WO2024070126A1 - Dispositif de génération de modèle de prévision de la demande - Google Patents

Dispositif de génération de modèle de prévision de la demande Download PDF

Info

Publication number
WO2024070126A1
WO2024070126A1 PCT/JP2023/025357 JP2023025357W WO2024070126A1 WO 2024070126 A1 WO2024070126 A1 WO 2024070126A1 JP 2023025357 W JP2023025357 W JP 2023025357W WO 2024070126 A1 WO2024070126 A1 WO 2024070126A1
Authority
WO
WIPO (PCT)
Prior art keywords
demand
forecasting model
data collection
parameter
demand forecasting
Prior art date
Application number
PCT/JP2023/025357
Other languages
English (en)
Japanese (ja)
Inventor
桂一 落合
弘之 佐藤
Original Assignee
株式会社Ｎｔｔドコモ
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 株式会社Ｎｔｔドコモ filed Critical 株式会社Ｎｔｔドコモ
Publication of WO2024070126A1 publication Critical patent/WO2024070126A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising

Definitions

  • This disclosure relates to a demand forecasting model generation device.
  • the demand forecasting model is generated by machine learning.
  • the demand forecasting model is a model for predicting the relationship between the price of a product and the sales volume of that product, or a model for predicting the relationship between the price of a paid service and the number of users of that service.
  • a demand forecasting model for a product is generated by machine learning using training data indicating a certain price and the actual sales volume of a product at that price.
  • Patent Document 1 discloses prior art related to the generation of a demand forecasting model by machine learning.
  • the learning data used to generate the demand forecasting model may indicate the price of the product and the actual sales volume of the product, as well as the degree of crowding at the store selling the product and the number of participants in a promotional campaign to promote the sale of the product.
  • the external environments of multiple stores are not necessarily the same.
  • the learning data may contain bias according to differences in the external environment.
  • One example of an external environment that causes bias in the learning data is the degree of crowding in the area in which the store from which the learning data was collected is located. This is because it is assumed that the higher the degree of crowding at a store, the higher the sales volume, and the degree of crowding at the store affects the degree of crowding in the area in which the store is located.
  • a promotional campaign is a campaign in which registered members of a point-granting service can participate.
  • points corresponding to a certain percentage of the purchase price of the product are returned to purchasers of the product.
  • a promotional campaign a higher return rate is set for registered members who have registered to use the point-granting service and who have also registered to participate in the promotional campaign.
  • An example of an external environment that affects the number of participants in a promotional campaign is the number of people registered for a points-granting service, which is a prerequisite for the promotional campaign.
  • a demand forecasting model generated based on learning data that includes biases due to differences in the external environment may not be able to make accurate demand forecasts.
  • a demand forecast model generating device is a demand forecast model generating device that generates a demand forecast model for predicting demand for an index in a target area, and includes an acquisition unit and a generation unit as described below.
  • the acquisition unit acquires at least one of a first parameter indicating the surrounding conditions of a data collection area from which data used to generate the demand forecast model was acquired and a second parameter related to the premise of promoting the demand.
  • the generation unit generates the demand forecast model using a loss function that assigns a weight based on at least one of the first parameter and the second parameter acquired by the acquisition unit to the error between the predicted value of the demand amount and the actual value of the demand amount.
  • a loss function is used that takes into account weights corresponding to differences in the external environment. Therefore, even if the data used to generate the demand forecasting model contains bias corresponding to differences in the external environment in the data collection area, the impact caused by the differences in the external environment is reduced, improving the prediction accuracy of the demand forecasting model.
  • FIG. 1 is a block diagram showing an example configuration of a demand forecasting model generating device 10 according to an embodiment of the present disclosure.
  • FIG. FIG. 2 is a diagram showing an example of the contents stored in a database DB1.
  • 13 is a flowchart showing the flow of a generation method executed by a processing device 140 of a demand forecasting model generation device 10 according to a program PR1.
  • a diagram showing the relationship between the number of visitors in a data collection area and the number of visitors in a first area including the data collection area, and the relationship between the number of participants in a sales promotion campaign and the number of registrants for a point-granting service.
  • FIG. 1 is a block diagram showing an example configuration of a demand forecasting model generating device 10 according to an embodiment of the present disclosure.
  • the demand forecasting model generating device 10 is a device that generates a demand forecasting model by machine learning.
  • the demand forecasting model predicts the sales quantity for the price of a product.
  • the demand forecasting model is generated for each store, such as a convenience store.
  • a store for which a demand forecasting model is generated is called a target area.
  • the target area in this embodiment is a single store. However, the target area in this disclosure may be a geographical range of a certain size in which multiple stores are located, i.e., a region.
