Prediction device, prediction method, and program

Abstract

A prediction device, a prediction method, and a program for predicting effective measures are provided. The prediction device (1) includes: a storage unit (14) that stores measure implementation information (144) indicating the measure effect in a first object that has completed implementing a measure; and a control unit (13) that predicts the measure effect in a second object that has not implemented the measure based on the measure implementation information. The control unit constructs a first graph (450A to 450D) composed of a plurality of nodes and a plurality of links that connect the nodes based on the similarity between the nodes. The plurality of nodes include at least one first node associated with the first object and at least one second node associated with the second object. Based on the measure implementation information, the degree of the measure effect in the first node is determined, and the degree of the measure effect is propagated from the first node as a base point to the second node in the first graph.

Description

Technical Field

[0001] The present disclosure relates to a prediction device, a prediction method, and a program for predicting the effect of a measure. Background Art

[0002] Patent Document 1 discloses a purchase prediction analysis system. The purchase prediction analysis system performs cluster analysis on regions based on multiple factors, and calculates the commodity purchase rate for each cluster. The purchase prediction analysis system determines whether to adopt or not adopt the calculated commodity purchase rate based on a specified criterion. The purchase prediction analysis system uses the commodity purchase rate of the region of the cluster for which the commodity purchase rate has been adopted as the target variable, and uses the factor score of the region as the explanatory variable to generate a calculation formula for multiple regression analysis, that is, a prediction model. The purchase prediction analysis system uses the generated calculation formula to calculate and predict the commodity purchase rate for all regions based on the explanatory variables of the regions. Thus, even in a region where the commodity purchase rate has not been appropriately obtained, an appropriate commodity purchase rate can be predicted.

[0003] Prior Art Documents

[0004] Patent Documents

[0005] Patent Document 1: JP-A-2016-6621 Summary of the Invention

[0006] -Problems to be Solved by the Invention-

[0007] The present disclosure provides a prediction device, a prediction method, and a program for predicting an effective measure.

[0008] -Means for Solving the Problems-

[0009] The prediction device of the present disclosure includes: a storage unit that stores measure implementation information indicating the measure effect in a first object in which a measure has been implemented; and a control unit that predicts the measure effect in a second object in which the measure has not been implemented based on the measure implementation information. The control unit constructs a first graph including a plurality of nodes and a plurality of links that connect the nodes based on the similarity between the nodes. The plurality of nodes include at least one first node associated with the first object and at least one second node associated with the second object. Based on the measure implementation information, the control unit determines the degree of the measure effect in the first node, and propagates the degree of the measure effect from the first node as a base point to the second node in the first graph.

[0010] These general and specific aspects can also be implemented by a system, a method, and a computer program, and combinations thereof.

[0011] -Effects of the Invention-

[0012] With the prediction device, prediction method, and program of the present disclosure, the degree of the effect of a measure is propagated using a graph including nodes and links, so that an effective measure can be predicted for an object to which the measure has not been implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a block diagram showing the structures of the prediction device and the terminal device.

[0014] Figure 2 It is a diagram showing the functional structure of the control unit during learning and the data stored in the storage unit.

[0015] Figure 3 It is a diagram showing an example of ID-POS data.

[0016] Figure 4 It is a diagram showing an example of customer data.

[0017] Figure 5 It is a diagram showing an example of area data.

[0018] Figure 6 It is a diagram showing an example of purchase data before the implementation of a measure.

[0019] Figure 7 It is a diagram showing an example of purchase data during the implementation of a measure.

[0020] Figure 8 It is a diagram schematically showing an example of prediction data.

[0021] Figure 9 It is a flowchart showing the generation operation of the prediction data of the prediction device.

[0022] Figure 10 It is a flowchart showing the setting process of the effect ranking for the completed implementation area.

[0023] Figure 11 It is a diagram for explaining the calculation of the measure effect value and the setting of the effect ranking for the completed implementation area.

[0024] Figure 12 It is a diagram for explaining the setting of the ranking probability for the completed implementation area.

[0025] Figure 13 It is a diagram for explaining the construction of the similarity graph for each area characteristic.

[0026] Figure 14 It is a flowchart showing the prediction process of the effect ranking for the unimplemented area.

[0027] Figure 15This is a diagram for explaining the construction of a prediction chart for ranking each effect.

[0028] Figure 16 This is a diagram for explaining the synthesis of a similarity chart based on importance.

[0029] Figure 17 This is a diagram schematically showing the propagation of effect rankings.

[0030] Figure 18 This is a diagram for explaining the prediction of effect rankings for unimplemented regions.

[0031] Figure 19 This is a diagram showing the functional structure of the control unit at the time of recommendation and the data stored in the storage unit.

[0032] Figure 20 This is a flowchart showing the recommendation operation of the prediction device.

[0033] Figure 21 This is a diagram for explaining the supplementation of purchase data in a new region.

[0034] Figure 22 This is a diagram for explaining the determination of effect rankings for business districts.

[0035] Figure 23 This is a diagram for explaining recommendations and non-recommendations corresponding to effect rankings.

[0036] Figure 24 This is a flowchart showing the update operation of the prediction data of the prediction device. Detailed implementation mode

[0037] (Research as the basis of the present disclosure)

[0038] It is desired to select and recommend effective measures from among multiple measure candidates for the objects of implementation measures. The objects of implementation measures are, for example, regions and stores. However, in the purchase prediction analysis system of Patent Document 1, when the number of regions where the effects of measures are clarified is small, it is difficult to predict the effects of measures for regions where the effects of measures are not clarified. Therefore, effective measures cannot be recommended for regions where measures have not been implemented.

[0039] The present disclosure provides a prediction device that can predict and recommend effective measures for other objects for which measures have not been implemented even when the number of objects for which measures have actually been implemented is small.

[0040] (Embodiment)

[0041] Hereinafter, embodiments will be described with reference to the drawings. In the present embodiment, the object of the implementation measures is a store, and examples of measures recommended for the store are described. In the present embodiment, the business area that can attract customers in the store, that is, the business district, includes one or more regions. The prediction device of the present embodiment uses the label propagation method, which is one of the types of machine learning, to propagate the effect ranking indicating the degree of the effect of each measure from the regions within the business district of the store where the implementation measures have been completed to the regions within the business district of the store where the measures have not been implemented. Thereby, the degree of the effect of each measure in the store where the measures have not been implemented is predicted, and measures with high effects are recommended to the store. For example, based on the actual results of measures within the business district of a small number of retail chain stores, measures predicted to have higher effects are recommended from among multiple measures to stores in other business districts. Measures implemented in the store are, for example, POP, island displays, LED signs, in-store visuals, receipt coupons, and point increases.

[0042] In this specification, the region within the business district of the store where the implementation measures have been completed is also referred to as the "completed implementation region". Excluding the completed implementation region, the region within the business district of the store where the measures have not been implemented is also referred to as the "unimplemented region".

[0043] 1. Structures of the Prediction Device and the Terminal Device

[0044] Figure 1 Shows the structures of the prediction device 1 and the terminal device 2. The prediction system 100 is constituted by the prediction device 1 and a plurality of terminal devices 2. The prediction system 100 uses data of stores where the measures have been implemented to predict and recommend effective measures for stores where the measures have not been implemented.

