised target identifier, and the attributes of the non-purchaser. A data processing device having, The trained model is generated by machine learning a plurality of training data that define the relationship between tag data, which is a combination of A data processing device characterized by the following features.

Images