  • target products products for which a demand forecasting model is generated are referred to as target products, etc.
  • the price of the target products, etc. in the target area is an example of an index in the target area.
  • the sales quantity of the target products relative to the price of the target products, etc. is an example of the demand for the index in the target area.
  • the target products, etc. are products.
  • the target products, etc. may also be services provided for a fee.
  • the fee for the service is an example of an index in the target area
  • the number of users of the service is an example of the demand for the index in the target area.
  • a service provided for a fee in the target area is an example of a first service in this disclosure.
  • the demand forecasting model generating device 10 generates a demand forecasting model by training a machine learning model using multiple learning data.
  • one learning data indicates at least the price of a target product, etc., and the actual sales quantity of the target product, etc. at that price.
  • the actual sales quantity of the target product, etc. is an actual value of the demand amount.
  • the learning data used to generate the demand forecasting model is collected in a store other than the store that is the target area.
  • the store where the learning data was collected is a store where the target product, etc. was actually sold at a certain price.
  • the store where the learning data was collected is called a data collection area.
  • the data collection area in this embodiment is different from the target area. However, the data collection area may be the same as the target area. Furthermore, the data collection area in this disclosure may be an area where multiple stores are located.
  • one piece of learning data indicates the degree of crowding in the data collection area (hereinafter, store crowding degree), the percentage of store visitors, and the number of CP accesses, in addition to the actual values of the price and sales volume of the target product, etc.
  • a point-granting service is provided to purchasers of products in the data collection area. Purchasers of products in the data collection area are registrants of the point-granting service. In the point-granting service, points corresponding to a certain percentage of the purchase price of the product are returned to purchasers of products in the data collection area. Among purchasers of products in the data collection area, those who have registered to participate in a promotional campaign can receive a point redemption rate higher than the point redemption rate in the point-granting service.
  • the point-granting service is an example of a second service in this disclosure.
  • the percentage of store visitors is the percentage of participants in the promotional campaign who visited the data collection area among those registrants of the point-granting service. Therefore, the number of participants in the promotional campaign who visited the data collection area can be obtained by multiplying the number of registrants of the point-granting service by the percentage of store visitors.
  • the number of CP accesses is the number of accesses to a web page on which information about the promotional campaign is posted.
  • one learning data set is composed of the actual sales volume value, which is the correct answer data, and four feature quantities (price, store congestion level, visitor ratio, and CP access count). However, the number of feature quantities constituting one learning data set may be two or more and three or less, or may be five or more.
  • the learning data may include bias according to differences in the external environment of the data collection area.
  • the external environment that causes bias in the learning data is, for example, the surrounding conditions of the data collection area and the conditions related to the premise of promoting demand in the data collection area.
  • the surrounding conditions of the data collection area are, for example, the congestion level (hereinafter, the surrounding congestion level) indicating the degree of crowding in a first area (for example, an area of 500 m x 500 m) including the data collection area. It is assumed that the sales volume of products in the data collection area is related to the number of customers who visit the data collection area. More specifically, it is assumed that the more customers visit the data collection area, the higher the sales volume.
  • An example of a situation related to the premise of promoting demand in a data collection area is the number of subscribers to a point-granting service, which is a premise of a promotional campaign. It is assumed that the more participants there are in a promotional campaign held in the data collection area, the higher the sales volume will be. The number of participants in a promotional campaign varies depending on the number of subscribers to the point-granting service. Conventional demand forecasting does not take into account differences in the circumstances related to the premise of promoting demand in a data collection area, making it difficult to make an accurate demand forecast.
  • the demand forecasting model generating device 10 of this embodiment is configured to generate a demand forecasting model by reducing the influence caused by the difference in the external environment, even if the learning data contains bias corresponding to the difference in the external environment in the data collection area.
  • the details are as follows.
  • the demand forecasting model generating device 10 includes a communication device 110, an input device 120, a storage device 130, a processing device 140, and a bus 150.
  • the communication device 110, the input device 120, the storage device 130, and the processing device 140 are interconnected by a bus 150 for mediating data exchange.
  • the bus 150 may be configured using a single bus, or may be configured using different buses between each element.