[0045] The prediction device 1 is a server. The terminal device 2 is various information processing devices such as a POS (Point of Sales) cash register, a personal computer, a tablet terminal, and a smartphone. For example, the prediction device 1 is a cloud server, and the terminal device 2 is installed in the store. In this case, the prediction device 1 and the terminal device 2 are connected via the Internet.

[0046] The prediction device 1 includes an input unit 11, a communication unit 12, a control unit 13, a storage unit 14, and a bus 15.

[0047] The input unit 11 is a user interface for inputting various operations based on the user. The input unit 11 can be implemented by a touch panel, a keyboard, buttons, switches, or a combination thereof.

[0048] The communication unit 12 includes a circuit that communicates with external devices according to a specified communication standard. The specified communication standard is, for example, LAN, Wi-Fi (registered trademark), Bluetooth (registered trademark), USB, HDMI (registered trademark). The communication unit 12 acquires data related to each store from the terminal devices 2 located in multiple stores. In the present embodiment, as the data related to the store, ID-POS data, customer data, and regional data are acquired. Additionally, the regional data can be acquired from the terminal device 2 or from other external devices. The communication unit 12 sends measure recommendation information indicating recommended measures to the terminal device 2.

[0049] The control unit 13 can be implemented by a semiconductor element or the like. The control unit 13 can include, for example, a microcomputer, CPU, MPU, GPU, DSP, FPGA, ASIC. The functions of the control unit 13 can be constituted by hardware or can be implemented by combining hardware and software. The control unit 13 reads the data and programs stored in the storage unit 14 and performs various arithmetic processes to implement specified functions.

[0050] The storage unit 14 is a storage medium that stores programs and data required to implement the functions of the prediction device 1. The storage unit 14 can be implemented by, for example, a hard disk drive (HDD), SSD, RAM, DRAM, ferroelectric memory, flash memory, magnetic disk, or a combination thereof.

[0051] The bus 15 is a signal line that electrically connects the input unit 11, communication unit 12, control unit 13, and storage unit 14.

[0052] The terminal device 2 includes an input unit 21, a communication unit 22, a control unit 23, a storage unit 24, a display unit 25, and a bus 26.

[0053] The terminal device 2 acquires ID-POS data, customer data, and regional data through the input unit 21 or the communication unit 22.

[0054] The input unit 21 can be implemented by a barcode reader, card reader, touch panel, keyboard, button, switch, or a combination thereof.

[0055] The communication unit 22 includes a circuit that communicates with external devices according to a specified communication standard. The specified communication standard is, for example, LAN, Wi-Fi (registered trademark), Bluetooth (registered trademark), USB, HDMI (registered trademark). The communication unit 22 sends ID-POS data, customer data, and regional data to the prediction device 1. The communication unit 22 acquires measure recommendation information from the prediction device 1.

[0056] The control unit 23 can be implemented by semiconductor elements or the like. For example, the control unit 23 can include a microcomputer, CPU, MPU, GPU, DSP, FPGA, ASIC. The functions of the control unit 23 can be constituted only by hardware, or can be implemented by combining hardware and software. The control unit 23 reads the data and programs stored in the storage unit 24 and performs various arithmetic processes to implement specified functions.

[0057] The storage unit 24 is a storage medium that stores the programs and data required to implement the functions of the terminal device 2. For example, the storage unit 24 can be implemented by a hard disk drive (HDD), SSD, RAM, DRAM, ferroelectric memory, flash memory, magnetic disk, or a combination thereof.

[0058] The display unit 25 is, for example, a liquid crystal display or an organic EL display. The display unit 25, for example, displays the recommended statements of the measures shown in the measure recommendation information.

[0059] The bus 26 is a signal line that electrically connects the input unit 21, communication unit 22, control unit 23, storage unit 24, and display unit 25.

[0060] 2. Operations of the prediction device during learning

[0061] 2.1 Functional structure of the prediction device during learning

[0062] Refer to Figures 2 to 8 , and describe the functions of the prediction device 1 during learning. Figure 2 Shows the functional structure of the control unit 13 of the prediction device 1 during learning and the data stored in the storage unit 14. Figure 3 Shows an example of the ID-POS data 141. Figure 4 Shows an example of the customer data 142. Figure 5 Shows an example of the area data 143. Figure 6 Shows an example of the purchase data 144A before implementing the measure. Figure 7 Shows an example of the purchase data 144B during implementing the measure. Figure 8 Schematically shows an example of the prediction data 145. In this embodiment, Figure 8 The areas a, b, c, etc. shown correspond to Figures 5 to 7 The areas "Moriguchi", "Minami-Makabe", "Kita-Makabe", etc. shown.

[0063] As Figure 2 shown, the control unit 13 of the prediction device 1 includes a data aggregation department 131, an effect ranking setting department 132, a prediction data generation department 133, and an update determination department 134.

[0064] In the storage unit 14 of the prediction device 1, ID-POS data 141, customer data 142, and area data 143 acquired from the terminal devices 2 in multiple stores via the communication unit 12 are stored.

[0065] The ID-POS data 141 is data indicating the sales amount of products. In Figure 3 the example, the ID-POS data 141 includes the time when the product was purchased, the ID of the customer who purchased the product, the purchased product, the unit price of the product and the number of units purchased, and the total amount of the purchased product.

[0066] The customer data 142 is data related to customers. In Figure 4 the example, the customer data 142 includes the customer ID, gender, the postal code of the customer's place of residence, and the birthday.

[0067] The area data 143 is data indicating the characteristics of the area. In Figure 5 the example, the area data 143 includes the postal code and area name of the area, the population, the number of households, and the male-female ratio of the population within the area.

[0068] The data aggregation headquarters 131 aggregates the ID-POS data 141, customer data 142, and area data 143 to generate purchase data 144. The data aggregation headquarters 131 stores the generated purchase data 144 in the storage unit 14.

[0069] The purchase data 144, as Figure 6 and Figure 7 shown, includes the purchase data 144A before the implementation of the measure and the purchase data 144B during the implementation of the measure. In Figure 6 and Figure 7 the example, the purchase data 144A and 144B include the postal code and area name of the area, the population, the number of households, the male-female ratio of the population, the sales amount within the area, and the customer unit price. The data aggregation headquarters 131 calculates, for example, the sales amount and customer unit price of the areas within the business district of the store based on the ID-POS data 141, customer data 142, and area data 143. In the generation of the purchase data 144, when each area is associated with the business districts of multiple stores, that is, when each area is associated with multiple stores, the sales amounts of all the stores associated with each area can also be aggregated. In the generation of the purchase data 144, each area can also be associated with only one store. The purchase data 144 is an example of the measure implementation information indicating the effect of the measure. For example, the difference in the sales amounts of the purchase data 144A and 144B represents the effect of the measure.

[0070] The effect ranking setting unit 132 determines the effect ranking of the measures for the areas within the business districts of the stores that have completed the implementation measures, i.e., the completed implementation areas. Specifically, the effect ranking setting unit 132 calculates the measure effect value based on the purchase data 144A before the implementation of the measures and the purchase data 144B during the implementation of the measures. The effect ranking setting unit 132 divides the degree of the effect of the measures into multiple effect rankings by comparing the measure effect value with a specified threshold value. The multiple effect rankings are, for example, rankings A, B, C, and D.