  • the communication device 110 is hardware (transmitting/receiving device) for communicating with other devices via an electric communication line such as the Internet.
  • the communication device 110 is also called, for example, a network device, a network controller, a network card, or a communication module.
  • Other devices that communicate with the communication device 110 via an electric communication line include, for example, a congestion sensor that detects the degree of store congestion in a data collection area, a sales management server that manages the sales results of products in each data collection area, and a service management server that manages the number of registered users of a point-granting service as well as the results of promotional campaigns held in each data collection area.
  • the results of promotional campaigns include the number of CP accesses and the percentage of visitors per store and per day.
  • the congestion sensor is, for example, a sensor that is installed in a data collection area and detects beacons in Wi-Fi or BLE (Bluetooth Low Energy). Note that Bluetooth is a registered trademark.
  • the communication device 110 communicates with a sales management server via telecommunications lines to collect an area identifier that uniquely indicates a data collection area, and the price of the target product, etc. and the actual sales volume of the target product, etc., for each day in the data collection area.
  • the communication device 110 also communicates with a service management server via telecommunications lines to collect the area identifier, the date of a promotional campaign held in the data collection area indicated by the area identifier, the percentage of store visitors, the number of CP accesses, and the number of registered users for the point-granting service.
  • the data collected by the communication device 110 from other devices via telecommunications lines is stored in database DB1 for each data collection area and for each day.
  • the input device 120 is a device for accepting input operations by a user of the demand forecasting model generating device 10.
  • the input device 120 is, for example, a keyboard with multiple operators, such as a numeric keypad, or a pointing device, such as a mouse.
  • an input operation such as pressing an operator
  • the input device 120 outputs operation content data indicating the content of the input operation to the processing device 140.
  • the operation content data is output from the input device 120 to the processing device 140, whereby the user's operation content is transmitted to the processing device 140.
  • the input operation on the input device 120 is, for example, an operation for specifying a data collection area, or an operation for specifying a period to which the dates of the learning data used to generate the demand forecasting model belong.
  • the storage device 130 is a recording medium that can be read by the processing device 140.
  • the storage device 130 may be composed of at least one of, for example, a ROM (Read Only Memory), an EPROM (Erasable Programmable ROM), an EEPROM (Electrically Erasable Programmable ROM), a RAM (Random Access Memory), etc.
  • a database DB1 and a program PR1 are pre-stored in the storage device 130.
  • Figure 2 is a diagram showing an example of database DB1.
  • Database DB1 stores an index, demand, store congestion level, visitor ratio, CP access count, surrounding congestion level, and number of registered users in association with an area identifier and a date.
  • the area identifier in this embodiment is a character string representing the name of the store that is the data collection area. However, the area identifier may also be a character string representing a serial number assigned to the store.
  • the index indicates the price of the target product, etc. sold in the data collection area indicated by the corresponding area identifier on the corresponding date.
  • the demand indicates the sales quantity of the target product, etc. actually sold in the data collection area indicated by the corresponding area identifier on the corresponding date. In other words, the demand indicates the actual sales quantity of the target product, etc.
  • the store congestion level is the congestion level of the data collection area indicated by the corresponding area identifier on the corresponding date.
  • the store congestion level is, for example, the number of beacons detected by a congestion level sensor in the data collection area indicated by the corresponding area identifier on the corresponding date. Therefore, the store congestion level indicates the total number of visitors to the data collection area on a day.
  • the visitor ratio indicates the ratio of the number of participants in a promotional campaign who visited the store to the number of registrants of a point-granting service in the data collection area indicated by the corresponding area identifier on the corresponding date.
  • the number of CP accesses indicates the number of accesses to a web page that contains information about a promotional campaign being held in the data collection area indicated by the corresponding area identifier on the corresponding date.
  • the degree of surrounding congestion is the degree of congestion of the first area including the data collection area indicated by the corresponding area identifier on the corresponding date.
  • the degree of surrounding congestion can be calculated based on the number of people staying in the first area, i.e., the number of visitors in the area.
  • the number of visitors in an area of 500 m x 500 m may be estimated by a mobile communication service provider.
  • the provider may estimate the number of visitors in the area based on the number of mobile communication terminals accommodated in the base station covering the area, the market share of the mobile communication service, and the penetration rate of mobile communication.
  • the calculation of the degree of surrounding congestion uses the number of visitors estimated by the mobile communication service provider.