[0071] The prediction data generation unit 133 predicts the effect ranking of the measures for the areas within the business districts of the stores that have not implemented the measures, i.e., the non-implemented areas. Specifically, the prediction data generation unit 133 constructs a prediction chart for each measure, and through the label propagation method, propagates the effect ranking from the nodes in the completed implementation areas to the nodes in the non-implemented areas. Thereby, the prediction data generation unit 133 generates prediction data 145 representing the effect ranking of each area for each measure and stores it in the storage unit 14.

[0072] The prediction data 145, for example, includes Figure 8 the information representing the prediction chart 450 as shown. The information representing the prediction chart 450, for example, includes the weights, i.e., the importance, when synthesizing multiple similarity charts. The details regarding the importance will be described later. The prediction chart 450 includes nodes 451, links 452 connecting the nodes 451, and labels 453 assigned to the nodes 451. In the present embodiment, the nodes 451 correspond to areas. The links 452 correspond to the similarity between areas. That is, the links 452 have a strength equivalent to the similarity between areas. For example, the prediction data 145 includes the information representing the strength of all the links 452 in the prediction chart 450. The labels 453 correspond to the effect rankings of the measures. The prediction data 145, for example, includes Figure 22 the data obtained by adding the effect rankings of each measure to the area data 143 as shown.

[0073] The update determination unit 134 determines whether to update the prediction data 145. For example, the update determination unit 134 determines whether to update the prediction data 145 based on the change in the effect ranking for the completed implementation areas set by the effect ranking setting unit 132. The update determination unit 134 may also determine whether to update the prediction data 145 based on at least one of the ID-POS data 141, customer data 142, area data 143, and purchase data 144. If the update determination unit 134 decides to update the prediction data 145, it instructs the prediction data generation unit 133 to update the prediction data 145. Thereby, in the prediction data generation unit 133, the prediction chart 450 is reconstructed and the prediction data 145 is updated.

[0074] 2.2 Overall operation during learning

[0075] Figure 9 Indicates the generation action of the prediction data by the control unit 13 of the prediction device 1.

[0076] The data collection headquarters 131 acquires the ID-POS data 141, customer data 142, and regional data 143 (S1). For example, the data collection headquarters 131 acquires the ID-POS data 141, customer data 142, and regional data 143 corresponding to a specified period before the implementation of the measures from the terminal devices 2 of multiple stores. Further, the data collection headquarters 131 acquires the ID-POS data 141, customer data 142, and regional data 143 corresponding to a specified period during the implementation of the measures from the terminal devices 2 of the stores where the measures have been implemented. The specified period is, for example, one month. The data collection headquarters 131 may also read the ID-POS data 141, customer data 142, and regional data 143 that have been previously acquired from the terminal device 2 and stored in the storage unit 14 from the storage unit 14 in step S1.

[0077] Based on the ID-POS data 141, customer data 142, and regional data 143, the data collection headquarters 131 generates the purchase data 144A before the implementation of the measures and the purchase data 144B during the implementation of the measures (S2).

[0078] The effect ranking setting unit 132 sets the effect ranking of the measures for the regions within the business districts of the stores where the measures have been completed (S3).

[0079] The prediction data generation unit 133 calculates the similarity between regions according to the characteristics of each region and generates a similarity chart (S4). The characteristics of the region are, for example, population, number of households, male-female ratio of the population, sales amount, and customer unit price. For example, a similarity chart of population, a similarity chart of the number of households, a similarity chart of the male-female ratio, a similarity chart of sales amount, and a similarity chart of customer unit price are respectively generated (refer to Figure 13 ).

[0080] The prediction data generation unit 133 synthesizes the similarity charts of the characteristics of each region and generates a prediction chart to predict the effect ranking of the regions within the business districts of the stores where the measures have not been implemented (S5). Thus, the prediction data 145 is generated.

[0081] The prediction device 1 performs the processes of steps S1 to S5 for each measure.

[0082] 2.3 Setting of the effect ranking for the regions where the implementation has been completed

[0083] Refer to Figures 10 to 12 , and the setting of the effect ranking for the regions within the business districts of the stores where the measures have been completed, that is, the completed implementation regions, will be described. Figure 10 Indicates the action of setting the effect ranking for the completed implementation regions (Figure 9 (Details of step S3). Figure 11 An example showing the measure effect value and the effect ranking. Figure 12 An example showing the ranking probability.

[0084] The effect ranking setting unit 132 calculates the measure effect value (S301) based on the purchase data 144A before the measure implementation and the purchase data 144B during the measure implementation. For example, the measure effect value is calculated by "Measure effect value = Sales during measure implementation / Sales before measure implementation × 100".

[0085] The effect ranking setting unit 132 determines the effect ranking (S302) based on the measure effect value. For example, the effect ranking setting unit 132 compares the measure effect value with three thresholds and assigns a ranking to any one of the rankings A, B, C, D.

[0086] The effect ranking setting unit 132 sets the probability of each effect ranking (S303) according to the determined effect ranking. Specifically, the probability of the effect ranking assigned with a ranking is set to 1.0, and the probability of the effect ranking not assigned with a ranking is set to 0. For example, as Figure 12 shown, for the Shiguchi determined to be the effect ranking A, the probability of ranking A is set to 1.0, and the probabilities of rankings B, C, D are set to 0.

[0087] 2.4 Construction of the similarity chart

[0088] Figure 13 is a diagram for explaining the generation of the similarity chart 45 of the characteristics of each region in step S4. The characteristics of the region are, for example, population, number of households, male-female ratio of the population, sales before measure implementation, and customer unit price. The prediction data generation unit 133 generates, in step S4, for example, the similarity chart 45a of the population, the similarity chart 45b of the number of households, the similarity chart 45c of the male-female ratio, the similarity chart 45d of the sales, and the similarity chart 45e of the customer unit price based on the purchase data 144. Without particularly distinguishing the similarity chart 45a of the population, the similarity chart 45b of the number of households, the similarity chart 45c of the male-female ratio, the similarity chart 45d of the sales, and the similarity chart 45e of the customer unit price, they are collectively referred to as the similarity chart 45. The nodes 45N of each similarity chart 45 are all regions. The links 45L of the similarity chart 45a of the population, the similarity chart 45b of the number of households, the similarity chart 45c of the male-female ratio, the similarity chart 45d of the sales, and the similarity chart 45e of the customer unit price have the similarities of the population, the number of households, the male-female ratio, the sales, and the customer unit price, that is, the link strengths, respectively.

[0089] Specifically, in step S4, the prediction data generation unit 133 calculates the link strength A in each similarity graph 45 by formula (1). kii .

[0090] [Formula 1]

[0091]

[0092] In formula (1), k is a characteristic of the region such as population, sales, and customer unit price, i and j are nodes 45N representing regions, and A kij is the strength of the link 45L between node i and node j of the similarity graph 45 related to the characteristic k of the region, v ki is the value of the characteristic k of node i, v kj is the value of the characteristic k of node j, and σ is a positive parameter. The link strength A kij is specifically the (i, j) component of matrix A k . In the calculation of the link strength A kij , normalization is performed so that A k 1 n = 1 n . Here, n is the total number of regions, and A k is the link strength of the similarity graph 45 for the characteristic k of the region. The link strength A k is specifically a matrix. That is, before the calculation of A kij , the process of normalizing v ki for k is performed. As a result, the sum of each row 161 representing the link strength of the similarity graph normalized to Figure 16 is 1.