  • the degree of surrounding congestion is an example of a first parameter indicating the situation of the first area including the data collection area.
  • the number of registered users indicates the number of registered users of the point-granting service in the data collection area indicated by the corresponding area identifier on the corresponding date.
  • the number of registered users is an example of a second parameter related to the premise of promoting demand in the data collection area.
  • one learning data set is formed by the indexes, demand volume, store congestion level, visitor ratio, and CP access count, which are associated with each other.
  • the surrounding congestion level and the number of registered users, which are associated with one learning data set are parameters that represent the external environment at the time the learning data set was collected.
  • the learning data set is daily data.
  • the learning data set may also be weekly data.
  • the date may be data that represents the week in which the learning data set and the parameters representing the external environment were collected.
  • the learning data set and the parameters representing the external environment set are store-based data.
  • the learning data set and the parameters representing the external environment set may also be area-based data in which multiple stores are located.
  • the processing device 140 includes one or more CPUs (Central Processing Units). When the demand forecasting model generating device 10 is powered on, the processing device 140 reads the program PR1 from the storage device 130. By executing the program PR1, the processing device 140 functions as the acquisition unit 140a and the generation unit 140b shown in FIG. 1.
  • the acquisition unit 140a and the generation unit 140b shown in FIG. 1 are software modules realized by operating a computer such as a CPU in accordance with software such as a program.
  • the functions performed by each of the acquisition unit 140a and the generation unit 140b are as follows:
  • the acquisition unit 140a reads out from the database DB1 N (N is an integer of 2 or more) pieces of learning data, and the surrounding congestion level and the number of registered users corresponding to each of the N pieces of learning data in response to an input operation on the input device 120. In this way, the acquisition unit 140a acquires from the database DB1 N pieces of learning data and parameters representing the external environment when each of the N pieces of learning data was collected.
  • the acquisition unit 140a reads out from the database DB1 a set of an index, a demand amount, a store congestion level, a percentage of the number of visitors, a CP access number, a surrounding congestion level, and a number of registered users that are associated with a date that belongs to the period.
  • the acquisition unit 140a reads out from the database DB1 a set of an index, a demand amount, a store congestion level, a percentage of the number of visitors, a CP access number, a surrounding congestion level, and a number of registered users that are associated with the input area identifier.
  • the generating unit 140b generates a demand forecasting model for predicting the sales volume for the sales price of a target product, etc., by having a machine learning model learn using the N pieces of learning data acquired by the acquiring unit 140a.
  • a statistical model based on Poisson regression represented by formula (1) is used as the machine learning model.
  • C1 is the store congestion degree in the target area.
  • C2 is the ratio of the number of visitors in the target area.
  • C3 is the number of CP accesses.
  • P is the sales price of the target product, etc.
  • DP is the predicted value of the sales volume.
  • e is the base of the natural logarithm.
  • ⁇ 0, ⁇ 1, ⁇ 2, ⁇ 3, and ⁇ 4 are parameters that characterize the demand forecasting model.
  • Generating a demand forecasting model means determining the parameters ⁇ 0, ⁇ 1, ⁇ 2, ⁇ 3, and ⁇ 4 by having a statistical model learn. Once the parameters ⁇ 0, ⁇ 1, ⁇ 2, ⁇ 3, and ⁇ 4 have been determined, the predicted values or past actual values of the price of the target goods, etc. sold in the target area, the store congestion level, the percentage of visitors, and the number of CP accesses are substituted into the right-hand side of equation (1), whereby a predicted value of the sales quantity of the target goods, etc. in the target area is obtained.
  • the generation unit 140b first calculates a predicted value of the sales volume for each of the N pieces of learning data acquired by the acquisition unit 140a according to formula (1).
  • the store congestion level is C1(i)
  • the percentage of the number of visitors is C2(i)
  • the number of CP accesses is C3(i)
  • the index is P(i).
  • the generation unit 140b substitutes C1(i) for C1, C2(i) for C2, C3(i) for C3, and P(i) for P, so that the left side of formula (1) represents the predicted value DP(i) of the sales volume based on the i-th learning data.
  • the generating unit 140b generates a loss function Loss1 of the formula (2) based on the predicted value DP(i) of the sales volume and the correct data in the i-th learning data (i.e., the actual value DR(i) of the demand volume).
  • the generating unit 140b determines the values of the parameters ⁇ 0, ⁇ 1, ⁇ 2, ⁇ 3, and ⁇ 4 so that the value of the loss function Loss1 is minimized.
  • C4(i) in the formula (2) is the degree of surrounding congestion corresponding to the i-th learning data
  • C5(i) is the number of registered users corresponding to the i-th learning data.
  • One method of estimating parameters is a method using maximum likelihood estimation. For the maximum likelihood estimation, existing technology may be appropriately adopted.
  • the generation unit 140b calculates the loss function of formula (2) using, among the N pieces of learning data, learning data different from both the learning data whose surrounding congestion degree C4(i) is 0 and the learning data whose number of registered users C5(i) is 0.
  • the acquisition unit 140a may acquire N pieces of learning data different from both the learning data whose surrounding congestion degree is 0 and the learning data whose number of registered users is 0.
  • the generation unit 140b calculates the loss function of formula (2) using the N pieces of learning data and the surrounding congestion degree and the number of registered users corresponding to each piece of learning data of the N pieces of learning data.
  • C4(i) in the denominator of the right side of formula (2) may be replaced with C4(i)+ ⁇
  • C5(i) in the denominator of the right side of formula (2) may be replaced with C5(i)+ ⁇ so that the loss function Loss1 can be calculated even when the surrounding congestion degree C4(i) or the number of registered users C5(i) is 0.
  • ⁇ is a very small number such as 0.001.
  • max(C4(i)) on the right-hand side of formula (2) is the maximum value of the surrounding congestion degree corresponding to the learning data used to generate the demand forecasting model.
  • max(C4(i))/C4(i) means the inverse of the surrounding congestion degree normalized by the maximum value. Normalizing the surrounding congestion degree by the maximum value means adjusting the value of the surrounding congestion degree corresponding to each learning data so that the maximum value of the surrounding congestion degree after normalization becomes 1.
  • max(C5(i)) on the right-hand side of formula (2) is the maximum value of the number of registered users corresponding to the learning data used to generate the demand forecasting model.
  • max(C5(i))/C5(i) means the inverse of the number of registered users normalized by the maximum value. Normalizing the number of registered users by the maximum value means adjusting the value of the number of registered users corresponding to each learning data so that the maximum value of the number of registered users after normalization becomes 1.
  • the loss function Loss2 shown in equation (3) has been used.
  • a weight according to the inverse of the surrounding congestion degree and a weight according to the inverse of the number of registered users of the point-granting service are assigned to the square of the difference between the predicted value DP(i) and the actual value DR(i).
  • the machine learning of this embodiment differs from conventional machine learning in that such a weighted loss function Loss1 is used.
  • the surrounding congestion degree of the data collection area and the number of registered users of the point-granting service are examples of external environments that affect the sales volume of products in the data collection area. Specifically, as shown in FIG.
  • the number of visitors in the first area including the data collection area is the upper limit of the number of visitors in the data collection area. Therefore, the surrounding congestion degree is the upper limit of the store congestion degree.
  • the number of registered users of the point-granting service is the upper limit of the number of participants in the sales promotion campaign. It is considered that the upper limit of the demand amount of the target product, etc. is a value according to the upper limit of the store congestion degree and the upper limit of the number of participants in the sales promotion campaign.
  • a loss function Loss1 weighted by the inverse of the surrounding congestion degree and the inverse of the number of registered users is used.
  • the demand forecasting model generating device 10 of this embodiment the loss function Loss1 shown in formula (2) is used. Therefore, even if the learning data used to generate the demand forecasting model includes bias according to the difference in the external environment in the data collection area (difference in the surrounding congestion degree and difference in the number of registered users), the influence caused by the difference in the external environment is reduced. Therefore, a demand forecasting model with improved prediction accuracy is generated according to the reduction in the influence.
  • this generation method includes the processes of an acquisition process SA100 and a generation process SA110.
  • the processing device 140 functions as an acquisition unit 140a.
  • the processing device 140 acquires N pieces of learning data and the surrounding congestion degree and number of registered users corresponding to each piece of learning data from the database DB1.
  • the processing device 140 functions as a generation unit 140b.
  • the processing device 140 first calculates a predicted value DP(i) of the sales quantity according to equation (1) based on each of the N pieces of learning data acquired in the acquisition process SA100.
  • a demand forecasting model is generated using the loss function Loss1 weighted using the "inverse of the surrounding congestion level" and the "inverse of the number of registered users of the point-granting service".
  • a demand forecasting model is generated using the above weighted loss function. Therefore, even if the learning data contains bias due to differences in the external environment in the data collection area, the influence caused by the difference is reduced. Therefore, a demand forecasting model with improved prediction accuracy is generated by the amount of the reduction in the influence.
  • the demand forecasting model can be generated while reducing the effects of these differences, and the prediction accuracy of the demand forecasting model improves in accordance with the reduction in these effects.
  • B Deformation
  • a value indicating an incentive given to participants in a promotional campaign held in the target area may be the index in the present disclosure, and the sales volume of the product or the number of participants in the promotional campaign may be the demand in the present disclosure.
  • the value indicating the incentive given to participants in the promotional campaign is, for example, a point redemption rate for participants in the promotional campaign, or a discount rate of the price or consideration for participants in the promotional campaign.