[0093] 2.5 Prediction of the probability of the effect ranking of unimplemented regions

[0094] Refer to Figures 14 to 18 to explain the prediction of the effect ranking of regions within the business district of stores where measures have not been implemented, that is, unimplemented regions. Figure 14 Represents the operation of predicting the effect ranking of unimplemented regions ( Figure 9 details of step S5). Figure 15 is a diagram for explaining the generation of the prediction graph for each effect ranking. Figure 16 is a diagram for explaining the synthesis of the similarity graph 45. Figure 17 Schematically represents the propagation of the effect ranking. In Figure 17 , the regions indicated by solid lines represent the regions where the effect ranking has been determined, and the regions indicated by dashed lines represent the regions where the effect ranking has not been determined. Figure 18 is a diagram for explaining the prediction of the effect ranking for unimplemented regions.

[0095] The prediction data generation unit 133 performs Figure 14 Steps S501 to S503 for each effect ranking. Thus, as Figure 15 shown, for example, for Measure 1, prediction charts 450A, 450B, 450C, and 450D with rankings A, B, C, and D as labels are respectively constructed. The nodes of the prediction charts 450A, 450B, 450C, and 450D are regions, and the links have strengths synthesized from the similarities of the respective characteristics of the regions. In the prediction charts 450A, 450B, 450C, and 450D, the effect rankings are propagated from the regions where the measure has been completed and implemented to the regions where it has not been implemented. Without particularly distinguishing the prediction charts 450A, 450B, 450C, and 450D, they are collectively referred to as the prediction chart 450.

[0096] Specifically, the prediction data generation unit 133 calculates the importance of each characteristic of the region (S501). The prediction data generation unit 133 calculates the importance using, for example, the EM (Expectation Maximization) algorithm according to Equation (2). The EM algorithm includes an E step and an M step. The E step calculates the possible importance, and the M step updates the probability of the effect ranking so that the expected value of the importance calculated in the E step is maximized.

[0097] (E step)

[0098] [Equation 2]

[0099]

[0100] In Equation (2),

[0101] is the updated importance of characteristic k, f is the predicted value of the probability of the effect ranking, L k is the Laplacian matrix of graph A k , ν, β net are positive parameters, and n is the total number of regions.

[0102] The Laplacian matrix L k can be obtained by Equation (3).

[0103] [Equation 3]

[0104] L k ≡ diag(A k 1 n ) - A k …(3)

[0105] The prediction data generation unit 133 is based on the importance u k, synthesize each similarity graph 45, and construct a prediction graph 450 (S502). Specifically, the prediction data generation unit 133 calculates the link strength A of the prediction graphs 450A to 450D through Equation (4). int The link strength A int Specifically, it is a matrix.

[0106] [Equation 4]

[0107]

[0108] Figure 16 To simplify the explanation, an example of calculating the link strength A of the prediction graph 450 based on the link strength A of the similarity graph 45 according to the population, the number of households, and the customer unit price is shown. k To calculate the link strength A of the prediction graph 450. int As shown in Equation (4) and Figure 16 shown, the link strength A of the similarity graph 45 k is multiplied by the importance u obtained according to the characteristics of each region k , and the sum is calculated to calculate the link strength A of the prediction graphs 450A to 450D respectively. int .

[0109] The prediction data generation unit 133 calculates the probability of the effect ranking of the nearby nodes of the node where the effect ranking is determined based on the link strength A of the prediction graphs 450A to 450D (S503). That is, the prediction data generation unit 133 propagates the effect ranking to the nearby nodes of the node where the effect ranking is determined. For example, in int the prediction graph 450A of the effect ranking A shown, the prediction data generation unit 133 calculates the probability of the effect ranking A at the nearby nodes of the node 451 where the effect ranking is determined based on the strength of the link 452. For example, calculate the probability of the effect ranking A of the nearby regions a and e of the region d, and the nearby regions f and g of the region h. Figure 17 shown.

[0110] Specifically, the prediction data generation unit 133 calculates the probability of the effect ranking through Equation (5).

[0111] (M step)

[0112] [Equation 5]

[0113]

[0114] In Equation (5),

[0115] is the updated predicted value of the probability of the effect ranking, f is the predicted value of the probability of the effect ranking before update, y is the probability of the effect ranking of the completed implementation region, G is the diagonal matrix for calculation, I nis the n-dimensional identity matrix, L int is the Laplacian matrix, β y 、β bias 、β net are positive parameters. The Laplacian matrix L int is calculated by Equation (6).

[0116] [Equation 6]

[0117] L int ≡ diag(A int 1 n ) - A int …(6)

[0118] The calculation diagonal matrix G is as follows. Here, l is the number of completed implementation regions, and n is the total number of regions.

[0119] [Equation 7]

[0120]

[0121] In the EM algorithm, using Equation (2) and Equation (5), the predicted values of the importance and the probability of the effect ranking are repeatedly calculated. When the change amount compared with the value before update is lower than the threshold, the predicted value of the probability of the effect ranking is determined.

[0122] 2.6 Determination of the Effect Ranking of Unimplemented Regions

[0123] The prediction data generation unit 133 determines whether the calculation of the probability of all effect rankings has ended for the neighboring nodes of the node for which the effect ranking has been determined (S504). For example, if the probabilities of all effect rankings A, B, C, and D have not been calculated, it returns to step S501, and performs steps S501 to S503 for the uncalculated effect rankings. If the probabilities of all effect rankings A, B, C, and D have been calculated, it proceeds to step S505.

[0124] The prediction data generation unit 133 determines the effect ranking based on the probabilities of each effect ranking for the neighboring nodes for which the probabilities of each effect ranking have been calculated (S505). For example, as Figure 18 shown, as the probabilities of the effect rankings A, B, C, and D for the Kyobashi region are calculated as "0.8", "0.3", "0.4", and "0.1", the effect ranking A with the highest probability of "0.8" is determined as the effect ranking of the Kyobashi region.

[0125] The prediction data generation unit 133 determines whether the effect rankings of all regions in the prediction chart 450 have been determined (S506). If the effect rankings of all regions have not been determined (No in S506), it returns to step S501. Thus, as Figure 17As shown, the effect ranking is propagated from the region where the effect ranking is determined to the region where the effect ranking is not determined.

[0126] If the effect rankings of all regions are determined (Yes in S506), the prediction data generation unit 133 stores the prediction data 145 in the storage unit 14 (S507). As described above, the prediction data 145 includes, for example, information representing Figure 8 the prediction charts 450 shown, that is, the information of prediction charts 450A, 450B, 450C, and 450D. Specifically, for example, the prediction data 145 includes the regions as nodes 451, the strength A of the links 452 int , the importance u k , and the effect ranking and the probability of the effect ranking of each region as the label 453.

[0127] As described above, the prediction device 1 performs the Figure 9 steps S1 to S5 shown for each measure. The prediction device 1 performs the Figure 14 steps S501 to S503 shown for each effect ranking. That is, the prediction device 1 generates prediction charts 450 equivalent to the number of "number of measures × number of rankings" based on the plurality of similarity charts 45, and calculates the probability of the effect ranking for each effect ranking in each measure. The prediction device 1 determines the effect ranking with the highest probability in each region as the effect ranking of the region.

[0128] 3. Actions of the prediction device during recommendation

[0129] 3.1 Functional structure of the prediction device during recommendation

[0130] Figure 19 Shows the functional structure of the control unit 13 of the prediction device 1 during recommendation and the data stored in the storage unit 14.

[0131] The control unit 13 of the prediction device 1 includes a business district setting unit 135, a purchase data supplementing unit 136, a prediction data updating unit 137, and a measure recommendation unit 138.

[0132] The business district setting unit 135 obtains store information of the store that is the object for which the measure effect is to be predicted from the input unit 11 or the communication unit 12 and sets the business district.

[0133] The purchase data supplementing unit 136 calculates the sales amount and customer unit price of a new store based on the sales amount and customer unit price of existing stores, and supplements the purchase data 144. The prediction data updating unit 137 uses the supplemented purchase data 144 to reconstruct the prediction chart 450 and updates the prediction data 145.

[0134] The measure recommendation unit 138 determines measures to be recommended to stores based on the prediction data 145. The measure recommendation unit 138 transmits measure recommendation information indicating the determined measures to the terminal device 2 via the communication unit 12.

[0135] 3.2 Actions during recommendation

[0136] Refer to Figures 20 to 23 and explain the recommendation of measures. Figure 20 Indicate the actions during recommendation by the control unit 13 of the prediction device 1. Figure 21 Indicate an example of the supplementation of the purchase data 144. Figure 22 It is a diagram for explaining the determination of the effect ranking of the business district. Figure 23 Indicate an example of the recommendation and non-recommendation of measures corresponding to the effect ranking.

[0137] In Figure 20 , if the business district setting unit 135 obtains store information of a store that is the object of predicting measure effects from the input unit 11 or the communication unit 12, it sets the business district of the store based on the store information (S601). For example, the business district setting unit 135 determines the residential areas of the visitors and sets the business district based on the ID-POS data 141 and customer data 142 obtained from the store to be predicted. The business district setting unit 135 may also set the area within a specified distance from the store to be predicted as the business district.

[0138] The business district setting unit 135 determines whether the areas within the set business district include new areas (S602). A new area is an area not included in the prediction data 145. For example, when the store to be predicted is a newly opened store, the effect ranking of the areas within the business district of the new store is not included in the prediction data 145 generated during learning. In this case, the areas within the business district set in step S601 include new areas. Thus, when new areas are included (Yes in S602), it proceeds to step S603. When the effect ranking of the areas within the set business district is included in the prediction data 145, that is, when new areas are not included (No in S602), it proceeds to step S606.

[0139] The purchase data supplementation unit 136 obtains the area data 143 of the new area (S603). For example, the purchase data supplementation unit 136 obtains the area data 143 of the new area from the terminal device 2 or other external devices and stores it in the storage unit 14. Or, the purchase data supplementation unit 136 reads the area data 143 of the new area that was previously obtained and stored in the storage unit 14 from the storage unit 14. The purchase data supplementation unit 136 supplements the purchase data 144 based on the area data 143 of the new area and the purchase data 144 of the existing area (S604). For example, as Figure 21As shown, the purchase data supplementing unit 136 adds the data 43 included in the regional data 143 of the new region to the purchase data 144, and calculates the sales amount and customer unit price 44B of the new region based on the sales amount and customer unit price 44A of the existing region. For example, the purchase data supplementing unit 136 sets the average value of the sales amount and customer unit price of the existing region as the sales amount and customer unit price of the new region. The purchase data supplementing unit 136 can also calculate the sales amount and customer unit price of the new region based on the sales amount and customer unit price of the existing region through regression analysis.

[0140] The prediction data updating unit 137 uses the supplemented purchase data 144 to reconstruct the prediction chart 450 of each measure (S605). The prediction data updating unit 137 updates the prediction data 145 in the storage unit 14 based on the reconstructed prediction chart 450. The step S605 of reconstructing the prediction chart 450 corresponds to Figure 9 steps S4 and S5.

[0141] The measure recommendation unit 138 predicts the effect ranking of all measures in the business district based on the prediction data 145 (S606). For example, when the business district set in step S601 includes multiple regions, for each measure, the population of the regions is aggregated according to each effect ranking, and the effect ranking with the largest population is set as the effect ranking of the business district. Specifically, for example, when the business district P including Figure 22 "Mikiguchi", "Minami Shin", "Kita Shin", "Hirakata", and "Kyobashi" shown in the figure is set in step S601, when determining the effect ranking of measure 1 in business district P, first calculate the total population of each effect ranking of measure 1. For measure 1, the population of effect ranking A is 1500 (=1200 + 300), the population of effect ranking B is 1100 (=1000 + 100), the population of effect ranking C is 0, and the population of effect ranking D is 600. Therefore, for measure 1, the effect ranking A with the largest population is set as the effect ranking of business district P. When determining the effect ranking of measure 2 in business district P, calculate the total population of each effect ranking of measure 2. For measure 2, the population of effect ranking A is 300, the population of effect ranking B is 1200, the population of effect ranking C is 1600 (=1000 + 600), and the population of effect ranking D is 100. In this case, for measure 2, the effect ranking C with the largest population is set as the effect ranking of business district P. When the business district set in step S601 includes only one region, for example, in Figure 22 the example where business district Q is set, the effect ranking of the Minoshima region included in business district Q is set as the effect ranking of business district Q.

[0142] The measure recommendation unit 138 determines the measure to be recommended based on the predicted effect ranking (S607). For example, asFigure 23 As shown, the measures with effect rankings A and B are determined as the recommended measures. The measure recommendation unit 138 sends measure recommendation information indicating the determined measures to the terminal device 2 of the store to be predicted specified in step S601 via the communication unit 12. For example, the measure recommendation information includes a recommendation statement for the measure. The terminal device 2 displays the recommendation statement on the display unit 25 based on the measure recommendation information.

[0143] As described above, the prediction device 1 propagates the effect rankings of the measures from the areas within the business districts of the stores that have completed implementing the measures to the areas within the business districts of the stores that have not implemented the measures by the label propagation method. The prediction device 1 predicts the effect rankings of all the measures in the business district of the store to be predicted based on the propagated effect rankings of the measures. The prediction device 1 determines the recommended measures based on the predicted effect rankings. Thus, even when there are few stores that have completed implementing the measures, effective measures can be recommended for the stores to be predicted.

[0144] 4. Update of Prediction Data

[0145] Figure 24 An example of the update operation of the prediction data 145 based on the control unit 13 of the prediction device 1 is shown. The prediction device 1 performs the Figure 9 processing shown when the prediction data 145 already exists in the storage unit 14. The prediction device 1 performs the Figure 24 processing shown. Figure 24 Steps S11, S12, S13, S16, and S17 of Figure 9 perform the same processing as steps S1, S2, S3, S4, and S5 of

[0146] The data collection headquarters 131 newly obtains ID-POS data 141, customer data 142, and regional data 143 (S11). Based on the newly obtained ID-POS data 141, customer data 142, and regional data 143, the data collection headquarters 131 generates purchase data 144A before the implementation of measures and purchase data 144B during the implementation of measures (S12). The effect ranking setting unit 132 sets the current effect ranking of the measures for the regions within the business districts of the stores that have completed the implementation of measures (S13). The update determination unit 134 compares the current effect ranking with the past effect ranking (S14). The update determination unit 134 determines whether the effect ranking has changed (S15). For example, the update determination unit 134 determines whether the average of the effect rankings of all regions is different between the past and the current. If the average of the effect rankings of all regions has changed, the update determination unit 134 determines that an update of the prediction data 145 is required and proceeds to step S16. The prediction data generation unit 133 calculates the similarity between regions according to the characteristics of each region, and generates a similarity chart 45 (S16). The prediction data generation unit 133 synthesizes the similarity charts 45 of the characteristics of each region and reconstructs the prediction chart 450 to predict the effect ranking of the regions within the business districts of the stores where the measures have not been implemented (S17). Thus, the prediction data 145 is updated.