  • the demand forecasting model is generated using a loss function weighted using the first parameter and the second parameter.
  • the demand forecasting model may be generated using a loss function weighted using either the first parameter or the second parameter.
  • a loss function weighted using only the surrounding congestion degree i.e., the first parameter
  • a loss function weighted using only the number of registered customers i.e., the second parameter
  • the demand forecasting model may be generated using a loss function weighted using at least one of the first parameter and the second parameter.
  • the first parameter represents the conditions surrounding the data collection area. Therefore, by using a loss function weighted using the first parameter, it is possible to generate a demand forecasting model that can predict demand regardless of the conditions surrounding the data collection area.
  • the second parameter represents the conditions regarding the assumptions for promoting demand in the data collection area. Therefore, by using a loss function weighted using the second parameter, it is possible to generate a demand forecasting model that can predict demand regardless of the conditions regarding the assumptions for promoting demand in the data collection area.
  • the program PR1 is stored in the storage device 130 of the demand forecast model generation device 10, but the program PR1 may be manufactured or sold separately.
  • the program PR1 may be provided to a purchaser, for example, by distributing a computer-readable recording medium such as a flash ROM on which the program PR1 is written, or by distributing the program PR1 by downloading it via a telecommunications line.
  • the acquisition unit 140a and the generation unit 140b are both software modules.
  • any one or all of the acquisition unit 140a and the generation unit 140b may be a hardware module.
  • the hardware module may be, for example, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), a field programmable gate array (FPGA), etc. Even if any one or all of the acquisition unit 140a and the generation unit 140b are hardware modules, the same effects as those of the above embodiment are achieved.
  • the database DB1 is stored in the storage device 130 of the demand forecasting model generating device 10.
  • the database DB1 may be stored in a storage device accessible from the processing device 140 via a communication network.
  • storage device 130 may also be a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory device (e.g., a card, a stick, a key drive), a CD-ROM (Compact Disc-ROM), a register, a removable disk, a hard disk, a floppy (registered trademark) disk, a magnetic strip, a database, a server, or other suitable storage medium.
  • a magneto-optical disk e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk
  • a smart card e.g., a flash memory device (e.g., a card, a stick, a key drive), a CD-ROM (Compact Disc-ROM), a register, a removable disk, a hard disk, a floppy (registered trademark) disk,
  • the information, signals, etc. described may be represented using any of a variety of different technologies.
  • data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.
  • the input and output information, etc. may be stored in a specific location (e.g., memory) or may be managed using a management table.
  • the input and output information, etc. may be overwritten, updated, or added to.
  • the output information, etc. may be deleted.
  • the input information, etc. may be transmitted to another device.
  • the determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a comparison of numerical values (e.g., a comparison with a predetermined value).
  • each function illustrated in FIG. 1 is realized by any combination of at least one of hardware and software. Furthermore, there are no particular limitations on the method of realizing each functional block. That is, each functional block may be realized using one device that is physically or logically coupled, or may be realized using two or more devices that are physically or logically separated and connected directly or indirectly (e.g., using wires, wirelessly, etc.). A functional block may be realized by combining software with the one device or the multiple devices.
  • the programs exemplified in the above embodiments should be broadly construed to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., regardless of whether they are called software, firmware, middleware, microcode, hardware description language, or by other names.
  • software, instructions, information, etc. may be transmitted and received via a transmission medium.
  • a transmission medium For example, if the software is transmitted from a website, server, or other remote source using at least one of wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and/or wireless technologies (such as infrared, microwave, etc.), then at least one of these wired and wireless technologies is included within the definition of a transmission medium.
  • wired technologies such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)
  • wireless technologies such as infrared, microwave, etc.
  • the information, parameters, etc. described in this disclosure may be expressed using absolute values, may be expressed using relative values from a predetermined value, or may be expressed using other corresponding information.
  • the mobile device may be a mobile station (MS).
  • MS mobile station
  • a mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other suitable term.
  • terms such as “mobile station,” “user terminal,” “user equipment (UE),” and “terminal” may be used interchangeably.
  • connection refers to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” to each other.
  • the coupling or connection between elements may be physical, logical, or a combination thereof.
  • “connected” may be read as "access”.
  • two elements may be considered to be “connected” or “coupled” to each other using at least one of one or more wires, cables, and printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and light (both visible and invisible) range, as some non-limiting and non-exhaustive examples.