[0147] Specifically, when the update determination unit 134 determines to update the prediction data 145, when the prediction data generation unit 133 reconstructs the prediction chart 450, it reapplies the calculation described in "2.5 Prediction of the Probability of the Effect Ranking of Regions Where Measures Have Not Been Implemented" above to update the predicted values of the importance and the probability of the effect ranking.

[0148] As described above, by updating the prediction data 145 when the effect ranking changes, for example, seasonal recommendations can be made.

[0149] 5. Effects and Supplements

[0150] The prediction device 1 of the present embodiment includes: a storage unit 14 that stores purchase data 144 indicating the measure effect in a store where a measure has been implemented, and a control unit 13 that predicts the measure effect in a store where the measure has not been implemented based on the purchase data 144. The store where the measure has been implemented is an example of the first object for which the measure has been implemented. The store where the measure has not been implemented is an example of the second object for which the measure has not been implemented. The purchase data 144 is an example of measure implementation information indicating the measure effect. The control unit 13 constructs prediction graphs 450A to 450D composed of a plurality of nodes and a plurality of links. The plurality of nodes include at least one regional node associated with the store of the first object and at least one regional node associated with the store of the second object. The plurality of links combine the nodes based on the similarity between the nodes. At least one regional node associated with the store of the first object is an example of the first node. At least one regional node associated with the store of the second object is an example of the second node. The prediction graphs 450A to 450D are an example of the first graph. The control unit 13 determines the degree of the measure effect in the nodes of the implemented area based on the purchase data 144, and propagates the degree of the measure effect from the nodes of the implemented area as a base point to the nodes of the unimplemented area in the prediction graphs 450A to 450D. Thus, even when there are few stores where the measure has been implemented, it is possible to predict effective measures in the unimplemented area. Since the degree of the measure effect is propagated sequentially from areas with higher similarity in the label propagation method, it is possible to predict effective measures for the unimplemented area even when the similarity between the implemented area and the unimplemented area is low.

[0151] Specifically, the degree of the measure effect includes a plurality of effect rankings. The control unit 13 propagates each effect ranking to the second node and calculates the probability of each effect ranking in the second node. The control unit 13 determines the effect ranking with the highest probability among the plurality of effect rankings as the effect ranking of the second node. Thus, it is possible to accurately predict the effect ranking of the measure for each region.

[0152] The plurality of nodes are associated with a plurality of characteristics. The plurality of characteristics are, for example, population, customer unit price, male-female ratio, and sales amount. The control unit 13 generates a similarity graph 45 including a plurality of nodes and a plurality of links for each characteristic. The plurality of links combine the nodes based on the similarity of each characteristic. The similarity graph 45 is an example of the second graph. The control unit 13 calculates the importance of each characteristic for each ranking, and synthesizes the similarity graphs 45 of each characteristic into one based on the importance to generate the prediction graphs 450A to 450D. Thus, since the effect ranking is propagated according to the characteristics of the region, it is possible to accurately predict the effect ranking of the measure for each region.

[0153] Based on the degree of the effect of the measures of the second node, the control unit 13 determines whether to recommend the measures to the stores of the second target. For example, the measures of A ranking and B ranking are recommended to the stores. Thereby, it is possible to recommend only effective measures to the stores.

[0154] When the store of the prediction target is associated with a business district including two or more regions, the control unit 13 determines the effect ranking of the business district according to the effect rankings of the respective regions in the business district. The business district is an example of a group. The control unit 13 determines whether to recommend the measures to the store of the prediction target according to the effect ranking of the business district. Thereby, it is possible to recommend effective measures to the store.

[0155] The control unit 13 generates prediction charts 450 for multiple measures respectively and spreads the degree of the effect of the measures, and determines the recommended measures from among the multiple measures based on the degree of the effect of the measures. Thereby, it is possible to recommend effective measures from among the multiple measures to the stores.

[0156] (Other embodiments)

[0157] As described above, the above embodiments have been described as examples of the technology disclosed in the present application. However, the technology in the present disclosure is not limited thereto, and can also be applied to embodiments with appropriate changes, replacements, additions, omissions, etc. Therefore, other embodiments are exemplified below.

[0158] In the above embodiment, an example in which the terminal device 2 is connected to the prediction device 1 via the Internet has been described. However, the prediction device 1 and the terminal device 2 may also be in each store and connected to the terminal device 2. In the above embodiment, the prediction system 100 is constituted by the prediction device 1 and the terminal device 2. However, all the functions of the prediction system 100 may also be realized by one device. Part of the functions of the prediction device 1 described in the above embodiment may also be performed by other prediction devices. For example, a prediction device having a data collection department 131, an effect ranking setting department 132, a prediction data generation department 133, and an update determination department 134 as functions during learning, and a prediction device having a business district setting department 135, a purchase data supplement department 136, a prediction data update department 137, and a measure recommendation department 138 as functions during recommendation may also be other devices.

[0159] In the above-described embodiment, the effect ranking of the implementation completion area is set based on the change in sales before and during the implementation of the measure. However, the effect ranking can also be set by other methods. For example, the effect ranking can also be set based on any one of sales, the number of visitors, the average customer price, the reach rate, the purchase rate, and the store visit rate before and during the implementation of the measure. Here, the reach rate = the number of people reaching the shelf / the number of visitors, the purchase rate = the number of purchasers of the product / the number of visitors, and the store visit rate = the number of visitors to the area / the population of the area. In addition, instead of using the values of sales, the number of visitors, the average customer price, etc. obtained from the store itself, values corrected by methods such as the seasonal adjustment method can be used. For example, the values can be corrected by the census station method, the MITI method, the monthly average method, the Parsons method, or the 12-month moving average method.

[0160] In the above-described embodiment, an example is described in which in the prediction chart 450 for the effect ranking of the dissemination measure, the node 451 is the area and the link 452 is the similarity of the population, the number of households, the average customer price, etc. of the area. However, the node 451 and the link 452 are not limited to the above-described embodiment. For example, the node 451 can be a store and the link 452 can be the similarity of the sales between stores or the sales ratio of each category. The node 451 can also be a customer and the link 452 can be the similarity of the purchase amount of the customer or the purchase ratio of each category.

[0161] In the above-described embodiment, at the time of recommendation, the total population of each effect ranking is calculated, and the effect ranking with the largest population is set as the effect ranking of the business district. However, the criterion for setting the effect ranking of the business district is not limited to the total population. For example, the number of households, the sales scale, etc. can also be totaled for each effect ranking, and the effect ranking with the maximum value can be set as the effect ranking of the business district.

[0162] In the above-described embodiment, the past effect ranking and the current effect ranking of the area where the measures have been implemented are compared to determine whether the prediction data 145 needs to be updated. However, the determination of the update is not limited to the above-described embodiment. The update determination unit 134 may also determine whether the prediction data 145 needs to be updated based on at least any one of the newly acquired ID-POS data 141, customer data 142, and area data 143. The update determination unit 134 may also determine whether the prediction data 145 needs to be updated based on the newly generated purchase data 144. For example, when the purchase ratio of a specific product or product category changes by a specified value or more, the update determination unit 134 may reconstruct the prediction chart 450 and update the prediction data 145. It may also be determined that an update is required when the measure effect value changes by a specified value or more. The update determination unit 134 may also generate data indicating the business situation, such as the sales volume, based on the purchase data 144. When a change in the business situation is detected, the prediction chart 450 may be reconstructed and the prediction data 145 may be updated. The update determination unit 134 may also exclude the areas where the purchased products are biased from the construction of the prediction chart 450. When the ratio of the occupations or the ratio of foreigners among the residents in the area changes, the prediction chart 450 may be reconstructed. It may also be excluded from the reconstruction of the prediction chart 450 according to the characteristics of the residents in the area, for example, the areas with a relatively large ratio of special occupations. It may also be determined whether an update is required based on the change amount of the usage time of each medium by the residents. The usage time of each medium is, for example, the average TV viewing time, the Internet usage time, the smartphone usage time, and the newspaper purchase rate. The update determination unit 134 may also determine whether an update is required based on the time information indicating the month, season, week, or year. For example, the prediction chart 450 may be reconstructed monthly. The update determination unit 134 may also update the prediction data 145 at a timing specified by the user via the input unit 11 or the communication unit 12. It is also possible to perform effect prediction such as the sales volume of seasonal products. When the prediction error is smaller than a specified threshold value, the reconstruction of the prediction chart 450 may be terminated. By reconstructing the prediction chart 450 in combination with the seasonality, it is possible to make recommendations that match the season.

[0163] In the above-described embodiment, an example of generating a prediction chart 450 for predicting the measure effects for each area and recommending measures to the store has been described. However, the structure of the prediction chart 450 and the content of the recommendation are not limited to the above-described embodiment. As shown below, it is also possible to generate prediction charts with other structures. It is also possible to recommend content different from the measures.

[0164] Modification Example 1: It is also possible to generate a prediction chart that targets customers, uses the similarity of the purchase amount or purchase ratio of each category per customer as a link, and uses the label as the product purchase trend. This prediction chart can also be used to recommend products in a retail store.

[0165] Modification Example 2: It is also possible to generate a prediction chart that targets products, uses similarities such as sales amount, sales quantity, and purchase rate as links, and uses the effect of measures for each product as tags. This prediction chart can also be used to recommend measures for category areas such as fruits and vegetables, and beverages in a retail store.

[0166] Modification Example 3: It is also possible to generate a prediction chart that targets a factory or a logistics base (logistics sorting site), uses similarities such as the demographics of employees' age, gender, etc., the construction year of the building, climate conditions, and equipment specifications as links, and uses the effect ranking resulting from changes in production efficiency or productivity as tags. This prediction chart can also be used to recommend measures for improving business efficiency in a factory or a logistics base. Examples of measures for improving business efficiency are layout changes and work systems.

[0167] Modification Example 4: It is also possible to generate a prediction chart that targets a deployment, uses similarities such as the age, gender, and years of service of employees within the deployment as links, and uses the effect ranking resulting from changes in job efficiency when the system is changed as tags. This prediction chart can also be used to recommend changes to information systems in an enterprise or a local government.

[0168] Modification Example 5: It is also possible to generate a prediction chart that targets an entertainment facility, uses similarities such as the age, gender, nationality, or number of visitors as links, and uses the effect ranking resulting from changes in the customer attraction rate or sales increase rate as tags. This prediction chart can also be used to recommend measures to entertainment facilities such as zoos and aquariums.

[0169] Modification Example 6: It is also possible to generate a prediction chart that targets residents or towns, uses similarities such as population, male-female ratio of the population, or the number of accidents as links, and uses the effect of accident prevention campaigns or crime prevention campaigns as tags. This prediction chart can also be used to recommend measures to an enterprise or a local government.

[0170] Other Modification Examples: Instead of measures, it is also possible to perform a) driving methods for drivers of automobiles, b) entrance examination schools, c) travel destinations or travel plans based on residential characteristics, d) news websites based on the utilization rate of applications, e) advertisement destinations or contents corresponding to viewer characteristics, f) advertisement destinations or contents corresponding to passerby characteristics, g) sports based on daily sports characteristics or geographical information, h) recommendations for actions such as diet, sleep, or walking corresponding to physical condition, biological signals, or the surrounding environment.

[0171] (Summary of the Embodiment)

[0172] (1) The prediction device of the present disclosure includes: a storage unit that stores measure implementation information indicating the measure effect in the first object where the measures have been implemented; and a control unit that predicts the measure effect in the second object where the measures have not been implemented based on the measure implementation information. The control unit constructs a first graph composed of a plurality of nodes and a plurality of links that combine the nodes based on the similarity between the nodes. The plurality of nodes include at least one first node associated with the first object and at least one second node associated with the second object. Based on the measure implementation information, the degree of the measure effect in the first node is determined, and the degree of the measure effect is propagated from the first node as a base point to the second node in the first graph.

[0173] Thus, even when there are few first objects in which the measures have been implemented, it is possible to predict effective measures in the second objects in which the measures have not been implemented.

[0174] (2) In the prediction device of (1), the control unit may also determine whether to recommend the measure to the second object based on the degree of the measure effect of the second node.

[0175] Thus, it is possible to recommend effective measures.

[0176] (3) In the prediction device of (1) or (2), the control unit may also determine whether an update of the prediction is required based on at least one of a plurality of characteristics. If it is determined to update, the link strength of the first graph is recalculated, and the degree of the measure effect is determined again.

[0177] (4) In any of the prediction devices of (1) to (3), the degree of the measure effect may include a plurality of rankings. The control unit propagates each ranking to the second node respectively, calculates the probability of each ranking in the second node, and determines the ranking with the highest probability among the plurality of rankings as the ranking of the second node.

[0178] Thus, it is possible to predict effective measures based on the rankings, and thus it is possible to recommend a plurality of measures.

[0179] (5) In the prediction device of (4), a plurality of nodes may be associated with a plurality of characteristics. The control unit generates a second graph composed of a plurality of nodes and a plurality of links that combine the nodes based on the similarity of each characteristic for each characteristic, calculates the importance of each characteristic for each ranking, and synthesizes the second graphs of each characteristic into one based on the importance to generate the first graph.

[0180] Thus, the rankings can be propagated based on the similarity of a plurality of characteristics.

[0181] (6) In the prediction device of (4), when the second object is related to a group including two or more nodes, the control unit determines the ranking of the group based on the rankings of the respective nodes within the group, and determines whether to recommend a measure to the second object based on the ranking of the group.

[0182] Thereby, an effective measure can be recommended when the second object is related to multiple nodes.

[0183] (7) In any one of the prediction devices of (1) to (4), the first object and the second object may be stores, and the multiple nodes correspond to any one of the store, the areas within the business district of the store, and the customers who come to the store.