  • the phrase “based on” does not mean “based only on,” unless otherwise specified. In other words, the phrase “based on” means both “based only on” and “based at least on.”
  • determining and “determining” as used in this disclosure may encompass a wide variety of actions. “Determining” and “determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, search, inquiry (e.g., searching in a table, database, or other data structure), and considering ascertaining as “judging” or “determining”. Also, “determining” and “determining” may include considering receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in memory) as “judging” or “determining”.
  • judgment and “decision” can include considering resolving, selecting, choosing, establishing, comparing, etc., to have been “judged” or “decided.” In other words, “judgment” and “decision” can include considering some action to have been “judged” or “decided.” Additionally, “judgment (decision)” can be interpreted as “assuming,” “expecting,” “considering,” etc.
  • a demand forecast model generating device is a demand forecast model generating device that generates a demand forecast model for predicting a demand amount for an index in a target area, and includes an acquisition unit and a generation unit as follows.
  • the acquisition unit acquires at least one of a first parameter indicating the surrounding conditions of a data collection area where data used to generate the demand forecast model was acquired and a second parameter related to the premise of the promotion of the demand amount.
  • the generation unit generates the demand forecast model using a loss function that assigns a weight based on at least one of the first parameter and the second parameter acquired by the acquisition unit to the error between the predicted value of the demand amount and the actual value of the demand amount.
  • the demand forecast model generating device of the first aspect even if the learning data contains a bias corresponding to a difference in the external environment represented by at least one of the first parameter and the second parameter, it is possible to generate a demand forecast model while reducing the influence caused by the difference, and the prediction accuracy of the demand forecast model improves in accordance with the reduction in the influence.
  • the index may be a value indicating the price of a product sold in the target area or an incentive given to participants in a promotional campaign for the product held in the target area
  • the demand may be the sales volume of the product in the target area.
  • a demand forecasting model that predicts the sales volume of a product in return for the price of the product sold in the target area can be generated while reducing the influence due to the difference in the external environment represented by at least one of the first parameter and the second parameter, and the prediction accuracy of the demand forecasting model improves in response to the reduction in the influence.
  • a demand forecasting model that predicts the sales volume of a product in return for the incentive given to participants in a promotional campaign held in the target area can be generated while reducing the influence due to the difference in the external environment represented by at least one of the first parameter and the second parameter, and the prediction accuracy of the demand forecasting model improves in response to the reduction in the influence.
  • the demand quantity may be a value indicating the price of a first service provided in the target area for a fee, or an incentive given to participants in a promotional campaign for the first service held in the target area, and the demand quantity may be the number of users of the first service in the target area.
  • a demand forecasting model that predicts the number of users of the first service in response to the price of the first service can be generated while reducing the influence caused by the difference in the external environment represented by at least one of the first parameter and the second parameter, and the prediction accuracy of the demand forecasting model improves in response to the reduction in the influence.
  • a demand forecasting model that predicts the number of users of the first service in response to an incentive given to participants in a promotional campaign held in the target area can be generated while reducing the influence caused by the difference in the external environment represented by at least one of the first parameter and the second parameter, and the prediction accuracy of the demand forecasting model improves in response to the reduction in the influence.
  • the first parameter may be a congestion degree indicating the degree of crowding in a first area including the data collection area.
  • the learning data contains a bias according to the degree of crowding in the first area including the data collection area, the influence caused by the difference in the degree of crowding can be reduced, and the prediction accuracy of the demand forecasting model improves in response to the reduction in the influence.
  • the second parameter may be the number of registered users who have registered to use a second service provided in association with the sale of a product in the data collection area or the provision of a first service provided in the data collection area for a fee.
  • the learning data contains a bias according to the number of registered users of the second service, it is possible to reduce the influence caused by the difference in the number of registered users, and the prediction accuracy of the demand forecasting model improves in accordance with the reduction in the influence.
  • 10...demand forecast model generating device 110...communication device, 120...input device, 130...storage device, 140...processing device, 140a...acquisition unit, 140b...generation unit, 150...bus, PR1...program, DB1...database.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