[0184] Thereby, an effective measure can be predicted for the store.

[0185] (8) In the prediction device of (7), the nodes may correspond to areas, and the similarity between the nodes is a similarity related to at least one of the population, the number of households, the male-female ratio of the population, the sales amount, and the customer unit price within the area.

[0186] (9) In the prediction device of (8), the control unit may determine the degree of the measure effect of the first node based on the difference between before and during the implementation of the measure for at least any one of the sales amount, the number of customers coming to the store, and the customer unit price.

[0187] (10) In the prediction device of (8), the business district of the store of the second object may include one or more areas, and the control unit determines the degree of the measure effect of the business district based on the degree of the measure effect of the areas included in the business district, and determines whether to recommend a measure to the store of the second object based on the degree of the measure effect of the business district.

[0188] (11) In the prediction device of (2), the control unit may generate a first chart for each of the multiple measures and disseminate the degree of the measure effect, and determine the recommended measure from among the multiple measures based on the degree of the measure effect.

[0189] Thereby, multiple measures can be recommended.

[0190] (12) In any one of the prediction devices of (1) to (4), the first object and the second object may be factories, and the multiple nodes correspond to any one of the factory, the construction year of the building, the meteorological conditions, the specifications of the equipment, and the employees who perform operations in the factory.

[0191] (13) In any one of the prediction devices of (1) to (4), the first object and the second object may be logistics bases, and the multiple nodes correspond to any one of the logistics base, the construction year of the building, the meteorological conditions, the specifications of the equipment, and the employees who perform operations in the logistics base.

[0192] (14) The prediction method of the present disclosure is based on measure implementation information representing the measure effect in the first object that has completed the implementation measures, and predicts the effect of the measures in the second object that has not implemented the measures through an arithmetic unit. The prediction method includes: a step of constructing a graph composed of a plurality of nodes and a plurality of links that combine the nodes based on the similarity between the nodes, where the plurality of nodes include at least one first node associated with the first object and at least one second node associated with the second object (S502); a step of determining the degree of the measure effect in the first node based on the measure implementation information (S3); and a step of propagating the degree of the measure effect from the first node as a base point to the second node in the graph (S503).

[0193] The prediction device and the prediction method according to all the claims of the present disclosure are implemented by cooperation with hardware resources such as a processor, a memory, and a program.

[0194] Industrial applicability

[0195] The prediction device of the present disclosure is useful, for example, as a device for recommending effective measures to stores that have not implemented the measures.

[0196] -Symbol description-

[0197] 1 Prediction device

[0198] 2 Terminal device

[0199] 11, 21 Input unit

[0200] 12, 22 Communication unit

[0201] 13, 23 Control unit

[0202] 14, 24 Storage unit

[0203] 15, 26 Bus

[0204] 25 Display unit

[0205] 100 Prediction system

[0206] 131 Data aggregation headquarters

[0207] 132 Effect ranking setting unit

[0208] 133 Prediction data generation unit

[0209] 134 Update determination unit

[0210] 135 Business district setting unit

[0211] 136 Purchase data supplement unit

[0212] 137 Prediction Data Update Unit

[0213] 138 Measure Recommendation Unit.

Claims

1. A prediction device, comprising: a storage unit that stores measure implementation information indicating the measure effect in a first object where a measure has been implemented; and a control unit that predicts the measure effect in a second object where the measure has not been implemented, based on the measure implementation information, wherein the control unit constructs a first graph composed of a plurality of nodes and a plurality of links that combine the nodes based on the similarity between the nodes, the plurality of nodes including at least one first node associated with the first object and at least one second node associated with the second object, determines the degree of the measure effect in the first node based on the measure implementation information, and propagates the degree of the measure effect from the first node as a base point to the second node in the first graph, wherein the degree of the measure effect includes a plurality of rankings, the control unit propagates each ranking to the second node respectively, calculates the probability of each ranking in the second node, and determines the ranking with the highest probability among the plurality of rankings as the ranking of the second node.

2. The prediction device according to claim 1, wherein the control unit determines whether to recommend the measure to the second object based on the degree of the measure effect of the second node.

3. The prediction device according to claim 1 or 2, wherein the control unit determines whether an update of the prediction is required based on at least one of a plurality of characteristics, and if it is determined to update, recalculates the link strength of the first graph and determines the degree of the measure effect again.

4. The prediction device according to claim 1, wherein the plurality of nodes are related to a plurality of characteristics, the control unit generates a second graph composed of the plurality of nodes and a plurality of links that combine the nodes based on the similarity of each characteristic for each characteristic, the control unit calculates the importance of each characteristic for each ranking, and synthesizes the second graphs of each characteristic into one based on the importance to generate the first graph.

5. The prediction device according to claim 1, wherein when the second object is related to a group including two or more nodes, the control unit determines the ranking of the group based on the rankings of the nodes within the group, and determines whether to recommend the measure to the second object based on the ranking of the group.

6. The prediction device according to claim 1 or 2, wherein the first object and the second object are stores, the plurality of nodes correspond to any one of the store, the area within the business district of the store, and the customers who come to the store.

7. The prediction device according to claim 6, wherein the plurality of nodes correspond to the area, the similarity between the nodes is a similarity related to at least one of the population, the number of households, the male-female ratio of the population, the sales amount, and the customer unit price within the area.

8. The prediction device according to claim 7, wherein the control unit determines the degree of the measure effect of the first node based on the difference before and after the implementation of the measure for at least any one of the sales amount, the number of customers coming to the store, and the customer unit price.

9. The prediction device according to claim 7, wherein, the business district of the store of the second object includes more than one area, the control unit determines the degree of the measure effect of the business district according to the degree of the measure effect of the areas included in the business district, and determines whether to recommend the measure to the store of the second object according to the degree of the measure effect of the business district.

10. The prediction device according to claim 2, wherein, the control unit respectively generates the first chart for a plurality of measures and disseminates the degree of the measure effect, and determines the recommended measure from among the plurality of measures based on the degree of the measure effect.

11. The prediction device according to claim 1 or 2, wherein, the first object and the second object are factories, the plurality of nodes correspond to any one of the factories, the construction years of the buildings, the meteorological conditions, the specifications of the equipment, and the employees working in the factories.

12. The prediction device according to claim 1 or 2, wherein, the first object and the second object are logistics bases, the plurality of nodes correspond to any one of the logistics bases, the construction years of the buildings, the meteorological conditions, the specifications of the equipment, and the employees working in the logistics bases.

13. A prediction method, based on measure implementation information representing the measure effect in a first object that has completed implementing a measure, the arithmetic unit predicts the effect of the measure in a second object that has not implemented the measure, the prediction method includes: a step of constructing a chart composed of a plurality of nodes and a plurality of links that combine the nodes based on the similarity between the nodes, the plurality of nodes including at least one first node associated with the first object and at least one second node associated with the second object; a step of determining the degree of the measure effect in the first node based on the measure implementation information; and a step of disseminating the degree of the measure effect from the first node as a base point to the second node in the chart, the degree of the measure effect includes a plurality of rankings, each ranking is respectively disseminated to the second node, the probability of each ranking in the second node is calculated, and the ranking with the highest probability among the plurality of rankings is determined as the ranking of the second node.

14. A program product, including a program, the program causes a computer to execute the prediction method according to claim 13.

Images