Ce dispositif de génération de modèle de prévision de la demande génère un modèle de prévision de la demande servant à prévoir une quantité de demande relative à un indicateur dans une zone cible. Le dispositif de génération de modèle de prévision de la demande comporte une unité d'acquisition et une unité de génération. L'unité d'acquisition acquiert un premier paramètre qui exprime les conditions autour d'une zone de collecte de données où ont été acquises les données à utiliser dans la génération du modèle de prévision de la demande, et/ou un second paramètre se rapportant à une condition préalable pour la promotion de ladite quantité de demande. L'unité de génération génère le modèle de prévision de la demande en utilisant une fonction de perte qui attribue un poids, sur la base du premier paramètre et/ou du second paramètre acquis par l'unité d'acquisition, à la différence entre la valeur prévue de la quantité de demande et la valeur réelle de la quantité de demande.
PCT/JP2023/025357 2022-09-27 2023-07-07 Dispositif de génération de modèle de prévision de la demande WO2024070126A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2022153854 2022-09-27
JP2022-153854 2022-09-27

Publications (1)

Publication Number Publication Date
WO2024070126A1 true WO2024070126A1 (fr) 2024-04-04

Family

ID=90476963

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2023/025357 WO2024070126A1 (fr) 2022-09-27 2023-07-07 Dispositif de génération de modèle de prévision de la demande

Country Status (1)

Country Link
WO (1) WO2024070126A1 (fr)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019159585A1 (fr) * 2018-02-14 2019-08-22 株式会社Ｎｔｔドコモ Système d'apprentissage, système d'estimation et modèle appris
WO2019187341A1 (fr) * 2018-03-30 2019-10-03 Ｎｅｃソリューションイノベータ株式会社 Dispositif de calcul d'indice, système de prédiction, procédé d'évaluation de prédiction de progression, et programme

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019159585A1 (fr) * 2018-02-14 2019-08-22 株式会社Ｎｔｔドコモ Système d'apprentissage, système d'estimation et modèle appris
WO2019187341A1 (fr) * 2018-03-30 2019-10-03 Ｎｅｃソリューションイノベータ株式会社 Dispositif de calcul d'indice, système de prédiction, procédé d'évaluation de prédiction de progression, et programme

Similar Documents

Publication Publication Date Title
US20190012683A1 (en) Method for predicting purchase probability based on behavior sequence of user and apparatus for the same
CN101517967B (zh) 网站的流量预测
US8732015B1 (en) Social media pricing engine
US20140351046A1 (en) System and Method for Predicting an Outcome By a User in a Single Score
US20150254709A1 (en) System and Method for Attributing Engagement Score Over a Channel
WO2009111166A2 (fr) Procédé et appareil pour un marketing fondé sur un réseau social, avec recommandations de promoteurs de produits
KR20070002741A (ko) 광고 수수료를 결정하는 방법 및 광고 수수료 결정 시스템
WO2009111167A2 (fr) Procédé et appareil pour un marketing fondé sur un réseau social, avec recommandations de consommateurs
US20110054999A1 (en) System and method for predicting user navigation within sponsored search advertisements
US20160210656A1 (en) System for marketing touchpoint attribution bias correction
CN104081392A (zh) 社会媒体配置文件的影响评分
JP2012048360A (ja) Ｉｄ価値評価装置、ｉｄ価値評価システム、及びｉｄ価値評価方法
CN111210335A (zh) 用户风险识别方法、装置及电子设备
CN109190028A (zh) 基于大数据的活动推荐方法、装置、电子设备及存储介质
CN110909222A (zh) 基于聚类的用户画像建立方法、装置、介质及电子设备
CN112348592A (zh) 广告推荐方法、装置、电子设备及介质
CN111210336A (zh) 用户风险模型生成方法、装置及电子设备
CN114443735B (zh) 酒店数据映射规则生成方法、装置、设备及存储介质
CN115935185A (zh) 一种推荐模型的训练方法及装置
KR20070070062A (ko) 서비스 평가 방법, 시스템 및 컴퓨터 프로그램 제품
CN109858756A (zh) 一种服务质量缺陷诊断方法和装置
CN112287208B (zh) 用户画像生成方法、装置、电子设备及存储介质
WO2024070126A1 (fr) Dispositif de génération de modèle de prévision de la demande
CN111523032A (zh) 确定用户偏好的方法、装置、介质和电子设备
CN112948701B (zh) 信息推荐装置、方法、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23871370

Country of ref document: EP

Kind code of ref document: A1